summaryrefslogtreecommitdiff
path: root/macros/generic/xint/xint.dtx
blob: a1ad014bc22535e9b67637bb9b546761591615c4 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595
4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
4622
4623
4624
4625
4626
4627
4628
4629
4630
4631
4632
4633
4634
4635
4636
4637
4638
4639
4640
4641
4642
4643
4644
4645
4646
4647
4648
4649
4650
4651
4652
4653
4654
4655
4656
4657
4658
4659
4660
4661
4662
4663
4664
4665
4666
4667
4668
4669
4670
4671
4672
4673
4674
4675
4676
4677
4678
4679
4680
4681
4682
4683
4684
4685
4686
4687
4688
4689
4690
4691
4692
4693
4694
4695
4696
4697
4698
4699
4700
4701
4702
4703
4704
4705
4706
4707
4708
4709
4710
4711
4712
4713
4714
4715
4716
4717
4718
4719
4720
4721
4722
4723
4724
4725
4726
4727
4728
4729
4730
4731
4732
4733
4734
4735
4736
4737
4738
4739
4740
4741
4742
4743
4744
4745
4746
4747
4748
4749
4750
4751
4752
4753
4754
4755
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769
4770
4771
4772
4773
4774
4775
4776
4777
4778
4779
4780
4781
4782
4783
4784
4785
4786
4787
4788
4789
4790
4791
4792
4793
4794
4795
4796
4797
4798
4799
4800
4801
4802
4803
4804
4805
4806
4807
4808
4809
4810
4811
4812
4813
4814
4815
4816
4817
4818
4819
4820
4821
4822
4823
4824
4825
4826
4827
4828
4829
4830
4831
4832
4833
4834
4835
4836
4837
4838
4839
4840
4841
4842
4843
4844
4845
4846
4847
4848
4849
4850
4851
4852
4853
4854
4855
4856
4857
4858
4859
4860
4861
4862
4863
4864
4865
4866
4867
4868
4869
4870
4871
4872
4873
4874
4875
4876
4877
4878
4879
4880
4881
4882
4883
4884
4885
4886
4887
4888
4889
4890
4891
4892
4893
4894
4895
4896
4897
4898
4899
4900
4901
4902
4903
4904
4905
4906
4907
4908
4909
4910
4911
4912
4913
4914
4915
4916
4917
4918
4919
4920
4921
4922
4923
4924
4925
4926
4927
4928
4929
4930
4931
4932
4933
4934
4935
4936
4937
4938
4939
4940
4941
4942
4943
4944
4945
4946
4947
4948
4949
4950
4951
4952
4953
4954
4955
4956
4957
4958
4959
4960
4961
4962
4963
4964
4965
4966
4967
4968
4969
4970
4971
4972
4973
4974
4975
4976
4977
4978
4979
4980
4981
4982
4983
4984
4985
4986
4987
4988
4989
4990
4991
4992
4993
4994
4995
4996
4997
4998
4999
5000
5001
5002
5003
5004
5005
5006
5007
5008
5009
5010
5011
5012
5013
5014
5015
5016
5017
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032
5033
5034
5035
5036
5037
5038
5039
5040
5041
5042
5043
5044
5045
5046
5047
5048
5049
5050
5051
5052
5053
5054
5055
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072
5073
5074
5075
5076
5077
5078
5079
5080
5081
5082
5083
5084
5085
5086
5087
5088
5089
5090
5091
5092
5093
5094
5095
5096
5097
5098
5099
5100
5101
5102
5103
5104
5105
5106
5107
5108
5109
5110
5111
5112
5113
5114
5115
5116
5117
5118
5119
5120
5121
5122
5123
5124
5125
5126
5127
5128
5129
5130
5131
5132
5133
5134
5135
5136
5137
5138
5139
5140
5141
5142
5143
5144
5145
5146
5147
5148
5149
5150
5151
5152
5153
5154
5155
5156
5157
5158
5159
5160
5161
5162
5163
5164
5165
5166
5167
5168
5169
5170
5171
5172
5173
5174
5175
5176
5177
5178
5179
5180
5181
5182
5183
5184
5185
5186
5187
5188
5189
5190
5191
5192
5193
5194
5195
5196
5197
5198
5199
5200
5201
5202
5203
5204
5205
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220
5221
5222
5223
5224
5225
5226
5227
5228
5229
5230
5231
5232
5233
5234
5235
5236
5237
5238
5239
5240
5241
5242
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254
5255
5256
5257
5258
5259
5260
5261
5262
5263
5264
5265
5266
5267
5268
5269
5270
5271
5272
5273
5274
5275
5276
5277
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289
5290
5291
5292
5293
5294
5295
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306
5307
5308
5309
5310
5311
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334
5335
5336
5337
5338
5339
5340
5341
5342
5343
5344
5345
5346
5347
5348
5349
5350
5351
5352
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364
5365
5366
5367
5368
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379
5380
5381
5382
5383
5384
5385
5386
5387
5388
5389
5390
5391
5392
5393
5394
5395
5396
5397
5398
5399
5400
5401
5402
5403
5404
5405
5406
5407
5408
5409
5410
5411
5412
5413
5414
5415
5416
5417
5418
5419
5420
5421
5422
5423
5424
5425
5426
5427
5428
5429
5430
5431
5432
5433
5434
5435
5436
5437
5438
5439
5440
5441
5442
5443
5444
5445
5446
5447
5448
5449
5450
5451
5452
5453
5454
5455
5456
5457
5458
5459
5460
5461
5462
5463
5464
5465
5466
5467
5468
5469
5470
5471
5472
5473
5474
5475
5476
5477
5478
5479
5480
5481
5482
5483
5484
5485
5486
5487
5488
5489
5490
5491
5492
5493
5494
5495
5496
5497
5498
5499
5500
5501
5502
5503
5504
5505
5506
5507
5508
5509
5510
5511
5512
5513
5514
5515
5516
5517
5518
5519
5520
5521
5522
5523
5524
5525
5526
5527
5528
5529
5530
5531
5532
5533
5534
5535
5536
5537
5538
5539
5540
5541
5542
5543
5544
5545
5546
5547
5548
5549
5550
5551
5552
5553
5554
5555
5556
5557
5558
5559
5560
5561
5562
5563
5564
5565
5566
5567
5568
5569
5570
5571
5572
5573
5574
5575
5576
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593
5594
5595
5596
5597
5598
5599
5600
5601
5602
5603
5604
5605
5606
5607
5608
5609
5610
5611
5612
5613
5614
5615
5616
5617
5618
5619
5620
5621
5622
5623
5624
5625
5626
5627
5628
5629
5630
5631
5632
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646
5647
5648
5649
5650
5651
5652
5653
5654
5655
5656
5657
5658
5659
5660
5661
5662
5663
5664
5665
5666
5667
5668
5669
5670
5671
5672
5673
5674
5675
5676
5677
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689
5690
5691
5692
5693
5694
5695
5696
5697
5698
5699
5700
5701
5702
5703
5704
5705
5706
5707
5708
5709
5710
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721
5722
5723
5724
5725
5726
5727
5728
5729
5730
5731
5732
5733
5734
5735
5736
5737
5738
5739
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752
5753
5754
5755
5756
5757
5758
5759
5760
5761
5762
5763
5764
5765
5766
5767
5768
5769
5770
5771
5772
5773
5774
5775
5776
5777
5778
5779
5780
5781
5782
5783
5784
5785
5786
5787
5788
5789
5790
5791
5792
5793
5794
5795
5796
5797
5798
5799
5800
5801
5802
5803
5804
5805
5806
5807
5808
5809
5810
5811
5812
5813
5814
5815
5816
5817
5818
5819
5820
5821
5822
5823
5824
5825
5826
5827
5828
5829
5830
5831
5832
5833
5834
5835
5836
5837
5838
5839
5840
5841
5842
5843
5844
5845
5846
5847
5848
5849
5850
5851
5852
5853
5854
5855
5856
5857
5858
5859
5860
5861
5862
5863
5864
5865
5866
5867
5868
5869
5870
5871
5872
5873
5874
5875
5876
5877
5878
5879
5880
5881
5882
5883
5884
5885
5886
5887
5888
5889
5890
5891
5892
5893
5894
5895
5896
5897
5898
5899
5900
5901
5902
5903
5904
5905
5906
5907
5908
5909
5910
5911
5912
5913
5914
5915
5916
5917
5918
5919
5920
5921
5922
5923
5924
5925
5926
5927
5928
5929
5930
5931
5932
5933
5934
5935
5936
5937
5938
5939
5940
5941
5942
5943
5944
5945
5946
5947
5948
5949
5950
5951
5952
5953
5954
5955
5956
5957
5958
5959
5960
5961
5962
5963
5964
5965
5966
5967
5968
5969
5970
5971
5972
5973
5974
5975
5976
5977
5978
5979
5980
5981
5982
5983
5984
5985
5986
5987
5988
5989
5990
5991
5992
5993
5994
5995
5996
5997
5998
5999
6000
6001
6002
6003
6004
6005
6006
6007
6008
6009
6010
6011
6012
6013
6014
6015
6016
6017
6018
6019
6020
6021
6022
6023
6024
6025
6026
6027
6028
6029
6030
6031
6032
6033
6034
6035
6036
6037
6038
6039
6040
6041
6042
6043
6044
6045
6046
6047
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061
6062
6063
6064
6065
6066
6067
6068
6069
6070
6071
6072
6073
6074
6075
6076
6077
6078
6079
6080
6081
6082
6083
6084
6085
6086
6087
6088
6089
6090
6091
6092
6093
6094
6095
6096
6097
6098
6099
6100
6101
6102
6103
6104
6105
6106
6107
6108
6109
6110
6111
6112
6113
6114
6115
6116
6117
6118
6119
6120
6121
6122
6123
6124
6125
6126
6127
6128
6129
6130
6131
6132
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144
6145
6146
6147
6148
6149
6150
6151
6152
6153
6154
6155
6156
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172
6173
6174
6175
6176
6177
6178
6179
6180
6181
6182
6183
6184
6185
6186
6187
6188
6189
6190
6191
6192
6193
6194
6195
6196
6197
6198
6199
6200
6201
6202
6203
6204
6205
6206
6207
6208
6209
6210
6211
6212
6213
6214
6215
6216
6217
6218
6219
6220
6221
6222
6223
6224
6225
6226
6227
6228
6229
6230
6231
6232
6233
6234
6235
6236
6237
6238
6239
6240
6241
6242
6243
6244
6245
6246
6247
6248
6249
6250
6251
6252
6253
6254
6255
6256
6257
6258
6259
6260
6261
6262
6263
6264
6265
6266
6267
6268
6269
6270
6271
6272
6273
6274
6275
6276
6277
6278
6279
6280
6281
6282
6283
6284
6285
6286
6287
6288
6289
6290
6291
6292
6293
6294
6295
6296
6297
6298
6299
6300
6301
6302
6303
6304
6305
6306
6307
6308
6309
6310
6311
6312
6313
6314
6315
6316
6317
6318
6319
6320
6321
6322
6323
6324
6325
6326
6327
6328
6329
6330
6331
6332
6333
6334
6335
6336
6337
6338
6339
6340
6341
6342
6343
6344
6345
6346
6347
6348
6349
6350
6351
6352
6353
6354
6355
6356
6357
6358
6359
6360
6361
6362
6363
6364
6365
6366
6367
6368
6369
6370
6371
6372
6373
6374
6375
6376
6377
6378
6379
6380
6381
6382
6383
6384
6385
6386
6387
6388
6389
6390
6391
6392
6393
6394
6395
6396
6397
6398
6399
6400
6401
6402
6403
6404
6405
6406
6407
6408
6409
6410
6411
6412
6413
6414
6415
6416
6417
6418
6419
6420
6421
6422
6423
6424
6425
6426
6427
6428
6429
6430
6431
6432
6433
6434
6435
6436
6437
6438
6439
6440
6441
6442
6443
6444
6445
6446
6447
6448
6449
6450
6451
6452
6453
6454
6455
6456
6457
6458
6459
6460
6461
6462
6463
6464
6465
6466
6467
6468
6469
6470
6471
6472
6473
6474
6475
6476
6477
6478
6479
6480
6481
6482
6483
6484
6485
6486
6487
6488
6489
6490
6491
6492
6493
6494
6495
6496
6497
6498
6499
6500
6501
6502
6503
6504
6505
6506
6507
6508
6509
6510
6511
6512
6513
6514
6515
6516
6517
6518
6519
6520
6521
6522
6523
6524
6525
6526
6527
6528
6529
6530
6531
6532
6533
6534
6535
6536
6537
6538
6539
6540
6541
6542
6543
6544
6545
6546
6547
6548
6549
6550
6551
6552
6553
6554
6555
6556
6557
6558
6559
6560
6561
6562
6563
6564
6565
6566
6567
6568
6569
6570
6571
6572
6573
6574
6575
6576
6577
6578
6579
6580
6581
6582
6583
6584
6585
6586
6587
6588
6589
6590
6591
6592
6593
6594
6595
6596
6597
6598
6599
6600
6601
6602
6603
6604
6605
6606
6607
6608
6609
6610
6611
6612
6613
6614
6615
6616
6617
6618
6619
6620
6621
6622
6623
6624
6625
6626
6627
6628
6629
6630
6631
6632
6633
6634
6635
6636
6637
6638
6639
6640
6641
6642
6643
6644
6645
6646
6647
6648
6649
6650
6651
6652
6653
6654
6655
6656
6657
6658
6659
6660
6661
6662
6663
6664
6665
6666
6667
6668
6669
6670
6671
6672
6673
6674
6675
6676
6677
6678
6679
6680
6681
6682
6683
6684
6685
6686
6687
6688
6689
6690
6691
6692
6693
6694
6695
6696
6697
6698
6699
6700
6701
6702
6703
6704
6705
6706
6707
6708
6709
6710
6711
6712
6713
6714
6715
6716
6717
6718
6719
6720
6721
6722
6723
6724
6725
6726
6727
6728
6729
6730
6731
6732
6733
6734
6735
6736
6737
6738
6739
6740
6741
6742
6743
6744
6745
6746
6747
6748
6749
6750
6751
6752
6753
6754
6755
6756
6757
6758
6759
6760
6761
6762
6763
6764
6765
6766
6767
6768
6769
6770
6771
6772
6773
6774
6775
6776
6777
6778
6779
6780
6781
6782
6783
6784
6785
6786
6787
6788
6789
6790
6791
6792
6793
6794
6795
6796
6797
6798
6799
6800
6801
6802
6803
6804
6805
6806
6807
6808
6809
6810
6811
6812
6813
6814
6815
6816
6817
6818
6819
6820
6821
6822
6823
6824
6825
6826
6827
6828
6829
6830
6831
6832
6833
6834
6835
6836
6837
6838
6839
6840
6841
6842
6843
6844
6845
6846
6847
6848
6849
6850
6851
6852
6853
6854
6855
6856
6857
6858
6859
6860
6861
6862
6863
6864
6865
6866
6867
6868
6869
6870
6871
6872
6873
6874
6875
6876
6877
6878
6879
6880
6881
6882
6883
6884
6885
6886
6887
6888
6889
6890
6891
6892
6893
6894
6895
6896
6897
6898
6899
6900
6901
6902
6903
6904
6905
6906
6907
6908
6909
6910
6911
6912
6913
6914
6915
6916
6917
6918
6919
6920
6921
6922
6923
6924
6925
6926
6927
6928
6929
6930
6931
6932
6933
6934
6935
6936
6937
6938
6939
6940
6941
6942
6943
6944
6945
6946
6947
6948
6949
6950
6951
6952
6953
6954
6955
6956
6957
6958
6959
6960
6961
6962
6963
6964
6965
6966
6967
6968
6969
6970
6971
6972
6973
6974
6975
6976
6977
6978
6979
6980
6981
6982
6983
6984
6985
6986
6987
6988
6989
6990
6991
6992
6993
6994
6995
6996
6997
6998
6999
7000
7001
7002
7003
7004
7005
7006
7007
7008
7009
7010
7011
7012
7013
7014
7015
7016
7017
7018
7019
7020
7021
7022
7023
7024
7025
7026
7027
7028
7029
7030
7031
7032
7033
7034
7035
7036
7037
7038
7039
7040
7041
7042
7043
7044
7045
7046
7047
7048
7049
7050
7051
7052
7053
7054
7055
7056
7057
7058
7059
7060
7061
7062
7063
7064
7065
7066
7067
7068
7069
7070
7071
7072
7073
7074
7075
7076
7077
7078
7079
7080
7081
7082
7083
7084
7085
7086
7087
7088
7089
7090
7091
7092
7093
7094
7095
7096
7097
7098
7099
7100
7101
7102
7103
7104
7105
7106
7107
7108
7109
7110
7111
7112
7113
7114
7115
7116
7117
7118
7119
7120
7121
7122
7123
7124
7125
7126
7127
7128
7129
7130
7131
7132
7133
7134
7135
7136
7137
7138
7139
7140
7141
7142
7143
7144
7145
7146
7147
7148
7149
7150
7151
7152
7153
7154
7155
7156
7157
7158
7159
7160
7161
7162
7163
7164
7165
7166
7167
7168
7169
7170
7171
7172
7173
7174
7175
7176
7177
7178
7179
7180
7181
7182
7183
7184
7185
7186
7187
7188
7189
7190
7191
7192
7193
7194
7195
7196
7197
7198
7199
7200
7201
7202
7203
7204
7205
7206
7207
7208
7209
7210
7211
7212
7213
7214
7215
7216
7217
7218
7219
7220
7221
7222
7223
7224
7225
7226
7227
7228
7229
7230
7231
7232
7233
7234
7235
7236
7237
7238
7239
7240
7241
7242
7243
7244
7245
7246
7247
7248
7249
7250
7251
7252
7253
7254
7255
7256
7257
7258
7259
7260
7261
7262
7263
7264
7265
7266
7267
7268
7269
7270
7271
7272
7273
7274
7275
7276
7277
7278
7279
7280
7281
7282
7283
7284
7285
7286
7287
7288
7289
7290
7291
7292
7293
7294
7295
7296
7297
7298
7299
7300
7301
7302
7303
7304
7305
7306
7307
7308
7309
7310
7311
7312
7313
7314
7315
7316
7317
7318
7319
7320
7321
7322
7323
7324
7325
7326
7327
7328
7329
7330
7331
7332
7333
7334
7335
7336
7337
7338
7339
7340
7341
7342
7343
7344
7345
7346
7347
7348
7349
7350
7351
7352
7353
7354
7355
7356
7357
7358
7359
7360
7361
7362
7363
7364
7365
7366
7367
7368
7369
7370
7371
7372
7373
7374
7375
7376
7377
7378
7379
7380
7381
7382
7383
7384
7385
7386
7387
7388
7389
7390
7391
7392
7393
7394
7395
7396
7397
7398
7399
7400
7401
7402
7403
7404
7405
7406
7407
7408
7409
7410
7411
7412
7413
7414
7415
7416
7417
7418
7419
7420
7421
7422
7423
7424
7425
7426
7427
7428
7429
7430
7431
7432
7433
7434
7435
7436
7437
7438
7439
7440
7441
7442
7443
7444
7445
7446
7447
7448
7449
7450
7451
7452
7453
7454
7455
7456
7457
7458
7459
7460
7461
7462
7463
7464
7465
7466
7467
7468
7469
7470
7471
7472
7473
7474
7475
7476
7477
7478
7479
7480
7481
7482
7483
7484
7485
7486
7487
7488
7489
7490
7491
7492
7493
7494
7495
7496
7497
7498
7499
7500
7501
7502
7503
7504
7505
7506
7507
7508
7509
7510
7511
7512
7513
7514
7515
7516
7517
7518
7519
7520
7521
7522
7523
7524
7525
7526
7527
7528
7529
7530
7531
7532
7533
7534
7535
7536
7537
7538
7539
7540
7541
7542
7543
7544
7545
7546
7547
7548
7549
7550
7551
7552
7553
7554
7555
7556
7557
7558
7559
7560
7561
7562
7563
7564
7565
7566
7567
7568
7569
7570
7571
7572
7573
7574
7575
7576
7577
7578
7579
7580
7581
7582
7583
7584
7585
7586
7587
7588
7589
7590
7591
7592
7593
7594
7595
7596
7597
7598
7599
7600
7601
7602
7603
7604
7605
7606
7607
7608
7609
7610
7611
7612
7613
7614
7615
7616
7617
7618
7619
7620
7621
7622
7623
7624
7625
7626
7627
7628
7629
7630
7631
7632
7633
7634
7635
7636
7637
7638
7639
7640
7641
7642
7643
7644
7645
7646
7647
7648
7649
7650
7651
7652
7653
7654
7655
7656
7657
7658
7659
7660
7661
7662
7663
7664
7665
7666
7667
7668
7669
7670
7671
7672
7673
7674
7675
7676
7677
7678
7679
7680
7681
7682
7683
7684
7685
7686
7687
7688
7689
7690
7691
7692
7693
7694
7695
7696
7697
7698
7699
7700
7701
7702
7703
7704
7705
7706
7707
7708
7709
7710
7711
7712
7713
7714
7715
7716
7717
7718
7719
7720
7721
7722
7723
7724
7725
7726
7727
7728
7729
7730
7731
7732
7733
7734
7735
7736
7737
7738
7739
7740
7741
7742
7743
7744
7745
7746
7747
7748
7749
7750
7751
7752
7753
7754
7755
7756
7757
7758
7759
7760
7761
7762
7763
7764
7765
7766
7767
7768
7769
7770
7771
7772
7773
7774
7775
7776
7777
7778
7779
7780
7781
7782
7783
7784
7785
7786
7787
7788
7789
7790
7791
7792
7793
7794
7795
7796
7797
7798
7799
7800
7801
7802
7803
7804
7805
7806
7807
7808
7809
7810
7811
7812
7813
7814
7815
7816
7817
7818
7819
7820
7821
7822
7823
7824
7825
7826
7827
7828
7829
7830
7831
7832
7833
7834
7835
7836
7837
7838
7839
7840
7841
7842
7843
7844
7845
7846
7847
7848
7849
7850
7851
7852
7853
7854
7855
7856
7857
7858
7859
7860
7861
7862
7863
7864
7865
7866
7867
7868
7869
7870
7871
7872
7873
7874
7875
7876
7877
7878
7879
7880
7881
7882
7883
7884
7885
7886
7887
7888
7889
7890
7891
7892
7893
7894
7895
7896
7897
7898
7899
7900
7901
7902
7903
7904
7905
7906
7907
7908
7909
7910
7911
7912
7913
7914
7915
7916
7917
7918
7919
7920
7921
7922
7923
7924
7925
7926
7927
7928
7929
7930
7931
7932
7933
7934
7935
7936
7937
7938
7939
7940
7941
7942
7943
7944
7945
7946
7947
7948
7949
7950
7951
7952
7953
7954
7955
7956
7957
7958
7959
7960
7961
7962
7963
7964
7965
7966
7967
7968
7969
7970
7971
7972
7973
7974
7975
7976
7977
7978
7979
7980
7981
7982
7983
7984
7985
7986
7987
7988
7989
7990
7991
7992
7993
7994
7995
7996
7997
7998
7999
8000
8001
8002
8003
8004
8005
8006
8007
8008
8009
8010
8011
8012
8013
8014
8015
8016
8017
8018
8019
8020
8021
8022
8023
8024
8025
8026
8027
8028
8029
8030
8031
8032
8033
8034
8035
8036
8037
8038
8039
8040
8041
8042
8043
8044
8045
8046
8047
8048
8049
8050
8051
8052
8053
8054
8055
8056
8057
8058
8059
8060
8061
8062
8063
8064
8065
8066
8067
8068
8069
8070
8071
8072
8073
8074
8075
8076
8077
8078
8079
8080
8081
8082
8083
8084
8085
8086
8087
8088
8089
8090
8091
8092
8093
8094
8095
8096
8097
8098
8099
8100
8101
8102
8103
8104
8105
8106
8107
8108
8109
8110
8111
8112
8113
8114
8115
8116
8117
8118
8119
8120
8121
8122
8123
8124
8125
8126
8127
8128
8129
8130
8131
8132
8133
8134
8135
8136
8137
8138
8139
8140
8141
8142
8143
8144
8145
8146
8147
8148
8149
8150
8151
8152
8153
8154
8155
8156
8157
8158
8159
8160
8161
8162
8163
8164
8165
8166
8167
8168
8169
8170
8171
8172
8173
8174
8175
8176
8177
8178
8179
8180
8181
8182
8183
8184
8185
8186
8187
8188
8189
8190
8191
8192
8193
8194
8195
8196
8197
8198
8199
8200
8201
8202
8203
8204
8205
8206
8207
8208
8209
8210
8211
8212
8213
8214
8215
8216
8217
8218
8219
8220
8221
8222
8223
8224
8225
8226
8227
8228
8229
8230
8231
8232
8233
8234
8235
8236
8237
8238
8239
8240
8241
8242
8243
8244
8245
8246
8247
8248
8249
8250
8251
8252
8253
8254
8255
8256
8257
8258
8259
8260
8261
8262
8263
8264
8265
8266
8267
8268
8269
8270
8271
8272
8273
8274
8275
8276
8277
8278
8279
8280
8281
8282
8283
8284
8285
8286
8287
8288
8289
8290
8291
8292
8293
8294
8295
8296
8297
8298
8299
8300
8301
8302
8303
8304
8305
8306
8307
8308
8309
8310
8311
8312
8313
8314
8315
8316
8317
8318
8319
8320
8321
8322
8323
8324
8325
8326
8327
8328
8329
8330
8331
8332
8333
8334
8335
8336
8337
8338
8339
8340
8341
8342
8343
8344
8345
8346
8347
8348
8349
8350
8351
8352
8353
8354
8355
8356
8357
8358
8359
8360
8361
8362
8363
8364
8365
8366
8367
8368
8369
8370
8371
8372
8373
8374
8375
8376
8377
8378
8379
8380
8381
8382
8383
8384
8385
8386
8387
8388
8389
8390
8391
8392
8393
8394
8395
8396
8397
8398
8399
8400
8401
8402
8403
8404
8405
8406
8407
8408
8409
8410
8411
8412
8413
8414
8415
8416
8417
8418
8419
8420
8421
8422
8423
8424
8425
8426
8427
8428
8429
8430
8431
8432
8433
8434
8435
8436
8437
8438
8439
8440
8441
8442
8443
8444
8445
8446
8447
8448
8449
8450
8451
8452
8453
8454
8455
8456
8457
8458
8459
8460
8461
8462
8463
8464
8465
8466
8467
8468
8469
8470
8471
8472
8473
8474
8475
8476
8477
8478
8479
8480
8481
8482
8483
8484
8485
8486
8487
8488
8489
8490
8491
8492
8493
8494
8495
8496
8497
8498
8499
8500
8501
8502
8503
8504
8505
8506
8507
8508
8509
8510
8511
8512
8513
8514
8515
8516
8517
8518
8519
8520
8521
8522
8523
8524
8525
8526
8527
8528
8529
8530
8531
8532
8533
8534
8535
8536
8537
8538
8539
8540
8541
8542
8543
8544
8545
8546
8547
8548
8549
8550
8551
8552
8553
8554
8555
8556
8557
8558
8559
8560
8561
8562
8563
8564
8565
8566
8567
8568
8569
8570
8571
8572
8573
8574
8575
8576
8577
8578
8579
8580
8581
8582
8583
8584
8585
8586
8587
8588
8589
8590
8591
8592
8593
8594
8595
8596
8597
8598
8599
8600
8601
8602
8603
8604
8605
8606
8607
8608
8609
8610
8611
8612
8613
8614
8615
8616
8617
8618
8619
8620
8621
8622
8623
8624
8625
8626
8627
8628
8629
8630
8631
8632
8633
8634
8635
8636
8637
8638
8639
8640
8641
8642
8643
8644
8645
8646
8647
8648
8649
8650
8651
8652
8653
8654
8655
8656
8657
8658
8659
8660
8661
8662
8663
8664
8665
8666
8667
8668
8669
8670
8671
8672
8673
8674
8675
8676
8677
8678
8679
8680
8681
8682
8683
8684
8685
8686
8687
8688
8689
8690
8691
8692
8693
8694
8695
8696
8697
8698
8699
8700
8701
8702
8703
8704
8705
8706
8707
8708
8709
8710
8711
8712
8713
8714
8715
8716
8717
8718
8719
8720
8721
8722
8723
8724
8725
8726
8727
8728
8729
8730
8731
8732
8733
8734
8735
8736
8737
8738
8739
8740
8741
8742
8743
8744
8745
8746
8747
8748
8749
8750
8751
8752
8753
8754
8755
8756
8757
8758
8759
8760
8761
8762
8763
8764
8765
8766
8767
8768
8769
8770
8771
8772
8773
8774
8775
8776
8777
8778
8779
8780
8781
8782
8783
8784
8785
8786
8787
8788
8789
8790
8791
8792
8793
8794
8795
8796
8797
8798
8799
8800
8801
8802
8803
8804
8805
8806
8807
8808
8809
8810
8811
8812
8813
8814
8815
8816
8817
8818
8819
8820
8821
8822
8823
8824
8825
8826
8827
8828
8829
8830
8831
8832
8833
8834
8835
8836
8837
8838
8839
8840
8841
8842
8843
8844
8845
8846
8847
8848
8849
8850
8851
8852
8853
8854
8855
8856
8857
8858
8859
8860
8861
8862
8863
8864
8865
8866
8867
8868
8869
8870
8871
8872
8873
8874
8875
8876
8877
8878
8879
8880
8881
8882
8883
8884
8885
8886
8887
8888
8889
8890
8891
8892
8893
8894
8895
8896
8897
8898
8899
8900
8901
8902
8903
8904
8905
8906
8907
8908
8909
8910
8911
8912
8913
8914
8915
8916
8917
8918
8919
8920
8921
8922
8923
8924
8925
8926
8927
8928
8929
8930
8931
8932
8933
8934
8935
8936
8937
8938
8939
8940
8941
8942
8943
8944
8945
8946
8947
8948
8949
8950
8951
8952
8953
8954
8955
8956
8957
8958
8959
8960
8961
8962
8963
8964
8965
8966
8967
8968
8969
8970
8971
8972
8973
8974
8975
8976
8977
8978
8979
8980
8981
8982
8983
8984
8985
8986
8987
8988
8989
8990
8991
8992
8993
8994
8995
8996
8997
8998
8999
9000
9001
9002
9003
9004
9005
9006
9007
9008
9009
9010
9011
9012
9013
9014
9015
9016
9017
9018
9019
9020
9021
9022
9023
9024
9025
9026
9027
9028
9029
9030
9031
9032
9033
9034
9035
9036
9037
9038
9039
9040
9041
9042
9043
9044
9045
9046
9047
9048
9049
9050
9051
9052
9053
9054
9055
9056
9057
9058
9059
9060
9061
9062
9063
9064
9065
9066
9067
9068
9069
9070
9071
9072
9073
9074
9075
9076
9077
9078
9079
9080
9081
9082
9083
9084
9085
9086
9087
9088
9089
9090
9091
9092
9093
9094
9095
9096
9097
9098
9099
9100
9101
9102
9103
9104
9105
9106
9107
9108
9109
9110
9111
9112
9113
9114
9115
9116
9117
9118
9119
9120
9121
9122
9123
9124
9125
9126
9127
9128
9129
9130
9131
9132
9133
9134
9135
9136
9137
9138
9139
9140
9141
9142
9143
9144
9145
9146
9147
9148
9149
9150
9151
9152
9153
9154
9155
9156
9157
9158
9159
9160
9161
9162
9163
9164
9165
9166
9167
9168
9169
9170
9171
9172
9173
9174
9175
9176
9177
9178
9179
9180
9181
9182
9183
9184
9185
9186
9187
9188
9189
9190
9191
9192
9193
9194
9195
9196
9197
9198
9199
9200
9201
9202
9203
9204
9205
9206
9207
9208
9209
9210
9211
9212
9213
9214
9215
9216
9217
9218
9219
9220
9221
9222
9223
9224
9225
9226
9227
9228
9229
9230
9231
9232
9233
9234
9235
9236
9237
9238
9239
9240
9241
9242
9243
9244
9245
9246
9247
9248
9249
9250
9251
9252
9253
9254
9255
9256
9257
9258
9259
9260
9261
9262
9263
9264
9265
9266
9267
9268
9269
9270
9271
9272
9273
9274
9275
9276
9277
9278
9279
9280
9281
9282
9283
9284
9285
9286
9287
9288
9289
9290
9291
9292
9293
9294
9295
9296
9297
9298
9299
9300
9301
9302
9303
9304
9305
9306
9307
9308
9309
9310
9311
9312
9313
9314
9315
9316
9317
9318
9319
9320
9321
9322
9323
9324
9325
9326
9327
9328
9329
9330
9331
9332
9333
9334
9335
9336
9337
9338
9339
9340
9341
9342
9343
9344
9345
9346
9347
9348
9349
9350
9351
9352
9353
9354
9355
9356
9357
9358
9359
9360
9361
9362
9363
9364
9365
9366
9367
9368
9369
9370
9371
9372
9373
9374
9375
9376
9377
9378
9379
9380
9381
9382
9383
9384
9385
9386
9387
9388
9389
9390
9391
9392
9393
9394
9395
9396
9397
9398
9399
9400
9401
9402
9403
9404
9405
9406
9407
9408
9409
9410
9411
9412
9413
9414
9415
9416
9417
9418
9419
9420
9421
9422
9423
9424
9425
9426
9427
9428
9429
9430
9431
9432
9433
9434
9435
9436
9437
9438
9439
9440
9441
9442
9443
9444
9445
9446
9447
9448
9449
9450
9451
9452
9453
9454
9455
9456
9457
9458
9459
9460
9461
9462
9463
9464
9465
9466
9467
9468
9469
9470
9471
9472
9473
9474
9475
9476
9477
9478
9479
9480
9481
9482
9483
9484
9485
9486
9487
9488
9489
9490
9491
9492
9493
9494
9495
9496
9497
9498
9499
9500
9501
9502
9503
9504
9505
9506
9507
9508
9509
9510
9511
9512
9513
9514
9515
9516
9517
9518
9519
9520
9521
9522
9523
9524
9525
9526
9527
9528
9529
9530
9531
9532
9533
9534
9535
9536
9537
9538
9539
9540
9541
9542
9543
9544
9545
9546
9547
9548
9549
9550
9551
9552
9553
9554
9555
9556
9557
9558
9559
9560
9561
9562
9563
9564
9565
9566
9567
9568
9569
9570
9571
9572
9573
9574
9575
9576
9577
9578
9579
9580
9581
9582
9583
9584
9585
9586
9587
9588
9589
9590
9591
9592
9593
9594
9595
9596
9597
9598
9599
9600
9601
9602
9603
9604
9605
9606
9607
9608
9609
9610
9611
9612
9613
9614
9615
9616
9617
9618
9619
9620
9621
9622
9623
9624
9625
9626
9627
9628
9629
9630
9631
9632
9633
9634
9635
9636
9637
9638
9639
9640
9641
9642
9643
9644
9645
9646
9647
9648
9649
9650
9651
9652
9653
9654
9655
9656
9657
9658
9659
9660
9661
9662
9663
9664
9665
9666
9667
9668
9669
9670
9671
9672
9673
9674
9675
9676
9677
9678
9679
9680
9681
9682
9683
9684
9685
9686
9687
9688
9689
9690
9691
9692
9693
9694
9695
9696
9697
9698
9699
9700
9701
9702
9703
9704
9705
9706
9707
9708
9709
9710
9711
9712
9713
9714
9715
9716
9717
9718
9719
9720
9721
9722
9723
9724
9725
9726
9727
9728
9729
9730
9731
9732
9733
9734
9735
9736
9737
9738
9739
9740
9741
9742
9743
9744
9745
9746
9747
9748
9749
9750
9751
9752
9753
9754
9755
9756
9757
9758
9759
9760
9761
9762
9763
9764
9765
9766
9767
9768
9769
9770
9771
9772
9773
9774
9775
9776
9777
9778
9779
9780
9781
9782
9783
9784
9785
9786
9787
9788
9789
9790
9791
9792
9793
9794
9795
9796
9797
9798
9799
9800
9801
9802
9803
9804
9805
9806
9807
9808
9809
9810
9811
9812
9813
9814
9815
9816
9817
9818
9819
9820
9821
9822
9823
9824
9825
9826
9827
9828
9829
9830
9831
9832
9833
9834
9835
9836
9837
9838
9839
9840
9841
9842
9843
9844
9845
9846
9847
9848
9849
9850
9851
9852
9853
9854
9855
9856
9857
9858
9859
9860
9861
9862
9863
9864
9865
9866
9867
9868
9869
9870
9871
9872
9873
9874
9875
9876
9877
9878
9879
9880
9881
9882
9883
9884
9885
9886
9887
9888
9889
9890
9891
9892
9893
9894
9895
9896
9897
9898
9899
9900
9901
9902
9903
9904
9905
9906
9907
9908
9909
9910
9911
9912
9913
9914
9915
9916
9917
9918
9919
9920
9921
9922
9923
9924
9925
9926
9927
9928
9929
9930
9931
9932
9933
9934
9935
9936
9937
9938
9939
9940
9941
9942
9943
9944
9945
9946
9947
9948
9949
9950
9951
9952
9953
9954
9955
9956
9957
9958
9959
9960
9961
9962
9963
9964
9965
9966
9967
9968
9969
9970
9971
9972
9973
9974
9975
9976
9977
9978
9979
9980
9981
9982
9983
9984
9985
9986
9987
9988
9989
9990
9991
9992
9993
9994
9995
9996
9997
9998
9999
10000
10001
10002
10003
10004
10005
10006
10007
10008
10009
10010
10011
10012
10013
10014
10015
10016
10017
10018
10019
10020
10021
10022
10023
10024
10025
10026
10027
10028
10029
10030
10031
10032
10033
10034
10035
10036
10037
10038
10039
10040
10041
10042
10043
10044
10045
10046
10047
10048
10049
10050
10051
10052
10053
10054
10055
10056
10057
10058
10059
10060
10061
10062
10063
10064
10065
10066
10067
10068
10069
10070
10071
10072
10073
10074
10075
10076
10077
10078
10079
10080
10081
10082
10083
10084
10085
10086
10087
10088
10089
10090
10091
10092
10093
10094
10095
10096
10097
10098
10099
10100
10101
10102
10103
10104
10105
10106
10107
10108
10109
10110
10111
10112
10113
10114
10115
10116
10117
10118
10119
10120
10121
10122
10123
10124
10125
10126
10127
10128
10129
10130
10131
10132
10133
10134
10135
10136
10137
10138
10139
10140
10141
10142
10143
10144
10145
10146
10147
10148
10149
10150
10151
10152
10153
10154
10155
10156
10157
10158
10159
10160
10161
10162
10163
10164
10165
10166
10167
10168
10169
10170
10171
10172
10173
10174
10175
10176
10177
10178
10179
10180
10181
10182
10183
10184
10185
10186
10187
10188
10189
10190
10191
10192
10193
10194
10195
10196
10197
10198
10199
10200
10201
10202
10203
10204
10205
10206
10207
10208
10209
10210
10211
10212
10213
10214
10215
10216
10217
10218
10219
10220
10221
10222
10223
10224
10225
10226
10227
10228
10229
10230
10231
10232
10233
10234
10235
10236
10237
10238
10239
10240
10241
10242
10243
10244
10245
10246
10247
10248
10249
10250
10251
10252
10253
10254
10255
10256
10257
10258
10259
10260
10261
10262
10263
10264
10265
10266
10267
10268
10269
10270
10271
10272
10273
10274
10275
10276
10277
10278
10279
10280
10281
10282
10283
10284
10285
10286
10287
10288
10289
10290
10291
10292
10293
10294
10295
10296
10297
10298
10299
10300
10301
10302
10303
10304
10305
10306
10307
10308
10309
10310
10311
10312
10313
10314
10315
10316
10317
10318
10319
10320
10321
10322
10323
10324
10325
10326
10327
10328
10329
10330
10331
10332
10333
10334
10335
10336
10337
10338
10339
10340
10341
10342
10343
10344
10345
10346
10347
10348
10349
10350
10351
10352
10353
10354
10355
10356
10357
10358
10359
10360
10361
10362
10363
10364
10365
10366
10367
10368
10369
10370
10371
10372
10373
10374
10375
10376
10377
10378
10379
10380
10381
10382
10383
10384
10385
10386
10387
10388
10389
10390
10391
10392
10393
10394
10395
10396
10397
10398
10399
10400
10401
10402
10403
10404
10405
10406
10407
10408
10409
10410
10411
10412
10413
10414
10415
10416
10417
10418
10419
10420
10421
10422
10423
10424
10425
10426
10427
10428
10429
10430
10431
10432
10433
10434
10435
10436
10437
10438
10439
10440
10441
10442
10443
10444
10445
10446
10447
10448
10449
10450
10451
10452
10453
10454
10455
10456
10457
10458
10459
10460
10461
10462
10463
10464
10465
10466
10467
10468
10469
10470
10471
10472
10473
10474
10475
10476
10477
10478
10479
10480
10481
10482
10483
10484
10485
10486
10487
10488
10489
10490
10491
10492
10493
10494
10495
10496
10497
10498
10499
10500
10501
10502
10503
10504
10505
10506
10507
10508
10509
10510
10511
10512
10513
10514
10515
10516
10517
10518
10519
10520
10521
10522
10523
10524
10525
10526
10527
10528
10529
10530
10531
10532
10533
10534
10535
10536
10537
10538
10539
10540
10541
10542
10543
10544
10545
10546
10547
10548
10549
10550
10551
10552
10553
10554
10555
10556
10557
10558
10559
10560
10561
10562
10563
10564
10565
10566
10567
10568
10569
10570
10571
10572
10573
10574
10575
10576
10577
10578
10579
10580
10581
10582
10583
10584
10585
10586
10587
10588
10589
10590
10591
10592
10593
10594
10595
10596
10597
10598
10599
10600
10601
10602
10603
10604
10605
10606
10607
10608
10609
10610
10611
10612
10613
10614
10615
10616
10617
10618
10619
10620
10621
10622
10623
10624
10625
10626
10627
10628
10629
10630
10631
10632
10633
10634
10635
10636
10637
10638
10639
10640
10641
10642
10643
10644
10645
10646
10647
10648
10649
10650
10651
10652
10653
10654
10655
10656
10657
10658
10659
10660
10661
10662
10663
10664
10665
10666
10667
10668
10669
10670
10671
10672
10673
10674
10675
10676
10677
10678
10679
10680
10681
10682
10683
10684
10685
10686
10687
10688
10689
10690
10691
10692
10693
10694
10695
10696
10697
10698
10699
10700
10701
10702
10703
10704
10705
10706
10707
10708
10709
10710
10711
10712
10713
10714
10715
10716
10717
10718
10719
10720
10721
10722
10723
10724
10725
10726
10727
10728
10729
10730
10731
10732
10733
10734
10735
10736
10737
10738
10739
10740
10741
10742
10743
10744
10745
10746
10747
10748
10749
10750
10751
10752
10753
10754
10755
10756
10757
10758
10759
10760
10761
10762
10763
10764
10765
10766
10767
10768
10769
10770
10771
10772
10773
10774
10775
10776
10777
10778
10779
10780
10781
10782
10783
10784
10785
10786
10787
10788
10789
10790
10791
10792
10793
10794
10795
10796
10797
10798
10799
10800
10801
10802
10803
10804
10805
10806
10807
10808
10809
10810
10811
10812
10813
10814
10815
10816
10817
10818
10819
10820
10821
10822
10823
10824
10825
10826
10827
10828
10829
10830
10831
10832
10833
10834
10835
10836
10837
10838
10839
10840
10841
10842
10843
10844
10845
10846
10847
10848
10849
10850
10851
10852
10853
10854
10855
10856
10857
10858
10859
10860
10861
10862
10863
10864
10865
10866
10867
10868
10869
10870
10871
10872
10873
10874
10875
10876
10877
10878
10879
10880
10881
10882
10883
10884
10885
10886
10887
10888
10889
10890
10891
10892
10893
10894
10895
10896
10897
10898
10899
10900
10901
10902
10903
10904
10905
10906
10907
10908
10909
10910
10911
10912
10913
10914
10915
10916
10917
10918
10919
10920
10921
10922
10923
10924
10925
10926
10927
10928
10929
10930
10931
10932
10933
10934
10935
10936
10937
10938
10939
10940
10941
10942
10943
10944
10945
10946
10947
10948
10949
10950
10951
10952
10953
10954
10955
10956
10957
10958
10959
10960
10961
10962
10963
10964
10965
10966
10967
10968
10969
10970
10971
10972
10973
10974
10975
10976
10977
10978
10979
10980
10981
10982
10983
10984
10985
10986
10987
10988
10989
10990
10991
10992
10993
10994
10995
10996
10997
10998
10999
11000
11001
11002
11003
11004
11005
11006
11007
11008
11009
11010
11011
11012
11013
11014
11015
11016
11017
11018
11019
11020
11021
11022
11023
11024
11025
11026
11027
11028
11029
11030
11031
11032
11033
11034
11035
11036
11037
11038
11039
11040
11041
11042
11043
11044
11045
11046
11047
11048
11049
11050
11051
11052
11053
11054
11055
11056
11057
11058
11059
11060
11061
11062
11063
11064
11065
11066
11067
11068
11069
11070
11071
11072
11073
11074
11075
11076
11077
11078
11079
11080
11081
11082
11083
11084
11085
11086
11087
11088
11089
11090
11091
11092
11093
11094
11095
11096
11097
11098
11099
11100
11101
11102
11103
11104
11105
11106
11107
11108
11109
11110
11111
11112
11113
11114
11115
11116
11117
11118
11119
11120
11121
11122
11123
11124
11125
11126
11127
11128
11129
11130
11131
11132
11133
11134
11135
11136
11137
11138
11139
11140
11141
11142
11143
11144
11145
11146
11147
11148
11149
11150
11151
11152
11153
11154
11155
11156
11157
11158
11159
11160
11161
11162
11163
11164
11165
11166
11167
11168
11169
11170
11171
11172
11173
11174
11175
11176
11177
11178
11179
11180
11181
11182
11183
11184
11185
11186
11187
11188
11189
11190
11191
11192
11193
11194
11195
11196
11197
11198
11199
11200
11201
11202
11203
11204
11205
11206
11207
11208
11209
11210
11211
11212
11213
11214
11215
11216
11217
11218
11219
11220
11221
11222
11223
11224
11225
11226
11227
11228
11229
11230
11231
11232
11233
11234
11235
11236
11237
11238
11239
11240
11241
11242
11243
11244
11245
11246
11247
11248
11249
11250
11251
11252
11253
11254
11255
11256
11257
11258
11259
11260
11261
11262
11263
11264
11265
11266
11267
11268
11269
11270
11271
11272
11273
11274
11275
11276
11277
11278
11279
11280
11281
11282
11283
11284
11285
11286
11287
11288
11289
11290
11291
11292
11293
11294
11295
11296
11297
11298
11299
11300
11301
11302
11303
11304
11305
11306
11307
11308
11309
11310
11311
11312
11313
11314
11315
11316
11317
11318
11319
11320
11321
11322
11323
11324
11325
11326
11327
11328
11329
11330
11331
11332
11333
11334
11335
11336
11337
11338
11339
11340
11341
11342
11343
11344
11345
11346
11347
11348
11349
11350
11351
11352
11353
11354
11355
11356
11357
11358
11359
11360
11361
11362
11363
11364
11365
11366
11367
11368
11369
11370
11371
11372
11373
11374
11375
11376
11377
11378
11379
11380
11381
11382
11383
11384
11385
11386
11387
11388
11389
11390
11391
11392
11393
11394
11395
11396
11397
11398
11399
11400
11401
11402
11403
11404
11405
11406
11407
11408
11409
11410
11411
11412
11413
11414
11415
11416
11417
11418
11419
11420
11421
11422
11423
11424
11425
11426
11427
11428
11429
11430
11431
11432
11433
11434
11435
11436
11437
11438
11439
11440
11441
11442
11443
11444
11445
11446
11447
11448
11449
11450
11451
11452
11453
11454
11455
11456
11457
11458
11459
11460
11461
11462
11463
11464
11465
11466
11467
11468
11469
11470
11471
11472
11473
11474
11475
11476
11477
11478
11479
11480
11481
11482
11483
11484
11485
11486
11487
11488
11489
11490
11491
11492
11493
11494
11495
11496
11497
11498
11499
11500
11501
11502
11503
11504
11505
11506
11507
11508
11509
11510
11511
11512
11513
11514
11515
11516
11517
11518
11519
11520
11521
11522
11523
11524
11525
11526
11527
11528
11529
11530
11531
11532
11533
11534
11535
11536
11537
11538
11539
11540
11541
11542
11543
11544
11545
11546
11547
11548
11549
11550
11551
11552
11553
11554
11555
11556
11557
11558
11559
11560
11561
11562
11563
11564
11565
11566
11567
11568
11569
11570
11571
11572
11573
11574
11575
11576
11577
11578
11579
11580
11581
11582
11583
11584
11585
11586
11587
11588
11589
11590
11591
11592
11593
11594
11595
11596
11597
11598
11599
11600
11601
11602
11603
11604
11605
11606
11607
11608
11609
11610
11611
11612
11613
11614
11615
11616
11617
11618
11619
11620
11621
11622
11623
11624
11625
11626
11627
11628
11629
11630
11631
11632
11633
11634
11635
11636
11637
11638
11639
11640
11641
11642
11643
11644
11645
11646
11647
11648
11649
11650
11651
11652
11653
11654
11655
11656
11657
11658
11659
11660
11661
11662
11663
11664
11665
11666
11667
11668
11669
11670
11671
11672
11673
11674
11675
11676
11677
11678
11679
11680
11681
11682
11683
11684
11685
11686
11687
11688
11689
11690
11691
11692
11693
11694
11695
11696
11697
11698
11699
11700
11701
11702
11703
11704
11705
11706
11707
11708
11709
11710
11711
11712
11713
11714
11715
11716
11717
11718
11719
11720
11721
11722
11723
11724
11725
11726
11727
11728
11729
11730
11731
11732
11733
11734
11735
11736
11737
11738
11739
11740
11741
11742
11743
11744
11745
11746
11747
11748
11749
11750
11751
11752
11753
11754
11755
11756
11757
11758
11759
11760
11761
11762
11763
11764
11765
11766
11767
11768
11769
11770
11771
11772
11773
11774
11775
11776
11777
11778
11779
11780
11781
11782
11783
11784
11785
11786
11787
11788
11789
11790
11791
11792
11793
11794
11795
11796
11797
11798
11799
11800
11801
11802
11803
11804
11805
11806
11807
11808
11809
11810
11811
11812
11813
11814
11815
11816
11817
11818
11819
11820
11821
11822
11823
11824
11825
11826
11827
11828
11829
11830
11831
11832
11833
11834
11835
11836
11837
11838
11839
11840
11841
11842
11843
11844
11845
11846
11847
11848
11849
11850
11851
11852
11853
11854
11855
11856
11857
11858
11859
11860
11861
11862
11863
11864
11865
11866
11867
11868
11869
11870
11871
11872
11873
11874
11875
11876
11877
11878
11879
11880
11881
11882
11883
11884
11885
11886
11887
11888
11889
11890
11891
11892
11893
11894
11895
11896
11897
11898
11899
11900
11901
11902
11903
11904
11905
11906
11907
11908
11909
11910
11911
11912
11913
11914
11915
11916
11917
11918
11919
11920
11921
11922
11923
11924
11925
11926
11927
11928
11929
11930
11931
11932
11933
11934
11935
11936
11937
11938
11939
11940
11941
11942
11943
11944
11945
11946
11947
11948
11949
11950
11951
11952
11953
11954
11955
11956
11957
11958
11959
11960
11961
11962
11963
11964
11965
11966
11967
11968
11969
11970
11971
11972
11973
11974
11975
11976
11977
11978
11979
11980
11981
11982
11983
11984
11985
11986
11987
11988
11989
11990
11991
11992
11993
11994
11995
11996
11997
11998
11999
12000
12001
12002
12003
12004
12005
12006
12007
12008
12009
12010
12011
12012
12013
12014
12015
12016
12017
12018
12019
12020
12021
12022
12023
12024
12025
12026
12027
12028
12029
12030
12031
12032
12033
12034
12035
12036
12037
12038
12039
12040
12041
12042
12043
12044
12045
12046
12047
12048
12049
12050
12051
12052
12053
12054
12055
12056
12057
12058
12059
12060
12061
12062
12063
12064
12065
12066
12067
12068
12069
12070
12071
12072
12073
12074
12075
12076
12077
12078
12079
12080
12081
12082
12083
12084
12085
12086
12087
12088
12089
12090
12091
12092
12093
12094
12095
12096
12097
12098
12099
12100
12101
12102
12103
12104
12105
12106
12107
12108
12109
12110
12111
12112
12113
12114
12115
12116
12117
12118
12119
12120
12121
12122
12123
12124
12125
12126
12127
12128
12129
12130
12131
12132
12133
12134
12135
12136
12137
12138
12139
12140
12141
12142
12143
12144
12145
12146
12147
12148
12149
12150
12151
12152
12153
12154
12155
12156
12157
12158
12159
12160
12161
12162
12163
12164
12165
12166
12167
12168
12169
12170
12171
12172
12173
12174
12175
12176
12177
12178
12179
12180
12181
12182
12183
12184
12185
12186
12187
12188
12189
12190
12191
12192
12193
12194
12195
12196
12197
12198
12199
12200
12201
12202
12203
12204
12205
12206
12207
12208
12209
12210
12211
12212
12213
12214
12215
12216
12217
12218
12219
12220
12221
12222
12223
12224
12225
12226
12227
12228
12229
12230
12231
12232
12233
12234
12235
12236
12237
12238
12239
12240
12241
12242
12243
12244
12245
12246
12247
12248
12249
12250
12251
12252
12253
12254
12255
12256
12257
12258
12259
12260
12261
12262
12263
12264
12265
12266
12267
12268
12269
12270
12271
12272
12273
12274
12275
12276
12277
12278
12279
12280
12281
12282
12283
12284
12285
12286
12287
12288
12289
12290
12291
12292
12293
12294
12295
12296
12297
12298
12299
12300
12301
12302
12303
12304
12305
12306
12307
12308
12309
12310
12311
12312
12313
12314
12315
12316
12317
12318
12319
12320
12321
12322
12323
12324
12325
12326
12327
12328
12329
12330
12331
12332
12333
12334
12335
12336
12337
12338
12339
12340
12341
12342
12343
12344
12345
12346
12347
12348
12349
12350
12351
12352
12353
12354
12355
12356
12357
12358
12359
12360
12361
12362
12363
12364
12365
12366
12367
12368
12369
12370
12371
12372
12373
12374
12375
12376
12377
12378
12379
12380
12381
12382
12383
12384
12385
12386
12387
12388
12389
12390
12391
12392
12393
12394
12395
12396
12397
12398
12399
12400
12401
12402
12403
12404
12405
12406
12407
12408
12409
12410
12411
12412
12413
12414
12415
12416
12417
12418
12419
12420
12421
12422
12423
12424
12425
12426
12427
12428
12429
12430
12431
12432
12433
12434
12435
12436
12437
12438
12439
12440
12441
12442
12443
12444
12445
12446
12447
12448
12449
12450
12451
12452
12453
12454
12455
12456
12457
12458
12459
12460
12461
12462
12463
12464
12465
12466
12467
12468
12469
12470
12471
12472
12473
12474
12475
12476
12477
12478
12479
12480
12481
12482
12483
12484
12485
12486
12487
12488
12489
12490
12491
12492
12493
12494
12495
12496
12497
12498
12499
12500
12501
12502
12503
12504
12505
12506
12507
12508
12509
12510
12511
12512
12513
12514
12515
12516
12517
12518
12519
12520
12521
12522
12523
12524
12525
12526
12527
12528
12529
12530
12531
12532
12533
12534
12535
12536
12537
12538
12539
12540
12541
12542
12543
12544
12545
12546
12547
12548
12549
12550
12551
12552
12553
12554
12555
12556
12557
12558
12559
12560
12561
12562
12563
12564
12565
12566
12567
12568
12569
12570
12571
12572
12573
12574
12575
12576
12577
12578
12579
12580
12581
12582
12583
12584
12585
12586
12587
12588
12589
12590
12591
12592
12593
12594
12595
12596
12597
12598
12599
12600
12601
12602
12603
12604
12605
12606
12607
12608
12609
12610
12611
12612
12613
12614
12615
12616
12617
12618
12619
12620
12621
12622
12623
12624
12625
12626
12627
12628
12629
12630
12631
12632
12633
12634
12635
12636
12637
12638
12639
12640
12641
12642
12643
12644
12645
12646
12647
12648
12649
12650
12651
12652
12653
12654
12655
12656
12657
12658
12659
12660
12661
12662
12663
12664
12665
12666
12667
12668
12669
12670
12671
12672
12673
12674
12675
12676
12677
12678
12679
12680
12681
12682
12683
12684
12685
12686
12687
12688
12689
12690
12691
12692
12693
12694
12695
12696
12697
12698
12699
12700
12701
12702
12703
12704
12705
12706
12707
12708
12709
12710
12711
12712
12713
12714
12715
12716
12717
12718
12719
12720
12721
12722
12723
12724
12725
12726
12727
12728
12729
12730
12731
12732
12733
12734
12735
12736
12737
12738
12739
12740
12741
12742
12743
12744
12745
12746
12747
12748
12749
12750
12751
12752
12753
12754
12755
12756
12757
12758
12759
12760
12761
12762
12763
12764
12765
12766
12767
12768
12769
12770
12771
12772
12773
12774
12775
12776
12777
12778
12779
12780
12781
12782
12783
12784
12785
12786
12787
12788
12789
12790
12791
12792
12793
12794
12795
12796
12797
12798
12799
12800
12801
12802
12803
12804
12805
12806
12807
12808
12809
12810
12811
12812
12813
12814
12815
12816
12817
12818
12819
12820
12821
12822
12823
12824
12825
12826
12827
12828
12829
12830
12831
12832
12833
12834
12835
12836
12837
12838
12839
12840
12841
12842
12843
12844
12845
12846
12847
12848
12849
12850
12851
12852
12853
12854
12855
12856
12857
12858
12859
12860
12861
12862
12863
12864
12865
12866
12867
12868
12869
12870
12871
12872
12873
12874
12875
12876
12877
12878
12879
12880
12881
12882
12883
12884
12885
12886
12887
12888
12889
12890
12891
12892
12893
12894
12895
12896
12897
12898
12899
12900
12901
12902
12903
12904
12905
12906
12907
12908
12909
12910
12911
12912
12913
12914
12915
12916
12917
12918
12919
12920
12921
12922
12923
12924
12925
12926
12927
12928
12929
12930
12931
12932
12933
12934
12935
12936
12937
12938
12939
12940
12941
12942
12943
12944
12945
12946
12947
12948
12949
12950
12951
12952
12953
12954
12955
12956
12957
12958
12959
12960
12961
12962
12963
12964
12965
12966
12967
12968
12969
12970
12971
12972
12973
12974
12975
12976
12977
12978
12979
12980
12981
12982
12983
12984
12985
12986
12987
12988
12989
12990
12991
12992
12993
12994
12995
12996
12997
12998
12999
13000
13001
13002
13003
13004
13005
13006
13007
13008
13009
13010
13011
13012
13013
13014
13015
13016
13017
13018
13019
13020
13021
13022
13023
13024
13025
13026
13027
13028
13029
13030
13031
13032
13033
13034
13035
13036
13037
13038
13039
13040
13041
13042
13043
13044
13045
13046
13047
13048
13049
13050
13051
13052
13053
13054
13055
13056
13057
13058
13059
13060
13061
13062
13063
13064
13065
13066
13067
13068
13069
13070
13071
13072
13073
13074
13075
13076
13077
13078
13079
13080
13081
13082
13083
13084
13085
13086
13087
13088
13089
13090
13091
13092
13093
13094
13095
13096
13097
13098
13099
13100
13101
13102
13103
13104
13105
13106
13107
13108
13109
13110
13111
13112
13113
13114
13115
13116
13117
13118
13119
13120
13121
13122
13123
13124
13125
13126
13127
13128
13129
13130
13131
13132
13133
13134
13135
13136
13137
13138
13139
13140
13141
13142
13143
13144
13145
13146
13147
13148
13149
13150
13151
13152
13153
13154
13155
13156
13157
13158
13159
13160
13161
13162
13163
13164
13165
13166
13167
13168
13169
13170
13171
13172
13173
13174
13175
13176
13177
13178
13179
13180
13181
13182
13183
13184
13185
13186
13187
13188
13189
13190
13191
13192
13193
13194
13195
13196
13197
13198
13199
13200
13201
13202
13203
13204
13205
13206
13207
13208
13209
13210
13211
13212
13213
13214
13215
13216
13217
13218
13219
13220
13221
13222
13223
13224
13225
13226
13227
13228
13229
13230
13231
13232
13233
13234
13235
13236
13237
13238
13239
13240
13241
13242
13243
13244
13245
13246
13247
13248
13249
13250
13251
13252
13253
13254
13255
13256
13257
13258
13259
13260
13261
13262
13263
13264
13265
13266
13267
13268
13269
13270
13271
13272
13273
13274
13275
13276
13277
13278
13279
13280
13281
13282
13283
13284
13285
13286
13287
13288
13289
13290
13291
13292
13293
13294
13295
13296
13297
13298
13299
13300
13301
13302
13303
13304
13305
13306
13307
13308
13309
13310
13311
13312
13313
13314
13315
13316
13317
13318
13319
13320
13321
13322
13323
13324
13325
13326
13327
13328
13329
13330
13331
13332
13333
13334
13335
13336
13337
13338
13339
13340
13341
13342
13343
13344
13345
13346
13347
13348
13349
13350
13351
13352
13353
13354
13355
13356
13357
13358
13359
13360
13361
13362
13363
13364
13365
13366
13367
13368
13369
13370
13371
13372
13373
13374
13375
13376
13377
13378
13379
13380
13381
13382
13383
13384
13385
13386
13387
13388
13389
13390
13391
13392
13393
13394
13395
13396
13397
13398
13399
13400
13401
13402
13403
13404
13405
13406
13407
13408
13409
13410
13411
13412
13413
13414
13415
13416
13417
13418
13419
13420
13421
13422
13423
13424
13425
13426
13427
13428
13429
13430
13431
13432
13433
13434
13435
13436
13437
13438
13439
13440
13441
13442
13443
13444
13445
13446
13447
13448
13449
13450
13451
13452
13453
13454
13455
13456
13457
13458
13459
13460
13461
13462
13463
13464
13465
13466
13467
13468
13469
13470
13471
13472
13473
13474
13475
13476
13477
13478
13479
13480
13481
13482
13483
13484
13485
13486
13487
13488
13489
13490
13491
13492
13493
13494
13495
13496
13497
13498
13499
13500
13501
13502
13503
13504
13505
13506
13507
13508
13509
13510
13511
13512
13513
13514
13515
13516
13517
13518
13519
13520
13521
13522
13523
13524
13525
13526
13527
13528
13529
13530
13531
13532
13533
13534
13535
13536
13537
13538
13539
13540
13541
13542
13543
13544
13545
13546
13547
13548
13549
13550
13551
13552
13553
13554
13555
13556
13557
13558
13559
13560
13561
13562
13563
13564
13565
13566
13567
13568
13569
13570
13571
13572
13573
13574
13575
13576
13577
13578
13579
13580
13581
13582
13583
13584
13585
13586
13587
13588
13589
13590
13591
13592
13593
13594
13595
13596
13597
13598
13599
13600
13601
13602
13603
13604
13605
13606
13607
13608
13609
13610
13611
13612
13613
13614
13615
13616
13617
13618
13619
13620
13621
13622
13623
13624
13625
13626
13627
13628
13629
13630
13631
13632
13633
13634
13635
13636
13637
13638
13639
13640
13641
13642
13643
13644
13645
13646
13647
13648
13649
13650
13651
13652
13653
13654
13655
13656
13657
13658
13659
13660
13661
13662
13663
13664
13665
13666
13667
13668
13669
13670
13671
13672
13673
13674
13675
13676
13677
13678
13679
13680
13681
13682
13683
13684
13685
13686
13687
13688
13689
13690
13691
13692
13693
13694
13695
13696
13697
13698
13699
13700
13701
13702
13703
13704
13705
13706
13707
13708
13709
13710
13711
13712
13713
13714
13715
13716
13717
13718
13719
13720
13721
13722
13723
13724
13725
13726
13727
13728
13729
13730
13731
13732
13733
13734
13735
13736
13737
13738
13739
13740
13741
13742
13743
13744
13745
13746
13747
13748
13749
13750
13751
13752
13753
13754
13755
13756
13757
13758
13759
13760
13761
13762
13763
13764
13765
13766
13767
13768
13769
13770
13771
13772
13773
13774
13775
13776
13777
13778
13779
13780
13781
13782
13783
13784
13785
13786
13787
13788
13789
13790
13791
13792
13793
13794
13795
13796
13797
13798
13799
13800
13801
13802
13803
13804
13805
13806
13807
13808
13809
13810
13811
13812
13813
13814
13815
13816
13817
13818
13819
13820
13821
13822
13823
13824
13825
13826
13827
13828
13829
13830
13831
13832
13833
13834
13835
13836
13837
13838
13839
13840
13841
13842
13843
13844
13845
13846
13847
13848
13849
13850
13851
13852
13853
13854
13855
13856
13857
13858
13859
13860
13861
13862
13863
13864
13865
13866
13867
13868
13869
13870
13871
13872
13873
13874
13875
13876
13877
13878
13879
13880
13881
13882
13883
13884
13885
13886
13887
13888
13889
13890
13891
13892
13893
13894
13895
13896
13897
13898
13899
13900
13901
13902
13903
13904
13905
13906
13907
13908
13909
13910
13911
13912
13913
13914
13915
13916
13917
13918
13919
13920
13921
13922
13923
13924
13925
13926
13927
13928
13929
13930
13931
13932
13933
13934
13935
13936
13937
13938
13939
13940
13941
13942
13943
13944
13945
13946
13947
13948
13949
13950
13951
13952
13953
13954
13955
13956
13957
13958
13959
13960
13961
13962
13963
13964
13965
13966
13967
13968
13969
13970
13971
13972
13973
13974
13975
13976
13977
13978
13979
13980
13981
13982
13983
13984
13985
13986
13987
13988
13989
13990
13991
13992
13993
13994
13995
13996
13997
13998
13999
14000
14001
14002
14003
14004
14005
14006
14007
14008
14009
14010
14011
14012
14013
14014
14015
14016
14017
14018
14019
14020
14021
14022
14023
14024
14025
14026
14027
14028
14029
14030
14031
14032
14033
14034
14035
14036
14037
14038
14039
14040
14041
14042
14043
14044
14045
14046
14047
14048
14049
14050
14051
14052
14053
14054
14055
14056
14057
14058
14059
14060
14061
14062
14063
14064
14065
14066
14067
14068
14069
14070
14071
14072
14073
14074
14075
14076
14077
14078
14079
14080
14081
14082
14083
14084
14085
14086
14087
14088
14089
14090
14091
14092
14093
14094
14095
14096
14097
14098
14099
14100
14101
14102
14103
14104
14105
14106
14107
14108
14109
14110
14111
14112
14113
14114
14115
14116
14117
14118
14119
14120
14121
14122
14123
14124
14125
14126
14127
14128
14129
14130
14131
14132
14133
14134
14135
14136
14137
14138
14139
14140
14141
14142
14143
14144
14145
14146
14147
14148
14149
14150
14151
14152
14153
14154
14155
14156
14157
14158
14159
14160
14161
14162
14163
14164
14165
14166
14167
14168
14169
14170
14171
14172
14173
14174
14175
14176
14177
14178
14179
14180
14181
14182
14183
14184
14185
14186
14187
14188
14189
14190
14191
14192
14193
14194
14195
14196
14197
14198
14199
14200
14201
14202
14203
14204
14205
14206
14207
14208
14209
14210
14211
14212
14213
14214
14215
14216
14217
14218
14219
14220
14221
14222
14223
14224
14225
14226
14227
14228
14229
14230
14231
14232
14233
14234
14235
14236
14237
14238
14239
14240
14241
14242
14243
14244
14245
14246
14247
14248
14249
14250
14251
14252
14253
14254
14255
14256
14257
14258
14259
14260
14261
14262
14263
14264
14265
14266
14267
14268
14269
14270
14271
14272
14273
14274
14275
14276
14277
14278
14279
14280
14281
14282
14283
14284
14285
14286
14287
14288
14289
14290
14291
14292
14293
14294
14295
14296
14297
14298
14299
14300
14301
14302
14303
14304
14305
14306
14307
14308
14309
14310
14311
14312
14313
14314
14315
14316
14317
14318
14319
14320
14321
14322
14323
14324
14325
14326
14327
14328
14329
14330
14331
14332
14333
14334
14335
14336
14337
14338
14339
14340
14341
14342
14343
14344
14345
14346
14347
14348
14349
14350
14351
14352
14353
14354
14355
14356
14357
14358
14359
14360
14361
14362
14363
14364
14365
14366
14367
14368
14369
14370
14371
14372
14373
14374
14375
14376
14377
14378
14379
14380
14381
14382
14383
14384
14385
14386
14387
14388
14389
14390
14391
14392
14393
14394
14395
14396
14397
14398
14399
14400
14401
14402
14403
14404
14405
14406
14407
14408
14409
14410
14411
14412
14413
14414
14415
14416
14417
14418
14419
14420
14421
14422
14423
14424
14425
14426
14427
14428
14429
14430
14431
14432
14433
14434
14435
14436
14437
14438
14439
14440
14441
14442
14443
14444
14445
14446
14447
14448
14449
14450
14451
14452
14453
14454
14455
14456
14457
14458
14459
14460
14461
14462
14463
14464
14465
14466
14467
14468
14469
14470
14471
14472
14473
14474
14475
14476
14477
14478
14479
14480
14481
14482
14483
14484
14485
14486
14487
14488
14489
14490
14491
14492
14493
14494
14495
14496
14497
14498
14499
14500
14501
14502
14503
14504
14505
14506
14507
14508
14509
14510
14511
14512
14513
14514
14515
14516
14517
14518
14519
14520
14521
14522
14523
14524
14525
14526
14527
14528
14529
14530
14531
14532
14533
14534
14535
14536
14537
14538
14539
14540
14541
14542
14543
14544
14545
14546
14547
14548
14549
14550
14551
14552
14553
14554
14555
14556
14557
14558
14559
14560
14561
14562
14563
14564
14565
14566
14567
14568
14569
14570
14571
14572
14573
14574
14575
14576
14577
14578
14579
14580
14581
14582
14583
14584
14585
14586
14587
14588
14589
14590
14591
14592
14593
14594
14595
14596
14597
14598
14599
14600
14601
14602
14603
14604
14605
14606
14607
14608
14609
14610
14611
14612
14613
14614
14615
14616
14617
14618
14619
14620
14621
14622
14623
14624
14625
14626
14627
14628
14629
14630
14631
14632
14633
14634
14635
14636
14637
14638
14639
14640
14641
14642
14643
14644
14645
14646
14647
14648
14649
14650
14651
14652
14653
14654
14655
14656
14657
14658
14659
14660
14661
14662
14663
14664
14665
14666
14667
14668
14669
14670
14671
14672
14673
14674
14675
14676
14677
14678
14679
14680
14681
14682
14683
14684
14685
14686
14687
14688
14689
14690
14691
14692
14693
14694
14695
14696
14697
14698
14699
14700
14701
14702
14703
14704
14705
14706
14707
14708
14709
14710
14711
14712
14713
14714
14715
14716
14717
14718
14719
14720
14721
14722
14723
14724
14725
14726
14727
14728
14729
14730
14731
14732
14733
14734
14735
14736
14737
14738
14739
14740
14741
14742
14743
14744
14745
14746
14747
14748
14749
14750
14751
14752
14753
14754
14755
14756
14757
14758
14759
14760
14761
14762
14763
14764
14765
14766
14767
14768
14769
14770
14771
14772
14773
14774
14775
14776
14777
14778
14779
14780
14781
14782
14783
14784
14785
14786
14787
14788
14789
14790
14791
14792
14793
14794
14795
14796
14797
14798
14799
14800
14801
14802
14803
14804
14805
14806
14807
14808
14809
14810
14811
14812
14813
14814
14815
14816
14817
14818
14819
14820
14821
14822
14823
14824
14825
14826
14827
14828
14829
14830
14831
14832
14833
14834
14835
14836
14837
14838
14839
14840
14841
14842
14843
14844
14845
14846
14847
14848
14849
14850
14851
14852
14853
14854
14855
14856
14857
14858
14859
14860
14861
14862
14863
14864
14865
14866
14867
14868
14869
14870
14871
14872
14873
14874
14875
14876
14877
14878
14879
14880
14881
14882
14883
14884
14885
14886
14887
14888
14889
14890
14891
14892
14893
14894
14895
14896
14897
14898
14899
14900
14901
14902
14903
14904
14905
14906
14907
14908
14909
14910
14911
14912
14913
14914
14915
14916
14917
14918
14919
14920
14921
14922
14923
14924
14925
14926
14927
14928
14929
14930
14931
14932
14933
14934
14935
14936
14937
14938
14939
14940
14941
14942
14943
14944
14945
14946
14947
14948
14949
14950
14951
14952
14953
14954
14955
14956
14957
14958
14959
14960
14961
14962
14963
14964
14965
14966
14967
14968
14969
14970
14971
14972
14973
14974
14975
14976
14977
14978
14979
14980
14981
14982
14983
14984
14985
14986
14987
14988
14989
14990
14991
14992
14993
14994
14995
14996
14997
14998
14999
15000
15001
15002
15003
15004
15005
15006
15007
15008
15009
15010
15011
15012
15013
15014
15015
15016
15017
15018
15019
15020
15021
15022
15023
15024
15025
15026
15027
15028
15029
15030
15031
15032
15033
15034
15035
15036
15037
15038
15039
15040
15041
15042
15043
15044
15045
15046
15047
15048
15049
15050
15051
15052
15053
15054
15055
15056
15057
15058
15059
15060
15061
15062
15063
15064
15065
15066
15067
15068
15069
15070
15071
15072
15073
15074
15075
15076
15077
15078
15079
15080
15081
15082
15083
15084
15085
15086
15087
15088
15089
15090
15091
15092
15093
15094
15095
15096
15097
15098
15099
15100
15101
15102
15103
15104
15105
15106
15107
15108
15109
15110
15111
15112
15113
15114
15115
15116
15117
15118
15119
15120
15121
15122
15123
15124
15125
15126
15127
15128
15129
15130
15131
15132
15133
15134
15135
15136
15137
15138
15139
15140
15141
15142
15143
15144
15145
15146
15147
15148
15149
15150
15151
15152
15153
15154
15155
15156
15157
15158
15159
15160
15161
15162
15163
15164
15165
15166
15167
15168
15169
15170
15171
15172
15173
15174
15175
15176
15177
15178
15179
15180
15181
15182
15183
15184
15185
15186
15187
15188
15189
15190
15191
15192
15193
15194
15195
15196
15197
15198
15199
15200
15201
15202
15203
15204
15205
15206
15207
15208
15209
15210
15211
15212
15213
15214
15215
15216
15217
15218
15219
15220
15221
15222
15223
15224
15225
15226
15227
15228
15229
15230
15231
15232
15233
15234
15235
15236
15237
15238
15239
15240
15241
15242
15243
15244
15245
15246
15247
15248
15249
15250
15251
15252
15253
15254
15255
15256
15257
15258
15259
15260
15261
15262
15263
15264
15265
15266
15267
15268
15269
15270
15271
15272
15273
15274
15275
15276
15277
15278
15279
15280
15281
15282
15283
15284
15285
15286
15287
15288
15289
15290
15291
15292
15293
15294
15295
15296
15297
15298
15299
15300
15301
15302
15303
15304
15305
15306
15307
15308
15309
15310
15311
15312
15313
15314
15315
15316
15317
15318
15319
15320
15321
15322
15323
15324
15325
15326
15327
15328
15329
15330
15331
15332
15333
15334
15335
15336
15337
15338
15339
15340
15341
15342
15343
15344
15345
15346
15347
15348
15349
15350
15351
15352
15353
15354
15355
15356
15357
15358
15359
15360
15361
15362
15363
15364
15365
15366
15367
15368
15369
15370
15371
15372
15373
15374
15375
15376
15377
15378
15379
15380
15381
15382
15383
15384
15385
15386
15387
15388
15389
15390
15391
15392
15393
15394
15395
15396
15397
15398
15399
15400
15401
15402
15403
15404
15405
15406
15407
15408
15409
15410
15411
15412
15413
15414
15415
15416
15417
15418
15419
15420
15421
15422
15423
15424
15425
15426
15427
15428
15429
15430
15431
15432
15433
15434
15435
15436
15437
15438
15439
15440
15441
15442
15443
15444
15445
15446
15447
15448
15449
15450
15451
15452
15453
15454
15455
15456
15457
15458
15459
15460
15461
15462
15463
15464
15465
15466
15467
15468
15469
15470
15471
15472
15473
15474
15475
15476
15477
15478
15479
15480
15481
15482
15483
15484
15485
15486
15487
15488
15489
15490
15491
15492
15493
15494
15495
15496
15497
15498
15499
15500
15501
15502
15503
15504
15505
15506
15507
15508
15509
15510
15511
15512
15513
15514
15515
15516
15517
15518
15519
15520
15521
15522
15523
15524
15525
15526
15527
15528
15529
15530
15531
15532
15533
15534
15535
15536
15537
15538
15539
15540
15541
15542
15543
15544
15545
15546
15547
15548
15549
15550
15551
15552
15553
15554
15555
15556
15557
15558
15559
15560
15561
15562
15563
15564
15565
15566
15567
15568
15569
15570
15571
15572
15573
15574
15575
15576
15577
15578
15579
15580
15581
15582
15583
15584
15585
15586
15587
15588
15589
15590
15591
15592
15593
15594
15595
15596
15597
15598
15599
15600
15601
15602
15603
15604
15605
15606
15607
15608
15609
15610
15611
15612
15613
15614
15615
15616
15617
15618
15619
15620
15621
15622
15623
15624
15625
15626
15627
15628
15629
15630
15631
15632
15633
15634
15635
15636
15637
15638
15639
15640
15641
15642
15643
15644
15645
15646
15647
15648
15649
15650
15651
15652
15653
15654
15655
15656
15657
15658
15659
15660
15661
15662
15663
15664
15665
15666
15667
15668
15669
15670
15671
15672
15673
15674
15675
15676
15677
15678
15679
15680
15681
15682
15683
15684
15685
15686
15687
15688
15689
15690
15691
15692
15693
15694
15695
15696
15697
15698
15699
15700
15701
15702
15703
15704
15705
15706
15707
15708
15709
15710
15711
15712
15713
15714
15715
15716
15717
15718
15719
15720
15721
15722
15723
15724
15725
15726
15727
15728
15729
15730
15731
15732
15733
15734
15735
15736
15737
15738
15739
15740
15741
15742
15743
15744
15745
15746
15747
15748
15749
15750
15751
15752
15753
15754
15755
15756
15757
15758
15759
15760
15761
15762
15763
15764
15765
15766
15767
15768
15769
15770
15771
15772
15773
15774
15775
15776
15777
15778
15779
15780
15781
15782
15783
15784
15785
15786
15787
15788
15789
15790
15791
15792
15793
15794
15795
15796
15797
15798
15799
15800
15801
15802
15803
15804
15805
15806
15807
15808
15809
15810
15811
15812
15813
15814
15815
15816
15817
15818
15819
15820
15821
15822
15823
15824
15825
15826
15827
15828
15829
15830
15831
15832
15833
15834
15835
15836
15837
15838
15839
15840
15841
15842
15843
15844
15845
15846
15847
15848
15849
15850
15851
15852
15853
15854
15855
15856
15857
15858
15859
15860
15861
15862
15863
15864
15865
15866
15867
15868
15869
15870
15871
15872
15873
15874
15875
15876
15877
15878
15879
15880
15881
15882
15883
15884
15885
15886
15887
15888
15889
15890
15891
15892
15893
15894
15895
15896
15897
15898
15899
15900
15901
15902
15903
15904
15905
15906
15907
15908
15909
15910
15911
15912
15913
15914
15915
15916
15917
15918
15919
15920
15921
15922
15923
15924
15925
15926
15927
15928
15929
15930
15931
15932
15933
15934
15935
15936
15937
15938
15939
15940
15941
15942
15943
15944
15945
15946
15947
15948
15949
15950
15951
15952
15953
15954
15955
15956
15957
15958
15959
15960
15961
15962
15963
15964
15965
15966
15967
15968
15969
15970
15971
15972
15973
15974
15975
15976
15977
15978
15979
15980
15981
15982
15983
15984
15985
15986
15987
15988
15989
15990
15991
15992
15993
15994
15995
15996
15997
15998
15999
16000
16001
16002
16003
16004
16005
16006
16007
16008
16009
16010
16011
16012
16013
16014
16015
16016
16017
16018
16019
16020
16021
16022
16023
16024
16025
16026
16027
16028
16029
16030
16031
16032
16033
16034
16035
16036
16037
16038
16039
16040
16041
16042
16043
16044
16045
16046
16047
16048
16049
16050
16051
16052
16053
16054
16055
16056
16057
16058
16059
16060
16061
16062
16063
16064
16065
16066
16067
16068
16069
16070
16071
16072
16073
16074
16075
16076
16077
16078
16079
16080
16081
16082
16083
16084
16085
16086
16087
16088
16089
16090
16091
16092
16093
16094
16095
16096
16097
16098
16099
16100
16101
16102
16103
16104
16105
16106
16107
16108
16109
16110
16111
16112
16113
16114
16115
16116
16117
16118
16119
16120
16121
16122
16123
16124
16125
16126
16127
16128
16129
16130
16131
16132
16133
16134
16135
16136
16137
16138
16139
16140
16141
16142
16143
16144
16145
16146
16147
16148
16149
16150
16151
16152
16153
16154
16155
16156
16157
16158
16159
16160
16161
16162
16163
16164
16165
16166
16167
16168
16169
16170
16171
16172
16173
16174
16175
16176
16177
16178
16179
16180
16181
16182
16183
16184
16185
16186
16187
16188
16189
16190
16191
16192
16193
16194
16195
16196
16197
16198
16199
16200
16201
16202
16203
16204
16205
16206
16207
16208
16209
16210
16211
16212
16213
16214
16215
16216
16217
16218
16219
16220
16221
16222
16223
16224
16225
16226
16227
16228
16229
16230
16231
16232
16233
16234
16235
16236
16237
16238
16239
16240
16241
16242
16243
16244
16245
16246
16247
16248
16249
16250
16251
16252
16253
16254
16255
16256
16257
16258
16259
16260
16261
16262
16263
16264
16265
16266
16267
16268
16269
16270
16271
16272
16273
16274
16275
16276
16277
16278
16279
16280
16281
16282
16283
16284
16285
16286
16287
16288
16289
16290
16291
16292
16293
16294
16295
16296
16297
16298
16299
16300
16301
16302
16303
16304
16305
16306
16307
16308
16309
16310
16311
16312
16313
16314
16315
16316
16317
16318
16319
16320
16321
16322
16323
16324
16325
16326
16327
16328
16329
16330
16331
16332
16333
16334
16335
16336
16337
16338
16339
16340
16341
16342
16343
16344
16345
16346
16347
16348
16349
16350
16351
16352
16353
16354
16355
16356
16357
16358
16359
16360
16361
16362
16363
16364
16365
16366
16367
16368
16369
16370
16371
16372
16373
16374
16375
16376
16377
16378
16379
16380
16381
16382
16383
16384
16385
16386
16387
16388
16389
16390
16391
16392
16393
16394
16395
16396
16397
16398
16399
16400
16401
16402
16403
16404
16405
16406
16407
16408
16409
16410
16411
16412
16413
16414
16415
16416
16417
16418
16419
16420
16421
16422
16423
16424
16425
16426
16427
16428
16429
16430
16431
16432
16433
16434
16435
16436
16437
16438
16439
16440
16441
16442
16443
16444
16445
16446
16447
16448
16449
16450
16451
16452
16453
16454
16455
16456
16457
16458
16459
16460
16461
16462
16463
16464
16465
16466
16467
16468
16469
16470
16471
16472
16473
16474
16475
16476
16477
16478
16479
16480
16481
16482
16483
16484
16485
16486
16487
16488
16489
16490
16491
16492
16493
16494
16495
16496
16497
16498
16499
16500
16501
16502
16503
16504
16505
16506
16507
16508
16509
16510
16511
16512
16513
16514
16515
16516
16517
16518
16519
16520
16521
16522
16523
16524
16525
16526
16527
16528
16529
16530
16531
16532
16533
16534
16535
16536
16537
16538
16539
16540
16541
16542
16543
16544
16545
16546
16547
16548
16549
16550
16551
16552
16553
16554
16555
16556
16557
16558
16559
16560
16561
16562
16563
16564
16565
16566
16567
16568
16569
16570
16571
16572
16573
16574
16575
16576
16577
16578
16579
16580
16581
16582
16583
16584
16585
16586
16587
16588
16589
16590
16591
16592
16593
16594
16595
16596
16597
16598
16599
16600
16601
16602
16603
16604
16605
16606
16607
16608
16609
16610
16611
16612
16613
16614
16615
16616
16617
16618
16619
16620
16621
16622
16623
16624
16625
16626
16627
16628
16629
16630
16631
16632
16633
16634
16635
16636
16637
16638
16639
16640
16641
16642
16643
16644
16645
16646
16647
16648
16649
16650
16651
16652
16653
16654
16655
16656
16657
16658
16659
16660
16661
16662
16663
16664
16665
16666
16667
16668
16669
16670
16671
16672
16673
16674
16675
16676
16677
16678
16679
16680
16681
16682
16683
16684
16685
16686
16687
16688
16689
16690
16691
16692
16693
16694
16695
16696
16697
16698
16699
16700
16701
16702
16703
16704
16705
16706
16707
16708
16709
16710
16711
16712
16713
16714
16715
16716
16717
16718
16719
16720
16721
16722
16723
16724
16725
16726
16727
16728
16729
16730
16731
16732
16733
16734
16735
16736
16737
16738
16739
16740
16741
16742
16743
16744
16745
16746
16747
16748
16749
16750
16751
16752
16753
16754
16755
16756
16757
16758
16759
16760
16761
16762
16763
16764
16765
16766
16767
16768
16769
16770
16771
16772
16773
16774
16775
16776
16777
16778
16779
16780
16781
16782
16783
16784
16785
16786
16787
16788
16789
16790
16791
16792
16793
16794
16795
16796
16797
16798
16799
16800
16801
16802
16803
16804
16805
16806
16807
16808
16809
16810
16811
16812
16813
16814
16815
16816
16817
16818
16819
16820
16821
16822
16823
16824
16825
16826
16827
16828
16829
16830
16831
16832
16833
16834
16835
16836
16837
16838
16839
16840
16841
16842
16843
16844
16845
16846
16847
16848
16849
16850
16851
16852
16853
16854
16855
16856
16857
16858
16859
16860
16861
16862
16863
16864
16865
16866
16867
16868
16869
16870
16871
16872
16873
16874
16875
16876
16877
16878
16879
16880
16881
16882
16883
16884
16885
16886
16887
16888
16889
16890
16891
16892
16893
16894
16895
16896
16897
16898
16899
16900
16901
16902
16903
16904
16905
16906
16907
16908
16909
16910
16911
16912
16913
16914
16915
16916
16917
16918
16919
16920
16921
16922
16923
16924
16925
16926
16927
16928
16929
16930
16931
16932
16933
16934
16935
16936
16937
16938
16939
16940
16941
16942
16943
16944
16945
16946
16947
16948
16949
16950
16951
16952
16953
16954
16955
16956
16957
16958
16959
16960
16961
16962
16963
16964
16965
16966
16967
16968
16969
16970
16971
16972
16973
16974
16975
16976
16977
16978
16979
16980
16981
16982
16983
16984
16985
16986
16987
16988
16989
16990
16991
16992
16993
16994
16995
16996
16997
16998
16999
17000
17001
17002
17003
17004
17005
17006
17007
17008
17009
17010
17011
17012
17013
17014
17015
17016
17017
17018
17019
17020
17021
17022
17023
17024
17025
17026
17027
17028
17029
17030
17031
17032
17033
17034
17035
17036
17037
17038
17039
17040
17041
17042
17043
17044
17045
17046
17047
17048
17049
17050
17051
17052
17053
17054
17055
17056
17057
17058
17059
17060
17061
17062
17063
17064
17065
17066
17067
17068
17069
17070
17071
17072
17073
17074
17075
17076
17077
17078
17079
17080
17081
17082
17083
17084
17085
17086
17087
17088
17089
17090
17091
17092
17093
17094
17095
17096
17097
17098
17099
17100
17101
17102
17103
17104
17105
17106
17107
17108
17109
17110
17111
17112
17113
17114
17115
17116
17117
17118
17119
17120
17121
17122
17123
17124
17125
17126
17127
17128
17129
17130
17131
17132
17133
17134
17135
17136
17137
17138
17139
17140
17141
17142
17143
17144
17145
17146
17147
17148
17149
17150
17151
17152
17153
17154
17155
17156
17157
17158
17159
17160
17161
17162
17163
17164
17165
17166
17167
17168
17169
17170
17171
17172
17173
17174
17175
17176
17177
17178
17179
17180
17181
17182
17183
17184
17185
17186
17187
17188
17189
17190
17191
17192
17193
17194
17195
17196
17197
17198
17199
17200
17201
17202
17203
17204
17205
17206
17207
17208
17209
17210
17211
17212
17213
17214
17215
17216
17217
17218
17219
17220
17221
17222
17223
17224
17225
17226
17227
17228
17229
17230
17231
17232
17233
17234
17235
17236
17237
17238
17239
17240
17241
17242
17243
17244
17245
17246
17247
17248
17249
17250
17251
17252
17253
17254
17255
17256
17257
17258
17259
17260
17261
17262
17263
17264
17265
17266
17267
17268
17269
17270
17271
17272
17273
17274
17275
17276
17277
17278
17279
17280
17281
17282
17283
17284
17285
17286
17287
17288
17289
17290
17291
17292
17293
17294
17295
17296
17297
17298
17299
17300
17301
17302
17303
17304
17305
17306
17307
17308
17309
17310
17311
17312
17313
17314
17315
17316
17317
17318
17319
17320
17321
17322
17323
17324
17325
17326
17327
17328
17329
17330
17331
17332
17333
17334
17335
17336
17337
17338
17339
17340
17341
17342
17343
17344
17345
17346
17347
17348
17349
17350
17351
17352
17353
17354
17355
17356
17357
17358
17359
17360
17361
17362
17363
17364
17365
17366
17367
17368
17369
17370
17371
17372
17373
17374
17375
17376
17377
17378
17379
17380
17381
17382
17383
17384
17385
17386
17387
17388
17389
17390
17391
17392
17393
17394
17395
17396
17397
17398
17399
17400
17401
17402
17403
17404
17405
17406
17407
17408
17409
17410
17411
17412
17413
17414
17415
17416
17417
17418
17419
17420
17421
17422
17423
17424
17425
17426
17427
17428
17429
17430
17431
17432
17433
17434
17435
17436
17437
17438
17439
17440
17441
17442
17443
17444
17445
17446
17447
17448
17449
17450
17451
17452
17453
17454
17455
17456
17457
17458
17459
17460
17461
17462
17463
17464
17465
17466
17467
17468
17469
17470
17471
17472
17473
17474
17475
17476
17477
17478
17479
17480
17481
17482
17483
17484
17485
17486
17487
17488
17489
17490
17491
17492
17493
17494
17495
17496
17497
17498
17499
17500
17501
17502
17503
17504
17505
17506
17507
17508
17509
17510
17511
17512
17513
17514
17515
17516
17517
17518
17519
17520
17521
17522
17523
17524
17525
17526
17527
17528
17529
17530
17531
17532
17533
17534
17535
17536
17537
17538
17539
17540
17541
17542
17543
17544
17545
17546
17547
17548
17549
17550
17551
17552
17553
17554
17555
17556
17557
17558
17559
17560
17561
17562
17563
17564
17565
17566
17567
17568
17569
17570
17571
17572
17573
17574
17575
17576
17577
17578
17579
17580
17581
17582
17583
17584
17585
17586
17587
17588
17589
17590
17591
17592
17593
17594
17595
17596
17597
17598
17599
17600
17601
17602
17603
17604
17605
17606
17607
17608
17609
17610
17611
17612
17613
17614
17615
17616
17617
17618
17619
17620
17621
17622
17623
17624
17625
17626
17627
17628
17629
17630
17631
17632
17633
17634
17635
17636
17637
17638
17639
17640
17641
17642
17643
17644
17645
17646
17647
17648
17649
17650
17651
17652
17653
17654
17655
17656
17657
17658
17659
17660
17661
17662
17663
17664
17665
17666
17667
17668
17669
17670
17671
17672
17673
17674
17675
17676
17677
17678
17679
17680
17681
17682
17683
17684
17685
17686
17687
17688
17689
17690
17691
17692
17693
17694
17695
17696
17697
17698
17699
17700
17701
17702
17703
17704
17705
17706
17707
17708
17709
17710
17711
17712
17713
17714
17715
17716
17717
17718
17719
17720
17721
17722
17723
17724
17725
17726
17727
17728
17729
17730
17731
17732
17733
17734
17735
17736
17737
17738
17739
17740
17741
17742
17743
17744
17745
17746
17747
17748
17749
17750
17751
17752
17753
17754
17755
17756
17757
17758
17759
17760
17761
17762
17763
17764
17765
17766
17767
17768
17769
17770
17771
17772
17773
17774
17775
17776
17777
17778
17779
17780
17781
17782
17783
17784
17785
17786
17787
17788
17789
17790
17791
17792
17793
17794
17795
17796
17797
17798
17799
17800
17801
17802
17803
17804
17805
17806
17807
17808
17809
17810
17811
17812
17813
17814
17815
17816
17817
17818
17819
17820
17821
17822
17823
17824
17825
17826
17827
17828
17829
17830
17831
17832
17833
17834
17835
17836
17837
17838
17839
17840
17841
17842
17843
17844
17845
17846
17847
17848
17849
17850
17851
17852
17853
17854
17855
17856
17857
17858
17859
17860
17861
17862
17863
17864
17865
17866
17867
17868
17869
17870
17871
17872
17873
17874
17875
17876
17877
17878
17879
17880
17881
17882
17883
17884
17885
17886
17887
17888
17889
17890
17891
17892
17893
17894
17895
17896
17897
17898
17899
17900
17901
17902
17903
17904
17905
17906
17907
17908
17909
17910
17911
17912
17913
17914
17915
17916
17917
17918
17919
17920
17921
17922
17923
17924
17925
17926
17927
17928
17929
17930
17931
17932
17933
17934
17935
17936
17937
17938
17939
17940
17941
17942
17943
17944
17945
17946
17947
17948
17949
17950
17951
17952
17953
17954
17955
17956
17957
17958
17959
17960
17961
17962
17963
17964
17965
17966
17967
17968
17969
17970
17971
17972
17973
17974
17975
17976
17977
17978
17979
17980
17981
17982
17983
17984
17985
17986
17987
17988
17989
17990
17991
17992
17993
17994
17995
17996
17997
17998
17999
18000
18001
18002
18003
18004
18005
18006
18007
18008
18009
18010
18011
18012
18013
18014
18015
18016
18017
18018
18019
18020
18021
18022
18023
18024
18025
18026
18027
18028
18029
18030
18031
18032
18033
18034
18035
18036
18037
18038
18039
18040
18041
18042
18043
18044
18045
18046
18047
18048
18049
18050
18051
18052
18053
18054
18055
18056
18057
18058
18059
18060
18061
18062
18063
18064
18065
18066
18067
18068
18069
18070
18071
18072
18073
18074
18075
18076
18077
18078
18079
18080
18081
18082
18083
18084
18085
18086
18087
18088
18089
18090
18091
18092
18093
18094
18095
18096
18097
18098
18099
18100
18101
18102
18103
18104
18105
18106
18107
18108
18109
18110
18111
18112
18113
18114
18115
18116
18117
18118
18119
18120
18121
18122
18123
18124
18125
18126
18127
18128
18129
18130
18131
18132
18133
18134
18135
18136
18137
18138
18139
18140
18141
18142
18143
18144
18145
18146
18147
18148
18149
18150
18151
18152
18153
18154
18155
18156
18157
18158
18159
18160
18161
18162
18163
18164
18165
18166
18167
18168
18169
18170
18171
18172
18173
18174
18175
18176
18177
18178
18179
18180
18181
18182
18183
18184
18185
18186
18187
18188
18189
18190
18191
18192
18193
18194
18195
18196
18197
18198
18199
18200
18201
18202
18203
18204
18205
18206
18207
18208
18209
18210
18211
18212
18213
18214
18215
18216
18217
18218
18219
18220
18221
18222
18223
18224
18225
18226
18227
18228
18229
18230
18231
18232
18233
18234
18235
18236
18237
18238
18239
18240
18241
18242
18243
18244
18245
18246
18247
18248
18249
18250
18251
18252
18253
18254
18255
18256
18257
18258
18259
18260
18261
18262
18263
18264
18265
18266
18267
18268
18269
18270
18271
18272
18273
18274
18275
18276
18277
18278
18279
18280
18281
18282
18283
18284
18285
18286
18287
18288
18289
18290
18291
18292
18293
18294
18295
18296
18297
18298
18299
18300
18301
18302
18303
18304
18305
18306
18307
18308
18309
18310
18311
18312
18313
18314
18315
18316
18317
18318
18319
18320
18321
18322
18323
18324
18325
18326
18327
18328
18329
18330
18331
18332
18333
18334
18335
18336
18337
18338
18339
18340
18341
18342
18343
18344
18345
18346
18347
18348
18349
18350
18351
18352
18353
18354
18355
18356
18357
18358
18359
18360
18361
18362
18363
18364
18365
18366
18367
18368
18369
18370
18371
18372
18373
18374
18375
18376
18377
18378
18379
18380
18381
18382
18383
18384
18385
18386
18387
18388
18389
18390
18391
18392
18393
18394
18395
18396
18397
18398
18399
18400
18401
18402
18403
18404
18405
18406
18407
18408
18409
18410
18411
18412
18413
18414
18415
18416
18417
18418
18419
18420
18421
18422
18423
18424
18425
18426
18427
18428
18429
18430
18431
18432
18433
18434
18435
18436
18437
18438
18439
18440
18441
18442
18443
18444
18445
18446
18447
18448
18449
18450
18451
18452
18453
18454
18455
18456
18457
18458
18459
18460
18461
18462
18463
18464
18465
18466
18467
18468
18469
18470
18471
18472
18473
18474
18475
18476
18477
18478
18479
18480
18481
18482
18483
18484
18485
18486
18487
18488
18489
18490
18491
18492
18493
18494
18495
18496
18497
18498
18499
18500
18501
18502
18503
18504
18505
18506
18507
18508
18509
18510
18511
18512
18513
18514
18515
18516
18517
18518
18519
18520
18521
18522
18523
18524
18525
18526
18527
18528
18529
18530
18531
18532
18533
18534
18535
18536
18537
18538
18539
18540
18541
18542
18543
18544
18545
18546
18547
18548
18549
18550
18551
18552
18553
18554
18555
18556
18557
18558
18559
18560
18561
18562
18563
18564
18565
18566
18567
18568
18569
18570
18571
18572
18573
18574
18575
18576
18577
18578
18579
18580
18581
18582
18583
18584
18585
18586
18587
18588
18589
18590
18591
18592
18593
18594
18595
18596
18597
18598
18599
18600
18601
18602
18603
18604
18605
18606
18607
18608
18609
18610
18611
18612
18613
18614
18615
18616
18617
18618
18619
18620
18621
18622
18623
18624
18625
18626
18627
18628
18629
18630
18631
18632
18633
18634
18635
18636
18637
18638
18639
18640
18641
18642
18643
18644
18645
18646
18647
18648
18649
18650
18651
18652
18653
18654
18655
18656
18657
18658
18659
18660
18661
18662
18663
18664
18665
18666
18667
18668
18669
18670
18671
18672
18673
18674
18675
18676
18677
18678
18679
18680
18681
18682
18683
18684
18685
18686
18687
18688
18689
18690
18691
18692
18693
18694
18695
18696
18697
18698
18699
18700
18701
18702
18703
18704
18705
18706
18707
18708
18709
18710
18711
18712
18713
18714
18715
18716
18717
18718
18719
18720
18721
18722
18723
18724
18725
18726
18727
18728
18729
18730
18731
18732
18733
18734
18735
18736
18737
18738
18739
18740
18741
18742
18743
18744
18745
18746
18747
18748
18749
18750
18751
18752
18753
18754
18755
18756
18757
18758
18759
18760
18761
18762
18763
18764
18765
18766
18767
18768
18769
18770
18771
18772
18773
18774
18775
18776
18777
18778
18779
18780
18781
18782
18783
18784
18785
18786
18787
18788
18789
18790
18791
18792
18793
18794
18795
18796
18797
18798
18799
18800
18801
18802
18803
18804
18805
18806
18807
18808
18809
18810
18811
18812
18813
18814
18815
18816
18817
18818
18819
18820
18821
18822
18823
18824
18825
18826
18827
18828
18829
18830
18831
18832
18833
18834
18835
18836
18837
18838
18839
18840
18841
18842
18843
18844
18845
18846
18847
18848
18849
18850
18851
18852
18853
18854
18855
18856
18857
18858
18859
18860
18861
18862
18863
18864
18865
18866
18867
18868
18869
18870
18871
18872
18873
18874
18875
18876
18877
18878
18879
18880
18881
18882
18883
18884
18885
18886
18887
18888
18889
18890
18891
18892
18893
18894
18895
18896
18897
18898
18899
18900
18901
18902
18903
18904
18905
18906
18907
18908
18909
18910
18911
18912
18913
18914
18915
18916
18917
18918
18919
18920
18921
18922
18923
18924
18925
18926
18927
18928
18929
18930
18931
18932
18933
18934
18935
18936
18937
18938
18939
18940
18941
18942
18943
18944
18945
18946
18947
18948
18949
18950
18951
18952
18953
18954
18955
18956
18957
18958
18959
18960
18961
18962
18963
18964
18965
18966
18967
18968
18969
18970
18971
18972
18973
18974
18975
18976
18977
18978
18979
18980
18981
18982
18983
18984
18985
18986
18987
18988
18989
18990
18991
18992
18993
18994
18995
18996
18997
18998
18999
19000
19001
19002
19003
19004
19005
19006
19007
19008
19009
19010
19011
19012
19013
19014
19015
19016
19017
19018
19019
19020
19021
19022
19023
19024
19025
19026
19027
19028
19029
19030
19031
19032
19033
19034
19035
19036
19037
19038
19039
19040
19041
19042
19043
19044
19045
19046
19047
19048
19049
19050
19051
19052
19053
19054
19055
19056
19057
19058
19059
19060
19061
19062
19063
19064
19065
19066
19067
19068
19069
19070
19071
19072
19073
19074
19075
19076
19077
19078
19079
19080
19081
19082
19083
19084
19085
19086
19087
19088
19089
19090
19091
19092
19093
19094
19095
19096
19097
19098
19099
19100
19101
19102
19103
19104
19105
19106
19107
19108
19109
19110
19111
19112
19113
19114
19115
19116
19117
19118
19119
19120
19121
19122
19123
19124
19125
19126
19127
19128
19129
19130
19131
19132
19133
19134
19135
19136
19137
19138
19139
19140
19141
19142
19143
19144
19145
19146
19147
19148
19149
19150
19151
19152
19153
19154
19155
19156
19157
19158
19159
19160
19161
19162
19163
19164
19165
19166
19167
19168
19169
19170
19171
19172
19173
19174
19175
19176
19177
19178
19179
19180
19181
19182
19183
19184
19185
19186
19187
19188
19189
19190
19191
19192
19193
19194
19195
19196
19197
19198
19199
19200
19201
19202
19203
19204
19205
19206
19207
19208
19209
19210
19211
19212
19213
19214
19215
19216
19217
19218
19219
19220
19221
19222
19223
19224
19225
19226
19227
19228
19229
19230
19231
19232
19233
19234
19235
19236
19237
19238
19239
19240
19241
19242
19243
19244
19245
19246
19247
19248
19249
19250
19251
19252
19253
19254
19255
19256
19257
19258
19259
19260
19261
19262
19263
19264
19265
19266
19267
19268
19269
19270
19271
19272
19273
19274
19275
19276
19277
19278
19279
19280
19281
19282
19283
19284
19285
19286
19287
19288
19289
19290
19291
19292
19293
19294
19295
19296
19297
19298
19299
19300
19301
19302
19303
19304
19305
19306
19307
19308
19309
19310
19311
19312
19313
19314
19315
19316
19317
19318
19319
19320
19321
19322
19323
19324
19325
19326
19327
19328
19329
19330
19331
19332
19333
19334
19335
19336
19337
19338
19339
19340
19341
19342
19343
19344
19345
19346
19347
19348
19349
19350
19351
19352
19353
19354
19355
19356
19357
19358
19359
19360
19361
19362
19363
19364
19365
19366
19367
19368
19369
19370
19371
19372
19373
19374
19375
19376
19377
19378
19379
19380
19381
19382
19383
19384
19385
19386
19387
19388
19389
19390
19391
19392
19393
19394
19395
19396
19397
19398
19399
19400
19401
19402
19403
19404
19405
19406
19407
19408
19409
19410
19411
19412
19413
19414
19415
19416
19417
19418
19419
19420
19421
19422
19423
19424
19425
19426
19427
19428
19429
19430
19431
19432
19433
19434
19435
19436
19437
19438
19439
19440
19441
19442
19443
19444
19445
19446
19447
19448
19449
19450
19451
19452
19453
19454
19455
19456
19457
19458
19459
19460
19461
19462
19463
19464
19465
19466
19467
19468
19469
19470
19471
19472
19473
19474
19475
19476
19477
19478
19479
19480
19481
19482
19483
19484
19485
19486
19487
19488
19489
19490
19491
19492
19493
19494
19495
19496
19497
19498
19499
19500
19501
19502
19503
19504
19505
19506
19507
19508
19509
19510
19511
19512
19513
19514
19515
19516
19517
19518
19519
19520
19521
19522
19523
19524
19525
19526
19527
19528
19529
19530
19531
19532
19533
19534
19535
19536
19537
19538
19539
19540
19541
19542
19543
19544
19545
19546
19547
19548
19549
19550
19551
19552
19553
19554
19555
19556
19557
19558
19559
19560
19561
19562
19563
19564
19565
19566
19567
19568
19569
19570
19571
19572
19573
19574
19575
19576
19577
19578
19579
19580
19581
19582
19583
19584
19585
19586
19587
19588
19589
19590
19591
19592
19593
19594
19595
19596
19597
19598
19599
19600
19601
19602
19603
19604
19605
19606
19607
19608
19609
19610
19611
19612
19613
19614
19615
19616
19617
19618
19619
19620
19621
19622
19623
19624
19625
19626
19627
19628
19629
19630
19631
19632
19633
19634
19635
19636
19637
19638
19639
19640
19641
19642
19643
19644
19645
19646
19647
19648
19649
19650
19651
19652
19653
19654
19655
19656
19657
19658
19659
19660
19661
19662
19663
19664
19665
19666
19667
19668
19669
19670
19671
19672
19673
19674
19675
19676
19677
19678
19679
19680
19681
19682
19683
19684
19685
19686
19687
19688
19689
19690
19691
19692
19693
19694
19695
19696
19697
19698
19699
19700
19701
19702
19703
19704
19705
19706
19707
19708
19709
19710
19711
19712
19713
19714
19715
19716
19717
19718
19719
19720
19721
19722
19723
19724
19725
19726
19727
19728
19729
19730
19731
19732
19733
19734
19735
19736
19737
19738
19739
19740
19741
19742
19743
19744
19745
19746
19747
19748
19749
19750
19751
19752
19753
19754
19755
19756
19757
19758
19759
19760
19761
19762
19763
19764
19765
19766
19767
19768
19769
19770
19771
19772
19773
19774
19775
19776
19777
19778
19779
19780
19781
19782
19783
19784
19785
19786
19787
19788
19789
19790
19791
19792
19793
19794
19795
19796
19797
19798
19799
19800
19801
19802
19803
19804
19805
19806
19807
19808
19809
19810
19811
19812
19813
19814
19815
19816
19817
19818
19819
19820
19821
19822
19823
19824
19825
19826
19827
19828
19829
19830
19831
19832
19833
19834
19835
19836
19837
19838
19839
19840
19841
19842
19843
19844
19845
19846
19847
19848
19849
19850
19851
19852
19853
19854
19855
19856
19857
19858
19859
19860
19861
19862
19863
19864
19865
19866
19867
19868
19869
19870
19871
19872
19873
19874
19875
19876
19877
19878
19879
19880
19881
19882
19883
19884
19885
19886
19887
19888
19889
19890
19891
19892
19893
19894
19895
19896
19897
19898
19899
19900
19901
19902
19903
19904
19905
19906
19907
19908
19909
19910
19911
19912
19913
19914
19915
19916
19917
19918
19919
19920
19921
19922
19923
19924
19925
19926
19927
19928
19929
19930
19931
19932
19933
19934
19935
19936
19937
19938
19939
19940
19941
19942
19943
19944
19945
19946
19947
19948
19949
19950
19951
19952
19953
19954
19955
19956
19957
19958
19959
19960
19961
19962
19963
19964
19965
19966
19967
19968
19969
19970
19971
19972
19973
19974
19975
19976
19977
19978
19979
19980
19981
19982
19983
19984
19985
19986
19987
19988
19989
19990
19991
19992
19993
19994
19995
19996
19997
19998
19999
20000
20001
20002
20003
20004
20005
20006
20007
20008
20009
20010
20011
20012
20013
20014
20015
20016
20017
20018
20019
20020
20021
20022
20023
20024
20025
20026
20027
20028
20029
20030
20031
20032
20033
20034
20035
20036
20037
20038
20039
20040
20041
20042
20043
20044
20045
20046
20047
20048
20049
20050
20051
20052
20053
20054
20055
20056
20057
20058
20059
20060
20061
20062
20063
20064
20065
20066
20067
20068
20069
20070
20071
20072
20073
20074
20075
20076
20077
20078
20079
20080
20081
20082
20083
20084
20085
20086
20087
20088
20089
20090
20091
20092
20093
20094
20095
20096
20097
20098
20099
20100
20101
20102
20103
20104
20105
20106
20107
20108
20109
20110
20111
20112
20113
20114
20115
20116
20117
20118
20119
20120
20121
20122
20123
20124
20125
20126
20127
20128
20129
20130
20131
20132
20133
20134
20135
20136
20137
20138
20139
20140
20141
20142
20143
20144
20145
20146
20147
20148
20149
20150
20151
20152
20153
20154
20155
20156
20157
20158
20159
20160
20161
20162
20163
20164
20165
20166
20167
20168
20169
20170
20171
20172
20173
20174
20175
20176
20177
20178
20179
20180
20181
20182
20183
20184
20185
20186
20187
20188
20189
20190
20191
20192
20193
20194
20195
20196
20197
20198
20199
20200
20201
20202
20203
20204
20205
20206
20207
20208
20209
20210
20211
20212
20213
20214
20215
20216
20217
20218
20219
20220
20221
20222
20223
20224
20225
20226
20227
20228
20229
20230
20231
20232
20233
20234
20235
20236
20237
20238
20239
20240
20241
20242
20243
20244
20245
20246
20247
20248
20249
20250
20251
20252
20253
20254
20255
20256
20257
20258
20259
20260
20261
20262
20263
20264
20265
20266
20267
20268
20269
20270
20271
20272
20273
20274
20275
20276
20277
20278
20279
20280
20281
20282
20283
20284
20285
20286
20287
20288
20289
20290
20291
20292
20293
20294
20295
20296
20297
20298
20299
20300
20301
20302
20303
20304
20305
20306
20307
20308
20309
20310
20311
20312
20313
20314
20315
20316
20317
20318
20319
20320
20321
20322
20323
20324
20325
20326
20327
20328
20329
20330
20331
20332
20333
20334
20335
20336
20337
20338
20339
20340
20341
20342
20343
20344
20345
20346
20347
20348
20349
20350
20351
20352
20353
20354
20355
20356
20357
20358
20359
20360
20361
20362
20363
20364
20365
20366
20367
20368
20369
20370
20371
20372
20373
20374
20375
20376
20377
20378
20379
20380
20381
20382
20383
20384
20385
20386
20387
20388
20389
20390
20391
20392
20393
20394
20395
20396
20397
20398
20399
20400
20401
20402
20403
20404
20405
20406
20407
20408
20409
20410
20411
20412
20413
20414
20415
20416
20417
20418
20419
20420
20421
20422
20423
20424
20425
20426
20427
20428
20429
20430
20431
20432
20433
20434
20435
20436
20437
20438
20439
20440
20441
20442
20443
20444
20445
20446
20447
20448
20449
20450
20451
20452
20453
20454
20455
20456
20457
20458
20459
20460
20461
20462
20463
20464
20465
20466
20467
20468
20469
20470
20471
20472
20473
20474
20475
20476
20477
20478
20479
20480
20481
20482
20483
20484
20485
20486
20487
20488
20489
20490
20491
20492
20493
20494
20495
20496
20497
20498
20499
20500
20501
20502
20503
20504
20505
20506
20507
20508
20509
20510
20511
20512
20513
20514
20515
20516
20517
20518
20519
20520
20521
20522
20523
20524
20525
20526
20527
20528
20529
20530
20531
20532
20533
20534
20535
20536
20537
20538
20539
20540
20541
20542
20543
20544
20545
20546
20547
20548
20549
20550
20551
20552
20553
20554
20555
20556
20557
20558
20559
20560
20561
20562
20563
20564
20565
20566
20567
20568
20569
20570
20571
20572
20573
20574
20575
20576
20577
20578
20579
20580
20581
20582
20583
20584
20585
20586
20587
20588
20589
20590
20591
20592
20593
20594
20595
20596
20597
20598
20599
20600
20601
20602
20603
20604
20605
20606
20607
20608
20609
20610
20611
20612
20613
20614
20615
20616
20617
20618
20619
20620
20621
20622
20623
20624
20625
20626
20627
20628
20629
20630
20631
20632
20633
20634
20635
20636
20637
20638
20639
20640
20641
20642
20643
20644
20645
20646
20647
20648
20649
20650
20651
20652
20653
20654
20655
20656
20657
20658
20659
20660
20661
20662
20663
20664
20665
20666
20667
20668
20669
20670
20671
20672
20673
20674
20675
20676
20677
20678
20679
20680
20681
20682
20683
20684
20685
20686
20687
20688
20689
20690
20691
20692
20693
20694
20695
20696
20697
20698
20699
20700
20701
20702
20703
20704
20705
20706
20707
20708
20709
20710
20711
20712
20713
20714
20715
20716
20717
20718
20719
20720
20721
20722
20723
20724
20725
20726
20727
20728
20729
20730
20731
20732
20733
20734
20735
20736
20737
20738
20739
20740
20741
20742
20743
20744
20745
20746
20747
20748
20749
20750
20751
20752
20753
20754
20755
20756
20757
20758
20759
20760
20761
20762
20763
20764
20765
20766
20767
20768
20769
20770
20771
20772
20773
20774
20775
20776
20777
20778
20779
20780
20781
20782
20783
20784
20785
20786
20787
20788
20789
20790
20791
20792
20793
20794
20795
20796
20797
20798
20799
20800
20801
20802
20803
20804
20805
20806
20807
20808
20809
20810
20811
20812
20813
20814
20815
20816
20817
20818
20819
20820
20821
20822
20823
20824
20825
20826
20827
20828
20829
20830
20831
20832
20833
20834
20835
20836
20837
20838
20839
20840
20841
20842
20843
20844
20845
20846
20847
20848
20849
20850
20851
20852
20853
20854
20855
20856
20857
20858
20859
20860
20861
20862
20863
20864
20865
20866
20867
20868
20869
20870
20871
20872
20873
20874
20875
20876
20877
20878
20879
20880
20881
20882
20883
20884
20885
20886
20887
20888
20889
20890
20891
20892
20893
20894
20895
20896
20897
20898
20899
20900
20901
20902
20903
20904
20905
20906
20907
20908
20909
20910
20911
20912
20913
20914
20915
20916
20917
20918
20919
20920
20921
20922
20923
20924
20925
20926
20927
20928
20929
20930
20931
20932
20933
20934
20935
20936
20937
20938
20939
20940
20941
20942
20943
20944
20945
20946
20947
20948
20949
20950
20951
20952
20953
20954
20955
20956
20957
20958
20959
20960
20961
20962
20963
20964
20965
20966
20967
20968
20969
20970
20971
20972
20973
20974
20975
20976
20977
20978
20979
20980
20981
20982
20983
20984
20985
20986
20987
20988
20989
20990
20991
20992
20993
20994
20995
20996
20997
20998
20999
21000
21001
21002
21003
21004
21005
21006
21007
21008
21009
21010
21011
21012
21013
21014
21015
21016
21017
21018
21019
21020
21021
21022
21023
21024
21025
21026
21027
21028
21029
21030
21031
21032
21033
21034
21035
21036
21037
21038
21039
21040
21041
21042
21043
21044
21045
21046
21047
21048
21049
21050
21051
21052
21053
21054
21055
21056
21057
21058
21059
21060
21061
21062
21063
21064
21065
21066
21067
21068
21069
21070
21071
21072
21073
21074
21075
21076
21077
21078
21079
21080
21081
21082
21083
21084
21085
21086
21087
21088
21089
21090
21091
21092
21093
21094
21095
21096
21097
21098
21099
21100
21101
21102
21103
21104
21105
21106
21107
21108
21109
21110
21111
21112
21113
21114
21115
21116
21117
21118
21119
21120
21121
21122
21123
21124
21125
21126
21127
21128
21129
21130
21131
21132
21133
21134
21135
21136
21137
21138
21139
21140
21141
21142
21143
21144
21145
21146
21147
21148
21149
21150
21151
21152
21153
21154
21155
21156
21157
21158
21159
21160
21161
21162
21163
21164
21165
21166
21167
21168
21169
21170
21171
21172
21173
21174
21175
21176
21177
21178
21179
21180
21181
21182
21183
21184
21185
21186
21187
21188
21189
21190
21191
21192
21193
21194
21195
21196
21197
21198
21199
21200
21201
21202
21203
21204
21205
21206
21207
21208
21209
21210
21211
21212
21213
21214
21215
21216
21217
21218
21219
21220
21221
21222
21223
21224
21225
21226
21227
21228
21229
21230
21231
21232
21233
21234
21235
21236
21237
21238
21239
21240
21241
21242
21243
21244
21245
21246
21247
21248
21249
21250
21251
21252
21253
21254
21255
21256
21257
21258
21259
21260
21261
21262
21263
21264
21265
21266
21267
21268
21269
21270
21271
21272
21273
21274
21275
21276
21277
21278
21279
21280
21281
21282
21283
21284
21285
21286
21287
21288
21289
21290
21291
21292
21293
21294
21295
21296
21297
21298
21299
21300
21301
21302
21303
21304
21305
21306
21307
21308
21309
21310
21311
21312
21313
21314
21315
21316
21317
21318
21319
21320
21321
21322
21323
21324
21325
21326
21327
21328
21329
21330
21331
21332
21333
21334
21335
21336
21337
21338
21339
21340
21341
21342
21343
21344
21345
21346
21347
21348
21349
21350
21351
21352
21353
21354
21355
21356
21357
21358
21359
21360
21361
21362
21363
21364
21365
21366
21367
21368
21369
21370
21371
21372
21373
21374
21375
21376
21377
21378
21379
21380
21381
21382
21383
21384
21385
21386
21387
21388
21389
21390
21391
21392
21393
21394
21395
21396
21397
21398
21399
21400
21401
21402
21403
21404
21405
21406
21407
21408
21409
21410
21411
21412
21413
21414
21415
21416
21417
21418
21419
21420
21421
21422
21423
21424
21425
21426
21427
21428
21429
21430
21431
21432
21433
21434
21435
21436
21437
21438
21439
21440
21441
21442
21443
21444
21445
21446
21447
21448
21449
21450
21451
21452
21453
21454
21455
21456
21457
21458
21459
21460
21461
21462
21463
21464
21465
21466
21467
21468
21469
21470
21471
21472
21473
21474
21475
21476
21477
21478
21479
21480
21481
21482
21483
21484
21485
21486
21487
21488
21489
21490
21491
21492
21493
21494
21495
21496
21497
21498
21499
21500
21501
21502
21503
21504
21505
21506
21507
21508
21509
21510
21511
21512
21513
21514
21515
21516
21517
21518
21519
21520
21521
21522
21523
21524
21525
21526
21527
21528
21529
21530
21531
21532
21533
21534
21535
21536
21537
21538
21539
21540
21541
21542
21543
21544
21545
21546
21547
21548
21549
21550
21551
21552
21553
21554
21555
21556
21557
21558
21559
21560
21561
21562
21563
21564
21565
21566
21567
21568
21569
21570
21571
21572
21573
21574
21575
21576
21577
21578
21579
21580
21581
21582
21583
21584
21585
21586
21587
21588
21589
21590
21591
21592
21593
21594
21595
21596
21597
21598
21599
21600
21601
21602
21603
21604
21605
21606
21607
21608
21609
21610
21611
21612
21613
21614
21615
21616
21617
21618
21619
21620
21621
21622
21623
21624
21625
21626
21627
21628
21629
21630
21631
21632
21633
21634
21635
21636
21637
21638
21639
21640
21641
21642
21643
21644
21645
21646
21647
21648
21649
21650
21651
21652
21653
21654
21655
21656
21657
21658
21659
21660
21661
21662
21663
21664
21665
21666
21667
21668
21669
21670
21671
21672
21673
21674
21675
21676
21677
21678
21679
21680
21681
21682
21683
21684
21685
21686
21687
21688
21689
21690
21691
21692
21693
21694
21695
21696
21697
21698
21699
21700
21701
21702
21703
21704
21705
21706
21707
21708
21709
21710
21711
21712
21713
21714
21715
21716
21717
21718
21719
21720
21721
21722
21723
21724
21725
21726
21727
21728
21729
21730
21731
21732
21733
21734
21735
21736
21737
21738
21739
21740
21741
21742
21743
21744
21745
21746
21747
21748
21749
21750
21751
21752
21753
21754
21755
21756
21757
21758
21759
21760
21761
21762
21763
21764
21765
21766
21767
21768
21769
21770
21771
21772
21773
21774
21775
21776
21777
21778
21779
21780
21781
21782
21783
21784
21785
21786
21787
21788
21789
21790
21791
21792
21793
21794
21795
21796
21797
21798
21799
21800
21801
21802
21803
21804
21805
21806
21807
21808
21809
21810
21811
21812
21813
21814
21815
21816
21817
21818
21819
21820
21821
21822
21823
21824
21825
21826
21827
21828
21829
21830
21831
21832
21833
21834
21835
21836
21837
21838
21839
21840
21841
21842
21843
21844
21845
21846
21847
21848
21849
21850
21851
21852
21853
21854
21855
21856
21857
21858
21859
21860
21861
21862
21863
21864
21865
21866
21867
21868
21869
21870
21871
21872
21873
21874
21875
21876
21877
21878
21879
21880
21881
21882
21883
21884
21885
21886
21887
21888
21889
21890
21891
21892
21893
21894
21895
21896
21897
21898
21899
21900
21901
21902
21903
21904
21905
21906
21907
21908
21909
21910
21911
21912
21913
21914
21915
21916
21917
21918
21919
21920
21921
21922
21923
21924
21925
21926
21927
21928
21929
21930
21931
21932
21933
21934
21935
21936
21937
21938
21939
21940
21941
21942
21943
21944
21945
21946
21947
21948
21949
21950
21951
21952
21953
21954
21955
21956
21957
21958
21959
21960
21961
21962
21963
21964
21965
21966
21967
21968
21969
21970
21971
21972
21973
21974
21975
21976
21977
21978
21979
21980
21981
21982
21983
21984
21985
21986
21987
21988
21989
21990
21991
21992
21993
21994
21995
21996
21997
21998
21999
22000
22001
22002
22003
22004
22005
22006
22007
22008
22009
22010
22011
22012
22013
22014
22015
22016
22017
22018
22019
22020
22021
22022
22023
22024
22025
22026
22027
22028
22029
22030
22031
22032
22033
22034
22035
22036
22037
22038
22039
22040
22041
22042
22043
22044
22045
22046
22047
22048
22049
22050
22051
22052
22053
22054
22055
22056
22057
22058
22059
22060
22061
22062
22063
22064
22065
22066
22067
22068
22069
22070
22071
22072
22073
22074
22075
22076
22077
22078
22079
22080
22081
22082
22083
22084
22085
22086
22087
22088
22089
22090
22091
22092
22093
22094
22095
22096
22097
22098
22099
22100
22101
22102
22103
22104
22105
22106
22107
22108
22109
22110
22111
22112
22113
22114
22115
22116
22117
22118
22119
22120
22121
22122
22123
22124
22125
22126
22127
22128
22129
22130
22131
22132
22133
22134
22135
22136
22137
22138
22139
22140
22141
22142
22143
22144
22145
22146
22147
22148
22149
22150
22151
22152
22153
22154
22155
22156
22157
22158
22159
22160
22161
22162
22163
22164
22165
22166
22167
22168
22169
22170
22171
22172
22173
22174
22175
22176
22177
22178
22179
22180
22181
22182
22183
22184
22185
22186
22187
22188
22189
22190
22191
22192
22193
22194
22195
22196
22197
22198
22199
22200
22201
22202
22203
22204
22205
22206
22207
22208
22209
22210
22211
22212
22213
22214
22215
22216
22217
22218
22219
22220
22221
22222
22223
22224
22225
22226
22227
22228
22229
22230
22231
22232
22233
22234
22235
22236
22237
22238
22239
22240
22241
22242
22243
22244
22245
22246
22247
22248
22249
22250
22251
22252
22253
22254
22255
22256
22257
22258
22259
22260
22261
22262
22263
22264
22265
22266
22267
22268
22269
22270
22271
22272
22273
22274
22275
22276
22277
22278
22279
22280
22281
22282
22283
22284
22285
22286
22287
22288
22289
22290
22291
22292
22293
22294
22295
22296
22297
22298
22299
22300
22301
22302
22303
22304
22305
22306
22307
22308
22309
22310
22311
22312
22313
22314
22315
22316
22317
22318
22319
22320
22321
22322
22323
22324
22325
22326
22327
22328
22329
22330
22331
22332
22333
22334
22335
22336
22337
22338
22339
22340
22341
22342
22343
22344
22345
22346
22347
22348
22349
22350
22351
22352
22353
22354
22355
22356
22357
22358
22359
22360
22361
22362
22363
22364
22365
22366
22367
22368
22369
22370
22371
22372
22373
22374
22375
22376
22377
22378
22379
22380
22381
22382
22383
22384
22385
22386
22387
22388
22389
22390
22391
22392
22393
22394
22395
22396
22397
22398
22399
22400
22401
22402
22403
22404
22405
22406
22407
22408
22409
22410
22411
22412
22413
22414
22415
22416
22417
22418
22419
22420
22421
22422
22423
22424
22425
22426
22427
22428
22429
22430
22431
22432
22433
22434
22435
22436
22437
22438
22439
22440
22441
22442
22443
22444
22445
22446
22447
22448
22449
22450
22451
22452
22453
22454
22455
22456
22457
22458
22459
22460
22461
22462
22463
22464
22465
22466
22467
22468
22469
22470
22471
22472
22473
22474
22475
22476
22477
22478
22479
22480
22481
22482
22483
22484
22485
22486
22487
22488
22489
22490
22491
22492
22493
22494
22495
22496
22497
22498
22499
22500
22501
22502
22503
22504
22505
22506
22507
22508
22509
22510
22511
22512
22513
22514
22515
22516
22517
22518
22519
22520
22521
22522
22523
22524
22525
22526
22527
22528
22529
22530
22531
22532
22533
22534
22535
22536
22537
22538
22539
22540
22541
22542
22543
22544
22545
22546
22547
22548
22549
22550
22551
22552
22553
22554
22555
22556
22557
22558
22559
22560
22561
22562
22563
22564
22565
22566
22567
22568
22569
22570
22571
22572
22573
22574
22575
22576
22577
22578
22579
22580
22581
22582
22583
22584
22585
22586
22587
22588
22589
22590
22591
22592
22593
22594
22595
22596
22597
22598
22599
22600
22601
22602
22603
22604
22605
22606
22607
22608
22609
22610
22611
22612
22613
22614
22615
22616
22617
22618
22619
22620
22621
22622
22623
22624
22625
22626
22627
22628
22629
22630
22631
22632
22633
22634
22635
22636
22637
22638
22639
22640
22641
22642
22643
22644
22645
22646
22647
22648
22649
22650
22651
22652
22653
22654
22655
22656
22657
22658
22659
22660
22661
22662
22663
22664
22665
22666
22667
22668
22669
22670
22671
22672
22673
22674
22675
22676
22677
22678
22679
22680
22681
22682
22683
22684
22685
22686
22687
22688
22689
22690
22691
22692
22693
22694
22695
22696
22697
22698
22699
22700
22701
22702
22703
22704
22705
22706
22707
22708
22709
22710
22711
22712
22713
22714
22715
22716
22717
22718
22719
22720
22721
22722
22723
22724
22725
22726
22727
22728
22729
22730
22731
22732
22733
22734
22735
22736
22737
22738
22739
22740
22741
22742
22743
22744
22745
22746
22747
22748
22749
22750
22751
22752
22753
22754
22755
22756
22757
22758
22759
22760
22761
22762
22763
22764
22765
22766
22767
22768
22769
22770
22771
22772
22773
22774
22775
22776
22777
22778
22779
22780
22781
22782
22783
22784
22785
22786
22787
22788
22789
22790
22791
22792
22793
22794
22795
22796
22797
22798
22799
22800
22801
22802
22803
22804
22805
22806
22807
22808
22809
22810
22811
22812
22813
22814
22815
22816
22817
22818
22819
22820
22821
22822
22823
22824
22825
22826
22827
22828
22829
22830
22831
22832
22833
22834
22835
22836
22837
22838
22839
22840
22841
22842
22843
22844
22845
22846
22847
22848
22849
22850
22851
22852
22853
22854
22855
22856
22857
22858
22859
22860
22861
22862
22863
22864
22865
22866
22867
22868
22869
22870
22871
22872
22873
22874
22875
22876
22877
22878
22879
22880
22881
22882
22883
22884
22885
22886
22887
22888
22889
22890
22891
22892
22893
22894
22895
22896
22897
22898
22899
22900
22901
22902
22903
22904
22905
22906
22907
22908
22909
22910
22911
22912
22913
22914
22915
22916
22917
22918
22919
22920
22921
22922
22923
22924
22925
22926
22927
22928
22929
22930
22931
22932
22933
22934
22935
22936
22937
22938
22939
22940
22941
22942
22943
22944
22945
22946
22947
22948
22949
22950
22951
22952
22953
22954
22955
22956
22957
22958
22959
22960
22961
22962
22963
22964
22965
22966
22967
22968
22969
22970
22971
22972
22973
22974
22975
22976
22977
22978
22979
22980
22981
22982
22983
22984
22985
22986
22987
22988
22989
22990
22991
22992
22993
22994
22995
22996
22997
22998
22999
23000
23001
23002
23003
23004
23005
23006
23007
23008
23009
23010
23011
23012
23013
23014
23015
23016
23017
23018
23019
23020
23021
23022
23023
23024
23025
23026
23027
23028
23029
23030
23031
23032
23033
23034
23035
23036
23037
23038
23039
23040
23041
23042
23043
23044
23045
23046
23047
23048
23049
23050
23051
23052
23053
23054
23055
23056
23057
23058
23059
23060
23061
23062
23063
23064
23065
23066
23067
23068
23069
23070
23071
23072
23073
23074
23075
23076
23077
23078
23079
23080
23081
23082
23083
23084
23085
23086
23087
23088
23089
23090
23091
23092
23093
23094
23095
23096
23097
23098
23099
23100
23101
23102
23103
23104
23105
23106
23107
23108
23109
23110
23111
23112
23113
23114
23115
23116
23117
23118
23119
23120
23121
23122
23123
23124
23125
23126
23127
23128
23129
23130
23131
23132
23133
23134
23135
23136
23137
23138
23139
23140
23141
23142
23143
23144
23145
23146
23147
23148
23149
23150
23151
23152
23153
23154
23155
23156
23157
23158
23159
23160
23161
23162
23163
23164
23165
23166
23167
23168
23169
23170
23171
23172
23173
23174
23175
23176
23177
23178
23179
23180
23181
23182
23183
23184
23185
23186
23187
23188
23189
23190
23191
23192
23193
23194
23195
23196
23197
23198
23199
23200
23201
23202
23203
23204
23205
23206
23207
23208
23209
23210
23211
23212
23213
23214
23215
23216
23217
23218
23219
23220
23221
23222
23223
23224
23225
23226
23227
23228
23229
23230
23231
23232
23233
23234
23235
23236
23237
23238
23239
23240
23241
23242
23243
23244
23245
23246
23247
23248
23249
23250
23251
23252
23253
23254
23255
23256
23257
23258
23259
23260
23261
23262
23263
23264
23265
23266
23267
23268
23269
23270
23271
23272
23273
23274
23275
23276
23277
23278
23279
23280
23281
23282
23283
23284
23285
23286
23287
23288
23289
23290
23291
23292
23293
23294
23295
23296
23297
23298
23299
23300
23301
23302
23303
23304
23305
23306
23307
23308
23309
23310
23311
23312
23313
23314
23315
23316
23317
23318
23319
23320
23321
23322
23323
23324
23325
23326
23327
23328
23329
23330
23331
23332
23333
23334
23335
23336
23337
23338
23339
23340
23341
23342
23343
23344
23345
23346
23347
23348
23349
23350
23351
23352
23353
23354
23355
23356
23357
23358
23359
23360
23361
23362
23363
23364
23365
23366
23367
23368
23369
23370
23371
23372
23373
23374
23375
23376
23377
23378
23379
23380
23381
23382
23383
23384
23385
23386
23387
23388
23389
23390
23391
23392
23393
23394
23395
23396
23397
23398
23399
23400
23401
23402
23403
23404
23405
23406
23407
23408
23409
23410
23411
23412
23413
23414
23415
23416
23417
23418
23419
23420
23421
23422
23423
23424
23425
23426
23427
23428
23429
23430
23431
23432
23433
23434
23435
23436
23437
23438
23439
23440
23441
23442
23443
23444
23445
23446
23447
23448
23449
23450
23451
23452
23453
23454
23455
23456
23457
23458
23459
23460
23461
23462
23463
23464
23465
23466
23467
23468
23469
23470
23471
23472
23473
23474
23475
23476
23477
23478
23479
23480
23481
23482
23483
23484
23485
23486
23487
23488
23489
23490
23491
23492
23493
23494
23495
23496
23497
23498
23499
23500
23501
23502
23503
23504
23505
23506
23507
23508
23509
23510
23511
23512
23513
23514
23515
23516
23517
23518
23519
23520
23521
23522
23523
23524
23525
23526
23527
23528
23529
23530
23531
23532
23533
23534
23535
23536
23537
23538
23539
23540
23541
23542
23543
23544
23545
23546
23547
23548
23549
23550
23551
23552
23553
23554
23555
23556
23557
23558
23559
23560
23561
23562
23563
23564
23565
23566
23567
23568
23569
23570
23571
23572
23573
23574
23575
23576
23577
23578
23579
23580
23581
23582
23583
23584
23585
23586
23587
23588
23589
23590
23591
23592
23593
23594
23595
23596
23597
23598
23599
23600
23601
23602
23603
23604
23605
23606
23607
23608
23609
23610
23611
23612
23613
23614
23615
23616
23617
23618
23619
23620
23621
23622
23623
23624
23625
23626
23627
23628
23629
23630
23631
23632
23633
23634
23635
23636
23637
23638
23639
23640
23641
23642
23643
23644
23645
23646
23647
23648
23649
23650
23651
23652
23653
23654
23655
23656
23657
23658
23659
23660
23661
23662
23663
23664
23665
23666
23667
23668
23669
23670
23671
23672
23673
23674
23675
23676
23677
23678
23679
23680
23681
23682
23683
23684
23685
23686
23687
23688
23689
23690
23691
23692
23693
23694
23695
23696
23697
23698
23699
23700
23701
23702
23703
23704
23705
23706
23707
23708
23709
23710
23711
23712
23713
23714
23715
23716
23717
23718
23719
23720
23721
23722
23723
23724
23725
23726
23727
23728
23729
23730
23731
23732
23733
23734
23735
23736
23737
23738
23739
23740
23741
23742
23743
23744
23745
23746
23747
23748
23749
23750
23751
23752
23753
23754
23755
23756
23757
23758
23759
23760
23761
23762
23763
23764
23765
23766
23767
23768
23769
23770
23771
23772
23773
23774
23775
23776
23777
23778
23779
23780
23781
23782
23783
23784
23785
23786
23787
23788
23789
23790
23791
23792
23793
23794
23795
23796
23797
23798
23799
23800
23801
23802
23803
23804
23805
23806
23807
23808
23809
23810
23811
23812
23813
23814
23815
23816
23817
23818
23819
23820
23821
23822
23823
23824
23825
23826
23827
23828
23829
23830
23831
23832
23833
23834
23835
23836
23837
23838
23839
23840
23841
23842
23843
23844
23845
23846
23847
23848
23849
23850
23851
23852
23853
23854
23855
23856
23857
23858
23859
23860
23861
23862
23863
23864
23865
23866
23867
23868
23869
23870
23871
23872
23873
23874
23875
23876
23877
23878
23879
23880
23881
23882
23883
23884
23885
23886
23887
23888
23889
23890
23891
23892
23893
23894
23895
23896
23897
23898
23899
23900
23901
23902
23903
23904
23905
23906
23907
23908
23909
23910
23911
23912
23913
23914
23915
23916
23917
23918
23919
23920
23921
23922
23923
23924
23925
23926
23927
23928
23929
23930
23931
23932
23933
23934
23935
23936
23937
23938
23939
23940
23941
23942
23943
23944
23945
23946
23947
23948
23949
23950
23951
23952
23953
23954
23955
23956
23957
23958
23959
23960
23961
23962
23963
23964
23965
23966
23967
23968
23969
23970
23971
23972
23973
23974
23975
23976
23977
23978
23979
23980
23981
23982
23983
23984
23985
23986
23987
23988
23989
23990
23991
23992
23993
23994
23995
23996
23997
23998
23999
24000
24001
24002
24003
24004
24005
24006
24007
24008
24009
24010
24011
24012
24013
24014
24015
24016
24017
24018
24019
24020
24021
24022
24023
24024
24025
24026
24027
24028
24029
24030
24031
24032
24033
24034
24035
24036
24037
24038
24039
24040
24041
24042
24043
24044
24045
24046
24047
24048
24049
24050
24051
24052
24053
24054
24055
24056
24057
24058
24059
24060
24061
24062
24063
24064
24065
24066
24067
24068
24069
24070
24071
24072
24073
24074
24075
24076
24077
24078
24079
24080
24081
24082
24083
24084
24085
24086
24087
24088
24089
24090
24091
24092
24093
24094
24095
24096
24097
24098
24099
24100
24101
24102
24103
24104
24105
24106
24107
24108
24109
24110
24111
24112
24113
24114
24115
24116
24117
24118
24119
24120
24121
24122
24123
24124
24125
24126
24127
24128
24129
24130
24131
24132
24133
24134
24135
24136
24137
24138
24139
24140
24141
24142
24143
24144
24145
24146
24147
24148
24149
24150
24151
24152
24153
24154
24155
24156
24157
24158
24159
24160
24161
24162
24163
24164
24165
24166
24167
24168
24169
24170
24171
24172
24173
24174
24175
24176
24177
24178
24179
24180
24181
24182
24183
24184
24185
24186
24187
24188
24189
24190
24191
24192
24193
24194
24195
24196
24197
24198
24199
24200
24201
24202
24203
24204
24205
24206
24207
24208
24209
24210
24211
24212
24213
24214
24215
24216
24217
24218
24219
24220
24221
24222
24223
24224
24225
24226
24227
24228
24229
24230
24231
24232
24233
24234
24235
24236
24237
24238
24239
24240
24241
24242
24243
24244
24245
24246
24247
24248
24249
24250
24251
24252
24253
24254
24255
24256
24257
24258
24259
24260
24261
24262
24263
24264
24265
24266
24267
24268
24269
24270
24271
24272
24273
24274
24275
24276
24277
24278
24279
24280
24281
24282
24283
24284
24285
24286
24287
24288
24289
24290
24291
24292
24293
24294
24295
24296
24297
24298
24299
24300
24301
24302
24303
24304
24305
24306
24307
24308
24309
24310
24311
24312
24313
24314
24315
24316
24317
24318
24319
24320
24321
24322
24323
24324
24325
24326
24327
24328
24329
24330
24331
24332
24333
24334
24335
24336
24337
24338
24339
24340
24341
24342
24343
24344
24345
24346
24347
24348
24349
24350
24351
24352
24353
24354
24355
24356
24357
24358
24359
24360
24361
24362
24363
24364
24365
24366
24367
24368
24369
24370
24371
24372
24373
24374
24375
24376
24377
24378
24379
24380
24381
24382
24383
24384
24385
24386
24387
24388
24389
24390
24391
24392
24393
24394
24395
24396
24397
24398
24399
24400
24401
24402
24403
24404
24405
24406
24407
24408
24409
24410
24411
24412
24413
24414
24415
24416
24417
24418
24419
24420
24421
24422
24423
24424
24425
24426
24427
24428
24429
24430
24431
24432
24433
24434
24435
24436
24437
24438
24439
24440
24441
24442
24443
24444
24445
24446
24447
24448
24449
24450
24451
24452
24453
24454
24455
24456
24457
24458
24459
24460
24461
24462
24463
24464
24465
24466
24467
24468
24469
24470
24471
24472
24473
24474
24475
24476
24477
24478
24479
24480
24481
24482
24483
24484
24485
24486
24487
24488
24489
24490
24491
24492
24493
24494
24495
24496
24497
24498
24499
24500
24501
24502
24503
24504
24505
24506
24507
24508
24509
24510
24511
24512
24513
24514
24515
24516
24517
24518
24519
24520
24521
24522
24523
24524
24525
24526
24527
24528
24529
24530
24531
24532
24533
24534
24535
24536
24537
24538
24539
24540
24541
24542
24543
24544
24545
24546
24547
24548
24549
24550
24551
24552
24553
24554
24555
24556
24557
24558
24559
24560
24561
24562
24563
24564
24565
24566
24567
24568
24569
24570
24571
24572
24573
24574
24575
24576
24577
24578
24579
24580
24581
24582
24583
24584
24585
24586
24587
24588
24589
24590
24591
24592
24593
24594
24595
24596
24597
24598
24599
24600
24601
24602
24603
24604
24605
24606
24607
24608
24609
24610
24611
24612
24613
24614
24615
24616
24617
24618
24619
24620
24621
24622
24623
24624
24625
24626
24627
24628
24629
24630
24631
24632
24633
24634
24635
24636
24637
24638
24639
24640
24641
24642
24643
24644
24645
24646
24647
24648
24649
24650
24651
24652
24653
24654
24655
24656
24657
24658
24659
24660
24661
24662
24663
24664
24665
24666
24667
24668
24669
24670
24671
24672
24673
24674
24675
24676
24677
24678
24679
24680
24681
24682
24683
24684
24685
24686
24687
24688
24689
24690
24691
24692
24693
24694
24695
24696
24697
24698
24699
24700
24701
24702
24703
24704
24705
24706
24707
24708
24709
24710
24711
24712
24713
24714
24715
24716
24717
24718
24719
24720
24721
24722
24723
24724
24725
24726
24727
24728
24729
24730
24731
24732
24733
24734
24735
24736
24737
24738
24739
24740
24741
24742
24743
24744
24745
24746
24747
24748
24749
24750
24751
24752
24753
24754
24755
24756
24757
24758
24759
24760
24761
24762
24763
24764
24765
24766
24767
24768
24769
24770
24771
24772
24773
24774
24775
24776
24777
24778
24779
24780
24781
24782
24783
24784
24785
24786
24787
24788
24789
24790
24791
24792
24793
24794
24795
24796
24797
24798
24799
24800
24801
24802
24803
24804
24805
24806
24807
24808
24809
24810
24811
24812
24813
24814
24815
24816
24817
24818
24819
24820
24821
24822
24823
24824
24825
24826
24827
24828
24829
24830
24831
24832
24833
24834
24835
24836
24837
24838
24839
24840
24841
24842
24843
24844
24845
24846
24847
24848
24849
24850
24851
24852
24853
24854
24855
24856
24857
24858
24859
24860
24861
24862
24863
24864
24865
24866
24867
24868
24869
24870
24871
24872
24873
24874
24875
24876
24877
24878
24879
24880
24881
24882
24883
24884
24885
24886
24887
24888
24889
24890
24891
24892
24893
24894
24895
24896
24897
24898
24899
24900
24901
24902
24903
24904
24905
24906
24907
24908
24909
24910
24911
24912
24913
24914
24915
24916
24917
24918
24919
24920
24921
24922
24923
24924
24925
24926
24927
24928
24929
24930
24931
24932
24933
24934
24935
24936
24937
24938
24939
24940
24941
24942
24943
24944
24945
24946
24947
24948
24949
24950
24951
24952
24953
24954
24955
24956
24957
24958
24959
24960
24961
24962
24963
24964
24965
24966
24967
24968
24969
24970
24971
24972
24973
24974
24975
24976
24977
24978
24979
24980
24981
24982
24983
24984
24985
24986
24987
24988
24989
24990
24991
24992
24993
24994
24995
24996
24997
24998
24999
25000
25001
25002
25003
25004
25005
25006
25007
25008
25009
25010
25011
25012
25013
25014
25015
25016
25017
25018
25019
25020
25021
25022
25023
25024
25025
25026
25027
25028
25029
25030
25031
25032
25033
25034
25035
25036
25037
25038
25039
25040
25041
25042
25043
25044
25045
25046
25047
25048
25049
25050
25051
25052
25053
25054
25055
25056
25057
25058
25059
25060
25061
25062
25063
25064
25065
25066
25067
25068
25069
25070
25071
25072
25073
25074
25075
25076
25077
25078
25079
25080
25081
25082
25083
25084
25085
25086
25087
25088
25089
25090
25091
25092
25093
25094
25095
25096
25097
25098
25099
25100
25101
25102
25103
25104
25105
25106
25107
25108
25109
25110
25111
25112
25113
25114
25115
25116
25117
25118
25119
25120
25121
25122
25123
25124
25125
25126
25127
25128
25129
25130
25131
25132
25133
25134
25135
25136
25137
25138
25139
25140
25141
25142
25143
25144
25145
25146
25147
25148
25149
25150
25151
25152
25153
25154
25155
25156
25157
25158
25159
25160
25161
25162
25163
25164
25165
25166
25167
25168
25169
25170
25171
25172
25173
25174
25175
25176
25177
25178
25179
25180
25181
25182
25183
25184
25185
25186
25187
25188
25189
25190
25191
25192
25193
25194
25195
25196
25197
25198
25199
25200
25201
25202
25203
25204
25205
25206
25207
25208
25209
25210
25211
25212
25213
25214
25215
25216
25217
25218
25219
25220
25221
25222
25223
25224
25225
25226
25227
25228
25229
25230
25231
25232
25233
25234
25235
25236
25237
25238
25239
25240
25241
25242
25243
25244
25245
25246
25247
25248
25249
25250
25251
25252
25253
25254
25255
25256
25257
25258
25259
25260
25261
25262
25263
25264
25265
25266
25267
25268
25269
25270
25271
25272
25273
25274
25275
25276
25277
25278
25279
25280
25281
25282
25283
25284
25285
25286
25287
25288
25289
25290
25291
25292
25293
25294
25295
25296
25297
25298
25299
25300
25301
25302
25303
25304
25305
25306
25307
25308
25309
25310
25311
25312
25313
25314
25315
25316
25317
25318
25319
25320
25321
25322
25323
25324
25325
25326
25327
25328
25329
25330
25331
25332
25333
25334
25335
25336
25337
25338
25339
25340
25341
25342
25343
25344
25345
25346
25347
25348
25349
25350
25351
25352
25353
25354
25355
25356
25357
25358
25359
25360
25361
25362
25363
25364
25365
25366
25367
25368
25369
25370
25371
25372
25373
25374
25375
25376
25377
25378
25379
25380
25381
25382
25383
25384
25385
25386
25387
25388
25389
25390
25391
25392
25393
25394
25395
25396
25397
25398
25399
25400
25401
25402
25403
25404
25405
25406
25407
25408
25409
25410
25411
25412
25413
25414
25415
25416
25417
25418
25419
25420
25421
25422
25423
25424
25425
25426
25427
25428
25429
25430
25431
25432
25433
25434
25435
25436
25437
25438
25439
25440
25441
25442
25443
25444
25445
25446
25447
25448
25449
25450
25451
25452
25453
25454
25455
25456
25457
25458
25459
25460
25461
25462
25463
25464
25465
25466
25467
25468
25469
25470
25471
25472
25473
25474
25475
25476
25477
25478
25479
25480
25481
25482
25483
25484
25485
25486
25487
25488
25489
25490
25491
25492
25493
25494
25495
25496
25497
25498
25499
25500
25501
25502
25503
25504
25505
25506
25507
25508
25509
25510
25511
25512
25513
25514
25515
25516
25517
25518
25519
25520
25521
25522
25523
25524
25525
25526
25527
25528
25529
25530
25531
25532
25533
25534
25535
25536
25537
25538
25539
25540
25541
25542
25543
25544
25545
25546
25547
25548
25549
25550
25551
25552
25553
25554
25555
25556
25557
25558
25559
25560
25561
25562
25563
25564
25565
25566
25567
25568
25569
25570
25571
25572
25573
25574
25575
25576
25577
25578
25579
25580
25581
25582
25583
25584
25585
25586
25587
25588
25589
25590
25591
25592
25593
25594
25595
25596
25597
25598
25599
25600
25601
25602
25603
25604
25605
25606
25607
25608
25609
25610
25611
25612
25613
25614
25615
25616
25617
25618
25619
25620
25621
25622
25623
25624
25625
25626
25627
25628
25629
25630
25631
25632
25633
25634
25635
25636
25637
25638
25639
25640
25641
25642
25643
25644
25645
25646
25647
25648
25649
25650
25651
25652
25653
25654
25655
25656
25657
25658
25659
25660
25661
25662
25663
25664
25665
25666
25667
25668
25669
25670
25671
25672
25673
25674
25675
25676
25677
25678
25679
25680
25681
25682
25683
25684
25685
25686
25687
25688
25689
25690
25691
25692
25693
25694
25695
25696
25697
25698
25699
25700
25701
25702
25703
25704
25705
25706
25707
25708
25709
25710
25711
25712
25713
25714
25715
25716
25717
25718
25719
25720
25721
25722
25723
25724
25725
25726
25727
25728
25729
25730
25731
25732
25733
25734
25735
25736
25737
25738
25739
25740
25741
25742
25743
25744
25745
25746
25747
25748
25749
25750
25751
25752
25753
25754
25755
25756
25757
25758
25759
25760
25761
25762
25763
25764
25765
25766
25767
25768
25769
25770
25771
25772
25773
25774
25775
25776
25777
25778
25779
25780
25781
25782
25783
25784
25785
25786
25787
25788
25789
25790
25791
25792
25793
25794
25795
25796
25797
25798
25799
25800
25801
25802
25803
25804
25805
25806
25807
25808
25809
25810
25811
25812
25813
25814
25815
25816
25817
25818
25819
25820
25821
25822
25823
25824
25825
25826
25827
25828
25829
25830
25831
25832
25833
25834
25835
25836
25837
25838
25839
25840
25841
25842
25843
25844
25845
25846
25847
25848
25849
25850
25851
25852
25853
25854
25855
25856
25857
25858
25859
25860
25861
25862
25863
25864
25865
25866
25867
25868
25869
25870
25871
25872
25873
25874
25875
25876
25877
25878
25879
25880
25881
25882
25883
25884
25885
25886
25887
25888
25889
25890
25891
25892
25893
25894
25895
25896
25897
25898
25899
25900
25901
25902
25903
25904
25905
25906
25907
25908
25909
25910
25911
25912
25913
25914
25915
25916
25917
25918
25919
25920
25921
25922
25923
25924
25925
25926
25927
25928
25929
25930
25931
25932
25933
25934
25935
25936
25937
25938
25939
25940
25941
25942
25943
25944
25945
25946
25947
25948
25949
25950
25951
25952
25953
25954
25955
25956
25957
25958
25959
25960
25961
25962
25963
25964
25965
25966
25967
25968
25969
25970
25971
25972
25973
25974
25975
25976
25977
25978
25979
25980
25981
25982
25983
25984
25985
25986
25987
25988
25989
25990
25991
25992
25993
25994
25995
25996
25997
25998
25999
26000
26001
26002
26003
26004
26005
26006
26007
26008
26009
26010
26011
26012
26013
26014
26015
26016
26017
26018
26019
26020
26021
26022
26023
26024
26025
26026
26027
26028
26029
26030
26031
26032
26033
26034
26035
26036
26037
26038
26039
26040
26041
26042
26043
26044
26045
26046
26047
26048
26049
26050
26051
26052
26053
26054
26055
26056
26057
26058
26059
26060
26061
26062
26063
26064
26065
26066
26067
26068
26069
26070
26071
26072
26073
26074
26075
26076
26077
26078
26079
26080
26081
26082
26083
26084
26085
26086
26087
26088
26089
26090
26091
26092
26093
26094
26095
26096
26097
26098
26099
26100
26101
26102
26103
26104
26105
26106
26107
26108
26109
26110
26111
26112
26113
26114
26115
26116
26117
26118
26119
26120
26121
26122
26123
26124
26125
26126
26127
26128
26129
26130
26131
26132
26133
26134
26135
26136
26137
26138
26139
26140
26141
26142
26143
26144
26145
26146
26147
26148
26149
26150
26151
26152
26153
26154
26155
26156
26157
26158
26159
26160
26161
26162
26163
26164
26165
26166
26167
26168
26169
26170
26171
26172
26173
26174
26175
26176
26177
26178
26179
26180
26181
26182
26183
26184
26185
26186
26187
26188
26189
26190
26191
26192
26193
26194
26195
26196
26197
26198
26199
26200
26201
26202
26203
26204
26205
26206
26207
26208
26209
26210
26211
26212
26213
26214
26215
26216
26217
26218
26219
26220
26221
26222
26223
26224
26225
26226
26227
26228
26229
26230
26231
26232
26233
26234
26235
26236
26237
26238
26239
26240
26241
26242
26243
26244
26245
26246
26247
26248
26249
26250
26251
26252
26253
26254
26255
26256
26257
26258
26259
26260
26261
26262
26263
26264
26265
26266
26267
26268
26269
26270
26271
26272
26273
26274
26275
26276
26277
26278
26279
26280
26281
26282
26283
26284
26285
26286
26287
26288
26289
26290
26291
26292
26293
26294
26295
26296
26297
26298
26299
26300
26301
26302
26303
26304
26305
26306
26307
26308
26309
26310
26311
26312
26313
26314
26315
26316
26317
26318
26319
26320
26321
26322
26323
26324
26325
26326
26327
26328
26329
26330
26331
26332
26333
26334
26335
26336
26337
26338
26339
26340
26341
26342
26343
26344
26345
26346
26347
26348
26349
26350
26351
26352
26353
26354
26355
26356
26357
26358
26359
26360
26361
26362
26363
26364
26365
26366
26367
26368
26369
26370
26371
26372
26373
26374
26375
26376
26377
26378
26379
26380
26381
26382
26383
26384
26385
26386
26387
26388
26389
26390
26391
26392
26393
26394
26395
26396
26397
26398
26399
26400
26401
26402
26403
26404
26405
26406
26407
26408
26409
26410
26411
26412
26413
26414
26415
26416
26417
26418
26419
26420
26421
26422
26423
26424
26425
26426
26427
26428
26429
26430
26431
26432
26433
26434
26435
26436
26437
26438
26439
26440
26441
26442
26443
26444
26445
26446
26447
26448
26449
26450
26451
26452
26453
26454
26455
26456
26457
26458
26459
26460
26461
26462
26463
26464
26465
26466
26467
26468
26469
26470
26471
26472
26473
26474
26475
26476
26477
26478
26479
26480
26481
26482
26483
26484
26485
26486
26487
26488
26489
26490
26491
26492
26493
26494
26495
26496
26497
26498
26499
26500
26501
26502
26503
26504
26505
26506
26507
26508
26509
26510
26511
26512
26513
26514
26515
26516
26517
26518
26519
26520
26521
26522
26523
26524
26525
26526
26527
26528
26529
26530
26531
26532
26533
26534
26535
26536
26537
26538
26539
26540
26541
26542
26543
26544
26545
26546
26547
26548
26549
26550
26551
26552
26553
26554
26555
26556
26557
26558
26559
26560
26561
26562
26563
26564
26565
26566
26567
26568
26569
26570
26571
26572
26573
26574
26575
26576
26577
26578
26579
26580
26581
26582
26583
26584
26585
26586
26587
26588
26589
26590
26591
26592
26593
26594
26595
26596
26597
26598
26599
26600
26601
26602
26603
26604
26605
26606
26607
26608
26609
26610
26611
26612
26613
26614
26615
26616
26617
26618
26619
26620
26621
26622
26623
26624
26625
26626
26627
26628
26629
26630
26631
26632
26633
26634
26635
26636
26637
26638
26639
26640
26641
26642
26643
26644
26645
26646
26647
26648
26649
26650
26651
26652
26653
26654
26655
26656
26657
26658
26659
26660
26661
26662
26663
26664
26665
26666
26667
26668
26669
26670
26671
26672
26673
26674
26675
26676
26677
26678
26679
26680
26681
26682
26683
26684
26685
26686
26687
26688
26689
26690
26691
26692
26693
26694
26695
26696
26697
26698
26699
26700
26701
26702
26703
26704
26705
26706
26707
26708
26709
26710
26711
26712
26713
26714
26715
26716
26717
26718
26719
26720
26721
26722
26723
26724
26725
26726
26727
26728
26729
26730
26731
26732
26733
26734
26735
26736
26737
26738
26739
26740
26741
26742
26743
26744
26745
26746
26747
26748
26749
26750
26751
26752
26753
26754
26755
26756
26757
26758
26759
26760
26761
26762
26763
26764
26765
26766
26767
26768
26769
26770
26771
26772
26773
26774
26775
26776
26777
26778
26779
26780
26781
26782
26783
26784
26785
26786
26787
26788
26789
26790
26791
26792
26793
26794
26795
26796
26797
26798
26799
26800
26801
26802
26803
26804
26805
26806
26807
26808
26809
26810
26811
26812
26813
26814
26815
26816
26817
26818
26819
26820
26821
26822
26823
26824
26825
26826
26827
26828
26829
26830
26831
26832
26833
26834
26835
26836
26837
26838
26839
26840
26841
26842
26843
26844
26845
26846
26847
26848
26849
26850
26851
26852
26853
26854
26855
26856
26857
26858
26859
26860
26861
26862
26863
26864
26865
26866
26867
26868
26869
26870
26871
26872
26873
26874
26875
26876
26877
26878
26879
26880
26881
26882
26883
26884
26885
26886
26887
26888
26889
26890
26891
26892
26893
26894
26895
26896
26897
26898
26899
26900
26901
26902
26903
26904
26905
26906
26907
26908
26909
26910
26911
26912
26913
26914
26915
26916
26917
26918
26919
26920
26921
26922
26923
26924
26925
26926
26927
26928
26929
26930
26931
26932
26933
26934
26935
26936
26937
26938
26939
26940
26941
26942
26943
26944
26945
26946
26947
26948
26949
26950
26951
26952
26953
26954
26955
26956
26957
26958
26959
26960
26961
26962
26963
26964
26965
26966
26967
26968
26969
26970
26971
26972
26973
26974
26975
26976
26977
26978
26979
26980
26981
26982
26983
26984
26985
26986
26987
26988
26989
26990
26991
26992
26993
26994
26995
26996
26997
26998
26999
27000
27001
27002
27003
27004
27005
27006
27007
27008
27009
27010
27011
27012
27013
27014
27015
27016
27017
27018
27019
27020
27021
27022
27023
27024
27025
27026
27027
27028
27029
27030
27031
27032
27033
27034
27035
27036
27037
27038
27039
27040
27041
27042
27043
27044
27045
27046
27047
27048
27049
27050
27051
27052
27053
27054
27055
27056
27057
27058
27059
27060
27061
27062
27063
27064
27065
27066
27067
27068
27069
27070
27071
27072
27073
27074
27075
27076
27077
27078
27079
27080
27081
27082
27083
27084
27085
27086
27087
27088
27089
27090
27091
27092
27093
27094
27095
27096
27097
27098
27099
27100
27101
27102
27103
27104
27105
27106
27107
27108
27109
27110
27111
27112
27113
27114
27115
27116
27117
27118
27119
27120
27121
27122
27123
27124
27125
27126
27127
27128
27129
27130
27131
27132
27133
27134
27135
27136
27137
27138
27139
27140
27141
27142
27143
27144
27145
27146
27147
27148
27149
27150
27151
27152
27153
27154
27155
27156
27157
27158
27159
27160
27161
27162
27163
27164
27165
27166
27167
27168
27169
27170
27171
27172
27173
27174
27175
27176
27177
27178
27179
27180
27181
27182
27183
27184
27185
27186
27187
27188
27189
27190
27191
27192
27193
27194
27195
27196
27197
27198
27199
27200
27201
27202
27203
27204
27205
27206
27207
27208
27209
27210
27211
27212
27213
27214
27215
27216
27217
27218
27219
27220
27221
27222
27223
27224
27225
27226
27227
27228
27229
27230
27231
27232
27233
27234
27235
27236
27237
27238
27239
27240
27241
27242
27243
27244
27245
27246
27247
27248
27249
27250
27251
27252
27253
27254
27255
27256
27257
27258
27259
27260
27261
27262
27263
27264
27265
27266
27267
27268
27269
27270
27271
27272
27273
27274
27275
27276
27277
27278
27279
27280
27281
27282
27283
27284
27285
27286
27287
27288
27289
27290
27291
27292
27293
27294
27295
27296
27297
27298
27299
27300
27301
27302
27303
27304
27305
27306
27307
27308
27309
27310
27311
27312
27313
27314
27315
27316
27317
27318
27319
27320
27321
27322
27323
27324
27325
27326
27327
27328
27329
27330
27331
27332
27333
27334
27335
27336
27337
27338
27339
27340
27341
27342
27343
27344
27345
27346
27347
27348
27349
27350
27351
27352
27353
27354
27355
27356
27357
27358
27359
27360
27361
27362
27363
27364
27365
27366
27367
27368
27369
27370
27371
27372
27373
27374
27375
27376
27377
27378
27379
27380
27381
27382
27383
27384
27385
27386
27387
27388
27389
27390
27391
27392
27393
27394
27395
27396
27397
27398
27399
27400
27401
27402
27403
27404
27405
27406
27407
27408
27409
27410
27411
27412
27413
27414
27415
27416
27417
27418
27419
27420
27421
27422
27423
27424
27425
27426
27427
27428
27429
27430
27431
27432
27433
27434
27435
27436
27437
27438
27439
27440
27441
27442
27443
27444
27445
27446
27447
27448
27449
27450
27451
27452
27453
27454
27455
27456
27457
27458
27459
27460
27461
27462
27463
27464
27465
27466
27467
27468
27469
27470
27471
27472
27473
27474
27475
27476
27477
27478
27479
27480
27481
27482
27483
27484
27485
27486
27487
27488
27489
27490
27491
27492
27493
27494
27495
27496
27497
27498
27499
27500
27501
27502
27503
27504
27505
27506
27507
27508
27509
27510
27511
27512
27513
27514
27515
27516
27517
27518
27519
27520
27521
27522
27523
27524
27525
27526
27527
27528
27529
27530
27531
27532
27533
27534
27535
27536
27537
27538
27539
27540
27541
27542
27543
27544
27545
27546
27547
27548
27549
27550
27551
27552
27553
27554
27555
27556
27557
27558
27559
27560
27561
27562
27563
27564
27565
27566
27567
27568
27569
27570
27571
27572
27573
27574
27575
27576
27577
27578
27579
27580
27581
27582
27583
27584
27585
27586
27587
27588
27589
27590
27591
27592
27593
27594
27595
27596
27597
27598
27599
27600
27601
27602
27603
27604
27605
27606
27607
27608
27609
27610
27611
27612
27613
27614
27615
27616
27617
27618
27619
27620
27621
27622
27623
27624
27625
27626
27627
27628
27629
27630
27631
27632
27633
27634
27635
27636
27637
27638
27639
27640
27641
27642
27643
27644
27645
27646
27647
27648
27649
27650
27651
27652
27653
27654
27655
27656
27657
27658
27659
27660
27661
27662
27663
27664
27665
27666
27667
27668
27669
27670
27671
27672
27673
27674
27675
27676
27677
27678
27679
27680
27681
27682
27683
27684
27685
27686
27687
27688
27689
27690
27691
27692
27693
27694
27695
27696
27697
27698
27699
27700
27701
27702
27703
27704
27705
27706
27707
27708
27709
27710
27711
27712
27713
27714
27715
27716
27717
27718
27719
27720
27721
27722
27723
27724
27725
27726
27727
27728
27729
27730
27731
27732
27733
27734
27735
27736
27737
27738
27739
27740
27741
27742
27743
27744
27745
27746
27747
27748
27749
27750
27751
27752
27753
27754
27755
27756
27757
27758
27759
27760
27761
27762
27763
27764
27765
27766
27767
27768
27769
27770
27771
27772
27773
27774
27775
27776
27777
27778
27779
27780
27781
27782
27783
27784
27785
27786
27787
27788
27789
27790
27791
27792
27793
27794
27795
27796
27797
27798
27799
27800
27801
27802
27803
27804
27805
27806
27807
27808
27809
27810
27811
27812
27813
27814
27815
27816
27817
27818
27819
27820
27821
27822
27823
27824
27825
27826
27827
27828
27829
27830
27831
27832
27833
27834
27835
27836
27837
27838
27839
27840
27841
27842
27843
27844
27845
27846
27847
27848
27849
27850
27851
27852
27853
27854
27855
27856
27857
27858
27859
27860
27861
27862
27863
27864
27865
27866
27867
27868
27869
27870
27871
27872
27873
27874
27875
27876
27877
27878
27879
27880
27881
27882
27883
27884
27885
27886
27887
27888
27889
27890
27891
27892
27893
27894
27895
27896
27897
27898
27899
27900
27901
27902
27903
27904
27905
27906
27907
27908
27909
27910
27911
27912
27913
27914
27915
27916
27917
27918
27919
27920
27921
27922
27923
27924
27925
27926
27927
27928
27929
27930
27931
27932
27933
27934
27935
27936
27937
27938
27939
27940
27941
27942
27943
27944
27945
27946
27947
27948
27949
27950
27951
27952
27953
27954
27955
27956
27957
27958
27959
27960
27961
27962
27963
27964
27965
27966
27967
27968
27969
27970
27971
27972
27973
27974
27975
27976
27977
27978
27979
27980
27981
27982
27983
27984
27985
27986
27987
27988
27989
27990
27991
27992
27993
27994
27995
27996
27997
27998
27999
28000
28001
28002
28003
28004
28005
28006
28007
28008
28009
28010
28011
28012
28013
28014
28015
28016
28017
28018
28019
28020
28021
28022
28023
28024
28025
28026
28027
28028
28029
28030
28031
28032
28033
28034
28035
28036
28037
28038
28039
28040
28041
28042
28043
28044
28045
28046
28047
28048
28049
28050
28051
28052
28053
28054
28055
28056
28057
28058
28059
28060
28061
28062
28063
28064
28065
28066
28067
28068
28069
28070
28071
28072
28073
28074
28075
28076
28077
28078
28079
28080
28081
28082
28083
28084
28085
28086
28087
28088
28089
28090
28091
28092
28093
28094
28095
28096
28097
28098
28099
28100
28101
28102
28103
28104
28105
28106
28107
28108
28109
28110
28111
28112
28113
28114
28115
28116
28117
28118
28119
28120
28121
28122
28123
28124
28125
28126
28127
28128
28129
28130
28131
28132
28133
28134
28135
28136
28137
28138
28139
28140
28141
28142
28143
28144
28145
28146
28147
28148
28149
28150
28151
28152
28153
28154
28155
28156
28157
28158
28159
28160
28161
28162
28163
28164
28165
28166
28167
28168
28169
28170
28171
28172
28173
28174
28175
28176
28177
28178
28179
28180
28181
28182
28183
28184
28185
28186
28187
28188
28189
28190
28191
28192
28193
28194
28195
28196
28197
28198
28199
28200
28201
28202
28203
28204
28205
28206
28207
28208
28209
28210
28211
28212
28213
28214
28215
28216
28217
28218
28219
28220
28221
28222
28223
28224
28225
28226
28227
28228
28229
28230
28231
28232
28233
28234
28235
28236
28237
28238
28239
28240
28241
28242
28243
28244
28245
28246
28247
28248
28249
28250
28251
28252
28253
28254
28255
28256
28257
28258
28259
28260
28261
28262
28263
28264
28265
28266
28267
28268
28269
28270
28271
28272
28273
28274
28275
28276
28277
28278
28279
28280
28281
28282
28283
28284
28285
28286
28287
28288
28289
28290
28291
28292
28293
28294
28295
28296
28297
28298
28299
28300
28301
28302
28303
28304
28305
28306
28307
28308
28309
28310
28311
28312
28313
28314
28315
28316
28317
28318
28319
28320
28321
28322
28323
28324
28325
28326
28327
28328
28329
28330
28331
28332
28333
28334
28335
28336
28337
28338
28339
28340
28341
28342
28343
28344
28345
28346
28347
28348
28349
28350
28351
28352
28353
28354
28355
28356
28357
28358
28359
28360
28361
28362
28363
28364
28365
28366
28367
28368
28369
28370
28371
28372
28373
28374
28375
28376
28377
28378
28379
28380
28381
28382
28383
28384
28385
28386
28387
28388
28389
28390
28391
28392
28393
28394
28395
28396
28397
28398
28399
28400
28401
28402
28403
28404
28405
28406
28407
28408
28409
28410
28411
28412
28413
28414
28415
28416
28417
28418
28419
28420
28421
28422
28423
28424
28425
28426
28427
28428
28429
28430
28431
28432
28433
28434
28435
28436
28437
28438
28439
28440
28441
28442
28443
28444
28445
28446
28447
28448
28449
28450
28451
28452
28453
28454
28455
28456
28457
28458
28459
28460
28461
28462
28463
28464
28465
28466
28467
28468
28469
28470
28471
28472
28473
28474
28475
28476
28477
28478
28479
28480
28481
28482
28483
28484
28485
28486
28487
28488
28489
28490
28491
28492
28493
28494
28495
28496
28497
28498
28499
28500
28501
28502
28503
28504
28505
28506
28507
28508
28509
28510
28511
28512
28513
28514
28515
28516
28517
28518
28519
28520
28521
28522
28523
28524
28525
28526
28527
28528
28529
28530
28531
28532
28533
28534
28535
28536
28537
28538
28539
28540
28541
28542
28543
28544
28545
28546
28547
28548
28549
28550
28551
28552
28553
28554
28555
28556
28557
28558
28559
28560
28561
28562
28563
28564
28565
28566
28567
28568
28569
28570
28571
28572
28573
28574
28575
28576
28577
28578
28579
28580
28581
28582
28583
28584
28585
28586
28587
28588
28589
28590
28591
28592
28593
28594
28595
28596
28597
28598
28599
28600
28601
28602
28603
28604
28605
28606
28607
28608
28609
28610
28611
28612
28613
28614
28615
28616
28617
28618
28619
28620
28621
28622
28623
28624
28625
28626
28627
28628
28629
28630
28631
28632
28633
28634
28635
28636
28637
28638
28639
28640
28641
28642
28643
28644
28645
28646
28647
28648
28649
28650
28651
28652
28653
28654
28655
28656
28657
28658
28659
28660
28661
28662
28663
28664
28665
28666
28667
28668
28669
28670
28671
28672
28673
28674
28675
28676
28677
28678
28679
28680
28681
28682
28683
28684
28685
28686
28687
28688
28689
28690
28691
28692
28693
28694
28695
28696
28697
28698
28699
28700
28701
28702
28703
28704
28705
28706
28707
28708
28709
28710
28711
28712
28713
28714
28715
28716
28717
28718
28719
28720
28721
28722
28723
28724
28725
28726
28727
28728
28729
28730
28731
28732
28733
28734
28735
28736
28737
28738
28739
28740
28741
28742
28743
28744
28745
28746
28747
28748
28749
28750
28751
28752
28753
28754
28755
28756
28757
28758
28759
28760
28761
28762
28763
28764
28765
28766
28767
28768
28769
28770
28771
28772
28773
28774
28775
28776
28777
28778
28779
28780
28781
28782
28783
28784
28785
28786
28787
28788
28789
28790
28791
28792
28793
28794
28795
28796
28797
28798
28799
28800
28801
28802
28803
28804
28805
28806
28807
28808
28809
28810
28811
28812
28813
28814
28815
28816
28817
28818
28819
28820
28821
28822
28823
28824
28825
28826
28827
28828
28829
28830
28831
28832
28833
28834
28835
28836
28837
28838
28839
28840
28841
28842
28843
28844
28845
28846
28847
28848
28849
28850
28851
28852
28853
28854
28855
28856
28857
28858
28859
28860
28861
28862
28863
28864
28865
28866
28867
28868
28869
28870
28871
28872
28873
28874
28875
28876
28877
28878
28879
28880
28881
28882
28883
28884
28885
28886
28887
28888
28889
28890
28891
28892
28893
28894
28895
28896
28897
28898
28899
28900
28901
28902
28903
28904
28905
28906
28907
28908
28909
28910
28911
28912
28913
28914
28915
28916
28917
28918
28919
28920
28921
28922
28923
28924
28925
28926
28927
28928
28929
28930
28931
28932
28933
28934
28935
28936
28937
28938
28939
28940
28941
28942
28943
28944
28945
28946
28947
28948
28949
28950
28951
28952
28953
28954
28955
28956
28957
28958
28959
28960
28961
28962
28963
28964
28965
28966
28967
28968
28969
28970
28971
28972
28973
28974
28975
28976
28977
28978
28979
28980
28981
28982
28983
28984
28985
28986
28987
28988
28989
28990
28991
28992
28993
28994
28995
28996
28997
28998
28999
29000
29001
29002
29003
29004
29005
29006
29007
29008
29009
29010
29011
29012
29013
29014
29015
29016
29017
29018
29019
29020
29021
29022
29023
29024
29025
29026
29027
29028
29029
29030
29031
29032
29033
29034
29035
29036
29037
29038
29039
29040
29041
29042
29043
29044
29045
29046
29047
29048
29049
29050
29051
29052
29053
29054
29055
29056
29057
29058
29059
29060
29061
29062
29063
29064
29065
29066
29067
29068
29069
29070
29071
29072
29073
29074
29075
29076
29077
29078
29079
29080
29081
29082
29083
29084
29085
29086
29087
29088
29089
29090
29091
29092
29093
29094
29095
29096
29097
29098
29099
29100
29101
29102
29103
29104
29105
29106
29107
29108
29109
29110
29111
29112
29113
29114
29115
29116
29117
29118
29119
29120
29121
29122
29123
29124
29125
29126
29127
29128
29129
29130
29131
29132
29133
29134
29135
29136
29137
29138
29139
29140
29141
29142
29143
29144
29145
29146
29147
29148
29149
29150
29151
29152
29153
29154
29155
29156
29157
29158
29159
29160
29161
29162
29163
29164
29165
29166
29167
29168
29169
29170
29171
29172
29173
29174
29175
29176
29177
29178
29179
29180
29181
29182
29183
29184
29185
29186
29187
29188
29189
29190
29191
29192
29193
29194
29195
29196
29197
29198
29199
29200
29201
29202
29203
29204
29205
29206
29207
29208
29209
29210
29211
29212
29213
29214
29215
29216
29217
29218
29219
29220
29221
29222
29223
29224
29225
29226
29227
29228
29229
29230
29231
29232
29233
29234
29235
29236
29237
29238
29239
29240
29241
29242
29243
29244
29245
29246
29247
29248
29249
29250
29251
29252
29253
29254
29255
29256
29257
29258
29259
29260
29261
29262
29263
29264
29265
29266
29267
29268
29269
29270
29271
29272
29273
29274
29275
29276
29277
29278
29279
29280
29281
29282
29283
29284
29285
29286
29287
29288
29289
29290
29291
29292
29293
29294
29295
29296
29297
29298
29299
29300
29301
29302
29303
29304
29305
29306
29307
29308
29309
29310
29311
29312
29313
29314
29315
29316
29317
29318
29319
29320
29321
29322
29323
29324
29325
29326
29327
29328
29329
29330
29331
29332
29333
29334
29335
29336
29337
29338
29339
29340
29341
29342
29343
29344
29345
29346
29347
29348
29349
29350
29351
29352
29353
29354
29355
29356
29357
29358
29359
29360
29361
29362
29363
29364
29365
29366
29367
29368
29369
29370
29371
29372
29373
29374
29375
29376
29377
29378
29379
29380
29381
29382
29383
29384
29385
29386
29387
29388
29389
29390
29391
29392
29393
29394
29395
29396
29397
29398
29399
29400
29401
29402
29403
29404
29405
29406
29407
29408
29409
29410
29411
29412
29413
29414
29415
29416
29417
29418
29419
29420
29421
29422
29423
29424
29425
29426
29427
29428
29429
29430
29431
29432
29433
29434
29435
29436
29437
29438
29439
29440
29441
29442
29443
29444
29445
29446
29447
29448
29449
29450
29451
29452
29453
29454
29455
29456
29457
29458
29459
29460
29461
29462
29463
29464
29465
29466
29467
29468
29469
29470
29471
29472
29473
29474
29475
29476
29477
29478
29479
29480
29481
29482
29483
29484
29485
29486
29487
29488
29489
29490
29491
29492
29493
29494
29495
29496
29497
29498
29499
29500
29501
29502
29503
29504
29505
29506
29507
29508
29509
29510
29511
29512
29513
29514
29515
29516
29517
29518
29519
29520
29521
29522
29523
29524
29525
29526
29527
29528
29529
29530
29531
29532
29533
29534
29535
29536
29537
29538
29539
29540
29541
29542
29543
29544
29545
29546
29547
29548
29549
29550
29551
29552
29553
29554
29555
29556
29557
29558
29559
29560
29561
29562
29563
29564
29565
29566
29567
29568
29569
29570
29571
29572
29573
29574
29575
29576
29577
29578
29579
29580
29581
29582
29583
29584
29585
29586
29587
29588
29589
29590
29591
29592
29593
29594
29595
29596
29597
29598
29599
29600
29601
29602
29603
29604
29605
29606
29607
29608
29609
29610
29611
29612
29613
29614
29615
29616
29617
29618
29619
29620
29621
29622
29623
29624
29625
29626
29627
29628
29629
29630
29631
29632
29633
29634
29635
29636
29637
29638
29639
29640
29641
29642
29643
29644
29645
29646
29647
29648
29649
29650
29651
29652
29653
29654
29655
29656
29657
29658
29659
29660
29661
29662
29663
29664
29665
29666
29667
29668
29669
29670
29671
29672
29673
29674
29675
29676
29677
29678
29679
29680
29681
29682
29683
29684
29685
29686
29687
29688
29689
29690
29691
29692
29693
29694
29695
29696
29697
29698
29699
29700
29701
29702
29703
29704
29705
29706
29707
29708
29709
29710
29711
29712
29713
29714
29715
29716
29717
29718
29719
29720
29721
29722
29723
29724
29725
29726
29727
29728
29729
29730
29731
29732
29733
29734
29735
29736
29737
29738
29739
29740
29741
29742
29743
29744
29745
29746
29747
29748
29749
29750
29751
29752
29753
29754
29755
29756
29757
29758
29759
29760
29761
29762
29763
29764
29765
29766
29767
29768
29769
29770
29771
29772
29773
29774
29775
29776
29777
29778
29779
29780
29781
29782
29783
29784
29785
29786
29787
29788
29789
29790
29791
29792
29793
29794
29795
29796
29797
29798
29799
29800
29801
29802
29803
29804
29805
29806
29807
29808
29809
29810
29811
29812
29813
29814
29815
29816
29817
29818
29819
29820
29821
29822
29823
29824
29825
29826
29827
29828
29829
29830
29831
29832
29833
29834
29835
29836
29837
29838
29839
29840
29841
29842
29843
29844
29845
29846
29847
29848
29849
29850
29851
29852
29853
29854
29855
29856
29857
29858
29859
29860
29861
29862
29863
29864
29865
29866
29867
29868
29869
29870
29871
29872
29873
29874
29875
29876
29877
29878
29879
29880
29881
29882
29883
29884
29885
29886
29887
29888
29889
29890
29891
29892
29893
29894
29895
29896
29897
29898
29899
29900
29901
29902
29903
29904
29905
29906
29907
29908
29909
29910
29911
29912
29913
29914
29915
29916
29917
29918
29919
29920
29921
29922
29923
29924
29925
29926
29927
29928
29929
29930
29931
29932
29933
29934
29935
29936
29937
29938
29939
29940
29941
29942
29943
29944
29945
29946
29947
29948
29949
29950
29951
29952
29953
29954
29955
29956
29957
29958
29959
29960
29961
29962
29963
29964
29965
29966
29967
29968
29969
29970
29971
29972
29973
29974
29975
29976
29977
29978
29979
29980
29981
29982
29983
29984
29985
29986
29987
29988
29989
29990
29991
29992
29993
29994
29995
29996
29997
29998
29999
30000
30001
30002
30003
30004
30005
30006
30007
30008
30009
30010
30011
30012
30013
30014
30015
30016
30017
30018
30019
30020
30021
30022
30023
30024
30025
30026
30027
30028
30029
30030
30031
30032
30033
30034
30035
30036
30037
30038
30039
30040
30041
30042
30043
30044
30045
30046
30047
30048
30049
30050
30051
30052
30053
30054
30055
30056
30057
30058
30059
30060
30061
30062
30063
30064
30065
30066
30067
30068
30069
30070
30071
30072
30073
30074
30075
30076
30077
30078
30079
30080
30081
30082
30083
30084
30085
30086
30087
30088
30089
30090
30091
30092
30093
30094
30095
30096
30097
30098
30099
30100
30101
30102
30103
30104
30105
30106
30107
30108
30109
30110
30111
30112
30113
30114
30115
30116
30117
30118
30119
30120
30121
30122
30123
30124
30125
30126
30127
30128
30129
30130
30131
30132
30133
30134
30135
30136
30137
30138
30139
30140
30141
30142
30143
30144
30145
30146
30147
30148
30149
30150
30151
30152
30153
30154
30155
30156
30157
30158
30159
30160
30161
30162
30163
30164
30165
30166
30167
30168
30169
30170
30171
30172
30173
30174
30175
30176
30177
30178
30179
30180
30181
30182
30183
30184
30185
30186
30187
30188
30189
30190
30191
30192
30193
30194
30195
30196
30197
30198
30199
30200
30201
30202
30203
30204
30205
30206
30207
30208
30209
30210
30211
30212
30213
30214
30215
30216
30217
30218
30219
30220
30221
30222
30223
30224
30225
30226
30227
30228
30229
30230
30231
30232
30233
30234
30235
30236
30237
30238
30239
30240
30241
30242
30243
30244
30245
30246
30247
30248
30249
30250
30251
30252
30253
30254
30255
30256
30257
30258
30259
30260
30261
30262
30263
30264
30265
30266
30267
30268
30269
30270
30271
30272
30273
30274
30275
30276
30277
30278
30279
30280
30281
30282
30283
30284
30285
30286
30287
30288
30289
30290
30291
30292
30293
30294
30295
30296
30297
30298
30299
30300
30301
30302
30303
30304
30305
30306
30307
30308
30309
30310
30311
30312
30313
30314
30315
30316
30317
30318
30319
30320
30321
30322
30323
30324
30325
30326
30327
30328
30329
30330
30331
30332
30333
30334
30335
30336
30337
30338
30339
30340
30341
30342
30343
30344
30345
30346
30347
30348
30349
30350
30351
30352
30353
30354
30355
30356
30357
30358
30359
30360
30361
30362
30363
30364
30365
30366
30367
30368
30369
30370
30371
30372
30373
30374
30375
30376
30377
30378
30379
30380
30381
30382
30383
30384
30385
30386
30387
30388
30389
30390
30391
30392
30393
30394
30395
30396
30397
30398
30399
30400
30401
30402
30403
30404
30405
30406
30407
30408
30409
30410
30411
30412
30413
30414
30415
30416
30417
30418
30419
30420
30421
30422
30423
30424
30425
30426
30427
30428
30429
30430
30431
30432
30433
30434
30435
30436
30437
30438
30439
30440
30441
30442
30443
30444
30445
30446
30447
30448
30449
30450
30451
30452
30453
30454
30455
30456
30457
30458
30459
30460
30461
30462
30463
30464
30465
30466
30467
30468
30469
30470
30471
30472
30473
30474
30475
30476
30477
30478
30479
30480
30481
30482
30483
30484
30485
30486
30487
30488
30489
30490
30491
30492
30493
30494
30495
30496
30497
30498
30499
30500
30501
30502
30503
30504
30505
30506
30507
30508
30509
30510
30511
30512
30513
30514
30515
30516
30517
30518
30519
30520
30521
30522
30523
30524
30525
30526
30527
30528
30529
30530
30531
30532
30533
30534
30535
30536
30537
30538
30539
30540
30541
30542
30543
30544
30545
30546
30547
30548
30549
30550
30551
30552
30553
30554
30555
30556
30557
30558
30559
30560
30561
30562
30563
30564
30565
30566
30567
30568
30569
30570
30571
30572
30573
30574
30575
30576
30577
30578
30579
30580
30581
30582
30583
30584
30585
30586
30587
30588
30589
30590
30591
30592
30593
30594
30595
30596
30597
30598
30599
30600
30601
30602
30603
30604
30605
30606
30607
30608
30609
30610
30611
30612
30613
30614
30615
30616
30617
30618
30619
30620
30621
30622
30623
30624
30625
30626
30627
30628
30629
30630
30631
30632
30633
30634
30635
30636
30637
30638
30639
30640
30641
30642
30643
30644
30645
30646
30647
30648
30649
30650
30651
30652
30653
30654
30655
30656
30657
30658
30659
30660
30661
30662
30663
30664
30665
30666
30667
30668
30669
30670
30671
30672
30673
30674
30675
30676
30677
30678
30679
30680
30681
30682
30683
30684
30685
30686
30687
30688
30689
30690
30691
30692
30693
30694
30695
30696
30697
30698
30699
30700
30701
30702
30703
30704
30705
30706
30707
30708
30709
30710
30711
30712
30713
30714
30715
30716
30717
30718
30719
30720
30721
30722
30723
30724
30725
30726
30727
30728
30729
30730
30731
30732
30733
30734
30735
30736
30737
30738
30739
30740
30741
30742
30743
30744
30745
30746
30747
30748
30749
30750
30751
30752
30753
30754
30755
30756
30757
30758
30759
30760
30761
30762
30763
30764
30765
30766
30767
30768
30769
30770
30771
30772
30773
30774
30775
30776
30777
30778
30779
30780
30781
30782
30783
30784
30785
30786
30787
30788
30789
30790
30791
30792
30793
30794
30795
30796
30797
30798
30799
30800
30801
30802
30803
30804
30805
30806
30807
30808
30809
30810
30811
30812
30813
30814
30815
30816
30817
30818
30819
30820
30821
30822
30823
30824
30825
30826
30827
30828
30829
30830
30831
30832
30833
30834
30835
30836
30837
30838
30839
30840
30841
30842
30843
30844
30845
30846
30847
30848
30849
30850
30851
30852
30853
30854
30855
30856
30857
30858
30859
30860
30861
30862
30863
30864
30865
30866
30867
30868
30869
30870
30871
30872
30873
30874
30875
30876
30877
30878
30879
30880
30881
30882
30883
30884
30885
30886
30887
30888
30889
30890
30891
30892
30893
30894
30895
30896
30897
30898
30899
30900
30901
30902
30903
30904
30905
30906
30907
30908
30909
30910
30911
30912
30913
30914
30915
30916
30917
30918
30919
30920
30921
30922
30923
30924
30925
30926
30927
30928
30929
30930
30931
30932
30933
30934
30935
30936
30937
30938
30939
30940
30941
30942
30943
30944
30945
30946
30947
30948
30949
30950
30951
30952
30953
30954
30955
30956
30957
30958
30959
30960
30961
30962
30963
30964
30965
30966
30967
30968
30969
30970
30971
30972
30973
30974
30975
30976
30977
30978
30979
30980
30981
30982
30983
30984
30985
30986
30987
30988
30989
30990
30991
30992
30993
30994
30995
30996
30997
30998
30999
31000
31001
31002
31003
31004
31005
31006
31007
31008
31009
31010
31011
31012
31013
31014
31015
31016
31017
31018
31019
31020
31021
31022
31023
31024
31025
31026
31027
31028
31029
31030
31031
31032
31033
31034
31035
31036
31037
31038
31039
31040
31041
31042
31043
31044
31045
31046
31047
31048
31049
31050
31051
31052
31053
31054
31055
31056
31057
31058
31059
31060
31061
31062
31063
31064
31065
31066
31067
31068
31069
31070
31071
31072
31073
31074
31075
31076
31077
31078
31079
31080
31081
31082
31083
31084
31085
31086
31087
31088
31089
31090
31091
31092
31093
31094
31095
31096
31097
31098
31099
31100
31101
31102
31103
31104
31105
31106
31107
31108
31109
31110
31111
31112
31113
31114
31115
31116
31117
31118
31119
31120
31121
31122
31123
31124
31125
31126
31127
31128
31129
31130
31131
31132
31133
31134
31135
31136
31137
31138
31139
31140
31141
31142
31143
31144
31145
31146
31147
31148
31149
31150
31151
31152
31153
31154
31155
31156
31157
31158
31159
31160
31161
31162
31163
31164
31165
31166
31167
31168
31169
31170
31171
31172
31173
31174
31175
31176
31177
31178
31179
31180
31181
31182
31183
31184
31185
31186
31187
31188
31189
31190
31191
31192
31193
31194
31195
31196
31197
31198
31199
31200
31201
31202
31203
31204
31205
31206
31207
31208
31209
31210
31211
31212
31213
31214
31215
31216
31217
31218
31219
31220
31221
31222
31223
31224
31225
31226
31227
31228
31229
31230
31231
31232
31233
31234
31235
31236
31237
31238
31239
31240
31241
31242
31243
31244
31245
31246
31247
31248
31249
31250
31251
31252
31253
31254
31255
31256
31257
31258
31259
31260
31261
31262
31263
31264
31265
31266
31267
31268
31269
31270
31271
31272
31273
31274
31275
31276
31277
31278
31279
31280
31281
31282
31283
31284
31285
31286
31287
31288
31289
31290
31291
31292
31293
31294
31295
31296
31297
31298
31299
31300
31301
31302
31303
31304
31305
31306
31307
31308
31309
31310
31311
31312
31313
31314
31315
31316
31317
31318
31319
31320
31321
31322
31323
31324
31325
31326
31327
31328
31329
31330
31331
31332
31333
31334
31335
31336
31337
31338
31339
31340
31341
31342
31343
31344
31345
31346
31347
31348
31349
31350
31351
31352
31353
31354
31355
31356
31357
31358
31359
31360
31361
31362
31363
31364
31365
31366
31367
31368
31369
31370
31371
31372
31373
31374
31375
31376
31377
31378
31379
31380
31381
31382
31383
31384
31385
31386
31387
31388
31389
31390
31391
31392
31393
31394
31395
31396
31397
31398
31399
31400
31401
31402
31403
31404
31405
31406
31407
31408
31409
31410
31411
31412
31413
31414
31415
31416
31417
31418
31419
31420
31421
31422
31423
31424
31425
31426
31427
31428
31429
31430
31431
31432
31433
31434
31435
31436
31437
31438
31439
31440
31441
31442
31443
31444
31445
31446
31447
31448
31449
31450
31451
31452
31453
31454
31455
31456
31457
31458
31459
31460
31461
31462
31463
31464
31465
31466
31467
31468
31469
31470
31471
31472
31473
31474
31475
31476
31477
31478
31479
31480
31481
31482
31483
31484
31485
31486
31487
31488
31489
31490
31491
31492
31493
31494
31495
31496
31497
31498
31499
31500
31501
31502
31503
31504
31505
31506
31507
31508
31509
31510
31511
31512
31513
31514
31515
31516
31517
31518
31519
31520
31521
31522
31523
31524
31525
31526
31527
31528
31529
31530
31531
31532
31533
31534
31535
31536
31537
31538
31539
31540
31541
31542
31543
31544
31545
31546
31547
31548
31549
31550
31551
31552
31553
31554
31555
31556
31557
31558
31559
31560
31561
31562
31563
31564
31565
31566
31567
31568
31569
31570
31571
31572
31573
31574
31575
31576
31577
31578
31579
31580
31581
31582
31583
31584
31585
31586
31587
31588
31589
31590
31591
31592
31593
31594
31595
31596
31597
31598
31599
31600
31601
31602
31603
31604
31605
31606
31607
31608
31609
31610
31611
31612
31613
31614
31615
31616
31617
31618
31619
31620
31621
31622
31623
31624
31625
31626
31627
31628
31629
31630
31631
31632
31633
31634
31635
31636
31637
31638
31639
31640
31641
31642
31643
31644
31645
31646
31647
31648
31649
31650
31651
31652
31653
31654
31655
31656
31657
31658
31659
31660
31661
31662
31663
31664
31665
31666
31667
31668
31669
31670
31671
31672
31673
31674
31675
31676
31677
31678
31679
31680
31681
31682
31683
31684
31685
31686
31687
31688
31689
31690
31691
31692
31693
31694
31695
31696
31697
31698
31699
31700
31701
31702
31703
31704
31705
31706
31707
31708
31709
31710
31711
31712
31713
31714
31715
31716
31717
31718
31719
31720
31721
31722
31723
31724
31725
31726
31727
31728
31729
31730
31731
31732
31733
31734
31735
31736
31737
31738
31739
31740
31741
31742
31743
31744
31745
31746
31747
31748
31749
31750
31751
31752
31753
31754
31755
31756
31757
31758
31759
31760
31761
31762
31763
31764
31765
31766
31767
31768
31769
31770
31771
31772
31773
31774
31775
31776
31777
31778
31779
31780
31781
31782
31783
31784
31785
31786
31787
31788
31789
31790
31791
31792
31793
31794
31795
31796
31797
31798
31799
31800
31801
31802
31803
31804
31805
31806
31807
31808
31809
31810
31811
31812
31813
31814
31815
31816
31817
31818
31819
31820
31821
31822
31823
31824
31825
31826
31827
31828
31829
31830
31831
31832
31833
31834
31835
31836
31837
31838
31839
31840
31841
31842
31843
31844
31845
31846
31847
31848
31849
31850
31851
31852
31853
31854
31855
31856
31857
31858
31859
31860
31861
31862
31863
31864
31865
31866
31867
31868
31869
31870
31871
31872
31873
31874
31875
31876
31877
31878
31879
31880
31881
31882
31883
31884
31885
31886
31887
31888
31889
31890
31891
31892
31893
31894
31895
31896
31897
31898
31899
31900
31901
31902
31903
31904
31905
31906
31907
31908
31909
31910
31911
31912
31913
31914
31915
31916
31917
31918
31919
31920
31921
31922
31923
31924
31925
31926
31927
31928
31929
31930
31931
31932
31933
31934
31935
31936
31937
31938
31939
31940
31941
31942
31943
31944
31945
31946
31947
31948
31949
31950
31951
31952
31953
31954
31955
31956
31957
31958
31959
31960
31961
31962
31963
31964
31965
31966
31967
31968
31969
31970
31971
31972
31973
31974
31975
31976
31977
31978
31979
31980
31981
31982
31983
31984
31985
31986
31987
31988
31989
31990
31991
31992
31993
31994
31995
31996
31997
31998
31999
32000
32001
32002
32003
32004
32005
32006
32007
32008
32009
32010
32011
32012
32013
32014
32015
32016
32017
32018
32019
32020
32021
32022
32023
32024
32025
32026
32027
32028
32029
32030
32031
32032
32033
32034
32035
32036
32037
32038
32039
32040
32041
32042
32043
32044
32045
32046
32047
32048
32049
32050
32051
32052
32053
32054
32055
32056
32057
32058
32059
32060
32061
32062
32063
32064
32065
32066
32067
32068
32069
32070
32071
32072
32073
32074
32075
32076
32077
32078
32079
32080
32081
32082
32083
32084
32085
32086
32087
32088
32089
32090
32091
32092
32093
32094
32095
32096
32097
32098
32099
32100
32101
32102
32103
32104
32105
32106
32107
32108
32109
32110
32111
32112
32113
32114
32115
32116
32117
32118
32119
32120
32121
32122
32123
32124
32125
32126
32127
32128
32129
32130
32131
32132
32133
32134
32135
32136
32137
32138
32139
32140
32141
32142
32143
32144
32145
32146
32147
32148
32149
32150
32151
32152
32153
32154
32155
32156
32157
32158
32159
32160
32161
32162
32163
32164
32165
32166
32167
32168
32169
32170
32171
32172
32173
32174
32175
32176
32177
32178
32179
32180
32181
32182
32183
32184
32185
32186
32187
32188
32189
32190
32191
32192
32193
32194
32195
32196
32197
32198
32199
32200
32201
32202
32203
32204
32205
32206
32207
32208
32209
32210
32211
32212
32213
32214
32215
32216
32217
32218
32219
32220
32221
32222
32223
32224
32225
32226
32227
32228
32229
32230
32231
32232
32233
32234
32235
32236
32237
32238
32239
32240
32241
32242
32243
32244
32245
32246
32247
32248
32249
32250
32251
32252
32253
32254
32255
32256
32257
32258
32259
32260
32261
32262
32263
32264
32265
32266
32267
32268
32269
32270
32271
32272
32273
32274
32275
32276
32277
32278
32279
32280
32281
32282
32283
32284
32285
32286
32287
32288
32289
32290
32291
32292
32293
32294
32295
32296
32297
32298
32299
32300
32301
32302
32303
32304
32305
32306
32307
32308
32309
32310
32311
32312
32313
32314
32315
32316
32317
32318
32319
32320
32321
32322
32323
32324
32325
32326
32327
32328
32329
32330
32331
32332
32333
32334
32335
32336
32337
32338
32339
32340
32341
32342
32343
32344
32345
32346
32347
32348
32349
32350
32351
32352
32353
32354
32355
32356
32357
32358
32359
32360
32361
32362
32363
32364
32365
32366
32367
32368
32369
32370
32371
32372
32373
32374
32375
32376
32377
32378
32379
32380
32381
32382
32383
32384
32385
32386
32387
32388
32389
32390
32391
32392
32393
32394
32395
32396
32397
32398
32399
32400
32401
32402
32403
32404
32405
32406
32407
32408
32409
32410
32411
32412
32413
32414
32415
32416
32417
32418
32419
32420
32421
32422
32423
32424
32425
32426
32427
32428
32429
32430
32431
32432
32433
32434
32435
32436
32437
32438
32439
32440
32441
32442
32443
32444
32445
32446
32447
32448
32449
32450
32451
32452
32453
32454
32455
32456
32457
32458
32459
32460
32461
32462
32463
32464
32465
32466
32467
32468
32469
32470
32471
32472
32473
32474
32475
32476
32477
32478
32479
32480
32481
32482
32483
32484
32485
32486
32487
32488
32489
32490
32491
32492
32493
32494
32495
32496
32497
32498
32499
32500
32501
32502
32503
32504
32505
32506
32507
32508
32509
32510
32511
32512
32513
32514
32515
32516
32517
32518
32519
32520
32521
32522
32523
32524
32525
32526
32527
32528
32529
32530
32531
32532
32533
32534
32535
32536
32537
32538
32539
32540
32541
32542
32543
32544
32545
32546
32547
32548
32549
32550
32551
32552
32553
32554
32555
32556
32557
32558
32559
32560
32561
32562
32563
32564
32565
32566
32567
32568
32569
32570
32571
32572
32573
32574
32575
32576
32577
32578
32579
32580
32581
32582
32583
32584
32585
32586
32587
32588
32589
32590
32591
32592
32593
32594
32595
32596
32597
32598
32599
32600
32601
32602
32603
32604
32605
32606
32607
32608
32609
32610
32611
32612
32613
32614
32615
32616
32617
32618
32619
32620
32621
32622
32623
32624
32625
32626
32627
32628
32629
32630
32631
32632
32633
32634
32635
32636
32637
32638
32639
32640
32641
32642
32643
32644
32645
32646
32647
32648
32649
32650
32651
32652
32653
32654
32655
32656
32657
32658
32659
32660
32661
32662
32663
32664
32665
32666
32667
32668
32669
32670
32671
32672
32673
32674
32675
32676
32677
32678
32679
32680
32681
32682
32683
32684
32685
32686
32687
32688
32689
32690
32691
32692
32693
32694
32695
32696
32697
32698
32699
32700
32701
32702
32703
32704
32705
32706
32707
32708
32709
32710
32711
32712
32713
32714
32715
32716
32717
32718
32719
32720
32721
32722
32723
32724
32725
32726
32727
32728
32729
32730
32731
32732
32733
32734
32735
32736
32737
32738
32739
32740
32741
32742
32743
32744
32745
32746
32747
32748
32749
32750
32751
32752
32753
32754
32755
32756
32757
32758
32759
32760
32761
32762
32763
32764
32765
32766
32767
32768
32769
32770
32771
32772
32773
32774
32775
32776
32777
32778
32779
32780
32781
32782
32783
32784
32785
32786
32787
32788
32789
32790
32791
32792
32793
32794
32795
32796
32797
32798
32799
32800
32801
32802
32803
32804
32805
32806
32807
32808
32809
32810
32811
32812
32813
32814
32815
32816
32817
32818
32819
32820
32821
32822
32823
32824
32825
32826
32827
32828
32829
32830
32831
32832
32833
32834
32835
32836
32837
32838
32839
32840
32841
32842
32843
32844
32845
32846
32847
32848
32849
32850
32851
32852
32853
32854
32855
32856
32857
32858
32859
32860
32861
32862
32863
32864
32865
32866
32867
32868
32869
32870
32871
32872
32873
32874
32875
32876
32877
32878
32879
32880
32881
32882
32883
32884
32885
32886
32887
32888
32889
32890
32891
32892
32893
32894
32895
32896
32897
32898
32899
32900
32901
32902
32903
32904
32905
32906
32907
32908
32909
32910
32911
32912
32913
32914
32915
32916
32917
32918
32919
32920
32921
32922
32923
32924
32925
32926
32927
32928
32929
32930
32931
32932
32933
32934
32935
32936
32937
32938
32939
32940
32941
32942
32943
32944
32945
32946
32947
32948
32949
32950
32951
32952
32953
32954
32955
32956
32957
32958
32959
32960
32961
32962
32963
32964
32965
32966
32967
32968
32969
32970
32971
32972
32973
32974
32975
32976
32977
32978
32979
32980
32981
32982
32983
32984
32985
32986
32987
32988
32989
32990
32991
32992
32993
32994
32995
32996
32997
32998
32999
33000
33001
33002
33003
33004
33005
33006
33007
33008
33009
33010
33011
33012
33013
33014
33015
33016
33017
33018
33019
33020
33021
33022
33023
33024
33025
33026
33027
33028
33029
33030
33031
33032
33033
33034
33035
33036
33037
33038
33039
33040
33041
33042
33043
33044
33045
33046
33047
33048
33049
33050
33051
33052
33053
33054
33055
33056
33057
33058
33059
33060
33061
33062
33063
33064
33065
33066
33067
33068
33069
33070
33071
33072
33073
33074
33075
33076
33077
33078
33079
33080
33081
33082
33083
33084
33085
33086
33087
33088
33089
33090
33091
33092
33093
33094
33095
33096
33097
33098
33099
33100
33101
33102
33103
33104
33105
33106
33107
33108
33109
33110
33111
33112
33113
33114
33115
33116
33117
33118
33119
33120
33121
33122
33123
33124
33125
33126
33127
33128
33129
33130
33131
33132
33133
33134
33135
33136
33137
33138
33139
33140
33141
33142
33143
33144
33145
33146
33147
33148
33149
33150
33151
33152
33153
33154
33155
33156
33157
33158
33159
33160
33161
33162
33163
33164
33165
33166
33167
33168
33169
33170
33171
33172
33173
33174
33175
33176
33177
33178
33179
33180
33181
33182
33183
33184
33185
33186
33187
33188
33189
33190
33191
33192
33193
33194
33195
33196
33197
33198
33199
33200
33201
33202
33203
33204
33205
33206
33207
33208
33209
33210
33211
33212
33213
33214
33215
33216
33217
33218
33219
33220
33221
33222
33223
33224
33225
33226
33227
33228
33229
33230
33231
33232
33233
33234
33235
33236
33237
33238
33239
33240
33241
33242
33243
33244
33245
33246
33247
33248
33249
33250
33251
33252
33253
33254
33255
33256
33257
33258
33259
33260
33261
33262
33263
33264
33265
33266
33267
33268
33269
33270
33271
33272
33273
33274
33275
33276
33277
33278
33279
33280
33281
33282
33283
33284
33285
33286
33287
33288
33289
33290
33291
33292
33293
33294
33295
33296
33297
33298
33299
33300
33301
33302
33303
33304
33305
33306
33307
33308
33309
33310
33311
33312
33313
33314
33315
33316
33317
33318
33319
33320
33321
33322
33323
33324
33325
33326
33327
33328
33329
33330
33331
33332
33333
33334
33335
33336
33337
33338
33339
33340
33341
33342
33343
33344
33345
33346
33347
33348
33349
33350
33351
33352
33353
33354
33355
33356
33357
33358
33359
33360
33361
33362
33363
33364
33365
33366
33367
33368
33369
33370
33371
33372
33373
33374
33375
33376
33377
33378
33379
33380
33381
33382
33383
33384
33385
33386
33387
33388
33389
33390
33391
33392
33393
33394
33395
33396
33397
33398
33399
33400
33401
33402
33403
33404
33405
33406
33407
33408
33409
33410
33411
33412
33413
33414
33415
33416
33417
33418
33419
33420
33421
33422
33423
33424
33425
33426
33427
33428
33429
33430
33431
33432
33433
33434
33435
33436
33437
33438
33439
33440
33441
33442
33443
33444
33445
33446
33447
33448
33449
33450
33451
33452
33453
33454
33455
33456
33457
33458
33459
33460
33461
33462
33463
33464
33465
33466
33467
33468
33469
33470
33471
33472
33473
33474
33475
33476
33477
33478
33479
33480
33481
33482
33483
33484
33485
33486
33487
33488
33489
33490
33491
33492
33493
33494
33495
33496
33497
33498
33499
33500
33501
33502
33503
33504
33505
33506
33507
33508
33509
33510
33511
33512
33513
33514
33515
33516
33517
33518
33519
33520
33521
33522
33523
33524
33525
33526
33527
33528
33529
33530
33531
33532
33533
33534
33535
33536
33537
33538
33539
33540
33541
33542
33543
33544
33545
33546
33547
33548
33549
33550
33551
33552
33553
33554
33555
33556
33557
33558
33559
33560
33561
33562
33563
33564
33565
33566
33567
33568
33569
33570
33571
33572
33573
33574
33575
33576
33577
33578
33579
33580
33581
33582
33583
33584
33585
33586
33587
33588
33589
33590
33591
33592
33593
33594
33595
33596
33597
33598
33599
33600
33601
33602
33603
33604
33605
33606
33607
33608
33609
33610
33611
33612
33613
33614
33615
33616
33617
33618
33619
33620
33621
33622
33623
33624
33625
33626
33627
33628
33629
33630
33631
33632
33633
33634
33635
33636
33637
33638
33639
33640
33641
33642
33643
33644
33645
33646
33647
33648
33649
33650
33651
33652
33653
33654
33655
33656
33657
33658
33659
33660
33661
33662
33663
33664
33665
33666
33667
33668
33669
33670
33671
33672
33673
33674
33675
33676
33677
33678
33679
33680
33681
33682
33683
33684
33685
33686
33687
33688
33689
33690
33691
33692
33693
33694
33695
33696
33697
33698
33699
33700
33701
33702
33703
33704
33705
33706
33707
33708
33709
33710
33711
33712
33713
33714
33715
33716
33717
33718
33719
33720
33721
33722
33723
33724
33725
33726
33727
33728
33729
33730
33731
33732
33733
33734
33735
33736
33737
33738
33739
33740
33741
33742
33743
33744
33745
33746
33747
33748
33749
33750
33751
33752
33753
33754
33755
33756
33757
33758
33759
33760
33761
33762
33763
33764
33765
33766
33767
33768
33769
33770
33771
33772
33773
33774
33775
33776
33777
33778
33779
33780
33781
33782
33783
33784
33785
33786
33787
33788
33789
33790
33791
33792
33793
33794
33795
33796
33797
33798
33799
33800
33801
33802
33803
33804
33805
33806
33807
33808
33809
33810
33811
33812
33813
33814
33815
33816
33817
33818
33819
33820
33821
33822
33823
33824
33825
33826
33827
33828
33829
33830
33831
33832
33833
33834
33835
33836
33837
33838
33839
33840
33841
33842
33843
33844
33845
33846
33847
33848
33849
33850
33851
33852
33853
33854
33855
33856
33857
33858
33859
33860
33861
33862
33863
33864
33865
33866
33867
33868
33869
33870
33871
33872
33873
33874
33875
33876
33877
33878
33879
33880
33881
33882
33883
33884
33885
33886
33887
33888
33889
33890
33891
33892
33893
33894
33895
33896
33897
33898
33899
33900
33901
33902
33903
33904
33905
33906
33907
33908
33909
33910
33911
33912
33913
33914
33915
33916
33917
33918
33919
33920
33921
33922
33923
33924
33925
33926
33927
33928
33929
33930
33931
33932
33933
33934
33935
33936
33937
33938
33939
33940
33941
33942
33943
33944
33945
33946
33947
33948
33949
33950
33951
33952
33953
33954
33955
33956
33957
33958
33959
33960
33961
33962
33963
33964
33965
33966
33967
33968
33969
33970
33971
33972
33973
33974
33975
33976
33977
33978
33979
33980
33981
33982
33983
33984
33985
33986
33987
33988
33989
33990
33991
33992
33993
33994
33995
33996
33997
33998
33999
34000
34001
34002
34003
34004
34005
34006
34007
34008
34009
34010
34011
34012
34013
34014
34015
34016
34017
34018
34019
34020
34021
34022
34023
34024
34025
34026
34027
34028
34029
34030
34031
34032
34033
34034
34035
34036
34037
34038
34039
34040
34041
34042
34043
34044
34045
34046
34047
34048
34049
34050
34051
34052
34053
34054
34055
34056
34057
34058
34059
34060
34061
34062
34063
34064
34065
34066
34067
34068
34069
34070
34071
34072
34073
34074
34075
34076
34077
34078
34079
34080
34081
34082
34083
34084
34085
34086
34087
34088
34089
34090
34091
34092
34093
34094
34095
34096
34097
34098
34099
34100
34101
34102
34103
34104
34105
34106
34107
34108
34109
34110
34111
34112
34113
34114
34115
34116
34117
34118
34119
34120
34121
34122
34123
34124
34125
34126
34127
34128
34129
34130
34131
34132
34133
34134
34135
34136
34137
34138
34139
34140
34141
34142
34143
34144
34145
34146
34147
34148
34149
34150
34151
34152
34153
34154
34155
34156
34157
34158
34159
34160
34161
34162
34163
34164
34165
34166
34167
34168
34169
34170
34171
34172
34173
34174
34175
34176
34177
34178
34179
34180
34181
34182
34183
34184
34185
34186
34187
34188
34189
34190
34191
34192
34193
34194
34195
34196
34197
34198
34199
34200
34201
34202
34203
34204
34205
34206
34207
34208
34209
34210
34211
34212
34213
34214
34215
34216
34217
34218
34219
34220
34221
34222
34223
34224
34225
34226
34227
34228
34229
34230
34231
34232
34233
34234
34235
34236
34237
34238
34239
34240
34241
34242
34243
34244
34245
34246
34247
34248
34249
34250
34251
34252
34253
34254
34255
34256
34257
34258
34259
34260
34261
34262
34263
34264
34265
34266
34267
34268
34269
34270
34271
34272
34273
34274
34275
34276
34277
34278
34279
34280
34281
34282
34283
34284
34285
34286
34287
34288
34289
34290
34291
34292
34293
34294
34295
34296
34297
34298
34299
34300
34301
34302
34303
34304
34305
34306
34307
34308
34309
34310
34311
34312
34313
34314
34315
34316
34317
34318
34319
34320
34321
34322
34323
34324
34325
34326
34327
34328
34329
34330
34331
34332
34333
34334
34335
34336
34337
34338
34339
34340
34341
34342
34343
34344
34345
34346
34347
34348
34349
34350
34351
34352
34353
34354
34355
34356
34357
34358
34359
34360
34361
34362
34363
34364
34365
34366
34367
34368
34369
34370
34371
34372
34373
34374
34375
34376
34377
34378
34379
34380
34381
34382
34383
34384
34385
34386
34387
34388
34389
34390
34391
34392
34393
34394
34395
34396
34397
34398
34399
34400
34401
34402
34403
34404
34405
34406
34407
34408
34409
34410
34411
34412
34413
34414
34415
34416
34417
34418
34419
34420
34421
34422
34423
34424
34425
34426
34427
34428
34429
34430
34431
34432
34433
34434
34435
34436
34437
34438
34439
34440
34441
34442
34443
34444
34445
34446
34447
34448
34449
34450
34451
34452
34453
34454
34455
34456
34457
34458
34459
34460
34461
34462
34463
34464
34465
34466
34467
34468
34469
34470
34471
34472
34473
34474
34475
34476
34477
34478
34479
34480
34481
34482
34483
34484
34485
34486
34487
34488
34489
34490
34491
34492
34493
34494
34495
34496
34497
34498
34499
34500
34501
34502
34503
34504
34505
34506
34507
34508
34509
34510
34511
34512
34513
34514
34515
34516
34517
34518
34519
34520
34521
34522
34523
34524
34525
34526
34527
34528
34529
34530
34531
34532
34533
34534
34535
34536
34537
34538
34539
34540
34541
34542
34543
34544
34545
34546
34547
34548
34549
34550
34551
34552
34553
34554
34555
34556
34557
34558
34559
34560
34561
34562
34563
34564
34565
34566
34567
34568
34569
34570
34571
34572
34573
34574
34575
34576
34577
34578
34579
34580
34581
34582
34583
34584
34585
34586
34587
34588
34589
34590
34591
34592
34593
34594
34595
34596
34597
34598
34599
34600
34601
34602
34603
34604
34605
34606
34607
34608
34609
34610
34611
34612
34613
34614
34615
34616
34617
34618
34619
34620
34621
34622
34623
34624
34625
34626
34627
34628
34629
34630
34631
34632
34633
34634
34635
34636
34637
34638
34639
34640
34641
34642
34643
34644
34645
34646
34647
34648
34649
34650
34651
34652
34653
34654
34655
34656
34657
34658
34659
34660
34661
34662
34663
34664
34665
34666
34667
34668
34669
34670
34671
34672
34673
34674
34675
34676
34677
34678
34679
34680
34681
34682
34683
34684
34685
34686
34687
34688
34689
34690
34691
34692
34693
34694
34695
34696
34697
34698
34699
34700
34701
34702
34703
34704
34705
34706
34707
34708
34709
34710
34711
34712
34713
34714
34715
34716
34717
34718
34719
34720
34721
34722
34723
34724
34725
34726
34727
34728
34729
34730
34731
34732
34733
34734
34735
34736
34737
34738
34739
34740
34741
34742
34743
34744
34745
34746
34747
34748
34749
34750
34751
34752
34753
34754
34755
34756
34757
34758
34759
34760
34761
34762
34763
34764
34765
34766
34767
34768
34769
34770
34771
34772
34773
34774
34775
34776
34777
34778
34779
34780
34781
34782
34783
34784
34785
34786
34787
34788
34789
34790
34791
34792
34793
34794
34795
34796
34797
34798
34799
34800
34801
34802
34803
34804
34805
34806
34807
34808
34809
34810
34811
34812
34813
34814
34815
34816
34817
34818
34819
34820
34821
34822
34823
34824
34825
34826
34827
34828
34829
34830
34831
34832
34833
34834
34835
34836
34837
34838
34839
34840
34841
34842
34843
34844
34845
34846
34847
34848
34849
34850
34851
34852
34853
34854
34855
34856
34857
34858
34859
34860
34861
34862
34863
34864
34865
34866
34867
34868
34869
34870
34871
34872
34873
34874
34875
34876
34877
34878
34879
34880
34881
34882
34883
34884
34885
34886
34887
34888
34889
34890
34891
34892
34893
34894
34895
34896
34897
34898
34899
34900
34901
34902
34903
34904
34905
34906
34907
34908
34909
34910
34911
34912
34913
34914
34915
34916
34917
34918
34919
34920
34921
34922
34923
34924
34925
34926
34927
34928
34929
34930
34931
34932
34933
34934
34935
34936
34937
34938
34939
34940
34941
34942
34943
34944
34945
34946
34947
34948
34949
34950
34951
34952
34953
34954
34955
34956
34957
34958
34959
34960
34961
34962
34963
34964
34965
34966
34967
34968
34969
34970
34971
34972
34973
34974
34975
34976
34977
34978
34979
34980
34981
34982
34983
34984
34985
34986
34987
34988
34989
34990
34991
34992
34993
34994
34995
34996
34997
34998
34999
35000
35001
35002
35003
35004
35005
35006
35007
35008
35009
35010
35011
35012
35013
35014
35015
35016
35017
35018
35019
35020
35021
35022
35023
35024
35025
35026
35027
35028
35029
35030
35031
35032
35033
35034
35035
35036
35037
35038
35039
35040
35041
35042
35043
35044
35045
35046
35047
35048
35049
35050
35051
35052
35053
35054
35055
35056
35057
35058
35059
35060
35061
35062
35063
35064
35065
35066
35067
35068
35069
35070
35071
35072
35073
35074
35075
35076
35077
35078
35079
35080
35081
35082
35083
35084
35085
35086
35087
35088
35089
35090
35091
35092
35093
35094
35095
35096
35097
35098
35099
35100
35101
35102
35103
35104
35105
35106
35107
35108
35109
35110
35111
35112
35113
35114
35115
35116
35117
35118
35119
35120
35121
35122
35123
35124
35125
35126
35127
35128
35129
35130
35131
35132
35133
35134
35135
35136
35137
35138
35139
35140
35141
35142
35143
35144
35145
35146
35147
35148
35149
35150
35151
35152
35153
35154
35155
35156
35157
35158
35159
35160
35161
35162
35163
35164
35165
35166
35167
35168
35169
35170
35171
35172
35173
35174
35175
35176
35177
35178
35179
35180
35181
35182
35183
35184
35185
35186
35187
35188
35189
35190
35191
35192
35193
35194
35195
35196
35197
35198
35199
35200
35201
35202
35203
35204
35205
35206
35207
35208
35209
35210
35211
35212
35213
35214
35215
35216
35217
35218
35219
35220
35221
35222
35223
35224
35225
35226
35227
35228
35229
35230
35231
35232
35233
35234
35235
35236
35237
35238
35239
35240
35241
35242
35243
35244
35245
35246
35247
35248
35249
35250
35251
35252
35253
35254
35255
35256
35257
35258
35259
35260
35261
35262
35263
35264
35265
35266
35267
35268
35269
35270
35271
35272
35273
35274
35275
35276
35277
35278
35279
35280
35281
35282
35283
35284
35285
35286
35287
35288
35289
35290
35291
35292
35293
35294
35295
35296
35297
35298
35299
35300
35301
35302
35303
35304
35305
35306
35307
35308
35309
35310
35311
35312
35313
35314
35315
35316
35317
35318
35319
35320
35321
35322
35323
35324
35325
35326
35327
35328
35329
35330
35331
35332
35333
35334
35335
35336
35337
35338
35339
35340
35341
35342
35343
35344
35345
35346
35347
35348
35349
35350
35351
35352
35353
35354
35355
35356
35357
35358
35359
35360
35361
35362
35363
35364
35365
35366
35367
35368
35369
35370
35371
35372
35373
35374
35375
35376
35377
35378
35379
35380
35381
35382
35383
35384
35385
35386
35387
35388
35389
35390
35391
35392
35393
35394
35395
35396
35397
35398
35399
35400
35401
35402
35403
35404
35405
35406
35407
35408
35409
35410
35411
35412
35413
35414
35415
35416
35417
35418
35419
35420
35421
35422
35423
35424
35425
35426
35427
35428
35429
35430
35431
35432
35433
35434
35435
35436
35437
35438
35439
35440
35441
35442
35443
35444
35445
35446
35447
35448
35449
35450
35451
35452
35453
35454
35455
35456
35457
35458
35459
35460
35461
35462
35463
35464
35465
35466
35467
35468
35469
35470
35471
35472
35473
35474
35475
35476
35477
35478
35479
35480
35481
35482
35483
35484
35485
35486
35487
35488
35489
35490
35491
35492
35493
35494
35495
35496
35497
35498
35499
35500
35501
35502
35503
35504
35505
35506
35507
35508
35509
35510
35511
35512
35513
35514
35515
35516
35517
35518
35519
35520
35521
35522
35523
35524
35525
35526
35527
35528
35529
35530
35531
35532
35533
35534
35535
35536
35537
35538
35539
35540
35541
35542
35543
35544
35545
35546
35547
35548
35549
35550
35551
35552
35553
35554
35555
35556
35557
35558
35559
35560
35561
35562
35563
35564
35565
35566
35567
35568
35569
35570
35571
35572
35573
35574
35575
35576
35577
35578
35579
35580
35581
35582
35583
35584
35585
35586
35587
35588
35589
35590
35591
35592
35593
35594
35595
35596
35597
35598
35599
35600
35601
35602
35603
35604
35605
35606
35607
35608
35609
35610
35611
35612
35613
35614
35615
35616
35617
35618
35619
35620
35621
35622
35623
35624
35625
35626
35627
35628
35629
35630
35631
35632
35633
35634
35635
35636
35637
35638
35639
35640
35641
35642
35643
35644
35645
35646
35647
35648
35649
35650
35651
35652
35653
35654
35655
35656
35657
35658
35659
35660
35661
35662
35663
35664
35665
35666
35667
35668
35669
35670
35671
35672
35673
35674
35675
35676
35677
35678
35679
35680
35681
35682
35683
35684
35685
35686
35687
35688
35689
35690
35691
35692
35693
35694
35695
35696
35697
35698
35699
35700
35701
35702
35703
35704
35705
35706
35707
35708
35709
35710
35711
35712
35713
35714
35715
35716
35717
35718
35719
35720
35721
35722
35723
35724
35725
35726
35727
35728
35729
35730
35731
35732
35733
35734
35735
35736
35737
35738
35739
35740
35741
35742
35743
35744
35745
35746
35747
35748
35749
35750
35751
35752
35753
35754
35755
35756
35757
35758
35759
35760
35761
35762
35763
35764
35765
35766
35767
35768
35769
35770
35771
35772
35773
35774
35775
35776
35777
35778
35779
35780
35781
35782
35783
35784
35785
35786
35787
35788
35789
35790
35791
35792
35793
35794
35795
35796
35797
35798
35799
35800
35801
35802
35803
35804
35805
35806
35807
35808
35809
35810
35811
35812
35813
35814
35815
35816
35817
35818
35819
35820
35821
35822
35823
35824
35825
35826
35827
35828
35829
35830
35831
35832
35833
35834
35835
35836
35837
35838
35839
35840
35841
35842
35843
35844
35845
35846
35847
35848
35849
35850
35851
35852
35853
35854
35855
35856
35857
35858
35859
35860
35861
35862
35863
35864
35865
35866
35867
35868
35869
35870
35871
35872
35873
35874
35875
35876
35877
35878
35879
35880
35881
35882
35883
35884
35885
35886
35887
35888
35889
35890
35891
35892
35893
35894
35895
35896
35897
35898
35899
35900
35901
35902
35903
35904
35905
35906
35907
35908
35909
35910
35911
35912
35913
35914
35915
35916
35917
35918
35919
35920
35921
35922
35923
35924
35925
35926
35927
35928
35929
35930
35931
35932
35933
35934
35935
35936
35937
35938
35939
35940
35941
35942
35943
35944
35945
35946
35947
35948
35949
35950
35951
35952
35953
35954
35955
35956
35957
35958
35959
35960
35961
35962
35963
35964
35965
35966
35967
35968
35969
35970
35971
35972
35973
35974
35975
35976
35977
35978
35979
35980
35981
35982
35983
35984
35985
35986
35987
35988
35989
35990
35991
35992
35993
35994
35995
35996
35997
35998
35999
36000
36001
36002
36003
36004
36005
36006
36007
36008
36009
36010
36011
36012
36013
36014
36015
36016
36017
36018
36019
36020
36021
36022
36023
36024
36025
36026
36027
36028
36029
36030
36031
36032
36033
36034
36035
36036
36037
36038
36039
36040
36041
36042
36043
36044
36045
36046
36047
36048
36049
36050
36051
36052
36053
36054
36055
36056
36057
36058
36059
36060
36061
36062
36063
36064
36065
36066
36067
36068
36069
36070
36071
36072
36073
36074
36075
36076
36077
36078
36079
36080
36081
36082
36083
36084
36085
36086
36087
36088
36089
36090
36091
36092
36093
36094
36095
36096
36097
36098
36099
36100
36101
36102
36103
36104
36105
36106
36107
36108
36109
36110
36111
36112
36113
36114
36115
36116
36117
36118
36119
36120
36121
36122
36123
36124
36125
36126
36127
36128
36129
36130
36131
36132
36133
36134
36135
36136
36137
36138
36139
36140
36141
36142
36143
36144
36145
36146
36147
36148
36149
36150
36151
36152
36153
36154
36155
36156
36157
36158
36159
36160
36161
36162
36163
36164
36165
36166
36167
36168
36169
36170
36171
36172
36173
36174
36175
36176
36177
36178
36179
36180
36181
36182
36183
36184
36185
36186
36187
36188
36189
36190
36191
36192
36193
36194
36195
36196
36197
36198
36199
36200
36201
36202
36203
36204
36205
36206
36207
36208
36209
36210
36211
36212
36213
36214
36215
36216
36217
36218
36219
36220
36221
36222
36223
36224
36225
36226
36227
36228
36229
36230
36231
36232
36233
36234
36235
36236
36237
36238
36239
36240
36241
36242
36243
36244
36245
36246
36247
36248
36249
36250
36251
36252
36253
36254
36255
36256
36257
36258
36259
36260
36261
36262
36263
36264
36265
36266
36267
36268
36269
36270
36271
36272
36273
36274
36275
36276
36277
36278
36279
36280
36281
36282
36283
36284
36285
36286
36287
36288
36289
36290
36291
36292
36293
36294
36295
36296
36297
36298
36299
36300
36301
36302
36303
36304
36305
36306
36307
36308
36309
36310
36311
36312
36313
36314
36315
36316
36317
36318
36319
36320
36321
36322
36323
36324
36325
36326
36327
36328
36329
36330
36331
36332
36333
36334
36335
36336
36337
36338
36339
36340
36341
36342
36343
36344
36345
36346
36347
36348
36349
36350
36351
36352
36353
36354
36355
36356
36357
36358
36359
36360
36361
36362
36363
36364
36365
36366
36367
36368
36369
36370
36371
36372
36373
36374
36375
36376
36377
36378
36379
36380
36381
36382
36383
36384
36385
36386
36387
36388
36389
36390
36391
36392
36393
36394
36395
36396
36397
36398
36399
36400
36401
36402
36403
36404
36405
36406
36407
36408
36409
36410
36411
36412
36413
36414
36415
36416
36417
36418
36419
36420
36421
36422
36423
36424
36425
36426
36427
36428
36429
36430
36431
36432
36433
36434
36435
36436
36437
36438
36439
36440
36441
36442
36443
36444
36445
36446
36447
36448
36449
36450
36451
36452
36453
36454
36455
36456
36457
36458
36459
36460
36461
36462
36463
36464
36465
36466
36467
36468
36469
36470
36471
36472
36473
36474
36475
36476
36477
36478
36479
36480
36481
36482
36483
36484
36485
36486
36487
36488
36489
36490
36491
36492
36493
36494
36495
36496
36497
36498
36499
36500
36501
36502
36503
36504
36505
36506
36507
36508
36509
36510
36511
36512
36513
36514
36515
36516
36517
36518
36519
36520
36521
36522
36523
36524
36525
36526
36527
36528
36529
36530
36531
36532
36533
36534
36535
36536
36537
36538
36539
36540
36541
36542
36543
36544
36545
36546
36547
36548
36549
36550
36551
36552
36553
36554
36555
36556
36557
36558
36559
36560
36561
36562
36563
36564
36565
36566
36567
36568
36569
36570
36571
36572
36573
36574
36575
36576
36577
36578
36579
36580
36581
36582
36583
36584
36585
36586
36587
36588
36589
36590
36591
36592
36593
36594
36595
36596
36597
36598
36599
36600
36601
36602
36603
36604
36605
36606
36607
36608
36609
36610
36611
36612
36613
36614
36615
36616
36617
36618
36619
36620
36621
36622
36623
36624
36625
36626
36627
36628
36629
36630
36631
36632
36633
36634
36635
36636
36637
36638
36639
36640
36641
36642
36643
36644
36645
36646
36647
36648
36649
36650
36651
36652
36653
36654
36655
36656
36657
36658
36659
36660
36661
36662
36663
36664
36665
36666
36667
36668
36669
36670
36671
36672
36673
36674
36675
36676
36677
36678
36679
36680
36681
36682
36683
36684
36685
36686
36687
36688
36689
36690
36691
36692
36693
36694
36695
36696
36697
36698
36699
36700
36701
36702
36703
36704
36705
36706
36707
36708
36709
36710
36711
36712
36713
36714
36715
36716
36717
36718
36719
36720
36721
36722
36723
36724
36725
36726
36727
36728
36729
36730
36731
36732
36733
36734
36735
36736
36737
36738
36739
36740
36741
36742
36743
36744
36745
36746
36747
36748
36749
36750
36751
36752
36753
36754
36755
36756
36757
36758
36759
36760
36761
36762
36763
36764
36765
36766
36767
36768
36769
36770
36771
36772
36773
36774
36775
36776
36777
36778
36779
36780
36781
36782
36783
36784
36785
36786
36787
36788
36789
36790
36791
36792
36793
36794
36795
36796
36797
36798
36799
36800
36801
36802
36803
36804
36805
36806
36807
36808
36809
36810
36811
36812
36813
36814
36815
36816
36817
36818
36819
36820
36821
36822
36823
36824
36825
36826
36827
36828
36829
36830
36831
36832
36833
36834
36835
36836
36837
36838
36839
36840
36841
36842
36843
36844
36845
36846
36847
36848
36849
36850
36851
36852
36853
36854
36855
36856
36857
36858
36859
36860
36861
36862
36863
36864
36865
36866
36867
36868
36869
36870
36871
36872
36873
36874
36875
36876
36877
36878
36879
36880
36881
36882
36883
36884
36885
36886
36887
36888
36889
36890
36891
36892
36893
36894
36895
36896
36897
36898
36899
36900
36901
36902
36903
36904
36905
36906
36907
36908
36909
36910
36911
36912
36913
36914
36915
36916
36917
36918
36919
36920
36921
36922
36923
36924
36925
36926
36927
36928
36929
36930
36931
36932
36933
36934
36935
36936
36937
36938
36939
36940
36941
36942
36943
36944
36945
36946
36947
36948
36949
36950
36951
36952
36953
36954
36955
36956
36957
36958
36959
36960
36961
36962
36963
36964
36965
36966
36967
36968
36969
36970
36971
36972
36973
36974
36975
36976
36977
36978
36979
36980
36981
36982
36983
36984
36985
36986
36987
36988
36989
36990
36991
36992
36993
36994
36995
36996
36997
36998
36999
37000
37001
37002
37003
37004
37005
37006
37007
37008
37009
37010
37011
37012
37013
37014
37015
37016
37017
37018
37019
37020
37021
37022
37023
37024
37025
37026
37027
37028
37029
37030
37031
37032
37033
37034
37035
37036
37037
37038
37039
37040
37041
37042
37043
37044
37045
37046
37047
37048
37049
37050
37051
37052
37053
37054
37055
37056
37057
37058
37059
37060
37061
37062
37063
37064
37065
37066
37067
37068
37069
37070
37071
37072
37073
37074
37075
37076
37077
37078
37079
37080
37081
37082
37083
37084
37085
37086
37087
37088
37089
37090
37091
37092
37093
37094
37095
37096
37097
37098
37099
37100
37101
37102
37103
37104
37105
37106
37107
37108
37109
37110
37111
37112
37113
37114
37115
37116
37117
37118
37119
37120
37121
37122
37123
37124
37125
37126
37127
37128
37129
37130
37131
37132
37133
37134
37135
37136
37137
37138
37139
37140
37141
37142
37143
37144
37145
37146
37147
37148
37149
37150
37151
37152
37153
37154
37155
37156
37157
37158
37159
37160
37161
37162
37163
37164
37165
37166
37167
37168
37169
37170
37171
37172
37173
37174
37175
37176
37177
37178
37179
37180
37181
37182
37183
37184
37185
37186
37187
37188
37189
37190
37191
37192
37193
37194
37195
37196
37197
37198
37199
37200
37201
37202
37203
37204
37205
37206
37207
37208
37209
37210
37211
37212
37213
37214
37215
37216
37217
37218
37219
37220
37221
37222
37223
37224
37225
37226
37227
37228
37229
37230
37231
37232
37233
37234
37235
37236
37237
37238
37239
37240
37241
37242
37243
37244
37245
37246
37247
37248
37249
37250
37251
37252
37253
37254
37255
37256
37257
37258
37259
37260
37261
37262
37263
37264
37265
37266
37267
37268
37269
37270
37271
37272
37273
37274
37275
37276
37277
37278
37279
37280
37281
37282
37283
37284
37285
37286
37287
37288
37289
37290
37291
37292
37293
37294
37295
37296
37297
37298
37299
37300
37301
37302
37303
37304
37305
37306
37307
37308
37309
37310
37311
37312
37313
37314
37315
37316
37317
37318
37319
37320
37321
37322
37323
37324
37325
37326
37327
37328
37329
37330
37331
37332
37333
37334
37335
37336
37337
37338
37339
37340
37341
37342
37343
37344
37345
37346
37347
37348
37349
37350
37351
37352
37353
37354
37355
37356
37357
37358
37359
37360
37361
37362
37363
37364
37365
37366
37367
37368
37369
37370
37371
37372
37373
37374
37375
37376
37377
37378
37379
37380
37381
37382
37383
37384
37385
37386
37387
37388
37389
37390
37391
37392
37393
37394
37395
37396
37397
37398
37399
37400
37401
37402
37403
37404
37405
37406
37407
37408
37409
37410
37411
37412
37413
37414
37415
37416
37417
37418
37419
37420
37421
37422
37423
37424
37425
37426
37427
37428
37429
37430
37431
37432
37433
37434
37435
37436
37437
37438
37439
37440
37441
37442
37443
37444
37445
37446
37447
37448
37449
37450
37451
37452
37453
37454
37455
37456
37457
37458
37459
37460
37461
37462
37463
37464
37465
37466
37467
37468
37469
37470
37471
37472
37473
37474
37475
37476
37477
37478
37479
37480
37481
37482
37483
37484
37485
37486
37487
37488
37489
37490
37491
37492
37493
37494
37495
37496
37497
37498
37499
37500
37501
37502
37503
37504
37505
37506
37507
37508
37509
37510
37511
37512
37513
37514
37515
37516
37517
37518
37519
37520
37521
37522
37523
37524
37525
37526
37527
37528
37529
37530
37531
37532
37533
37534
37535
37536
37537
37538
37539
37540
37541
37542
37543
37544
37545
37546
37547
37548
37549
37550
37551
37552
37553
37554
37555
37556
37557
37558
37559
37560
37561
37562
37563
37564
37565
37566
37567
37568
37569
37570
37571
37572
37573
37574
37575
37576
37577
37578
37579
37580
37581
37582
37583
37584
37585
37586
37587
37588
37589
37590
37591
37592
37593
37594
37595
37596
37597
37598
37599
37600
37601
37602
37603
37604
37605
37606
37607
37608
37609
37610
37611
37612
37613
37614
37615
37616
37617
37618
37619
37620
37621
37622
37623
37624
37625
37626
37627
37628
37629
37630
37631
37632
37633
37634
37635
37636
37637
37638
37639
37640
37641
37642
37643
37644
37645
37646
37647
37648
37649
37650
37651
37652
37653
37654
37655
37656
37657
37658
37659
37660
37661
37662
37663
37664
37665
37666
37667
37668
37669
37670
37671
37672
37673
37674
37675
37676
37677
37678
37679
37680
37681
37682
37683
37684
37685
37686
37687
37688
37689
37690
37691
37692
37693
37694
37695
37696
37697
37698
37699
37700
37701
37702
37703
37704
37705
37706
37707
37708
37709
37710
37711
37712
37713
37714
37715
37716
37717
37718
37719
37720
37721
37722
37723
37724
37725
37726
37727
37728
37729
37730
37731
37732
37733
37734
37735
37736
37737
37738
37739
37740
37741
37742
37743
37744
37745
37746
37747
37748
37749
37750
37751
37752
37753
37754
37755
37756
37757
37758
37759
37760
37761
37762
37763
37764
37765
37766
37767
37768
37769
37770
37771
37772
37773
37774
37775
37776
37777
37778
37779
37780
37781
37782
37783
37784
37785
37786
37787
37788
37789
37790
37791
37792
37793
37794
37795
37796
37797
37798
37799
37800
37801
37802
37803
37804
37805
37806
37807
37808
37809
37810
37811
37812
37813
37814
37815
37816
37817
37818
37819
37820
37821
37822
37823
37824
37825
37826
37827
37828
37829
37830
37831
37832
37833
37834
37835
37836
37837
37838
37839
37840
37841
37842
37843
37844
37845
37846
37847
37848
37849
37850
37851
37852
37853
37854
37855
37856
37857
37858
37859
37860
37861
37862
37863
37864
37865
37866
37867
37868
37869
37870
37871
37872
37873
37874
37875
37876
37877
37878
37879
37880
37881
37882
37883
37884
37885
37886
37887
37888
37889
37890
37891
37892
37893
37894
37895
37896
37897
37898
37899
37900
37901
37902
37903
37904
37905
37906
37907
37908
37909
37910
37911
37912
37913
37914
37915
37916
37917
37918
37919
37920
37921
37922
37923
37924
37925
37926
37927
37928
37929
37930
37931
37932
37933
37934
37935
37936
37937
37938
37939
37940
37941
37942
37943
37944
37945
37946
37947
37948
37949
37950
37951
37952
37953
37954
37955
37956
37957
37958
37959
37960
37961
37962
37963
37964
37965
37966
37967
37968
37969
37970
37971
37972
37973
37974
37975
37976
37977
37978
37979
37980
37981
37982
37983
37984
37985
37986
37987
37988
37989
37990
37991
37992
37993
37994
37995
37996
37997
37998
37999
38000
38001
38002
38003
38004
38005
38006
38007
38008
38009
38010
38011
38012
38013
38014
38015
38016
38017
38018
38019
38020
38021
38022
38023
38024
38025
38026
38027
38028
38029
38030
38031
38032
38033
38034
38035
38036
38037
38038
38039
38040
38041
38042
38043
38044
38045
38046
38047
38048
38049
38050
38051
38052
38053
38054
38055
38056
38057
38058
38059
38060
38061
38062
38063
38064
38065
38066
38067
38068
38069
38070
38071
38072
38073
38074
38075
38076
38077
38078
38079
38080
38081
38082
38083
38084
38085
38086
38087
38088
38089
38090
38091
38092
38093
38094
38095
38096
38097
38098
38099
38100
38101
38102
38103
38104
38105
38106
38107
38108
38109
38110
38111
38112
38113
38114
38115
38116
38117
38118
38119
38120
38121
38122
38123
38124
38125
38126
38127
38128
38129
38130
38131
38132
38133
38134
38135
38136
38137
38138
38139
38140
38141
38142
38143
38144
38145
38146
38147
38148
38149
38150
38151
38152
38153
38154
38155
38156
38157
38158
38159
38160
38161
38162
38163
38164
38165
38166
38167
38168
38169
38170
38171
38172
38173
38174
38175
38176
38177
38178
38179
38180
38181
38182
38183
38184
38185
38186
38187
38188
38189
38190
38191
38192
38193
38194
38195
38196
38197
38198
38199
38200
38201
38202
38203
38204
38205
38206
38207
38208
38209
38210
38211
38212
38213
38214
38215
38216
38217
38218
38219
38220
38221
38222
38223
38224
38225
38226
38227
38228
38229
38230
38231
38232
38233
38234
38235
38236
38237
38238
38239
38240
38241
38242
38243
38244
38245
38246
38247
38248
38249
38250
38251
38252
38253
38254
38255
38256
38257
38258
38259
38260
38261
38262
38263
38264
38265
38266
38267
38268
38269
38270
38271
38272
38273
38274
38275
38276
38277
38278
38279
38280
38281
38282
38283
38284
38285
38286
38287
38288
38289
38290
38291
38292
38293
38294
38295
38296
38297
38298
38299
38300
38301
38302
38303
38304
38305
38306
38307
38308
38309
38310
38311
38312
38313
38314
38315
38316
38317
38318
38319
38320
38321
38322
38323
38324
38325
38326
38327
38328
38329
38330
38331
38332
38333
38334
38335
38336
38337
38338
38339
38340
38341
38342
38343
38344
38345
38346
38347
38348
38349
38350
38351
38352
38353
38354
38355
38356
38357
38358
38359
38360
38361
38362
38363
38364
38365
38366
38367
38368
38369
38370
38371
38372
38373
38374
38375
38376
38377
38378
38379
38380
38381
38382
38383
38384
38385
38386
38387
38388
38389
38390
38391
38392
38393
38394
38395
38396
38397
38398
38399
38400
38401
38402
38403
38404
38405
38406
38407
38408
38409
38410
38411
38412
38413
38414
38415
38416
38417
38418
38419
38420
38421
38422
38423
38424
38425
38426
38427
38428
38429
38430
38431
38432
38433
38434
38435
38436
38437
38438
38439
38440
38441
38442
38443
38444
38445
38446
38447
38448
38449
38450
38451
38452
38453
38454
38455
38456
38457
38458
38459
38460
38461
38462
38463
38464
38465
38466
38467
38468
38469
38470
38471
38472
38473
38474
38475
38476
38477
38478
38479
38480
38481
38482
38483
38484
38485
38486
38487
38488
38489
38490
38491
38492
38493
38494
38495
38496
38497
38498
38499
38500
38501
38502
38503
38504
38505
38506
38507
38508
38509
38510
38511
38512
38513
38514
38515
38516
38517
38518
38519
38520
38521
38522
38523
38524
38525
38526
38527
38528
38529
38530
38531
38532
38533
38534
38535
38536
38537
38538
38539
38540
38541
38542
38543
38544
38545
38546
38547
38548
38549
38550
38551
38552
38553
38554
38555
38556
38557
38558
38559
38560
38561
38562
38563
38564
38565
38566
38567
38568
38569
38570
38571
38572
38573
38574
38575
38576
38577
38578
38579
38580
38581
38582
38583
38584
38585
38586
38587
38588
38589
38590
38591
38592
38593
38594
38595
38596
38597
38598
38599
38600
38601
38602
38603
38604
38605
38606
38607
38608
38609
38610
38611
38612
38613
38614
38615
38616
38617
38618
38619
38620
38621
38622
38623
38624
38625
38626
38627
38628
38629
38630
38631
38632
38633
38634
38635
38636
38637
38638
38639
38640
38641
38642
38643
38644
38645
38646
38647
38648
38649
38650
38651
38652
38653
38654
38655
38656
38657
38658
38659
38660
38661
38662
38663
38664
38665
38666
38667
38668
38669
38670
38671
38672
38673
38674
38675
38676
38677
38678
38679
38680
38681
38682
38683
38684
38685
38686
38687
38688
38689
38690
38691
38692
38693
38694
38695
38696
38697
38698
38699
38700
38701
38702
38703
38704
38705
38706
38707
38708
38709
38710
38711
38712
38713
38714
38715
38716
38717
38718
38719
38720
38721
38722
38723
38724
38725
38726
38727
38728
38729
38730
38731
38732
38733
38734
38735
38736
38737
38738
38739
38740
38741
38742
38743
38744
38745
38746
38747
38748
38749
38750
38751
38752
38753
38754
38755
38756
38757
38758
38759
38760
38761
38762
38763
38764
38765
38766
38767
38768
38769
38770
38771
38772
38773
38774
38775
38776
38777
38778
38779
38780
38781
38782
38783
38784
38785
38786
38787
38788
38789
38790
38791
38792
38793
38794
38795
38796
38797
38798
38799
38800
38801
38802
38803
38804
38805
38806
38807
38808
38809
38810
38811
38812
38813
38814
38815
38816
38817
38818
38819
38820
38821
38822
38823
38824
38825
38826
38827
38828
38829
38830
38831
38832
38833
38834
38835
38836
38837
38838
38839
38840
38841
38842
38843
38844
38845
38846
38847
38848
38849
38850
38851
38852
38853
38854
38855
38856
38857
38858
38859
38860
38861
38862
38863
38864
38865
38866
38867
38868
38869
38870
38871
38872
38873
38874
38875
38876
38877
38878
38879
38880
38881
38882
38883
38884
38885
38886
38887
38888
38889
38890
38891
38892
38893
38894
38895
38896
38897
38898
38899
38900
38901
38902
38903
38904
38905
38906
38907
38908
38909
38910
38911
38912
38913
38914
38915
38916
38917
38918
38919
38920
38921
38922
38923
38924
38925
38926
38927
38928
38929
38930
38931
38932
38933
38934
38935
38936
38937
38938
38939
38940
38941
38942
38943
38944
38945
38946
38947
38948
38949
38950
38951
38952
38953
38954
38955
38956
38957
38958
38959
38960
38961
38962
38963
38964
38965
38966
38967
38968
38969
38970
38971
38972
38973
38974
38975
38976
38977
38978
38979
38980
38981
38982
38983
38984
38985
38986
38987
38988
38989
38990
38991
38992
38993
38994
38995
38996
38997
38998
38999
39000
39001
39002
39003
39004
39005
39006
39007
39008
39009
39010
39011
39012
39013
39014
39015
39016
39017
39018
39019
39020
39021
39022
39023
39024
39025
39026
39027
39028
39029
39030
39031
39032
39033
39034
39035
39036
39037
39038
39039
39040
39041
39042
39043
39044
39045
39046
39047
39048
39049
39050
39051
39052
39053
39054
39055
39056
39057
39058
39059
39060
39061
39062
39063
39064
39065
39066
39067
39068
39069
39070
39071
39072
39073
39074
39075
39076
39077
39078
39079
39080
39081
39082
39083
39084
39085
39086
39087
39088
39089
39090
39091
39092
39093
39094
39095
39096
39097
39098
39099
39100
39101
39102
39103
39104
39105
39106
39107
39108
39109
39110
39111
39112
39113
39114
39115
39116
39117
39118
39119
39120
39121
39122
39123
39124
39125
39126
39127
39128
39129
39130
39131
39132
39133
39134
39135
39136
39137
39138
39139
39140
39141
39142
39143
39144
39145
39146
39147
39148
39149
39150
39151
39152
39153
39154
39155
39156
39157
39158
39159
39160
39161
39162
39163
39164
39165
39166
39167
39168
39169
39170
39171
39172
39173
39174
39175
39176
39177
39178
39179
39180
39181
39182
39183
39184
39185
39186
39187
39188
39189
39190
39191
39192
39193
39194
39195
39196
39197
39198
39199
39200
39201
39202
39203
39204
39205
39206
39207
39208
39209
39210
39211
39212
39213
39214
39215
39216
39217
39218
39219
39220
39221
39222
39223
39224
39225
39226
39227
39228
39229
39230
39231
39232
39233
39234
39235
39236
39237
39238
39239
39240
39241
39242
39243
39244
39245
39246
39247
39248
39249
39250
39251
39252
39253
39254
39255
39256
39257
39258
39259
39260
39261
39262
39263
39264
39265
39266
39267
39268
39269
39270
39271
39272
39273
39274
39275
39276
39277
39278
39279
39280
39281
39282
39283
39284
39285
39286
39287
39288
39289
39290
39291
39292
39293
39294
39295
39296
39297
39298
39299
39300
39301
39302
39303
39304
39305
39306
39307
39308
39309
39310
39311
39312
39313
39314
39315
39316
39317
39318
39319
39320
39321
39322
39323
39324
39325
39326
39327
39328
39329
39330
39331
39332
39333
39334
39335
39336
39337
39338
39339
39340
39341
39342
39343
39344
39345
39346
39347
39348
39349
39350
39351
39352
39353
39354
39355
39356
39357
39358
39359
39360
39361
39362
39363
39364
39365
39366
39367
39368
39369
39370
39371
39372
39373
39374
39375
39376
39377
39378
39379
39380
39381
39382
39383
39384
39385
39386
39387
39388
39389
39390
39391
39392
39393
39394
39395
39396
39397
39398
39399
39400
39401
39402
39403
39404
39405
39406
39407
39408
39409
39410
39411
39412
39413
39414
39415
39416
39417
39418
39419
39420
39421
39422
39423
39424
39425
39426
39427
39428
39429
39430
39431
39432
39433
39434
39435
39436
39437
39438
39439
39440
39441
39442
39443
39444
39445
39446
39447
39448
39449
39450
39451
39452
39453
39454
39455
39456
39457
39458
39459
39460
39461
39462
39463
39464
39465
39466
39467
39468
39469
39470
39471
39472
39473
39474
39475
39476
39477
39478
39479
39480
39481
39482
39483
39484
39485
39486
39487
39488
39489
39490
39491
39492
39493
39494
39495
39496
39497
39498
39499
39500
39501
39502
39503
39504
39505
39506
39507
39508
39509
39510
39511
39512
39513
39514
39515
39516
39517
39518
39519
39520
39521
39522
39523
39524
39525
39526
39527
39528
39529
39530
39531
39532
39533
39534
39535
39536
39537
39538
39539
39540
39541
39542
39543
39544
39545
39546
39547
39548
39549
39550
39551
39552
39553
39554
39555
39556
39557
39558
39559
39560
39561
39562
39563
39564
39565
39566
39567
39568
39569
39570
39571
39572
39573
39574
39575
39576
39577
39578
39579
39580
39581
39582
39583
39584
39585
39586
39587
39588
39589
39590
39591
39592
39593
39594
39595
39596
39597
39598
39599
39600
39601
39602
39603
39604
39605
39606
39607
39608
39609
39610
39611
39612
39613
39614
39615
39616
39617
39618
39619
39620
39621
39622
39623
39624
39625
39626
39627
39628
39629
39630
39631
39632
39633
39634
39635
39636
39637
39638
39639
39640
39641
39642
39643
39644
39645
39646
39647
39648
39649
39650
39651
39652
39653
39654
39655
39656
39657
39658
39659
39660
39661
39662
39663
39664
39665
39666
39667
39668
39669
39670
39671
39672
39673
39674
39675
39676
39677
39678
39679
39680
39681
39682
39683
39684
39685
39686
39687
39688
39689
39690
39691
39692
39693
39694
39695
39696
39697
39698
39699
39700
39701
39702
39703
39704
39705
39706
39707
39708
39709
39710
39711
39712
39713
39714
39715
39716
39717
39718
39719
39720
39721
39722
39723
39724
39725
39726
39727
39728
39729
39730
39731
39732
39733
39734
39735
39736
39737
39738
39739
39740
39741
39742
39743
39744
39745
39746
39747
39748
39749
39750
39751
39752
39753
39754
39755
39756
39757
39758
39759
39760
39761
39762
39763
39764
39765
39766
39767
39768
39769
39770
39771
39772
39773
39774
39775
39776
39777
39778
39779
39780
39781
39782
39783
39784
39785
39786
39787
39788
39789
39790
39791
39792
39793
39794
39795
39796
39797
39798
39799
39800
39801
39802
39803
39804
39805
39806
39807
39808
39809
39810
39811
39812
39813
39814
39815
39816
39817
39818
39819
39820
39821
39822
39823
39824
39825
39826
39827
39828
39829
39830
39831
39832
39833
39834
39835
39836
39837
39838
39839
39840
39841
39842
39843
39844
39845
39846
39847
39848
39849
39850
39851
39852
39853
39854
39855
39856
39857
39858
39859
39860
39861
39862
39863
39864
39865
39866
39867
39868
39869
39870
39871
39872
39873
39874
39875
39876
39877
39878
39879
39880
39881
39882
39883
39884
39885
39886
39887
39888
39889
39890
39891
39892
39893
39894
39895
39896
39897
39898
39899
39900
39901
39902
39903
39904
39905
39906
39907
39908
39909
39910
39911
39912
39913
39914
39915
39916
39917
39918
39919
39920
39921
39922
39923
39924
39925
39926
39927
39928
39929
39930
39931
39932
39933
39934
39935
39936
39937
39938
39939
39940
39941
39942
39943
39944
39945
39946
39947
39948
39949
39950
39951
39952
39953
39954
39955
39956
39957
39958
39959
39960
39961
39962
39963
39964
39965
39966
39967
39968
39969
39970
39971
39972
39973
39974
39975
39976
39977
39978
39979
39980
39981
39982
39983
39984
39985
39986
39987
39988
39989
39990
39991
39992
39993
39994
39995
39996
39997
39998
39999
40000
40001
40002
40003
40004
40005
40006
40007
40008
40009
40010
40011
40012
40013
40014
40015
40016
40017
40018
40019
40020
40021
40022
40023
40024
40025
40026
40027
40028
40029
40030
40031
40032
40033
40034
40035
40036
40037
40038
40039
40040
40041
40042
40043
40044
40045
40046
40047
40048
40049
40050
40051
40052
40053
40054
40055
40056
40057
40058
40059
40060
40061
40062
40063
40064
40065
40066
40067
40068
40069
40070
40071
40072
40073
40074
40075
40076
40077
40078
40079
40080
40081
40082
40083
40084
40085
40086
40087
40088
40089
40090
40091
40092
40093
40094
40095
40096
40097
40098
40099
40100
40101
40102
40103
40104
40105
40106
40107
40108
40109
40110
40111
40112
40113
40114
40115
40116
40117
40118
40119
40120
40121
40122
40123
40124
40125
40126
40127
40128
40129
40130
40131
40132
40133
40134
40135
40136
40137
40138
40139
40140
40141
40142
40143
40144
40145
40146
40147
40148
40149
40150
40151
40152
40153
40154
40155
40156
40157
40158
40159
40160
40161
40162
40163
40164
40165
40166
40167
40168
40169
40170
40171
40172
40173
40174
40175
40176
40177
40178
40179
40180
40181
40182
40183
40184
40185
40186
40187
40188
40189
40190
40191
40192
40193
40194
40195
40196
40197
40198
40199
40200
40201
40202
40203
40204
40205
40206
40207
40208
40209
40210
40211
40212
40213
40214
40215
40216
40217
40218
40219
40220
40221
40222
40223
40224
40225
40226
40227
40228
40229
40230
40231
40232
40233
40234
40235
40236
40237
40238
40239
40240
40241
40242
40243
40244
40245
40246
40247
40248
40249
40250
40251
40252
40253
40254
40255
40256
40257
40258
40259
40260
40261
40262
40263
40264
40265
40266
40267
40268
40269
40270
40271
40272
40273
40274
40275
40276
40277
40278
40279
40280
40281
40282
40283
40284
40285
40286
40287
40288
40289
40290
40291
40292
40293
40294
40295
40296
40297
40298
40299
40300
40301
40302
40303
40304
40305
40306
40307
40308
40309
40310
40311
40312
40313
40314
40315
40316
40317
40318
40319
40320
40321
40322
40323
40324
40325
40326
40327
40328
40329
40330
40331
40332
40333
40334
40335
40336
40337
40338
40339
40340
40341
40342
40343
40344
40345
40346
40347
40348
40349
40350
40351
40352
40353
40354
40355
40356
40357
40358
40359
40360
40361
40362
40363
40364
40365
40366
40367
40368
40369
40370
40371
40372
40373
40374
40375
40376
40377
40378
40379
40380
40381
40382
40383
40384
40385
40386
40387
40388
40389
40390
40391
40392
40393
40394
40395
40396
40397
40398
40399
40400
40401
40402
40403
40404
40405
40406
40407
40408
40409
40410
40411
40412
40413
40414
40415
40416
40417
40418
40419
40420
40421
40422
40423
40424
40425
40426
40427
40428
40429
40430
40431
40432
40433
40434
40435
40436
40437
40438
40439
40440
40441
40442
40443
40444
40445
40446
40447
40448
40449
40450
40451
40452
40453
40454
40455
40456
40457
40458
40459
40460
40461
40462
40463
40464
40465
40466
40467
40468
40469
40470
40471
40472
40473
40474
40475
40476
40477
40478
40479
40480
40481
40482
40483
40484
40485
40486
40487
40488
40489
40490
40491
40492
40493
40494
40495
40496
40497
40498
40499
40500
40501
40502
40503
40504
40505
40506
40507
40508
40509
40510
40511
40512
40513
40514
40515
40516
40517
40518
40519
40520
40521
40522
40523
40524
40525
40526
40527
40528
40529
40530
40531
40532
40533
40534
40535
40536
40537
40538
40539
40540
40541
40542
40543
40544
40545
40546
40547
40548
40549
40550
40551
40552
40553
40554
40555
40556
40557
40558
40559
40560
40561
40562
40563
40564
40565
40566
40567
40568
40569
40570
40571
40572
40573
40574
40575
40576
40577
40578
40579
40580
40581
40582
40583
40584
40585
40586
40587
40588
40589
40590
40591
40592
40593
40594
40595
40596
40597
40598
40599
40600
40601
40602
40603
40604
40605
40606
40607
40608
40609
40610
40611
40612
40613
40614
40615
40616
40617
40618
40619
40620
40621
40622
40623
40624
40625
40626
40627
40628
40629
40630
40631
40632
40633
40634
40635
40636
40637
40638
40639
40640
40641
40642
40643
40644
40645
40646
40647
40648
40649
40650
40651
40652
40653
40654
40655
40656
40657
40658
40659
40660
40661
40662
40663
40664
40665
40666
40667
40668
40669
40670
40671
40672
40673
40674
40675
40676
40677
40678
40679
40680
40681
40682
40683
40684
40685
40686
40687
40688
40689
40690
40691
40692
40693
40694
40695
40696
40697
40698
40699
40700
40701
40702
40703
40704
40705
40706
40707
40708
40709
40710
40711
40712
40713
40714
40715
40716
40717
40718
40719
40720
40721
40722
40723
40724
40725
40726
40727
40728
40729
40730
40731
40732
40733
40734
40735
40736
40737
40738
40739
40740
40741
40742
40743
40744
40745
40746
40747
40748
40749
40750
40751
40752
40753
40754
40755
40756
40757
40758
40759
40760
40761
40762
40763
40764
40765
40766
40767
40768
40769
40770
40771
40772
40773
40774
40775
40776
40777
40778
40779
40780
40781
40782
40783
40784
40785
40786
40787
40788
40789
40790
40791
40792
40793
40794
40795
40796
40797
40798
40799
40800
40801
40802
40803
40804
40805
40806
40807
40808
40809
40810
40811
40812
40813
40814
40815
40816
40817
40818
40819
40820
40821
40822
40823
40824
40825
40826
40827
40828
40829
40830
40831
40832
40833
40834
40835
40836
40837
40838
40839
40840
40841
40842
40843
40844
40845
40846
40847
40848
40849
40850
40851
40852
40853
40854
40855
40856
40857
40858
40859
40860
40861
40862
40863
40864
40865
40866
40867
40868
40869
40870
40871
40872
40873
40874
40875
40876
40877
40878
40879
40880
40881
40882
40883
40884
40885
40886
40887
40888
40889
40890
40891
40892
40893
40894
40895
40896
40897
40898
40899
40900
40901
40902
40903
40904
40905
40906
40907
40908
40909
40910
40911
40912
40913
40914
40915
40916
40917
40918
40919
40920
40921
40922
40923
40924
40925
40926
40927
40928
40929
40930
40931
40932
40933
40934
40935
40936
40937
40938
40939
40940
40941
40942
40943
40944
40945
40946
40947
40948
40949
40950
40951
40952
40953
40954
40955
40956
40957
40958
40959
40960
40961
40962
40963
40964
40965
40966
40967
40968
40969
40970
40971
40972
40973
40974
40975
40976
40977
40978
40979
40980
40981
40982
40983
40984
40985
40986
40987
40988
40989
40990
40991
40992
40993
40994
40995
40996
40997
40998
40999
41000
41001
41002
41003
41004
41005
41006
41007
41008
41009
41010
41011
41012
41013
41014
41015
41016
41017
41018
41019
41020
41021
41022
41023
41024
41025
41026
41027
41028
41029
41030
41031
41032
41033
41034
41035
41036
41037
41038
41039
41040
41041
41042
41043
41044
41045
41046
41047
41048
41049
41050
41051
41052
41053
41054
41055
41056
41057
41058
41059
41060
41061
41062
41063
41064
41065
41066
41067
41068
41069
41070
41071
41072
41073
41074
41075
41076
41077
41078
41079
41080
41081
41082
41083
41084
41085
41086
41087
41088
41089
41090
41091
41092
41093
41094
41095
41096
41097
41098
41099
41100
41101
41102
41103
41104
41105
41106
41107
41108
41109
41110
41111
41112
41113
41114
41115
41116
41117
41118
41119
41120
41121
41122
41123
41124
41125
41126
41127
41128
41129
41130
41131
41132
41133
41134
41135
41136
41137
41138
41139
41140
41141
41142
41143
41144
41145
41146
41147
41148
41149
41150
41151
41152
41153
41154
41155
41156
41157
41158
41159
41160
41161
41162
41163
41164
41165
41166
41167
41168
41169
41170
41171
41172
41173
41174
41175
41176
41177
41178
41179
41180
41181
41182
41183
41184
41185
41186
41187
41188
41189
41190
41191
41192
41193
41194
41195
41196
41197
41198
41199
41200
41201
41202
41203
41204
41205
41206
41207
41208
41209
41210
41211
41212
41213
41214
41215
41216
41217
41218
41219
41220
41221
41222
41223
41224
41225
41226
41227
41228
41229
41230
41231
41232
41233
41234
41235
41236
41237
41238
41239
41240
41241
41242
41243
41244
41245
41246
41247
41248
41249
41250
41251
41252
41253
41254
41255
41256
41257
41258
41259
41260
41261
41262
41263
41264
41265
41266
41267
41268
41269
41270
41271
41272
41273
41274
41275
41276
41277
41278
41279
41280
41281
41282
41283
41284
41285
41286
41287
41288
41289
41290
41291
41292
41293
41294
41295
41296
41297
41298
41299
41300
41301
41302
41303
41304
41305
41306
41307
41308
41309
41310
41311
41312
41313
41314
41315
41316
41317
41318
41319
41320
41321
41322
41323
41324
41325
41326
41327
41328
41329
41330
41331
41332
41333
41334
41335
41336
41337
41338
41339
41340
41341
41342
41343
41344
41345
41346
41347
41348
41349
41350
41351
41352
41353
41354
41355
41356
41357
41358
41359
41360
41361
41362
41363
41364
41365
41366
41367
41368
41369
41370
41371
41372
41373
41374
41375
41376
41377
41378
41379
41380
41381
41382
41383
41384
41385
41386
41387
41388
41389
41390
41391
41392
41393
41394
41395
41396
41397
41398
41399
41400
41401
41402
41403
41404
41405
41406
41407
41408
41409
41410
41411
41412
41413
41414
41415
41416
41417
41418
41419
41420
41421
41422
41423
41424
41425
41426
41427
41428
41429
41430
41431
41432
41433
41434
41435
41436
41437
41438
41439
41440
41441
41442
41443
41444
41445
41446
41447
41448
41449
41450
41451
41452
41453
41454
41455
41456
41457
41458
41459
41460
41461
41462
41463
41464
41465
41466
41467
41468
41469
41470
41471
41472
41473
41474
41475
41476
41477
41478
41479
41480
41481
41482
41483
41484
41485
41486
41487
41488
41489
41490
41491
41492
41493
41494
41495
41496
41497
41498
41499
41500
41501
41502
41503
41504
41505
41506
41507
41508
41509
41510
41511
41512
41513
41514
41515
41516
41517
41518
41519
41520
41521
41522
41523
41524
41525
41526
41527
41528
41529
41530
41531
41532
41533
41534
41535
41536
41537
41538
41539
41540
41541
41542
41543
41544
41545
41546
41547
41548
41549
41550
41551
41552
41553
41554
41555
41556
41557
41558
41559
41560
41561
41562
41563
41564
41565
41566
41567
41568
41569
41570
41571
41572
41573
41574
41575
41576
41577
41578
41579
41580
41581
41582
41583
41584
41585
41586
41587
41588
41589
41590
41591
41592
41593
41594
41595
41596
41597
41598
41599
41600
41601
41602
41603
41604
41605
41606
41607
41608
41609
41610
41611
41612
41613
41614
41615
41616
41617
41618
41619
41620
41621
41622
41623
41624
41625
41626
41627
41628
41629
41630
41631
41632
41633
41634
41635
41636
41637
41638
41639
41640
41641
41642
41643
41644
41645
41646
41647
41648
41649
41650
41651
41652
41653
41654
41655
41656
41657
41658
41659
41660
41661
41662
41663
41664
41665
41666
41667
41668
41669
41670
41671
41672
41673
41674
41675
41676
41677
41678
41679
41680
41681
41682
41683
41684
41685
41686
41687
41688
41689
41690
41691
41692
41693
41694
41695
41696
41697
41698
41699
41700
41701
41702
41703
41704
41705
41706
41707
41708
41709
41710
41711
41712
41713
41714
41715
41716
41717
41718
41719
41720
41721
41722
41723
41724
41725
41726
41727
41728
41729
41730
41731
41732
41733
41734
41735
41736
41737
41738
41739
41740
41741
41742
41743
41744
41745
41746
41747
41748
41749
41750
41751
41752
41753
41754
41755
41756
41757
41758
41759
41760
41761
41762
41763
41764
41765
41766
41767
41768
41769
41770
41771
41772
41773
41774
41775
41776
41777
41778
41779
41780
41781
41782
41783
41784
41785
41786
41787
41788
41789
41790
41791
41792
41793
41794
41795
41796
41797
41798
41799
41800
41801
41802
41803
41804
41805
41806
41807
41808
41809
41810
41811
41812
41813
41814
41815
41816
41817
41818
41819
41820
41821
41822
41823
41824
41825
41826
41827
41828
41829
41830
41831
41832
41833
41834
41835
41836
41837
41838
41839
41840
41841
41842
41843
41844
41845
41846
41847
41848
41849
41850
41851
41852
41853
41854
41855
41856
41857
41858
41859
41860
41861
41862
41863
41864
41865
41866
41867
41868
41869
41870
41871
41872
41873
41874
41875
41876
41877
41878
41879
41880
41881
41882
41883
41884
41885
41886
41887
41888
41889
41890
41891
41892
41893
41894
41895
41896
41897
41898
41899
41900
41901
41902
41903
41904
41905
41906
41907
41908
41909
41910
41911
41912
41913
41914
41915
41916
41917
41918
41919
41920
41921
41922
41923
41924
41925
41926
41927
41928
41929
41930
41931
41932
41933
41934
41935
41936
41937
41938
41939
41940
41941
41942
41943
41944
41945
41946
41947
41948
41949
41950
41951
41952
41953
41954
41955
41956
41957
41958
41959
41960
41961
41962
41963
41964
41965
41966
41967
41968
41969
41970
41971
41972
41973
41974
41975
41976
41977
41978
41979
41980
41981
41982
41983
41984
41985
41986
41987
41988
41989
41990
41991
41992
41993
41994
41995
41996
41997
41998
41999
42000
42001
42002
42003
42004
42005
42006
42007
42008
42009
42010
42011
42012
42013
42014
42015
42016
42017
42018
42019
42020
42021
42022
42023
42024
42025
42026
42027
42028
42029
42030
42031
42032
42033
42034
42035
42036
42037
42038
42039
42040
42041
42042
42043
42044
42045
42046
42047
42048
42049
42050
42051
42052
42053
42054
42055
42056
42057
42058
42059
42060
42061
42062
42063
42064
42065
42066
42067
42068
42069
42070
42071
42072
42073
42074
42075
42076
42077
42078
42079
42080
42081
42082
42083
42084
42085
42086
42087
42088
42089
42090
42091
42092
42093
42094
42095
42096
42097
42098
42099
42100
42101
42102
42103
42104
42105
42106
42107
42108
42109
42110
42111
42112
42113
42114
42115
42116
42117
42118
42119
42120
42121
42122
42123
42124
42125
42126
42127
42128
42129
42130
42131
42132
42133
42134
42135
42136
42137
42138
42139
42140
42141
42142
42143
42144
42145
42146
42147
42148
42149
42150
42151
42152
42153
42154
42155
42156
42157
42158
42159
42160
42161
42162
42163
42164
42165
42166
42167
42168
42169
42170
42171
42172
42173
42174
42175
42176
42177
42178
42179
42180
42181
42182
42183
42184
42185
42186
42187
42188
42189
42190
42191
42192
42193
42194
42195
42196
42197
42198
42199
42200
42201
42202
42203
42204
42205
42206
42207
42208
42209
42210
42211
42212
42213
42214
42215
42216
42217
42218
42219
42220
42221
42222
42223
42224
42225
42226
42227
42228
42229
42230
42231
42232
42233
42234
42235
42236
42237
42238
42239
42240
42241
42242
42243
42244
42245
42246
42247
42248
42249
42250
42251
42252
42253
42254
42255
42256
42257
42258
42259
42260
42261
42262
42263
42264
42265
42266
42267
42268
42269
42270
42271
42272
42273
42274
42275
42276
42277
42278
42279
42280
42281
42282
42283
42284
42285
42286
42287
42288
42289
42290
42291
42292
42293
42294
42295
42296
42297
42298
42299
42300
42301
42302
42303
42304
42305
42306
42307
42308
42309
42310
42311
42312
42313
42314
42315
42316
42317
42318
42319
42320
42321
42322
42323
42324
42325
42326
42327
42328
42329
42330
42331
42332
42333
42334
42335
42336
42337
42338
42339
42340
42341
42342
42343
42344
42345
42346
42347
42348
42349
42350
42351
42352
42353
42354
42355
42356
42357
42358
42359
42360
42361
42362
42363
42364
42365
42366
42367
42368
42369
42370
42371
42372
42373
42374
42375
42376
42377
42378
42379
42380
42381
42382
42383
42384
42385
42386
42387
42388
42389
42390
42391
42392
42393
42394
42395
42396
42397
42398
42399
42400
42401
42402
42403
42404
42405
42406
42407
42408
42409
42410
42411
42412
42413
42414
42415
42416
42417
42418
42419
42420
42421
42422
42423
42424
42425
42426
42427
42428
42429
42430
42431
42432
42433
42434
42435
42436
42437
42438
42439
42440
42441
42442
42443
42444
42445
42446
42447
42448
42449
42450
42451
42452
42453
42454
42455
42456
42457
42458
42459
42460
42461
42462
42463
42464
42465
42466
42467
42468
42469
42470
42471
42472
42473
42474
42475
42476
42477
42478
42479
42480
42481
42482
42483
42484
42485
42486
42487
42488
42489
42490
42491
42492
42493
42494
42495
42496
42497
42498
42499
42500
42501
42502
42503
42504
42505
42506
42507
42508
42509
42510
42511
42512
42513
42514
42515
42516
42517
42518
42519
42520
42521
42522
42523
42524
42525
42526
42527
42528
42529
42530
42531
42532
42533
42534
42535
42536
42537
42538
42539
42540
42541
42542
42543
42544
42545
42546
42547
42548
42549
42550
42551
42552
42553
42554
42555
42556
42557
42558
42559
42560
42561
42562
42563
42564
42565
42566
42567
42568
42569
42570
42571
42572
42573
42574
42575
42576
42577
42578
42579
42580
42581
42582
42583
42584
42585
42586
42587
42588
42589
42590
42591
42592
42593
42594
42595
42596
42597
42598
42599
42600
42601
42602
42603
42604
42605
42606
42607
42608
42609
42610
42611
42612
42613
42614
42615
42616
42617
42618
42619
42620
42621
42622
42623
42624
42625
42626
42627
42628
42629
42630
42631
42632
42633
42634
42635
42636
42637
42638
42639
42640
42641
42642
42643
42644
42645
42646
42647
42648
42649
42650
42651
42652
42653
42654
42655
42656
42657
42658
42659
42660
42661
42662
42663
42664
42665
42666
42667
42668
42669
42670
42671
42672
42673
42674
42675
42676
42677
42678
42679
42680
42681
42682
42683
42684
42685
42686
42687
42688
42689
42690
42691
42692
42693
42694
42695
42696
42697
42698
42699
42700
42701
42702
42703
42704
42705
42706
42707
42708
42709
42710
42711
42712
42713
42714
42715
42716
42717
42718
42719
42720
42721
42722
42723
42724
42725
42726
42727
42728
42729
42730
42731
42732
42733
42734
42735
42736
42737
42738
42739
42740
42741
42742
42743
42744
42745
42746
42747
42748
42749
42750
42751
42752
42753
42754
42755
42756
42757
42758
42759
42760
42761
42762
42763
42764
42765
42766
42767
42768
42769
42770
42771
42772
42773
42774
42775
42776
42777
42778
42779
42780
42781
42782
42783
42784
42785
42786
42787
42788
42789
42790
42791
42792
42793
42794
42795
42796
42797
42798
42799
42800
42801
42802
42803
42804
42805
42806
42807
42808
42809
42810
42811
42812
42813
42814
42815
42816
42817
42818
42819
42820
42821
42822
42823
42824
42825
42826
42827
42828
42829
42830
42831
42832
42833
42834
42835
42836
42837
42838
42839
42840
42841
42842
42843
42844
42845
42846
42847
42848
42849
42850
42851
42852
42853
42854
42855
42856
42857
42858
42859
42860
42861
42862
42863
42864
42865
42866
42867
42868
42869
42870
42871
42872
42873
42874
42875
42876
42877
42878
42879
42880
42881
42882
42883
42884
42885
42886
42887
42888
42889
42890
42891
42892
42893
42894
42895
42896
42897
42898
42899
42900
42901
42902
42903
42904
42905
42906
42907
42908
42909
42910
42911
42912
42913
42914
42915
42916
42917
42918
42919
42920
42921
42922
42923
42924
42925
42926
42927
42928
42929
42930
42931
42932
42933
42934
42935
42936
42937
42938
42939
42940
42941
42942
42943
42944
42945
42946
42947
42948
42949
42950
42951
42952
42953
42954
42955
42956
42957
42958
42959
42960
42961
42962
42963
42964
42965
42966
42967
42968
42969
42970
42971
42972
42973
42974
42975
42976
42977
42978
42979
42980
42981
42982
42983
42984
42985
42986
42987
42988
42989
42990
42991
42992
42993
42994
42995
42996
42997
42998
42999
43000
43001
43002
43003
43004
43005
43006
43007
43008
43009
43010
43011
43012
43013
43014
43015
43016
43017
43018
43019
43020
43021
43022
43023
43024
43025
43026
43027
43028
43029
43030
43031
43032
43033
43034
43035
43036
43037
43038
43039
43040
43041
43042
43043
43044
43045
43046
43047
43048
43049
43050
43051
43052
43053
43054
43055
43056
43057
43058
43059
43060
43061
43062
43063
43064
43065
43066
43067
43068
43069
43070
43071
43072
43073
43074
43075
43076
43077
43078
43079
43080
43081
43082
43083
43084
43085
43086
43087
43088
43089
43090
43091
43092
43093
43094
43095
43096
43097
43098
43099
43100
43101
43102
43103
43104
43105
43106
43107
43108
43109
43110
43111
43112
43113
43114
43115
43116
43117
43118
43119
43120
43121
43122
43123
43124
43125
43126
43127
43128
43129
43130
43131
43132
43133
43134
43135
43136
43137
43138
43139
43140
43141
43142
43143
43144
43145
43146
43147
43148
43149
43150
43151
43152
43153
43154
43155
43156
43157
43158
43159
43160
43161
43162
43163
43164
43165
43166
43167
43168
43169
43170
43171
43172
43173
43174
43175
43176
43177
43178
43179
43180
43181
43182
43183
43184
43185
43186
43187
43188
43189
43190
43191
43192
43193
43194
43195
43196
43197
43198
43199
43200
43201
43202
43203
43204
43205
43206
43207
43208
43209
43210
43211
43212
43213
43214
43215
43216
43217
43218
43219
43220
43221
43222
43223
43224
43225
43226
43227
43228
43229
43230
43231
43232
43233
43234
43235
43236
43237
43238
43239
43240
43241
43242
43243
43244
43245
43246
43247
43248
43249
43250
43251
43252
43253
43254
43255
43256
43257
43258
43259
43260
43261
43262
43263
43264
43265
43266
43267
43268
43269
43270
43271
43272
43273
43274
43275
43276
43277
43278
43279
43280
43281
43282
43283
43284
43285
43286
43287
43288
43289
43290
43291
43292
43293
43294
43295
43296
43297
43298
43299
43300
43301
43302
43303
43304
43305
43306
43307
43308
43309
43310
43311
43312
43313
43314
43315
43316
43317
43318
43319
43320
43321
43322
43323
43324
43325
43326
43327
43328
43329
43330
43331
43332
43333
43334
43335
43336
43337
43338
43339
43340
43341
43342
43343
43344
43345
43346
43347
43348
43349
43350
43351
43352
43353
43354
43355
43356
43357
43358
43359
43360
43361
43362
43363
43364
43365
43366
43367
43368
43369
43370
43371
43372
43373
43374
43375
43376
43377
43378
43379
43380
43381
43382
43383
43384
43385
43386
43387
43388
43389
43390
43391
43392
43393
43394
43395
43396
43397
43398
43399
43400
43401
43402
43403
43404
43405
43406
43407
43408
43409
43410
43411
43412
43413
43414
43415
43416
43417
43418
43419
43420
43421
43422
43423
43424
43425
43426
43427
43428
43429
43430
43431
43432
43433
43434
43435
43436
43437
43438
43439
43440
43441
43442
43443
43444
43445
43446
43447
43448
43449
43450
43451
43452
43453
43454
43455
43456
43457
43458
43459
43460
43461
43462
43463
43464
43465
43466
43467
43468
43469
43470
43471
43472
43473
43474
43475
43476
43477
43478
43479
43480
43481
43482
43483
43484
43485
43486
43487
43488
43489
43490
43491
43492
43493
43494
43495
43496
43497
43498
43499
43500
43501
43502
43503
43504
43505
43506
43507
43508
43509
43510
43511
43512
43513
43514
43515
43516
43517
43518
43519
43520
43521
43522
43523
43524
43525
43526
43527
43528
43529
43530
43531
43532
43533
43534
43535
43536
43537
43538
43539
43540
43541
43542
43543
43544
43545
43546
43547
43548
43549
43550
43551
43552
43553
43554
43555
43556
43557
43558
43559
43560
43561
43562
43563
43564
43565
43566
43567
43568
43569
43570
43571
43572
43573
43574
43575
43576
43577
43578
43579
43580
43581
43582
43583
43584
43585
43586
43587
43588
43589
43590
43591
43592
43593
43594
43595
43596
43597
43598
43599
43600
43601
43602
43603
43604
43605
43606
43607
43608
43609
43610
43611
43612
43613
43614
43615
43616
43617
43618
43619
43620
43621
43622
43623
43624
43625
43626
43627
43628
43629
43630
43631
43632
43633
43634
43635
43636
43637
43638
43639
43640
43641
43642
43643
43644
43645
43646
43647
43648
43649
43650
43651
43652
43653
43654
43655
43656
43657
43658
43659
43660
43661
43662
43663
43664
43665
43666
43667
43668
43669
43670
43671
43672
43673
43674
43675
43676
43677
43678
43679
43680
43681
43682
43683
43684
43685
43686
43687
43688
43689
43690
43691
43692
43693
43694
43695
43696
43697
43698
43699
43700
43701
43702
43703
43704
43705
43706
43707
43708
43709
43710
43711
43712
43713
43714
43715
43716
43717
43718
43719
43720
43721
43722
43723
43724
43725
43726
43727
43728
43729
43730
43731
43732
43733
43734
43735
43736
43737
43738
43739
43740
43741
43742
43743
43744
43745
43746
43747
43748
43749
43750
43751
43752
43753
43754
43755
43756
43757
43758
43759
43760
43761
43762
43763
43764
43765
43766
43767
43768
43769
43770
43771
43772
43773
43774
43775
43776
43777
43778
43779
43780
43781
43782
43783
43784
43785
43786
43787
43788
43789
43790
43791
43792
43793
43794
43795
43796
43797
43798
43799
43800
43801
43802
43803
43804
43805
43806
43807
43808
43809
43810
43811
43812
43813
43814
43815
43816
43817
43818
43819
43820
43821
43822
43823
43824
43825
43826
43827
43828
43829
43830
43831
43832
43833
43834
43835
43836
43837
43838
43839
43840
43841
43842
43843
43844
43845
43846
43847
43848
43849
43850
43851
43852
43853
43854
43855
43856
43857
43858
43859
43860
43861
43862
43863
43864
43865
43866
43867
43868
43869
43870
43871
43872
43873
43874
43875
43876
43877
43878
43879
43880
43881
43882
43883
43884
43885
43886
43887
43888
43889
43890
43891
43892
43893
43894
43895
43896
43897
43898
43899
43900
43901
43902
43903
43904
43905
43906
43907
43908
43909
43910
43911
43912
43913
43914
43915
43916
43917
43918
43919
43920
43921
43922
43923
43924
43925
43926
43927
43928
43929
43930
43931
43932
43933
43934
43935
43936
43937
43938
43939
43940
43941
43942
43943
43944
43945
43946
43947
43948
43949
43950
43951
43952
43953
43954
43955
43956
43957
43958
43959
43960
43961
43962
43963
43964
43965
43966
43967
43968
43969
43970
43971
43972
43973
43974
43975
43976
43977
43978
43979
43980
43981
43982
43983
43984
43985
43986
43987
43988
43989
43990
43991
43992
43993
43994
43995
43996
43997
43998
43999
44000
44001
44002
44003
44004
44005
44006
44007
44008
44009
44010
44011
44012
44013
44014
44015
44016
44017
44018
44019
44020
44021
44022
44023
44024
44025
44026
44027
44028
44029
44030
44031
44032
44033
44034
44035
44036
44037
44038
44039
44040
44041
44042
44043
44044
44045
44046
44047
44048
44049
44050
44051
44052
44053
44054
44055
44056
44057
44058
44059
44060
44061
44062
44063
44064
44065
44066
44067
44068
44069
44070
44071
44072
44073
44074
44075
44076
44077
44078
44079
44080
44081
44082
44083
44084
44085
44086
44087
44088
44089
44090
44091
44092
44093
44094
44095
44096
44097
44098
44099
44100
44101
44102
44103
44104
44105
44106
44107
44108
44109
44110
44111
44112
44113
44114
44115
44116
44117
44118
44119
44120
44121
44122
44123
44124
44125
44126
44127
44128
44129
44130
44131
44132
44133
44134
44135
44136
44137
44138
44139
44140
44141
44142
44143
44144
44145
44146
44147
44148
44149
44150
44151
44152
44153
44154
44155
44156
44157
44158
44159
44160
44161
44162
44163
44164
44165
44166
44167
44168
44169
44170
44171
44172
44173
44174
44175
44176
44177
44178
44179
44180
44181
44182
44183
44184
44185
44186
44187
44188
44189
44190
44191
44192
44193
44194
44195
44196
44197
44198
44199
44200
44201
44202
44203
44204
44205
44206
44207
44208
44209
44210
44211
44212
44213
44214
44215
44216
44217
44218
44219
44220
44221
44222
44223
44224
44225
44226
44227
44228
44229
44230
44231
44232
44233
44234
44235
44236
44237
44238
44239
44240
44241
44242
44243
44244
44245
44246
44247
44248
44249
44250
44251
44252
44253
44254
44255
44256
44257
44258
44259
44260
44261
44262
44263
44264
44265
44266
44267
44268
44269
44270
44271
44272
44273
44274
44275
44276
44277
44278
44279
44280
44281
44282
44283
44284
44285
44286
44287
44288
44289
44290
44291
44292
44293
44294
44295
44296
44297
44298
44299
44300
44301
44302
44303
44304
44305
44306
44307
44308
44309
44310
44311
44312
44313
44314
44315
44316
44317
44318
44319
44320
44321
44322
44323
44324
44325
44326
44327
44328
44329
44330
44331
44332
44333
44334
44335
44336
44337
44338
44339
44340
44341
44342
44343
44344
44345
44346
44347
44348
44349
44350
44351
44352
44353
44354
44355
44356
44357
44358
44359
44360
44361
44362
44363
44364
44365
44366
44367
44368
44369
44370
44371
44372
44373
44374
44375
44376
44377
44378
44379
44380
44381
44382
44383
44384
44385
44386
44387
44388
44389
44390
44391
44392
44393
44394
44395
44396
44397
44398
44399
44400
44401
44402
44403
44404
44405
44406
44407
44408
44409
44410
44411
44412
44413
44414
44415
44416
44417
44418
44419
44420
44421
44422
44423
44424
44425
44426
44427
44428
44429
44430
44431
44432
44433
44434
44435
44436
44437
44438
44439
44440
44441
44442
44443
44444
44445
44446
44447
44448
44449
44450
44451
44452
44453
44454
44455
44456
44457
44458
44459
44460
44461
44462
44463
44464
44465
44466
44467
44468
44469
44470
44471
44472
44473
44474
44475
44476
44477
44478
44479
44480
44481
44482
44483
44484
44485
44486
44487
44488
44489
44490
44491
44492
44493
44494
44495
44496
44497
44498
44499
44500
44501
44502
44503
44504
44505
44506
44507
44508
44509
44510
44511
44512
44513
44514
44515
44516
44517
44518
44519
44520
44521
44522
44523
44524
44525
44526
44527
44528
44529
44530
44531
44532
44533
44534
44535
44536
44537
44538
44539
44540
44541
44542
44543
44544
44545
44546
44547
44548
44549
44550
44551
44552
44553
44554
44555
44556
44557
44558
44559
44560
44561
44562
44563
44564
44565
44566
44567
44568
44569
44570
44571
44572
44573
44574
44575
44576
44577
44578
44579
44580
44581
44582
44583
44584
44585
44586
44587
44588
44589
44590
44591
44592
44593
44594
44595
44596
44597
44598
44599
44600
44601
44602
44603
44604
44605
44606
44607
44608
44609
44610
44611
44612
44613
44614
44615
44616
44617
44618
44619
44620
44621
44622
44623
44624
44625
44626
44627
44628
44629
44630
44631
44632
44633
44634
44635
44636
44637
44638
44639
44640
44641
44642
44643
44644
44645
44646
44647
44648
44649
44650
44651
44652
44653
44654
44655
44656
44657
44658
44659
44660
44661
44662
44663
44664
44665
44666
44667
44668
44669
44670
44671
44672
44673
44674
44675
44676
44677
44678
44679
44680
44681
44682
44683
44684
44685
44686
44687
44688
44689
44690
44691
44692
44693
44694
44695
44696
44697
44698
44699
44700
44701
44702
44703
44704
44705
44706
44707
44708
44709
44710
44711
44712
44713
44714
44715
44716
44717
44718
44719
44720
44721
44722
44723
44724
44725
44726
44727
44728
44729
44730
44731
44732
44733
44734
44735
44736
44737
44738
44739
44740
44741
44742
44743
44744
44745
44746
44747
44748
44749
44750
44751
44752
44753
44754
44755
44756
44757
44758
44759
44760
44761
44762
44763
44764
44765
44766
44767
44768
44769
44770
44771
44772
44773
44774
44775
44776
44777
44778
44779
44780
44781
44782
44783
44784
44785
44786
44787
44788
44789
44790
44791
44792
44793
44794
44795
44796
44797
44798
44799
44800
44801
44802
44803
44804
44805
44806
44807
44808
44809
44810
44811
44812
44813
44814
44815
44816
44817
44818
44819
44820
44821
44822
44823
44824
44825
44826
44827
44828
44829
44830
44831
44832
44833
44834
44835
44836
44837
44838
44839
44840
44841
44842
44843
44844
44845
44846
44847
44848
44849
44850
44851
44852
44853
44854
44855
44856
44857
44858
44859
44860
44861
44862
44863
44864
44865
44866
44867
44868
44869
44870
44871
44872
44873
44874
44875
44876
44877
44878
44879
44880
44881
44882
44883
44884
44885
44886
44887
44888
44889
44890
44891
44892
44893
44894
44895
44896
44897
44898
44899
44900
44901
44902
44903
44904
44905
44906
44907
44908
44909
44910
44911
44912
44913
44914
44915
44916
44917
44918
44919
44920
44921
44922
44923
44924
44925
44926
44927
44928
44929
44930
44931
44932
44933
44934
44935
44936
44937
44938
44939
44940
44941
44942
44943
44944
44945
44946
44947
44948
44949
44950
44951
44952
44953
44954
44955
44956
44957
44958
44959
44960
44961
44962
44963
44964
44965
44966
44967
44968
44969
44970
44971
44972
44973
44974
44975
44976
44977
44978
44979
44980
44981
44982
44983
44984
44985
44986
44987
44988
44989
44990
44991
44992
44993
44994
44995
44996
44997
44998
44999
45000
45001
45002
45003
45004
45005
45006
45007
45008
45009
45010
45011
45012
45013
45014
45015
45016
45017
45018
45019
45020
45021
45022
45023
45024
45025
45026
45027
45028
45029
45030
45031
45032
45033
45034
45035
45036
45037
45038
45039
45040
45041
45042
45043
45044
45045
45046
45047
45048
45049
45050
45051
45052
45053
45054
45055
45056
45057
45058
45059
45060
45061
45062
45063
45064
45065
45066
45067
45068
45069
45070
45071
45072
45073
45074
45075
45076
45077
45078
45079
45080
45081
45082
45083
45084
45085
45086
45087
45088
45089
45090
45091
45092
45093
45094
45095
45096
45097
45098
45099
45100
45101
45102
45103
45104
45105
45106
45107
45108
45109
45110
45111
45112
45113
45114
45115
45116
45117
45118
45119
45120
45121
45122
45123
45124
45125
45126
45127
45128
45129
45130
45131
45132
45133
45134
45135
45136
45137
45138
45139
45140
45141
45142
45143
45144
45145
45146
45147
45148
45149
45150
45151
45152
45153
45154
45155
45156
45157
45158
45159
45160
45161
45162
45163
45164
45165
45166
45167
45168
45169
45170
45171
45172
45173
45174
45175
45176
45177
45178
45179
45180
45181
45182
45183
45184
45185
45186
45187
45188
45189
45190
45191
45192
45193
45194
45195
45196
45197
45198
45199
45200
45201
45202
45203
45204
45205
45206
45207
45208
45209
45210
45211
45212
45213
45214
45215
45216
45217
45218
45219
45220
45221
45222
45223
45224
45225
45226
45227
45228
45229
45230
45231
45232
45233
45234
45235
45236
45237
45238
45239
45240
45241
45242
45243
45244
45245
45246
45247
45248
45249
45250
45251
45252
45253
45254
45255
45256
45257
45258
45259
45260
45261
45262
45263
45264
45265
45266
45267
45268
45269
45270
45271
45272
45273
45274
45275
45276
45277
45278
45279
45280
45281
45282
45283
45284
45285
45286
45287
45288
45289
45290
45291
45292
45293
45294
45295
45296
45297
45298
45299
45300
45301
45302
45303
45304
45305
45306
45307
45308
45309
45310
45311
45312
45313
45314
45315
45316
45317
45318
45319
45320
45321
45322
45323
45324
45325
45326
45327
45328
45329
45330
45331
45332
45333
45334
45335
45336
45337
45338
45339
45340
45341
45342
45343
45344
45345
45346
45347
45348
45349
45350
45351
45352
45353
45354
45355
45356
45357
45358
45359
45360
45361
45362
45363
45364
45365
45366
45367
45368
45369
45370
45371
45372
45373
45374
45375
45376
45377
45378
45379
45380
45381
45382
45383
45384
45385
45386
45387
45388
45389
45390
45391
45392
45393
45394
45395
45396
45397
45398
45399
45400
45401
45402
45403
45404
45405
45406
45407
45408
45409
45410
45411
45412
45413
45414
45415
45416
45417
45418
45419
45420
45421
45422
45423
45424
45425
45426
45427
45428
45429
45430
45431
45432
45433
45434
45435
45436
45437
45438
45439
45440
45441
45442
45443
45444
45445
45446
45447
45448
45449
45450
45451
45452
45453
45454
45455
45456
45457
45458
45459
45460
45461
45462
45463
45464
45465
45466
45467
45468
45469
45470
45471
45472
45473
45474
45475
45476
45477
45478
45479
45480
45481
45482
45483
45484
45485
45486
45487
45488
45489
45490
45491
45492
45493
45494
45495
45496
45497
45498
45499
45500
45501
45502
45503
45504
45505
45506
45507
45508
45509
45510
45511
45512
45513
45514
45515
45516
45517
45518
45519
45520
45521
45522
45523
45524
45525
45526
45527
45528
45529
45530
45531
45532
45533
45534
45535
45536
45537
45538
45539
45540
45541
45542
45543
45544
45545
45546
45547
45548
45549
45550
45551
45552
45553
45554
45555
45556
45557
45558
45559
45560
45561
45562
45563
45564
45565
45566
45567
45568
45569
45570
45571
45572
45573
45574
45575
45576
45577
45578
45579
45580
45581
45582
45583
45584
45585
45586
45587
45588
45589
45590
45591
45592
45593
45594
45595
45596
45597
45598
45599
45600
45601
45602
45603
45604
45605
45606
45607
45608
45609
45610
45611
45612
45613
45614
45615
45616
45617
45618
45619
45620
45621
45622
45623
45624
45625
45626
45627
45628
45629
45630
45631
45632
45633
45634
45635
45636
45637
45638
45639
45640
45641
45642
45643
45644
45645
45646
45647
45648
45649
45650
45651
45652
45653
45654
45655
45656
45657
45658
45659
45660
45661
45662
45663
45664
45665
45666
45667
45668
45669
45670
45671
45672
45673
45674
45675
45676
45677
45678
45679
45680
45681
45682
45683
45684
45685
45686
45687
45688
45689
45690
45691
45692
45693
45694
45695
45696
45697
45698
45699
45700
45701
45702
45703
45704
45705
45706
45707
45708
45709
45710
45711
45712
45713
45714
45715
45716
45717
45718
45719
45720
45721
45722
45723
45724
45725
45726
45727
45728
45729
45730
45731
45732
45733
45734
45735
45736
45737
45738
45739
45740
45741
45742
45743
45744
45745
45746
45747
45748
45749
45750
45751
45752
45753
45754
45755
45756
45757
45758
45759
45760
45761
45762
45763
45764
45765
45766
45767
45768
45769
45770
45771
45772
45773
45774
45775
45776
45777
45778
45779
45780
45781
45782
45783
45784
45785
45786
45787
45788
45789
45790
45791
45792
45793
45794
45795
45796
45797
45798
45799
45800
45801
45802
45803
45804
45805
45806
45807
45808
45809
45810
45811
45812
45813
45814
45815
45816
45817
45818
45819
45820
45821
45822
45823
45824
45825
45826
45827
45828
45829
45830
45831
45832
45833
45834
45835
45836
45837
45838
45839
45840
45841
45842
45843
45844
45845
45846
45847
45848
45849
45850
45851
45852
45853
45854
45855
45856
45857
45858
45859
45860
45861
45862
45863
45864
45865
45866
45867
45868
45869
45870
45871
45872
45873
45874
45875
45876
45877
45878
45879
45880
45881
45882
45883
45884
45885
45886
45887
45888
45889
45890
45891
45892
45893
45894
45895
45896
45897
45898
45899
45900
45901
45902
45903
45904
45905
45906
45907
45908
45909
45910
45911
45912
45913
45914
45915
45916
45917
45918
45919
45920
45921
45922
45923
45924
45925
45926
45927
45928
45929
45930
45931
45932
45933
45934
45935
45936
45937
45938
45939
45940
45941
45942
45943
45944
45945
45946
45947
45948
45949
45950
45951
45952
45953
45954
45955
45956
45957
45958
45959
45960
45961
45962
45963
45964
45965
45966
45967
45968
45969
45970
45971
45972
45973
45974
45975
45976
45977
45978
45979
45980
45981
45982
45983
45984
45985
45986
45987
45988
45989
45990
45991
45992
45993
45994
45995
45996
45997
45998
45999
46000
46001
46002
46003
46004
46005
46006
46007
46008
46009
46010
46011
46012
46013
46014
46015
46016
46017
46018
46019
46020
46021
46022
46023
46024
46025
46026
46027
46028
46029
46030
46031
46032
46033
46034
46035
46036
46037
46038
46039
46040
46041
46042
46043
46044
46045
46046
46047
46048
46049
46050
46051
46052
46053
46054
46055
46056
46057
46058
46059
46060
46061
46062
46063
46064
46065
46066
46067
46068
46069
46070
46071
46072
46073
46074
46075
46076
46077
46078
46079
46080
46081
46082
46083
46084
46085
46086
46087
46088
46089
46090
46091
46092
46093
46094
46095
46096
46097
46098
46099
46100
46101
46102
46103
46104
46105
46106
46107
46108
46109
46110
46111
46112
46113
46114
46115
46116
46117
46118
46119
46120
46121
46122
46123
46124
46125
46126
46127
46128
46129
46130
46131
46132
46133
46134
46135
46136
46137
46138
46139
46140
46141
46142
46143
46144
46145
46146
46147
46148
46149
46150
46151
46152
46153
46154
46155
46156
46157
46158
46159
46160
46161
46162
46163
46164
46165
46166
46167
46168
46169
46170
46171
46172
46173
46174
46175
46176
46177
46178
46179
46180
46181
46182
46183
46184
46185
46186
46187
46188
46189
46190
46191
46192
46193
46194
46195
46196
46197
46198
46199
46200
46201
46202
46203
46204
46205
46206
46207
46208
46209
46210
46211
46212
46213
46214
46215
46216
46217
46218
46219
46220
46221
46222
46223
46224
46225
46226
46227
46228
46229
46230
46231
46232
46233
46234
46235
46236
46237
46238
46239
46240
46241
46242
46243
46244
46245
46246
46247
46248
46249
46250
46251
46252
46253
46254
46255
46256
46257
46258
46259
46260
46261
46262
46263
46264
46265
46266
46267
46268
46269
46270
46271
46272
46273
46274
46275
46276
46277
46278
46279
46280
46281
46282
46283
46284
46285
46286
46287
46288
46289
46290
46291
46292
46293
46294
46295
46296
46297
46298
46299
46300
46301
46302
46303
46304
46305
46306
46307
46308
46309
46310
46311
46312
46313
46314
46315
46316
46317
46318
46319
46320
46321
46322
46323
46324
46325
46326
46327
46328
46329
46330
46331
46332
46333
46334
46335
46336
46337
46338
46339
46340
46341
46342
46343
46344
46345
46346
46347
46348
46349
46350
46351
46352
46353
46354
46355
46356
46357
46358
46359
46360
46361
46362
46363
46364
46365
46366
46367
46368
46369
46370
46371
46372
46373
46374
46375
46376
46377
46378
46379
46380
46381
46382
46383
46384
46385
46386
46387
46388
46389
46390
46391
46392
46393
46394
46395
46396
46397
46398
46399
46400
46401
46402
46403
46404
46405
46406
46407
46408
46409
46410
46411
46412
46413
46414
46415
46416
46417
46418
46419
46420
46421
46422
46423
46424
46425
46426
46427
46428
46429
46430
46431
46432
46433
46434
46435
46436
46437
46438
46439
46440
46441
46442
46443
46444
46445
46446
46447
46448
46449
46450
46451
46452
46453
46454
46455
46456
46457
46458
46459
46460
46461
46462
46463
46464
46465
46466
46467
46468
46469
46470
46471
46472
46473
46474
46475
46476
46477
46478
46479
46480
46481
46482
46483
46484
46485
46486
46487
46488
46489
46490
46491
46492
46493
46494
46495
46496
46497
46498
46499
46500
46501
46502
46503
46504
46505
46506
46507
46508
46509
46510
46511
46512
46513
46514
46515
46516
46517
46518
46519
46520
46521
46522
46523
46524
46525
46526
46527
46528
46529
46530
46531
46532
46533
46534
46535
46536
46537
46538
46539
46540
46541
46542
46543
46544
46545
46546
46547
46548
46549
46550
46551
46552
46553
46554
46555
46556
46557
46558
46559
46560
46561
46562
46563
46564
46565
46566
46567
46568
46569
46570
46571
46572
46573
46574
46575
46576
46577
46578
46579
46580
46581
46582
46583
46584
46585
46586
46587
46588
46589
46590
46591
46592
46593
46594
46595
46596
46597
46598
46599
46600
46601
46602
46603
46604
46605
46606
46607
46608
46609
46610
46611
46612
46613
46614
46615
46616
46617
46618
46619
46620
46621
46622
46623
46624
46625
46626
46627
46628
46629
46630
46631
46632
46633
46634
46635
46636
46637
46638
46639
46640
46641
46642
46643
46644
46645
46646
46647
46648
46649
46650
46651
46652
46653
46654
46655
46656
46657
46658
46659
46660
46661
46662
46663
46664
46665
46666
46667
46668
46669
46670
46671
46672
46673
46674
46675
46676
46677
46678
46679
46680
46681
46682
46683
46684
46685
46686
46687
46688
46689
46690
46691
46692
46693
46694
46695
46696
46697
46698
46699
46700
46701
46702
46703
46704
46705
46706
46707
46708
46709
46710
46711
46712
46713
46714
46715
46716
46717
46718
46719
46720
46721
46722
46723
46724
46725
46726
46727
46728
46729
46730
46731
46732
46733
46734
46735
46736
46737
46738
46739
46740
46741
46742
46743
46744
46745
46746
46747
46748
46749
46750
46751
46752
46753
46754
46755
46756
46757
46758
46759
46760
46761
46762
46763
46764
46765
46766
46767
46768
46769
46770
46771
46772
46773
46774
46775
46776
46777
46778
46779
46780
46781
46782
46783
46784
46785
46786
46787
46788
46789
46790
46791
46792
46793
46794
46795
46796
46797
46798
46799
46800
46801
46802
46803
46804
46805
46806
46807
46808
46809
46810
46811
46812
46813
46814
46815
46816
46817
46818
46819
46820
46821
46822
46823
46824
46825
46826
46827
46828
46829
46830
46831
46832
46833
46834
46835
46836
46837
46838
46839
46840
46841
46842
46843
46844
46845
46846
46847
46848
46849
46850
46851
46852
46853
46854
46855
46856
46857
46858
46859
46860
46861
46862
46863
46864
46865
46866
46867
46868
46869
46870
46871
46872
46873
46874
46875
46876
46877
46878
46879
46880
46881
46882
46883
46884
46885
46886
46887
46888
46889
46890
46891
46892
46893
46894
46895
46896
46897
46898
46899
46900
46901
46902
46903
46904
46905
46906
46907
46908
46909
46910
46911
46912
46913
46914
46915
46916
46917
46918
46919
46920
46921
46922
46923
46924
46925
46926
46927
46928
46929
46930
46931
46932
46933
46934
46935
46936
46937
46938
46939
46940
46941
46942
46943
46944
46945
46946
46947
46948
46949
46950
46951
46952
46953
46954
46955
46956
46957
46958
46959
46960
46961
46962
46963
46964
46965
46966
46967
46968
46969
46970
46971
46972
46973
46974
46975
46976
46977
46978
46979
46980
46981
46982
46983
46984
46985
46986
46987
46988
46989
46990
46991
46992
46993
46994
46995
46996
46997
46998
46999
47000
47001
47002
47003
47004
47005
47006
47007
47008
47009
47010
47011
47012
47013
47014
47015
47016
47017
47018
47019
47020
47021
47022
47023
47024
47025
47026
47027
47028
47029
47030
47031
47032
47033
47034
47035
47036
47037
47038
47039
47040
47041
47042
47043
47044
47045
47046
47047
47048
47049
47050
47051
47052
47053
47054
47055
47056
47057
47058
47059
47060
47061
47062
47063
47064
47065
47066
47067
47068
47069
47070
47071
47072
47073
47074
47075
47076
47077
47078
47079
47080
47081
47082
47083
47084
47085
47086
47087
47088
47089
47090
47091
47092
47093
47094
47095
47096
47097
47098
47099
47100
47101
47102
47103
47104
47105
47106
47107
47108
47109
47110
47111
47112
47113
47114
47115
47116
47117
47118
47119
47120
47121
47122
47123
47124
47125
47126
47127
47128
47129
47130
47131
47132
47133
47134
47135
47136
47137
47138
47139
47140
47141
47142
47143
47144
47145
47146
47147
47148
47149
47150
47151
47152
47153
47154
47155
47156
47157
47158
47159
47160
47161
47162
47163
47164
47165
47166
47167
47168
47169
47170
47171
47172
47173
47174
47175
47176
47177
47178
47179
47180
47181
47182
47183
47184
47185
47186
47187
47188
47189
47190
47191
47192
47193
47194
47195
47196
47197
47198
47199
47200
47201
47202
47203
47204
47205
47206
47207
47208
47209
47210
47211
47212
47213
47214
47215
47216
47217
47218
47219
47220
47221
47222
47223
47224
47225
47226
47227
47228
47229
47230
47231
47232
47233
47234
47235
47236
47237
47238
47239
47240
47241
47242
47243
47244
47245
47246
47247
47248
47249
47250
47251
47252
47253
47254
47255
47256
47257
47258
47259
47260
47261
47262
47263
47264
47265
47266
47267
47268
47269
47270
47271
47272
47273
47274
47275
47276
47277
47278
47279
47280
47281
47282
47283
47284
47285
47286
47287
47288
47289
47290
47291
47292
47293
47294
47295
47296
47297
47298
47299
47300
47301
47302
47303
47304
47305
47306
47307
47308
47309
47310
47311
47312
47313
47314
47315
47316
47317
47318
47319
47320
47321
47322
47323
47324
47325
47326
47327
47328
47329
47330
47331
47332
47333
47334
47335
47336
47337
47338
47339
47340
47341
47342
47343
47344
47345
47346
47347
47348
47349
47350
47351
47352
47353
47354
47355
47356
47357
47358
47359
47360
47361
47362
47363
47364
47365
47366
47367
47368
47369
47370
47371
47372
47373
47374
47375
47376
47377
47378
47379
47380
47381
47382
47383
47384
47385
47386
47387
47388
47389
47390
47391
47392
47393
47394
47395
47396
47397
47398
47399
47400
47401
47402
47403
47404
47405
47406
47407
47408
47409
47410
47411
47412
47413
47414
47415
47416
47417
47418
47419
47420
47421
47422
47423
47424
47425
47426
47427
47428
47429
47430
47431
47432
47433
47434
47435
47436
47437
47438
47439
47440
47441
47442
47443
47444
47445
47446
47447
47448
47449
47450
47451
47452
47453
47454
47455
47456
47457
47458
47459
47460
47461
47462
47463
47464
47465
47466
47467
47468
47469
47470
47471
47472
47473
47474
47475
47476
47477
47478
47479
47480
47481
47482
47483
47484
47485
47486
47487
47488
47489
47490
47491
47492
47493
47494
47495
47496
47497
47498
47499
47500
47501
47502
47503
47504
47505
47506
47507
47508
47509
47510
47511
47512
47513
47514
47515
47516
47517
47518
47519
47520
47521
47522
47523
47524
47525
47526
47527
47528
47529
47530
47531
47532
47533
47534
47535
47536
47537
47538
47539
47540
47541
47542
47543
47544
47545
47546
47547
47548
47549
47550
47551
47552
47553
47554
47555
47556
47557
47558
47559
47560
47561
47562
47563
47564
47565
47566
47567
47568
47569
47570
47571
47572
47573
47574
47575
47576
47577
47578
47579
47580
47581
47582
47583
47584
47585
47586
47587
47588
47589
47590
47591
47592
47593
47594
47595
47596
47597
47598
47599
47600
47601
47602
47603
47604
47605
47606
47607
47608
47609
47610
47611
47612
47613
47614
47615
47616
47617
47618
47619
47620
47621
47622
47623
47624
47625
47626
47627
47628
47629
47630
47631
47632
47633
47634
47635
47636
47637
47638
47639
47640
47641
47642
47643
47644
47645
47646
47647
47648
47649
47650
47651
47652
47653
47654
47655
47656
47657
47658
47659
47660
47661
47662
47663
47664
47665
47666
47667
47668
47669
47670
47671
47672
47673
47674
47675
47676
47677
47678
47679
47680
47681
47682
47683
47684
47685
47686
47687
47688
47689
47690
47691
47692
47693
47694
47695
47696
47697
47698
47699
47700
47701
47702
47703
47704
47705
47706
47707
47708
47709
47710
47711
47712
47713
47714
47715
47716
47717
47718
47719
47720
47721
47722
47723
47724
47725
47726
47727
47728
47729
47730
47731
47732
47733
47734
47735
47736
47737
47738
47739
47740
47741
47742
47743
47744
47745
47746
47747
47748
47749
47750
47751
47752
47753
47754
47755
47756
47757
47758
47759
47760
47761
47762
47763
47764
47765
47766
47767
47768
47769
47770
47771
47772
47773
47774
47775
47776
47777
47778
47779
47780
47781
47782
47783
47784
47785
47786
47787
47788
47789
47790
47791
47792
47793
47794
47795
47796
47797
47798
47799
47800
47801
47802
47803
47804
47805
47806
47807
47808
47809
47810
47811
47812
47813
47814
47815
47816
47817
47818
47819
47820
47821
47822
47823
47824
47825
47826
47827
47828
47829
47830
47831
47832
47833
47834
47835
47836
47837
47838
47839
47840
47841
47842
47843
47844
47845
47846
47847
47848
47849
47850
47851
47852
47853
47854
47855
47856
47857
47858
47859
47860
47861
47862
47863
47864
47865
47866
47867
47868
47869
47870
47871
47872
47873
47874
47875
47876
47877
47878
47879
47880
47881
47882
47883
47884
47885
47886
47887
47888
47889
47890
47891
47892
47893
47894
47895
47896
47897
47898
47899
47900
47901
47902
47903
47904
47905
47906
47907
47908
47909
47910
47911
47912
47913
47914
47915
47916
47917
47918
47919
47920
47921
47922
47923
47924
47925
47926
47927
47928
47929
47930
47931
47932
47933
47934
47935
47936
47937
47938
47939
47940
47941
47942
47943
47944
47945
47946
47947
47948
47949
47950
47951
47952
47953
47954
47955
47956
47957
47958
47959
47960
47961
47962
47963
47964
47965
47966
47967
47968
47969
47970
47971
47972
47973
47974
47975
47976
47977
47978
47979
47980
47981
47982
47983
47984
47985
47986
47987
47988
47989
47990
47991
47992
47993
47994
47995
47996
47997
47998
47999
48000
48001
48002
48003
48004
48005
48006
48007
48008
48009
48010
48011
% -*- coding: utf-8; time-stamp-format: "%02d-%02m-%:y at %02H:%02M:%02S %Z" -*-
% This file: xint.dtx. Proudly produced by xint-dtxbuild.sh.
% Extract all files via "etex xint.dtx" and do "make help"
% or follow instructions from extracted README.md.
%<*dtx>
\def\xintdtxtimestamp {Time-stamp: <13-07-2021 at 21:50:14 CEST>}
%</dtx>
%<*drv>
%% ---------------------------------------------------------------
\def\xintdocdate {2021/07/13}
\def\xintbndldate{2021/07/13}
\def\xintbndlversion {1.4j}
%</drv>
%<readme>% README
%<changes>% CHANGE LOG
%<readme|changes>% xint 1.4j
%<readme|changes>% 2021/07/13
%<readme|changes>
%<readme|changes>    Source:  xint.dtx 1.4j 2021/07/13 (doc 2021/07/13)
%<readme|changes>    Author:  Jean-Francois Burnol
%<readme|changes>    Info:    Expandable operations on big integers, decimals, fractions
%<readme|changes>    License: LPPL 1.3c
%<readme|changes>
%<*!readme&!changes&!dohtmlsh&!makefile>
%% ---------------------------------------------------------------
%% The xint bundle 1.4j 2021/07/13
%% Copyright (C) 2013-2021 by Jean-Francois Burnol
%<xintkernel>%% xintkernel: Paraphernalia for the xint packages
%<xinttools>%% xinttools: Expandable and non-expandable utilities
%<xintcore>%% xintcore: Expandable arithmetic on big integers
%<xint>%% xint: Expandable operations on big integers
%<xintfrac>%% xintfrac: Expandable operations on fractions
%<xintexpr>%% xintexpr: Expandable expression parser
%<xintbinhex>%% xintbinhex: Expandable binary and hexadecimal conversions
%<xintgcd>%% xintgcd: Euclidean algorithm with xint package
%<xintseries>%% xintseries: Expandable partial sums with xint package
%<xintcfrac>%% xintcfrac: Expandable continued fractions with xint package
%<xinttrig>%% xinttrig: Trigonometry for the xintexpr package
%<xintlog>%% xintlog: Logarithms and exponentials for xintexpr
%% ---------------------------------------------------------------
%</!readme&!changes&!dohtmlsh&!makefile>
%<*dtx>
\bgroup\catcode2 0 \catcode`\\ 12 ^^Biffalse
%</dtx>
%<*readme>--------------------------------------------------------

Description
===========

It is possible to use the package both with Plain (`\input xintexpr.sty`)
or with the LaTeX macro format (`\usepackage{xintexpr}`).

The basic aim is provide *expandable* computations on (arbitrarily big)
integers, fractions, and floating point numbers (at a user chosen
precision).  The four operations and the square-root extraction achieve
the *correct rounding* for the given arbitrary precision.  Exponential
(natural and to the base ten), logarithm (also to the base 10),
fractional powers, direct and inverse trigonometrical functions are
available up to 62 digits of precision.  The syntax supports dummy
variables (to generate sequences of values) and nested structures.
Support for user-declared functions and variables is implemented.

Usage on the command line
=========================

One can use `xintexpr` as an interactive calculator on the command line.
See the [xintsession](http://ctan.org/pkg/xintsession) package.

    >>> 2^100;
    @_1     1267650600228229401496703205376
    >>> cos(1);
    @_2     0.5403023058681397
    >>> &fp=32
    (/usr/local/texlive/2021/texmf-dist/tex/generic/xint/xintlog.sty)
    (/usr/local/texlive/2021/texmf-dist/tex/generic/xint/xinttrig.sty)
    fp mode (log and trig reloaded at Digits=32)
    >>> cos(1);
    @_3     0.54030230586813971740093660744298
    >>> 3^1000;
    @_4     1.3220708194808066368904552597521e477
    >>> &exact
    exact mode (floating point evaluations use 32 digits)
    >>> 3^1000;
    @_5     132207081948080663689045525975... (trimmed for this README)

Installation
============

`xint` is included in [TeXLive](http://tug.org/texlive/) (hence
[MacTeX](http://tug.org/mactex/) also) and in
[MikTeX](http://www.miktex.org/). Thus, use the package manager to
update your distribution.

Alternatives:

- download
   [`xint.tds.zip`](http://mirror.ctan.org/install/macros/generic/xint.tds.zip)
   and install in a suitable TDS-compliant repertory via `unzip`. "admin"
   privilges might be needed, as well as a file database rebuild (`texhash`).
   For example, on macos x, installation into user home folder (no `sudo`,
   and no `texhash` as it is recommended to not have a ls-R file there)

        unzip xint.tds.zip -d  ~/Library/texmf

- all files can be extracted using `etex xint.dtx`, or `make` if the
  `Makefile` included in the CTAN upload is present; see the file `INSTALL`,
  if present, else read the help in extracted file `Makefile.mk`.

Documentation
=============

`README.md`: this file

`CHANGES.html`: change log as relevant to end users
  (`texdoc --list xint`)

`xint.pdf`: user manual

`sourcexint.pdf`: commented source code
   (`texdoc --list xint` or `texdoc sourcexint`)

Requirements
============

Since release `1.4`, `xintexpr` requires the `\expanded` primitive. This
is a functionality of all major TeX engines since TeXLive 2019.

License
=======

Copyright (C) 2013-2021 by Jean-Francois Burnol

This Work may be distributed and/or modified under the
conditions of the LaTeX Project Public License version 1.3c.
This version of this license is in

> <http://www.latex-project.org/lppl/lppl-1-3c.txt>

and version 1.3 or later is part of all distributions of
LaTeX version 2005/12/01 or later.

This Work has the LPPL maintenance status `author-maintained`.

The Author of this Work is `Jean-Francois Burnol`.

This Work consists of the files `Makefile`, `INSTALL`, and `xint.dtx`
and its extracted and derived files inclusive of the documentation
files `xint.pdf`, `sourcexint.pdf` and `CHANGES.html`.

See `xint.pdf` for contact information.

%</readme>--------------------------------------------------------
%<*changes>-------------------------------------------------------

`1.4j (2021/07/13)`
----

### Bug fixes

 - **xinttools**: a brace removal bug affected the venerable `\xintSeq`
   if producing a single number (e.g. `\xintSeq{10}{10}` expanded to
   `10` not `{10}`).  Thanks to Christophe Poulain for report.

`1.4i (2021/06/11)`
----

### New features

 - **xintexpr**: the concept of simultaneous assignments is extended: in
   case of more variables than values the extraneous variables do not
   cause an error message but are simply set to the `nil` value; in case
   of more values than variables, the last variable is defined to be the
   ople concatenating all the extra values.

 - **xintexpr**: built-in functions usable with arbitrarily many
   arguments such as `max()`, `gcd()`, or `len()` are now again usable
   with a single numeric argument: since `1.4` a lone argument had to be
   a `nutple` (which was automatically unpacked).  It can now again be a
   number.

### Bug fixes

 - **xintexpr**: simultaneous assignments via `\xintdefvar` to 10 or
   more variables was broken if the right hand side was an ople (of
   length at least 10, as it had prior to this release to match the
   number of variables).

 - **xintexpr**: the mechanism which allows to define variables with
   names already in use for pre-existing functions was broken for some
   built-in functions: those handling syntax with dummy variables
   (currently `subs()`, `subsm()`, `subsn()`, `seq()`, `add()`, `mul()`,
   `ndseq()`, `ndmap()`, `ndfillraw()`) and the so-called "pseudo"
   functions (currently `bool()`, `togl()`, `protect()`, `qint()`,
   `qfrac()`, `qfloat()`, `qraw()`, `random()`, `qrand()`, `rbit()`).
   For example the function `seq()` was broken if the user had defined a
   variable `seq`.

`1.4h (2021/05/27)`
----

### Bug fixes

 - **xintexpr**: the recent `1.4g` introduced a bug breaking input of
   the type `<operator><space token><macro>`.

 - **xintexpr**: since `1.4` (`2020/01/31`) the `omit` and `abort`
   keywords were broken if used inside a substitution, itself nested in
   a `seq()` or similar construct.

 - **xintexpr**: since `1.4c` the `\xintthespaceseparated` (added at
   `1.4a`) inserted two, not one, spaces at one specific location near
   the end of its output.  A bit cosmetic problem, fixed nevertheless.

`1.4g (2021/05/25)`
----

### Breaking changes

 - **xintexpr**: the parsing of power operators `**` and `^` now
   proceeds in a right associative way: `2^3^4==2^(3^4)`.

 - **xintexpr**: single-character operators `&`, `|`, and `=`
   (deprecated since `1.1`) have been removed.  Use `&&`, `||` and `==`
   respectively.

### Deprecated

 - **xintfrac**: old typesetting macros dating back to `1.03` and `1.04`
   releases `\xintFrac`, `\xintSignedFrac`, `\xintFwOver`,
   `\xintSignedFwOver` are deprecated. Please use the new names
   `\xintTeXFrac`, `\xintTeXsignedFrac`, `\xintTeXOver`,
   `\xintTeXsignedOver`.  The old names will emit warnings.

### New features

 - Messages written to the terminal and log file during the handling of
   recovery from exceptions use a new mechanism; when in interactive
   mode, the user is prompted only once, not thrice, to enter `<return>`
   for recovery and continued processing.

 - **xintfrac**: `\xintTeXfromSci` (added about 7 or 8 years too late)
   to help typesetting values output by `\xintfloateval` in a more
   useful way than `\xintTeXFrac` (formerly `\xintFrac`).

### Bug fixes

 - **xintexpr**: with **xintbinhex** loaded, `"` is recognized as prefix
   for hexadecimal input; but a bug (present ever since this support for
   hexadecimal input was added) caused syntax such as `"\macro` to break
   the parser.  Also, leading zeros such as in `"0000A` where not properly
   trimmed since `1.2m`.

 - **xintexpr**: authorize `x! == y` without parentheses.  Formerly the
   parser mis-interpreted `!` as first character of the `!=`
   not-equal-to comparison operator, subsequently causing breakage when
   finding the second `=`.

 - **xintexpr**: various error situations were badly handled.
   - `\xinteval{1/0}` did not recover gracefully due to
     longstanding bug in **xintfrac**.
   - `\xintfloateval{1/0}` did not recover gracefully either... due to some
     other reason.
   - recovery from encountering an unknown variable name was
     broken since `1.4`; also, with an unknown function name if in
     `\xintfloatexpr`.
   - `\xinteval{_4}` caused an infinite loop (with underscore `_` having
     its normal catcode).  This was unexpected, as the `_` is documented
     to be ignored if in-between digits (and is ignored if last after
     some digits).  It will be now ignored also if encountered upfront.
     A similar problem existed with variable names starting with a `_`
     (of normal catcode), which however are clearly not legal.  The
     (normal catcode) `_` will then be ignored as well if encountered in
     front of a variable or function name (but not inside of course).
   - ``\xinteval{\catcode`@}`` crashed (one had to use an added `\the`
     or `\number`).
   - illegal ``\xinteval{/3}`` was not intercepted and ultimately
     caused low-level errors (same with `^`).

   Surely, further situations remain where bad input will crash parser.
   
`1.4f (2021/05/10)`
----

### Breaking changes

 - **xintexpr**: `\xintieval{[-D]...}`, which rounds to a multiple of
   `1eD` for `D` positive now *does not insert the trailing zeros* (as
   done at `1.4e`) *nor a scientific part* `eD` (as prior to `1.4e`).
   The use case envisioned is for the quantized value to be used with an
   appropriate unit, for example `k` for `D=3` or `M` for `D=6`
   etc... Sorry for the very long process which was needed to reach this
   final decision.

 - **xintexpr**: for Digits beyond the officially supported range for
   accurate math functions, i.e. for `D>62`, computations were still
   done and printed with full number of digits, but the extra digits
   were meaningless; they now operate on and output mantissas limited to
   `min(D,64)` digits.
 
 - **xintexpr**: for powers `a^b` with Digits at most `8`, the number
   `a` is now float-rounded to Digits before computation, as is done for
   `Digits>8`; previously `9` significant digits were kept.

 - **xintexpr**: further changes in the computation of powers, see the
   bug fixes below.

 - **xintexpr**: the `float_()` function got renamed into `float_dgt()`.
 
### Bug fixes

 - **xintexpr**: the documentation said `float_()` function had been
   renamed `float_dgt()` but actually that was not yet the case.

 - **xintexpr**: powers `a^b` (with exponent `b` neither integer nor
   half-integer) stopped being accurate regarding the last digits for
   `|b|` about `1000` and beyond.  Except for `0.8<a<1.25` for which
   accuracy was maintained up to about `|b|=10^7`.  Fixed via keeping
   the same suitable number of extra digits for internal computations of
   logarithms, independently of whether `a` is close to `1` or not.

 - **xintexpr**: powers `a^b` in `\xinteval` with `|b|` an integer at
   least `10000` were paradoxically computed with less accuracy than in
   `\xintfloateval`... although they were documented to be handled
   exactly the same in the two parsers.

 - **xintexpr**: powers `a^b` for Digits at most `8` unconditionally
   used `log10()/pow10()` approach but due to the limited accuracy of
   the poorman logarithm (`9` fractional digits) this was inaccurate
   already for `b` about `100`.  So, the handling is now as for Digits at
   least `9`, i.e. integer and half-integer exponents are handled via
   the legacy `\xintFloatPower/\xintFloatSqrt` allowing arbitrarily big
   exponents.  It is advised to split big powers with non integer non
   half integer exponents into a product; this is not done internally
   to avoid costly overhead for possibly rare use cases.

 - **xintexpr**: the invalid input `(-1)^2.5` triggered an undefined
   control sequence error if Digits was at most `8`.

`1.4e (2021/05/05)`
----

### Breaking changes

 - **xintlog**: `\poormanloghack` now a no-op.

 - **xinttrig**: loading the package does not define left-over variables
   holding the values of the inverse factorials used in the sine and
   cosine series.
 
 - **xintexpr**: the output format of `\xinteval`, which uses
   `\xintFracToSci,` has changed.  In particular, if the number has a
   power of ten part, it is not output with an integer mantissa, but
   with a scientific mantissa `d.d...` with always at least one digit
   after the decimal mark (possibly `0`) and trailing zeros are trimmed
   out.  This is the same output format as used for `\xintfloateval`,
   apart of course from the fact that the mantissa lengths are not
   limited.
   
 - **xintexpr**: the output format of `\xintfloateval`, which uses
   `\xintPFloat`, changed.  The `\xintfloatexprPrintOne` macro has
   changed its signature to `[#1]{#2}` i.e. its first argument
   will be within brackets not braces.

 - **xintexpr**: when using `\xintieval{[D]...}` optional `[D]` with a
   negative `D`, which triggers quantization to a positive power of ten,
   the output (if not the zero value) will be an integer with `N=abs(D)`
   explicit trailing zeros, not an integer mantissa followed by `eN`.

 - **xintexpr**: `\xinteval` will not compute powers `a^N` exactly if
   `N` in absolute value is at least the (rounded) quotient of `10000`
   by the number of digits of `a`; it will then use the
   logarithm/exponential (in base 10) approach, according to the
   prevailing Digits setting (at `1.4f` this got modified again and integer
   exponents large enough to trigger floating point evaluation are handled
   exactly as in `\xintfloateval`, i.e. using the legacy **xintfrac**
   `\xintFloatPower`,
   not the logarithm/exponential approach which loses accuracy for exponents
   of the order of `100000000` and beyond).

 - **xintexpr**: `\xintdeffloatvar` now always rounds the assigned value
   to the target precision.  Formerly, inputs actually involving no
   float operations, such as for example a sub-expression `\xintexpr1/20!\relax`
   (in
   contrast to `1/20!`), or an explicit single number having more digits
   than the precision, got stored "as is" in the defined variable,
   without pre-rounding to the Digits precision.

 - **xintfrac**:`\xintPFloat` and `\xintFracToSci` have both been
   modified.  The macro `\xintFracToSciE` does not exist anymore, as
   `\xintFracToSci` in the case of scientific exponents hands over the
   process to `\xintPFloat` (without the rounding to Digits, of course).

### New features

 - **xintlog**: now working up to `62` digits. The legacy faster
   `poormanlog`-based macros are kept for computations with Digits up to
   `8`.  Fractional powers are now available both in `xinteval` and
   `xintfloateval`, per default.

 - **xinttrig**: now working up to `62` digits and with increased
   accuracy.  Special faster (or not as slow if you prefer) mode at `8`
   digits or less.

 - **xintexpr**: the constraints on the `\xintexprPrintOne` replacement
   macro, which defaults to `\xintFracToSci`, have been much simplified.

 - **xintexpr**: `\xintiexprPrintOne` (defaults to `\xintDecToString`)

 - **xintfrac**: `\xintPFloatE` may be redefined as a macro which
   fetches the scientific exponent as a mandatory argument delimited
   with a dot, and outputs a suitable formatted result (f-expandably),
   also delimited with a dot which will be removed by internal
   processing.  The default however simply expands to the letter `e`.
 
 - **xintfrac**: `\xintDecToStringREZ` is like `\xintDecToString` but
   starts by removing trailing zeroes.

### Bug fixes

 - **xintfrac**, **xintexpr**: it was not possible to use
   `\xinttheDigits` in the right hand side of an `\xintDigits`
   assignment.  For example: `\xintDigits*:=\numexpr\xinttheDigits+4;`.
   This is now allowed, and the same applies to the macro interface,
   for example `\xintSetDigits*{\xinttheDigits+4}`.


`1.4d (2021/03/29)`
----

### Breaking changes

 - `quo()` and `rem()` in `\xintiiexpr/\xintiieval` renamed to
   `iquo()` and `irem()`.

 - The output of `gcd()` and `lcm()` as applied to fractions is now
   always in lowest terms.

 - The log message to report a variable creation (if `\xintverbosetrue`)
   does not use (double) quotes anymore around the name.  By the way,
   quotes were never used for function names.

### Bug fixes

 - Ever since `1.3` the `quo()` and `rem()` functions in `\xintexpr`
   (not the ones in `\xintiiexpr`) were broken as their (officially
   deprecated) support macros had been removed!  They had somewhat
   useless definitions anyway.  They have now been officially removed
   from the syntax.  Their siblings in `\xintiieval` were renamed to
   `iquo()` and `irem()`.

 - Sadly, `gcd()` was broken in `\xintexpr` since `1.4`, if the first
   argument vanished.  And `gcd()` was broken in `\xintiiexpr` since
   `1.3d` if *any* argument vanished.  I did have a unit test! (which
   obviously was too limited ...)

   Further, the `\xintGCDof` and `\xintLCMof` **xintfrac** macros were
   added at `1.4` but did not behave like other **xintfrac** macros with
   respect to parsing their arguments: e.g. `\xintGCDof{2}{03}` gave an
   unexpected non-numeric result.

 - The `first()` and `last()` functions, if used as arguments to
   numerical functions such as `sqr()` inside an `\xintdeffunc`
   caused the defined function to be broken.

`1.4c (2021/02/20)`
----

### Bug fixes

 - Fix `1.4` regression which broke syntax `varname(...)` which supposedly
   is allowed and inserts a tacit multiplication.

`1.4b (2020/02/25)`
----

All changes regard the **xintexpr** module.

### Future

 - `&`, `|`, (as Boolean operators) and `=` (as equality test) have long
   been deprecated in favour of `&&`, `||` and `==`.  They will be
   removed at next major release.

 - At next major release the power operators `**` and `^` will turn from
   left to right associative.  I.e. `2**2**3` will give `256`, not `64`.
   This is to match with Python and l3fp.

 - `\thexintexpr` et al. (introduced at `1.2h` but not documented
   anymore for some time) will be removed at next major release.  The
   original `\xinttheexpr` et al. have always been so much better names.
   Besides, since `1.4`, `\xintexpr` can be used directly in typesetting
   flow.

### New features

 - Function `zip()` is modeled on Python's function of the same name.
 
 - Function `flat()` removes all nesting to produce a "one-dimensional"
   list having the exact same leaves (some possibly empty) as the
   original (in the same order).
 
 - Chaining of comparison operators (e.g. `x<y<z`) as in Python (but all
   comparisons are done even if one is found false) and l3fp.

 - It was possible since `1.4`'s `\xintFracToSciE` to configure the
   separator between mantissas and exponents in the output of
   `\xinteval` but strangely there was no way to customize the output of
   `\xintfloateval`.  The added `\xintPFloatE` fixes this.

### Bug fixes

 - `\xintieval{[D]...}` with a negative `D` (a feature added at `1.4a`)
   used erroneously a catcode 12 `e` in output, which moreover remained
   immuned to the `\xintFracToSciE` setting.

`1.4a (2020/02/19)`
----

All changes regard the **xintexpr** module.

### Breaking changes

 - The macros implementing customization of
   `\xintthealign` have modified meanings and names.

### New features

 - `\xintthespaceseparated` (serves to provide
   suitable input to PS-Tricks `\listplot`).

 - The optional argument `[D]` to `\xintieval/\xintiexpr`
   can be negative, with the same meaning as the non-negative case, i.e.
   rounding to an integer multiple of `10^(-D)`.

    The same applies to the functions `trunc()` and `round()`.  And to
   the `\xintTrunc`, `\xintRound`, `\xintiTrunc`, and `\xintiRound` macros
   of **xintfrac**.

### Bug fixes

 - Usage of `round()` and `trunc()` within `\xintdeffunc`
   got broken at `1.4`.

 - `add()` and `mul()` were supposedly accepting the
   `omit`, `abort` and `break()` keywords since `1.4` but this was
   broken.
   

`1.4 (2020/01/31)`
----

### Breaking changes

Please note that this list may still be incomplete.  If not otherly
specified all items regard the **xintexpr** module.

 - The `\expanded` primitive (TeXLive 2019) is **required**.  This does
   not affect the macro layer **xintcore**, **xint**, **xintfrac**,
   **xinttools** (yet).

 - Formerly square brackets `[...]` were, on their own, not different
   from parentheses (and thus disappeared from the output), but they are
   now a genuine constructor of nested lists.  For example `\xinteval{1,
   [2, [3, 4]], 5}` produces `1, [2, [3, 4]], 5` (recall this is free
   bloatware).

 - The output of `\xinteval` has changed (besides containing brackets).
   It does not use anymore the so-called *raw* **xintfrac** format,
   i.e. things such as `A/B[N]` (which can still be used in input but
   are discouraged in **xintexpr** context), but scientific notation
   `AeN/B`.  As formerly, the denominator is printed only if `B>1` and
   the scientific part is dropped if the exponent vanishes.  In this way
   the output of `\xinteval` can be pasted to alternative software.

 - The output format of `\xinthe\xintboolexpr` also has changed.  It uses
   `True` and `False` (which are accepted on input), and this can
   easily be configured otherwise (also `true` and `false` are accepted
   on input).

 - The "broadcasting" (as it turned out, à la `NumPy`) of scalar
   operations on one-dimensional "lists", e.g `3*[1,3,5,7]+10` acting
   itemwise is **dropped**.  It is hoped to implement such operations
   again in stronger form in future releases.  Pre-existing alternative
   syntax is available, also to produce the bracketed (cf. next item)
   `[13,19,25,31]` which will be the output in future.

 - The `divmod()` function now produces on output such a bracketed
   pair, but simultaneous assignment such as `\xintdefvar xq, xr =
   divmod(a,b);` will work transparently.
   
 - The syntax for using conditionals in function declarations has
   changed.  Now, one *must* use the `?` and `??` short-circuit boolean
   branching operators whereas in the past it was explained that the
   syntax had to use the `if()` and `ifsgn()` functions.

 - Macros `\xintGCD`, `\xintLCM`, `\xintGCDof` and `\xintLCMof` formerly
   provided by **xintgcd** got moved to **xintfrac** (which is not
   loaded by **xintgcd**).  Moreover, they were extended to handle
   general fractions on input but this also means that their output is
   now obiding by the raw **xintfrac** format.  The integer only
   `\xintiiGCD`, `\xintiiLCM` also got moved out of **xintgcd**, but to
   **xint** which is now loaded automatically by **xintgcd**.  The few
   remaining macros of **xintgcd** at least do not need other imports as
   **xintgcd** now loads also automatically **xinttools** which is a
   dependency for two of them.
   
### Improvements and new features

Please note that this list is currently incomplete.  For more
information look at the user manual and the documented source code
`sourcexint.pdf`.

Unless otherwise specified all changes commented upon here regard
**xintexpr**.  Important: all the new syntax is to be considered
experimental.  The author may change some names in future release, or
even the interface (whether to use semi-colons or colons etc...).
 
 - The `\csname` encapsulation technique used since **xintexpr** initial
   release (`1.07 2013/05/25`) to move around possibly large data during
   expansion-only operations is replaced with methods based on the
   `\expanded` engine primitive.  The latter is available in all major
   engines since TeXLive 2019.

    Formerly, and with default memory settings, one would typically
   saturate the string pool memory after about of the order of 50,000
   independent floating point evaluations of expressions of average
   complexity on 16-digits numbers.

    There is thus no string pool memory impact at all but one can
   now hit TeX's main memory limit (which typically stands at 5,000,000
   words) from defining large variables or generating on the fly large
   data.  TeX distributions have a configuration file allowing to enlarge
   TeX memory parameters and regenerate the (eTeX based) formats.

 - The package supports input and output of arbitrarily *nested lists*,
   a.k.a. *oples* or *nlists*, with `[...]` as the constructor of
   *bracketed lists*, a.k.a *nut-ples*.  Operations on these objects (as
   briefly surveyed in later items) are inspired from syntax and
   functionalities of `NumPy`'s *ndarrays*.  Our *oples* (hence also
   their packaged form *nut-ples*) may have *leaves* at varying depths
   rather than obeying an N-dimensional hyperrectangular shape.  But the
   syntax does provide specific constructors for *ndlists*
   (i.e. hyperrectangular *oples* or *nut-ples*).

     In a (distant?) future, perhaps **xintexpr** itself or a
   third-party package will provide an interface, say `\xintstorearray`,
   `\xintgetarray`, to store (which can not be expandable) and retrieve
   (which can be expandable and thus be embedded inside expressions
   parsed by `\xintexpr`, `\xintiiexpr` or `\xintfloatexpr`) such
   *ndlists* from TeX memory.  This is why the package does not use the
   word *ndarray* and reserves it for such memory stored objects.

 - The `*` serves as *unpacking* operator on *nut-ples*, i.e. reversing
   the `[]` bracketing of an *ople*.

 - *oples* have no exact equivalent in `Python`. For example
   **xintexpr** allows `foo(Var1, x)` if `foo` is a function of 4
   variables and `Var1` is a variable producing a length 3 *ople*, or
   `foo(Var2)` if `Var2` is a variable producing a length 4 *ople*.
   Python would require here to use explicitly the `*`-unpacking notation
   on some "packed" objects.

    Variable and function values may be *oples* (even *nil*), but in
   function declarations variables must stand for *one-ples*, i.e. either
   *numbers* or *nut-ples* (as there is no non-ambiguous way to split
   e.g. 5 arguments into two separate *oples*).

 - Simultaneous assignment to at least two variables via `\xintdefvar`
   et al. automatically unpacks the assigned value if it is a *one-ple*.
   If this value was in fact a *number*, low-level errors will result
   shortly afterwards as no check is done if the unpacking was illicit.
   (update: this last remark does not apply since the `1.4i` extension to
   the concept of simultaneous assignments)
   
 - The `NumPy` concept and syntax for nested slicing and item selection
   are implemented.  Currently *stepping* and the *Ellipsis object* are
   not yet available.  Only so-called basic slicing is currently
   supported.  (The author has not yet read the section of `NumPy`
   documentation on so-called *advanced indexing*).

 - The *broadcasting* of scalar operations, such as itemwise addition or
   multiplication of *nut-ples* of the same shape is **not yet implemented**.

 - Slicing and indexing apply also at top level to the *oples* with
   behaviour conforming to intuitive expectations (see user manual); if
   it turns out the *ople* is in fact a *nut-ple*, the top-level
   slicing/indexing switches to the `Python/NumPy` conventions, i.e. it
   operates inside the brackets for slicing and removes brackets if
   indexing.

 - The syntax `ndseq(expression in x, y, ..., x = values; y = values;
   ...)` constructs a (bracketed) *ndlist* by evaluation the expression
   on all possible Cartesian n-uples, where the first variable indexes
   the first axis, the second the next, etc...

 - The `ndmap(foo, values1; values2; ...; valuesN)` syntax constructs a
   (bracketed) *ndlist* by evaluating the function `foo` on all elements
   of the cartesian product of the given (one-dimensional) value lists.

 - The two concepts of `\xintdeffunc` (for recursive definitions) and
   `\xintdefefunc` (for functions which expand immediately in other
   function declarations) have been merged. The `\xintdefefunc` et al.
   are deprecated and kept as aliases for `\xintdeffunc` et al.

 - `\xintdefufunc` allows to define so-called *universal functions*,
   i.e. functions `foo` such that `foo(myople)` will apply itemwise at
   arbitrary depth in the nested structure. The function `foo` is
   allowed to produce from a scalar an *ople*...

 - The variables in function declarations can now be multi-letter words.

 - The last positional variable in a function declaration can be prefixed
   with a `*` meaning exactly as in Python (*variadic* function
   argument) that it stands for a one-dimensional *nut-ple* receiving all
   remaining arguments from the function call beyond the first
   positional ones.  It is thus an optional argument, but syntax for
   named optional arguments with default values is not yet implemented.

 - Dummy variables used in constructors can also be multi-letter words,
   if they have been declared as such.

 - In variable and function declarations, if the expression contains
   inner semi-colons, it is not needed anymore to brace them to avoid
   mis-interpretation as the final semi-colon which is mandated by the
   syntax to serve as expression terminator.

 - `subsm(expression, var1 = value1; var2 = value2; ...)` provides a leaner
   syntax for multiple substitutions; they must be independent, though.

 - `subsn(expression, var1 = value1; var2 = value2; ...)` provides a
   leaner syntax for nested substitutions, i.e., each `valueJ` may be an
   expression using the dummy variables `varK` with `K>J`.  And finally
   of course the evaluated expression can refer to all variables.

 - `\xintthealign\xintexpr...\relax` (or with `\xintfloatexpr` or
   `\xintiiexpr` or `\xintboolexpr`...) will use a TeX alignment to
   display *oples*. The output (for regular N-dimensional lists) looks
   very similar to what `Python/NumPy` produces in interactive session.
   This is entirely configurable and can also be set-up to be used for
   writing into external files.

    Attention that `\xintthealign` only works if followed by `\xintexpr`
   et al., not by `\xinteval{}`.

 - It is now possible to use `\xintexpr...\relax` directly for
   typesetting.  The syntax `\xinteval{...}` or
   `\xintthe\xintexpr...\relax` is needed only if one wants the
   expansion to give the explicit digits, but `\xintexpr...\relax` by
   itself will typeset as would have the other ones.  Further it can be
   used in so-called moving arguments, because when output to an
   external file it uses only characters with standard catcodes (and
   produces the same protected and re-tokenizable result it would in an
   `\edef`.)

    As formerly, `\xintexpr...\relax` is the preferred way to include an
   expression into another one. Using `\xinteval` is a waste because it
   forces the outer parser to re-digest all the digits (or now also the
   square brackets).

 - The output format of `\xintfloateval` with scientific notation has
   not changed (apart from possible presence of bracketed lists), but
   the author hesitates because the *prettifying* it does by default is
   not really adapted to display of arrays (see `\xintthealign`).
   Anyway, this is configurable by the user.  It is possible to
   specify whether to use `e` or `E`.

 - Function declarations are able to parse a much wider part of the
   syntax, but some severe limitations remain.  Refer to the user manual for
   related information.

 - We have made an effort on some error messages, and when working
   interactively in a shell it may even be sometimes possible to insert
   for example a correct variable or function name in place of the not
   recognized one.  But don't expect miracles when trying to intervene
   in the midst of a purely expandable expansion...

### Bug fixes

Bugs? Those identified in `1.3f` were almost features.  As per `1.4` the
code base of **xintexpr** received multiple successive core refactorings
and added numerous new features, and our test suite although
significantly enlarged is not yet extensive enough.  Please report bugs
by mail.

### TODO

 - The long delayed overhaul of how floating point numbers are handled
   is delayed again.  It has remained basically identical to its initial
   provisory version from `1.07 2013/05/25` (which was based upon what
   was originally only a set of expandable macros for computations with
   big integers), and suffers from the author lack of knowledge of the
   notion of "data type" in modern programming.  Indeed, he never took a
   CS class, and disables JavaScript in his browser (or allows only
   select non-tracking scripts, a rare beast in modern days).

 - Prior to integrating all of `NumPy`, it is envisioned to start with
   matrix algebra first.


`1.3f (2019/09/10)`
----

### Improvements and new features

 - **xintfrac**: `\xintDigits = P;` syntax (i.e. without a colon) is now
   accepted in addition to `\xintDigits := P;`.

     Document that the ending semi-colon can not be an active character
   and that it has always been allowed to use in its place a
   non-expanding token e.g. `\xintDigits := 32\relax`.

     Add `\xintSetDigits`.

 - **xintexpr**: add starred variants `\xintDigits*` and `\xintSetDigits*`
   which execute `\xintreloadxinttrig`.

     Revert 1.3e ban on usage of `\xinteval` et al. inside expressions by
   `\xintdeffunc`.  And make them usable also inside macro definitions via
   `\xintNewExpr`.

### Bug fixes

 - **xintexpr**: fix bug preventing usage of `\xintdefefunc` to define a
   function without variables.

     Fix some issue with `\xintfloatexpr[D]..\relax` if used inside an
   expression parsed by `\xintdeffunc` et al.


`1.3e (2019/04/05)`
----

### Breaking changes

 - (_reverted at 1.3f_) When defining functions, sub-expressions can only
   use the `\xint(float)expr...\relax` syntax. One can not use there the
   `\xint(float)eval` wrappers.

### Improvements and new features

 - The **xinttrig** library is automatically loaded by **xintexpr**. It
   provides direct and inverse trigonometrical functions using either
   degrees or radians with a precision of up to (a bit less than) 60
   digits. It is for the most part implemented using high level user
   interface, but will probably get some optimizations in future (and
   perhaps extension to more digits).

 - The **xintlog** library is automatically loaded by **xintexpr**. It
   uses [poormanlog](http://ctan.org/pkg/poormanlog) to provide
   logarithms and exponentials with almost 9 digits of precision.
   Extended precision is for a future release.

 - **xintexpr**: `\xintdefefunc`, `\xintdeffloatefunc`, `\xintdefiiefunc`
   define functions which are not protected against expansion in the
   definition of other functions; refer to `xint.pdf` for the related
   explanations.

   Notice that whole area of `\xintdef(e)func`, `\xintNewExpr`,
   `\xintNewFunction` is complex and to be considered still as work in
   progress as it has a number of shortcomings.

 - **xintexpr**: `inv()`, `ilog10()`, `sfloat()`, behaviour of
   `qfloat()` slightly modified.

 - **xintexpr**: `\xintensuredummy`, `\xintrestorelettervar`.

 - The optional argument of `\xintfloatexpr` or `\xintfloateval` (it
   must be at start of braced argument) can be negative; it then means
   to trim (and round) from the output at float precision that many
   least significant digits.

### Bug fixes

 - Some bugfixes related to user functions with no variables at all;
   they were dysfunctional.

`1.3d (2019/01/06)`
----

### Breaking changes

 - **xintexpr**: the `gcd()` and `lcm()` functions formerly converted
   their arguments to integers via `\xintNum`. They now handle general
   input with no such modification.

 - **xintexpr**: former `\xinteval`, `\xintieval`, `\xintiieval`, and
   `\xintfloateval` renamed to `\xintexpro`, `\xintiexpro`,
   `\xintiiexpro`, and `\xintfloatexpro`.

### Improvements and new features

 - **xintexpr**: the `gcd()` and `lcm()` multi-arguments functions have
   been refactored to handle general fractions. The dependency on
   **xintgcd** is removed.

 - **xintexpr**: three-way branching `\xintifsgnexpr`,
   `\xintifsgnfloatexpr`, `\xintifsgniiexpr` conditional macros.

 - **xintexpr**: `\xintunassignexprfunc`, `\xintunassigniiexprfunc`,
   `\xintunassignfloatexprfunc` to "undefine" functions.

 - **xintexpr**: `\xintunassignvar` really makes the (multi-letter) variable
   unknown (formerly, it only gave it value zero),

 - **xintexpr**: functions `isone()` and `isint()`.

 - **xintexpr**: `\xinteval`, `\xintieval`, `\xintiieval`, and
   `\xintfloateval` as synonyms to `\xinttheexpr...\relax` etc..., but
   with the (comma-separated) expression as a usual braced macro
   argument.

### Bug fixes

 - **xintcore**, **xintexpr** : division in `\xintiiexpr` was broken for
   a zero dividend and a one-digit divisor (e.g. ``0//7``) since `1.2p`
   due to a bug in `\xintiiDivMod` for such arguments. The bug was
   signaled (thanks to Kpym for report) and fixed shortly after `1.3c`
   release but I then completely forgot to upload a bugfix release to
   CTAN at that time, apologies for that.

`1.3c (2018/06/17)`
----

### Improvements and new features

 - **xintexpr**: with `\xintglobaldefstrue`, `\xintdefvar`,
   `\xintdeffunc`, `\xintNewExpr` et al. make definitions with global
   scope.

 - **xintexpr**: `qraw()` for fast input of (very many) comma separated
    numbers (in suitable raw format).

 - **xintexpr**: the colon in the `:=` part of the syntax for
   `\xintdefvar` and variants is now optional; and if present it may be
   an active character or have any (reasonable) catcode.
   
 - **xintexpr**: `\xintdefvar`, `\xintdeffunc` and their variants try to
   set the catcode of the semi-colon which delimits their arguments; of
   course this will not work if that catcode is already frozen.

 - `\xintUniformDeviate` is better documented and `sourcexint.pdf` is better
   hyperlinked and includes indices for the macros defined by each package.

### Bug fixes

 - **xintfrac**: since `1.3` release, it loaded **xintgcd** in
   contradiction to what the documentation says (hence also **xintexpr**
   loaded **xintgcd** automatically). There is no actual dependency so
   the loading is removed for now.

`1.3b (2018/05/18)`
----

### Improvements and new features

All additions related to randomness are marked as work-in-progress. They
require an engine providing the `\(pdf)uniformdeviate` primitive.

 - **xintkernel**: `\xintUniformDeviate`.

 - **xint**: `\xintRandomDigits`, `\xintXRandomDigits`, `\xintiiRandRange`,
   `\xintiiRandRangeAtoB`.

 - **xintfrac**: support macros (not public, mainly because internal
   format for floats is surely not final) for `random()` and `qrand()`.

 - **xintexpr**: `random()`, `qrand()`, and `randrange(A[, B])`.

 - **xintexpr**: when a function `foo()` is declared via `\xintdeffunc`
   (et al.) to be parameter-less, it can be used as `foo()`; formerly
   `foo(nil)` syntax was required.

 - The usual provision of user manual "improvements".


`1.3a (2018/03/07)`
----

### Removed

 - **xintcore**, **xint**, **xintfrac**: removal of the internal macros
   which were used at `1.2o` to add a deprecation mechanism; all
   deprecated macros have been removed at `1.3` so there was no reason
   to keep the code used for deprecating them.

### Improvements and new features

 - **xintexpr**: new conditionals `ifone()` and `ifint()`.

 - **xintfrac**: `\xintREZ` is faster on inputs having one hundred
   digits or more.

 - Added to the user manual mention of macros such as `\xintDivFloor`,
   `\xintMod`, `\xintModTrunc`, which had been left out so far.

### Bug fixes

 - **xintexpr**: the mechanism for adjunction to the expression parsers
   of user defined functions was refactored and improved at previous
   release `1.3`: in particular recursive definitions became possible.
   But an oversight made these recursive functions quite inefficient (to
   remain polite.) This release fixes the problem.


`1.3 (2018/03/01)`
----

### Breaking changes

 - **xintcore**, **xint**, **xintfrac**: all macros deprecated at `1.2o`
   got removed.

 - **xintfrac**: addition and subtraction of `a/b` and `c/d` now use the
   l.c.m. of the denominators. Similarly the macro supporting the modulo
   operator `/:` uses a l.c.m. for the denominator of the result.

 - **xintexpr**: the addition, subtraction, modulo `/:`, and the
   `mod()` and `divmod()` functions produce generally smaller denominators
   (see previous item).

 - **xintexpr**: formerly, the internal macros which are internally
   associated to user-declared functions were using comma separated
   parameter texts. They now do not use such commas (their meanings,
   which may again change in future, are written for information to the
   log under `\xintverbosetrue`).

### Improvements and new features

 - **xintexpr**: user-defined functions may now be of a recursive
   nature. This was made possible by a refactoring of the `\xintNewExpr`
   mechanism. It became both leaner and more extensive than formerly.

 - **xintfrac**: new macros `\xintPIrr` and `\xintDecToString`. The
   latter is a backport of a `polexpr 0.4` utility, and it is to be
   considered unstable.

 - **xintexpr**: new function `preduce()` associated with `\xintPIrr`.


`1.2q (2018/02/06)`
----

### Improvements and new features

 - **xintexpr**: tacit multiplication extended to cases such as `3!4!5!`
   or `(1+2)3`.

### Bug fixes

 - **xintcore**: sadly, refactoring at `1.2l` of subtraction left an
   extra character in an inner macro causing breakage in some rare
   circumstances. This should not have escaped our test suite!


`1.2p (2017/12/05)`
----

### Breaking changes

 - **xintgcd**: `\xintBezout{a}{b}`'s output consists of `{u}{v}{d}`
   with `u*a+v*b==d`, with `d` the GCD. Formerly it was
   `{a}{b}{u}{v}{d}`, and with `u*a-v*b==d`.

 - **xintgcd**: `\xintBezout{0}{0}` expands to `{0}{0}{0}`. Formerly
   (since `1.2l`) it raised `InvalidOperation`.
 
 - **xintcore**: `\xintiiMod` is now associated with floored division.
   The former meaning (associated with truncated division) is available
   as `\xintiiModTrunc`.
 
 - **xintfrac**: `\xintMod` is now associated with floored division. The
   former meaning is available as `\xintModTrunc`.

 - **xintexpr**: the ``//`` operator and its associated modulo ``'mod'``
   (or ``/:``) now correspond to floored division, like the Python
   language `//`, `%`, and `divmod(x, y)`. Formerly they had been
   associated to truncated division. This is breaking change for
   operands of opposite signs.

### Improvements and new features

 - **xinttools**: `\xintListWithSep`, which had remained unchanged since
   its introduction at `1.04 (2013/04/25)`, was rewritten for increased
   speed.

 - **xintexpr**: `\xintdefvar`'s syntax is extended to allow
   simultaneous assignments. Examples:
   `\xintdefvar x1, x2, x3 := 1, 3**10, 3**20;` or
   `\xintdefiivar A, B := B, A 'mod' B;`
   for already defined variables `A` and `B`.

 - **xintexpr**: added `divmod()` to the built-in functions. It is
   associated with floored division, like the Python language `divmod()`.
   Related support macros added to **xintcore**, and **xintfrac**.

### Bug fixes

 - **xintgcd**: `\xintBezout{6}{3}` (for example) expanded to
   `{6}{3}{-0}{-1}{3}`, but the `-0` should have been `0`.

 - **xintgcd**: it still used macro `\xintiAbs` although the latter had
   been deprecated from **xintcore**.

 - **xintexpr**: in float expressions the `//` and `/:` (aka `'mod'`)
   operators did not round their operands to the float precision prior
   to computing with them, contrarily to other infix arithmetic
   operators and to the `mod(f,g)` function; thus, `mod(f,g)` and
   `f 'mod' g` were not completely equivalent.

 - various documentation fixes; in particular, the partial dependency of
   **xintcfrac** on **xinttools** had not been mentioned.


`1.2o (2017/08/29)`
----

### Breaking changes

 - **xint**: `\xintAND`, `\xintOR`, ... and similar Boolean logic macros do
   not apply anymore `\xintNum` (or `\xintRaw` if **xintfrac** is loaded), to
   their arguments (often, from internal usage of `\xintSgn`), but only
   f-expand them (using e.g. `\xintiiSgn`). This is kept un-modified even if
   loading **xintfrac**.

### Deprecated

Deprecated macros raise an error but, generally, then expand as in former
releases. They will all get removed at some future release.

 - **xintcore**: `\xintiOpp`, `\xintiAbs`, `\xintiAdd`, `\xintiSub`,
   `\xintiMul`, `\xintiDivision`, `\xintiQuo`, `\xintiRem`, `\xintiDivRound`,
   `\xintiDivTrunc`, `\xintiMod`, `\xintiSqr`, `\xintiPow`, and `\xintiFac`
   are deprecated. Only the `ii`-named variants get defined.

 - **xintcore**: `\xintCmp` and `\xintSgn` are deprecated from **xintcore**
   (which only defines `\xintiiCmp` and `\xintiiSgn`) as they actually belong
   to **xintfrac**.

 - **xintcore**: `\xintiiFDg`, resp. `\xintiiLDg`, are renamed `\xintFDg`,
   resp. `\xintLDg`. Former denominations are deprecated.

 - **xint**: `\xintMON`, `\xintMMON`, `\xintiMax`,
   `\xintiMin`, `\xintiMaxof`, `\xintiMinof`, `\xintiSquareRoot`,
   `\xintiSqrt`, `\xintiSqrtR`, `\xintiBinomial`, and `\xintiPFactorial` are
   deprecated. Only `ii`-named variants get defined.

 - **xint**: `\xintEq`, `\xintGeq`, `\xintGt`, `\xintLt`, `\xintGtorEq`,
   `\xintLtorEq`, `\xintIsZero`, `\xintIsNotZero`, `\xintIsOne`,
   `\xintOdd`, `\xintEven`, `\xintifSgn`,
   `\xintifCmp`, `\xintifEq`, `\xintifGt`, `\xintifLt`, `\xintifZero`,
   `\xintifNotZero`, `\xintifOne`, `\xintifOdd`, are deprecated. These macros
   belong to **xintfrac**. Package **xint** defines only the `ii`-named
   variants.

 - **xint**: `\xintNeq` was renamed to `\xintNotEq` which however is only
   provided by **xintfrac**. Package **xint** defines `\xintiiNotEq`, and
   `\xintNeq` is deprecated.

 - **xint**: `\xintNot` was renamed to `\xintNOT`, former denomination is
   deprecated. See also item about Boolean logic macros in the *Incompatible
   Changes* section.


`1.2n (2017/08/06)`
----

### Breaking changes

 - **xintbinhex** does not load package **xintcore** anymore, but only
   **xintkernel**.

### Improvements and new features

 - **xintbinhex** has only **xintkernel** as dependency.

 - Macros of **xintbinhex** have been improved for speed and increased maximal
   sizes of allowable inputs.
 

`1.2m (2017/07/31)`
----

### Breaking changes

 - **xintbinhex**: the length of the input is now limited. The maximum
   size depends on the macro and ranges from about `4000` to about
   `19900` digits.

 - **xintbinhex**: `\xintCHexToBin` is now the variant of
   `\xintHexToBin` which does not remove leading binary zeroes: `N`
   hex-digits give on output exactly `4N` binary digits.

### Improvements and new features

 - **xintbinhex**: all macros have been rewritten using techniques from
   the 1.2 release (they had remained unmodified since `1.08` of
   `2013/06/07`.) The new macros are faster but limited to a few
   thousand digits. The `1.08` routines could handle tens of thousands
   of digits, but not in a reasonable time.

### Bug fixes

 - user manual: the `Changes` section wrongly stated at `1.2l` that the
   macros of **xintbinhex** had been made robust against non terminated
   input such as ``\number\mathcode`\-``. Unfortunately the author fell
   into the trap of believing his own documentation and he forgot to
   actually implement the change. Now done.

 - user manual: the PDF bookmarks were messed up.

 - **xint**, **xintfrac**: `\xintGeq`, `\xintMax`, `\xintMin`, suffered
   from some extra overhead. This was caused by use of some auxiliaries
   from the very early days which got redefined at some stage. This is
   fixed here with some additional efficiency improvements and pruning
   of old code.
   

`1.2l (2017/07/26)`
----

### Removed

 - `\xintiiSumExpr`, `\xintiiPrdExpr` (**xint**) and `\xintSumExpr`,
   `\xintPrdExpr` (**xintfrac**). They had not been formally deprecated,
   but had been left un-documented since `1.09d (2013/10/22)`.

 - internal macro `\xint_gob_til_xint_relax` removed.

### Improvements and new features

 - the underscore character `_` is accepted by the **xintexpr** parsers
   as a digit separator (the space character already could be used for
   improved readability of big numbers). It is not allowed as *first*
   character of a number, as it would then be mis-interpreted as the
   start of a possible variable name.

 - some refactoring in **xintcore** auxiliary routines and in
   `\xintiiSub` and `\xintiiCmp` for some small efficiency gains.

 - code comments in **xintcore** are better formatted, but remain
   sparse.

 - **xintcore**, **xint**, **xintfrac**, ... : some macros were not
   robust against arguments whose expansion looks forward for some
   termination (e.g. ``\number\mathcode`\-``), and particularly, most
   were fragile against inputs using non-terminated ``\numexpr`` (such
   as `\xintiiAdd{\the\numexpr1}{2}` or `\xintRaw{\numexpr1}`). This was
   not a bug per se, as the user manual did not claim such inputs were
   legal, but it was slightly inconvenient. Most macros (particularly
   those of **xintfrac**) have now been made robust against such inputs.
   Some macros from **xintcore** primarily destined to internal usage
   still accept only properly terminated arguments such as
   ``\the\mathcode`\-<space>`` or ``\the\numexpr1\relax``.

     The situation with expressions is unchanged: syntax such as
   `\xintexpr \numexpr1+2\relax` is illegal as the ending `\relax` token
   will get swallowed by the `\numexpr`; but it is needed by the
   ``xintexpr``-ession parser, hence the parser will expand forward and
   presumably end with in an "illegal token" error, or provoke some
   low-level TeX error (N.B.: a closing brace `}` for example can not
   terminate an ``xintexpr``-ession, the parser must find a `\relax`
   token at some point). Thus there must be in this example a second
   `\relax`.

 - experimental code for error conditions; there is no complete user
   interface yet, it is done in preparation for next major release and
   is completely unstable and undocumented.
 
### Bug fixes

 - **xintbinhex**: since `1.2 (2015/10/10)`, `\xintHexToDec` was
   broken due to an undefined macro (it was in `xint.sty`, but the
   module by itself is supposedly dependent only upon `xintcore.sty`).

 - **xintgcd**: macro `\xintBezout` produced partially wrong output if
   one of its two arguments was zero.

 - **xintfrac**: the manual said one could use directly `\numexpr`
   compatible expressions in arithmetic macros (without even a
   `\numexpr` encapsulation) if they were expressed with up to 8 tokens.
   There was a bug if these 8 tokens evaluated to zero. The bug has been
   fixed, and up to 9 tokens are now accepted. But it is simpler to use
   `\the\numexpr` prefix and not to worry about the token count... The
   ending `\relax` is now un-needed.


`1.2k (2017/01/06)`
----

### Breaking changes

 - macro `\xintFloat` which rounds its input to a floating point number
   does _not_ print anymore `10.0...0eN` to signal an upwards rounding
   to the next power of ten. The mantissa has in all cases except the
   zero input exactly one digit before the decimal mark.

 - some floating point computations may differ in the least significant
   digits, due to a change in the rounding algorithm applied to macro
   arguments expressed as fractions and to an improvement in precision
   regarding half-integer powers in expressions. See next.

### Improvements and new features

 - the initial rounding to the target precision `P` which is applied by
   the floating point macros from **xintfrac** to their arguments
   achieves the _exact (aka correct) rounding_ even for inputs which are
   fractions with more than `P+2` digits in their numerators and
   denominators (`>1`.) Hence the computed values depend only on the
   arguments as rational numbers and not upon their representatives.
   This is not relevant to _expressions_ (**xintexpr**), because the
   `\xintfloatexpr` parser sees there `/` as an operator and does not
   (apart from special constructs) get to manipulate fractions as such.

 - `\xintnewdummy` is public interface to a `1.2e` macro which serves to
   declare any given catcode 11 character as a dummy variable for
   expressions (**xintexpr**). This is useful for Unicode engines (the
   Latin letters being already all pre-declared as dummy variables.)

 - added `\xintiSqrtR`, there was only `\xintiiSqrtR` alongside
   `\xintiSqrt` and `\xintiiSqrt` (**xint**).

 - added non public `\xintLastItem:f:csv` to **xinttools** for faster
   `last()` function, and improved `\xintNewExpr` compatibility. Also
   `\xintFirstItem:f:csv`.

### Bug fixes

 - the `1.2f` half-integer powers computed within `\xintfloatexpr` had a
   silly rounding to the target precision just _before_ the final
   square-root extraction, thus possibly losing some precision. The
   `1.2k` implementation keeps guard digits for this final square root
   extraction. As for integer exponents, it is guaranteed that the
   computed value differs from the exact one by less than `0.52 ulp`
   (for inputs having at most `\xinttheDigits` digits.)
   
 - more regressions from `1.2i` were fixed: `\xintLen` (**xint**,
   **xintfrac**) and `\xintDouble` (**xintcore**) had forgotten that
   their argument was allowed to be negative. A regression test suite is
   now in place and is being slowly expanded to cover more macros.

 - `\xintiiSquareRoot{0}` now produces `{1}{1}`, which fits better the
   general documented behaviour of this macro than `11`.


`1.2j (2016/12/22)`
----

### Improvements and new features

 - **xinttools** and **xintexpr**:

    1. slightly improves the speed of `\xintTrim`.

    2. speed gains for the handlers of comma separated lists
       implementing Python-like slicing and item extraction. Relevant
       non (user) documented macros better documented in
       `sourcexint.pdf`.

 - significant documentations tweaks (inclusive of suppressing things!),
   and among them two beautiful hyperlinked tables with both horizontal
   and vertical rules which bring the documentation of the **xintexpr**
   syntax to a kind of awe-inspiring perfection... except that
   implementation of some math functions is still lacking.
   
### Bug fixes

 - fix two `1.2i` regressions caused by undefined macros (`\xintNthElt`
   in certain branches and `[list][N]` item extraction in certain
   cases.) The test files existed but were not executed prior to
   release. Automation in progress.


`1.2i (2016/12/13)`
----

### Breaking changes

 - `\xintDecSplit` second argument must have no sign (former code
   replaced it with its absolute value, a sign now may cause an error.)

### Removed

 - deprecated macros `\xintifTrue`, `\xintifTrueFalse`, `\xintQuo`,
   `\xintRem`, `\xintquo`, `\xintrem`.

### Improvements and new features

 - **xintkernel**: `\xintLength` is faster. New macros:

    - `\xintLastItem` to fetch the last item from its argument,

    - `\romannumeral\xintgobble` for gobbling many (up to 531440)
      upstream braced items or tokens.

    - `\romannumeral\xintreplicate` which is copied over from the expl3
      `\prg_replicate:nn` with some minor changes.

 - **xinttools**: general token list handling routines `\xintKeep`,
   `\xintTrim` and `\xintNthElt` are faster; but the novel `\xintTrim`
   can only remove up to a maximum of 531440 items.


     Also, `\xintFor` partially improves on some issues which are
   reported upon in the documentation.

 - some old macros have been rewritten entirely or partially using
   techniques which **xint** started using in release `1.2`:

    - **xintcore**: `\xintDouble`, `\xintHalf`, `\xintInc`, `\xintDec`,
      `\xintiiLDg`, `\xintDSR` (originally from **xint**), a novel
      `\xintDSRr`.

    - **xint**: `\xintDSH`, `\xintDSx`, `\xintDecSplit`, `\xintiiE`.

    - **xintfrac**: as a result of the above `\xintTrunc`, `\xintRound`
      and `\xintXTrunc` got faster. But the main improvement for them is
      with decimal inputs which formerly had not been treated separately
      from the general fraction case. Also, `\xintXTrunc` does not
      anymore create a dependency of **xintfrac** on **xinttools**.

 - the documentation has again been (slightly) re-organized; it has a
   new sub-section on the Miller-Rabin primality test, to illustrate
   some use of `\xintNewFunction` for recursive definitions.

 - the documentation has dropped the LaTeX "command" terminology (which
   had been used initially in 2013 for some forgotten reasons and should
   have been removed long ago) and uses only the more apt "macro", as
   after all, all of **xint** is about expansion of macros (plus the use
   of `\numexpr`).

### Bug fixes

 - `\xintDecSplitL` and `\xintDecSplitR` from **xint** produced their
   output in a spurious brace pair (bug introduced in `1.2f`).


`1.2h (2016/11/20)`
----

### Improvements and new features

 - new macro `\xintNewFunction` in **xintexpr** which allows to extend
   the parser syntax with functions in situations where `\xintdeffunc`
   is not usable (typically, because dummy variables are used over a not
   yet determined range of values because it depends on the variables).

 - after three years of strict obedience to `xint` prefix, now
   `\thexintexpr`, `\thexintiexpr`, `\thexintfloatexpr`, and
   `\thexintiiexpr` are provided as synonyms to `\xinttheexpr`, etc...

### Bug fixes

 - the `(cond)?{foo}{bar}` operator from **xintexpr** mis-behaved in
   certain circumstances (such as an empty `foo`).

 - the **xintexpr** `1.2f` `binomial` function (which uses
   `\xintiiBinomial` from **xint.sty** or `\xintFloatBinomial` from
   **xintfrac.sty**) deliberately raised an error for `binomial(x,y)`
   with `y<0` or `x<y`. This was unfortunate, and it now simply
   evaluates to zero in such cases.

 - similarly the `pfactorial` function was very strict and
   `pfactorial(x,y)` deliberately raised an out-of-range error if not
   used with non-negative integers with `x` less than `y`. It now avoids
   doing that and allows negative arguments.

 - the `add` and `mul` from **xintexpr**, which work with dummy
   variables since `1.1`, raised an error since `1.2c 2015/11/16` when
   the dummy variable was given an empty range (or list) of values,
   rather than producing respectively `0` and `1` as formerly.


`1.2g (2016/03/19)`
----

### Breaking changes

 - inside expressions, list item selector `[L][n]` counts starting at
   zero, not at one. This is more coherent with `[L][a:b]` which was
   already exactly like in Python since its introduction. A function
   len(L) replaces earlier `[L][0]`.

 - former `iter` keyword now called `iterr`. Indeed it matched with
   `rrseq`, the new `iter` (which was somehow missing from `1.1`) is the
   one matching `rseq`. Allows to iterate more easily with a "list"
   variable.

### Improvements and new features

 - in **xintexpr.sty**: list selectors `[L][n]` and `[L][a:b]` are more
   efficient: the earlier `1.1` routines did back and forth conversions
   from comma separated values to braced tokens, the `1.2g` routines use
   macros from **xinttools.sty** handling directly the encountered lists
   of comma separated values.

 - in **xinttools.sty**: slight improvements in the efficiency of the
   `\xintNthElt`, `\xintKeep`, `\xintTrim` routines and new routines
   handling directly comma separated values. The latter are not included
   in the user manual (they are not `\long`, they don't make efforts to
   preserve some braces, do not worry about spaces, all those worries
   being irrelevant to the use in expressions for list selectors).

 - a slight speed improvement to `\xintFloatSqrt` in its quest of
   correct rounding.

 - float multiplication and division handle more swiftly operands
   (non-fractional) with few digits, when the float precision is large.

 - the syntax of expressions is described in a devoted chapter of the
   documentation; an example shows how to implement (expandably) the
   Brent-Salamin algorithm for computation of Pi using `iter` in a float
   expression.


`1.2f (2016/03/12)`
----

### Breaking changes

 - no more `\xintFac` macro but `\xintiFac/\xintiiFac/\xintFloatFac`.

### Improvements and new features

 - functions `binomial`, `pfactorial` and `factorial` in both integer
   and float versions.

 - macros `\xintiiBinomial`, `\xintiiPFactorial`
   (**xint.sty**) and `\xintFloatBinomial`, `\xintFloatPFactorial`
   (**xintfrac.sty**). Improvements to `\xintFloatFac`.

 - faster implementation and increased accuracy of float power macros.
   Half-integer exponents are now accepted inside float expressions.

 - faster implementation of both integral and float square root macros.

 - the float square root achieves
   *correct* (aka *exact*) rounding in arbitrary precision.

 - modified behaviour for the `\xintPFloat` macro, used by
   `\xintthefloatexpr` to prettify its output. It now opts for decimal
   notation if and only if scientific notation would use an exponent between
   `-5` and `5` inclusive. The zero value is printed `0.` with a dot.

 - the float macros for addition, subtraction, multiplication, division now
   first round their two operands to P, not P+2, significant places before
   doing the actual computation (P being the target precision). The same
   applies to the power macros and to the square root macro.

 - the documentation offers a more precise (and accurate) discussion of
   floating point issues.

 - various under-the-hood code improvements; the floatexpr operations are
   chained in a faster way, from skipping some unneeded parsing on results of
   earlier computations. The absence of a real inner data structure for floats
   (incorporating their precisions, for one) is however still a bit hair
   raising: currently the lengths of the mantissas of the operands are computed
   again by each float macro or expression operation.

 - (TeXperts only) the macros defined (internally) from `\xintdeffunc` et al.
   constructs do not incorporate an initial `\romannumeral` anymore.

 - renewed desperate efforts at improving the documentation by random
   shuffling of sections and well thought additions; cuts were considered and
   even performed.

### Bug fixes

 - squaring macro `\xintSqr` from **xintfrac.sty** was broken due to a
   misspelled sub-macro name. Dates back to `1.1` release of `2014/10/28`
   `:-((`.

 - `1.2c`'s fix to the subtraction bug from `1.2` introduced another bug,
   which in some cases could create leading zeroes in the output, or even
   worse. This could invalidate other routines using subtractions, like
   `\xintiiSquareRoot`.

 - the comparison operators were not recognized by `\xintNewIIExpr` and
   `\xintdefiifunc` constructs.


`1.2e (2015/11/22)`
----

### Improvements and new features

 - macro `\xintunassignvar`.

 - slight modifications of the logged messages in case of `\xintverbosetrue`.

 - a space in  `\xintdeffunc f(x)<space>:= expression ;` is now accepted.

 - documentation enhancements: the _Quick Sort_ section with its included
   code samples has been entirely re-written;  the _Commands of the xintexpr
   package_ section has been extended and reviewed entirely.

### Bug fixes

 - in **xintfrac**: the `\xintFloatFac` from release `1.2` parsed its
   argument only through `\numexpr` but it should have used `\xintNum`.

 - in **xintexpr**: release `1.2d` had broken the recognition of
   sub-expressions immediately after variable names (with tacit
   multiplication).

 - in **xintexpr**: contrarily to what `1.2d` documentation said, tacit
   multiplication was not yet always done with enhanced precedence. Now
   yes.


`1.2d (2015/11/18)`
----

### Improvements and new features

 - the function definitions done by `\xintdeffunc` et al., as well as
   the macro declarations by `\xintNewExpr` et al. now have only local
   scope.

 - tacit multiplication applies to more cases, for example (x+y)z, and
   always ties more than standard * infix operator, e.g. x/2y is like
   x/(2*y).

 - some documentation enhancements, particularly in the chapter on
   xintexpr.sty, and also in the code source comments.

### Bug fixes

 - in **xintcore**: release `1.2c` had inadvertently broken the
   `\xintiiDivRound` macro.


`1.2c (2015/11/16)`
----

### Improvements and new features

 - macros `\xintdeffunc`, `\xintdefiifunc`, `\xintdeffloatfunc` and
   boolean `\ifxintverbose`.

 - on-going code improvements and documentation enhancements, but
   stopped in order to issue this bugfix release.

### Bug fixes

 - in **xintcore**: recent release `1.2` introduced a bug in the
   subtraction (happened when 00000001 was found under certain
   circumstances at certain mod 8 locations).


`1.2b (2015/10/29)`
----

### Bug fixes

 - in **xintcore**: recent release `1.2` introduced a bug in the division
   macros, causing a crash when the divisor started with 99999999 (it was
   attempted to use with 1+99999999 a subroutine expecting only 8-digits
   numbers).


`1.2a (2015/10/19)`
----

### Improvements and new features

 - added `\xintKeepUnbraced`, `\xintTrimUnbraced` (**xinttools**) and fixed
   documentation of `\xintKeep` and `\xintTrim` regarding brace stripping.

 - added `\xintiiMaxof/\xintiiMinof` (**xint**).

 - TeX hackers only: replaced all code uses of ``\romannumeral-`0``
   by the quicker ``\romannumeral`&&@`` (`^` being used as letter,
   had to find another character usable with catcode 7).

### Bug fixes

 - in **xintexpr**: recent release `1.2` introduced a bad bug in the
   parsing of decimal numbers and as a result `\xinttheexpr 0.01\relax`
   expanded to `0` ! (sigh...)


`1.2 (2015/10/10)`
----

### Removed

 - the macros `\xintAdd`, `\xintSub`, `\xintMul`, `\xintMax`,
   `\xintMin`, `\xintMaxof`, `\xintMinof` are removed from package
   **xint**, and only exist in the versions from **xintfrac**. With only
   **xintcore** or **xint** loaded, one _must_ use `\xintiiAdd`,
   `\xintiiSub`, ..., or `\xintiAdd`, `\xintiSub`, etc...

### Improvements and new features

 - the basic arithmetic implemented in **xintcore** has been entirely
   rewritten. The mathematics remains the elementary school one, but the
   `TeX` implementation achieves higher speed (except, regarding
   addition/subtraction, for numbers up to about thirty digits), the
   gains becoming quite significant for numbers with hundreds of digits.

 - the inputs must have less than 19959 digits. But computations with
   thousands of digits take time.

 - a previously standing limitation of `\xintexpr`, `\xintiiexpr`, and
   of `\xintfloatexpr` to numbers of less than 5000 digits has been
   lifted.

 - a *qint* function is provided to help the parser gather huge integers
   in one-go, as an exception to its normal mode of operation which
   expands token by token.

 - `\xintFloatFac` macro for computing the factorials of integers as
   floating point numbers to a given precision. The `!` postfix operator
   inside `\xintfloatexpr` maps to this new macro rather than to the
   exact factorial as used by `\xintexpr` and `\xintiiexpr`.

 - there is more flexibility in the parsing done by the macros from
   **xintfrac** on fractional input: the decimal parts of both the
   numerator and the denominator may arise from a separate expansion via
   ``\romannumeral-`0``. Also the strict `A/B[N]` format is a bit
   relaxed: `N` may be anything understood by `\numexpr` (it could even
   be empty but that possibility has been removed by later `1.2f`
   release.)

 - on the other hand an isolated dot `.` is not legal syntax anymore
   inside the expression parsers: there must be digits either before or
   after. It remains legal input for the macros of **xintfrac**.

 - added `\ht`, `\dp`, `\wd`, `\fontcharht`, etc... to the tokens
   recognized by the parsers and expanded by `\number`.

 - an obscure bug in package **xintkernel** has been fixed, regarding
   the sanitization of catcodes: under certain circumstances (which
   could not occur in a normal `LaTeX` context), unusual catcodes could
   end up being propagated to the external world.

 - an effort at randomly shuffling around various pieces of the
   documentation has been done.


`1.1c (2015/09/12)`
----

 - bugfix regarding macro `\xintAssign` from **xinttools** which did
   not behave correctly in some circumstances (if there was a space
   before `\to`, in particular).

 - very minor code improvements, and correction of some issues
   regarding the source code formatting in `sourcexint.pdf`, and
   minor issues in `Makefile.mk`.


`1.1b (2015/08/31)`
----

 - bugfix: some macros needed by the integer division routine from
   **xintcore** had been left in **xint.sty** since release `1.1`. This
   for example broke the `\xintGCD` from **xintgcd** if package **xint**
   was not loaded.

 - Slight enhancements to the documentation, particularly in the
   `Read this first` section.


`1.1a (2014/11/07)`
----

 - fixed a bug which prevented `\xintNewExpr` from producing correctly working
   macros from a comma separated replacement text.

 - `\xintiiSqrtR` for rounded integer square root; former `\xintiiSqrt`
   already produced truncated integer square root; corresponding function
   `sqrtr` added to `\xintiiexpr..\relax` syntax.

 - use of straight quotes in the documentation for better legibility.

 - added `\xintiiIsOne`, `\xintiiifOne`, `\xintiiifCmp`, `\xintiiifEq`,
    `\xintiiifGt`, `\xintiiifLt`, `\xintiiifOdd`, `\xintiiCmp`, `\xintiiEq`,
    `\xintiiGt`, `\xintiiLt`, `\xintiiLtorEq`, `\xintiiGtorEq`, `\xintiiNeq`,
    mainly for efficiency of `\xintiiexpr`.

 - for the same reason, added `\xintiiGCD` and `\xintiiLCM`.

 - added the previously mentioned `ii` macros, and some others from `1.1`, to
   the user manual. But their main usage is internal to `\xintiiexpr`, to skip
   unnecessary overheads.

 - various typographical fixes throughout the documentation, and a bit
   of clean up of the code comments. Improved `\Factors` example of nested
   `subs`, `rseq`, `iter` in `\xintiiexpr`.


`1.1 (2014/10/28)`
----

### Breaking changes

 - in `\xintiiexpr`, `/` does _rounded_ division, rather than the
   Euclidean division (for positive arguments, this is truncated division).
   The `//` operator does truncated division,

 - the `:` operator for three-way branching is gone, replaced with `??`,

 - `1e(3+5)` is now illegal. The number parser identifies `e` and `E`
   in the same way it does for the decimal mark, earlier versions treated
   `e` as `E` rather as infix operators of highest precedence,

 - the `add` and `mul` have a new syntax, old syntax is with `` `+` `` and
   `` `*` `` (left quotes mandatory), `sum` and `prd` are gone,

 - no more special treatment for encountered brace pairs `{..}` by the
   number scanner, `a/b[N]` notation can be used without use of braces (the
   `N` will end up as is in a `\numexpr`, it is not parsed by the
   `\xintexpr`-ession scanner),

 - in earlier releases, place holders for `\xintNewExpr` could either
   be denoted `#1`, `#2`, ... or also `$1`, `$2`, ...
   Only the usual `#` form is now accepted and the special cases previously
   treated via the second form are now managed via a `protect(...)` function.

 - **xintfrac**: `\xintFloor` and `\xintCeil` add a trailing `/1[0]` to their
   (integer) output. New `\xintiFloor` and `\xintiCeil` do not.

### Removed

 - `\xintnumexpr`, `\xintthenumexpr`, `\xintNewNumExpr`: use
   `\xintiexpr`, `\xinttheiexpr`, `\xintNewIExpr`.

### Deprecated

 - `\xintDivision`, `\xintQuo`, `\xintRem`: use `\xintiDivision`,
   `\xintiQuo`, `\xintiRem`.

 - `\xintMax`, `\xintMin`, `\xintAdd`, `\xintSub`, `\xintMul`
   (**xint**): their usage without **xintfrac** is deprecated; use
   `\xintiMax`, `\xintiMin`, `\xintiAdd`, `\xintiSub`, `\xintiMul`.

 - the `&` and `|` as Boolean operators in `xintexpr`-essions are
   deprecated in favour of `&&` and `||`. The single letter operators
   might be assigned some other meaning in some later release (bitwise
   operations, perhaps). Do not use them.

### Improvements and new features

  * new package **xintcore** has been split off **xint**. It contains the
      core arithmetic macros (it is loaded by LaTeX package **bnumexpr**),

  * neither **xint** nor **xintfrac** load **xinttools**. Only
      **xintexpr** does,

  * whenever some portion of code has been revised, often use has been made of
      the `\xint_dothis` and `\xint_orthat` pair of macros for expandably
      branching,

  * these tiny helpful macros, and a few others are in package
      **xintkernel** which contains also the catcode and loading order
      management code, initially inspired by code found in Heiko Oberdiek's
      packages,

  * the source code, which was suppressed from `xint.pdf` in release
      `1.09n`, is now compiled into a separate file `sourcexint.pdf`,

  * faster handling by `\xintAdd`, `\xintSub`, `\xintMul`, ... of the case
      where one of the arguments is zero,

  * the `\xintAdd` and `\xintSub` macros from package **xintfrac** check if
      one of the denominators is a multiple of the other, and only if this is
      not the case do they multiply the denominators. But systematic reduction
      would be too costly,

  * this naturally will be also the case for the `+` and `-` operations
      in `\xintexpr`,

  * **xint** added `\xintiiDivRound`, `\xintiiDivTrunc`, `\xintiiMod`
    for rounded and truncated division of big integers (next to
    `\xintiiQuo` and `\xintiiRem`),

  * with **xintfrac** loaded, the `\xintNum` macro does `\xintTTrunc`
      (which is truncation to an integer, same as `\xintiTrunc {0}`),

  * added `\xintMod` to **xintfrac** for modulo operation with
      fractional numbers,

  * added `\xintiFloor` and `\xintiCeil` to **xintfrac**,

  * `\xintiexpr`, `\xinttheiexpr` admit an optional argument within brackets
      `[d]`, they round the computation result (or results, if comma separated)
      to `d` digits after decimal mark, (the whole computation is done exactly,
      as in `xintexpr`),

  * `\xintfloatexpr`, `\xintthefloatexpr` similarly admit an optional
      argument which serves to keep only `d` digits of precision, getting rid
      of cumulated uncertainties in the last digits (the whole computation is
      done according to the precision set via `\xintDigits`),

  * `\xinttheexpr` and `\xintthefloatexpr` _pretty-print_ if possible, the
      former removing unit denominator or `[0]` brackets, the latter avoiding
      scientific notation if decimal notation is practical,

  * the `//` does truncated division and `/:` is the associated modulo,

  * multi-character operators `&&`, `||`, `==`, `<=`, `>=`, `!=`,
      `**`,

  * multi-letter infix binary words `'and'`, `'or'`, `'xor'`, `'mod'`
      (straight quotes mandatory),

  * functions `even`, `odd`,

  * `\xintdefvar A3:=3.1415;` for variable definitions (non expandable,
    naturally), usable in subsequent expressions; variable names may contain
    letters, digits, underscores. They should not start with a digit, the `@`
    is reserved, and single lowercase and uppercase Latin letters are
    predefined to work as dummy variables (see next),

  * generation of comma separated lists `a..b`, `a..[d]..b`,

  * Python syntax-like list extractors `[list][n:]`, `[list][:n]`,
    `[list][a:b]` allowing negative indices, but no optional step argument,
    and `[list][n]` (`n=0` for the number of items in the list),

  * functions `first`, `last`, `reversed`,

  * itemwise operations on comma separated lists `a*[list]`, etc.., possible
    on both sides `a*[list]^b`, and obeying the same precedence rules as with
    numbers,

  * `add` and `mul` must use a dummy variable: `add(x(x+1)(x-1), x=-10..10)`,

  * variable substitutions with `subs`:
    `subs(subs(add(x^2+y^2,x=1..y),y=t),t=20)`,

  * sequence generation using `seq` with a dummy variable: `seq(x^3,
    x=-10..10)`,

  * simple recursive lists with `rseq`, with `@` given the last value,
     `rseq(1;2@+1,i=1..10)`,

  * higher recursion with `rrseq`, `@1`, `@2`, `@3`, `@4`, and `@@(n)`
     for earlier values, up to `n=K` where `K` is the number of terms of the
     initial stretch `rrseq(0,1;@1+@2,i=2..100)`,

  * iteration with `iter` which is like `rrseq` but outputs only the
     last `K` terms, where `K` was the number of initial terms,

  * inside `seq`, `rseq`, `rrseq`, `iter`, possibility to use `omit`,
     `abort` and `break` to control termination,

  * `n++` potentially infinite index generation for `seq`, `rseq`,
     `rrseq`, and `iter`, it is advised to use `abort` or `break(..)` at
     some point,

  * the `add`, `mul`, `seq`, ... are nestable,

  * `\xintthecoords` converts a comma separated list of an even number
     of items to the format expected by the `TikZ` `coordinates` syntax,

  * completely new version `\xintNewExpr`, `protect` function to handle
     external macros. The dollar sign
     `$` for place holders is not accepted anymore, only the standard macro
     parameter `#`. Not all constructs are compatible with `\xintNewExpr`.
% $ this docstripped line for emacs buffer fontification issues in doctex-mode

### Bug fixes

  - `\xintZapFirstSpaces` hence also `\xintZapSpaces` from package **xinttools**
      were buggy when used with  an argument either empty or containing only
      space tokens.

  - `\xintiiexpr` did not strip leading zeroes, hence
    `\xinttheiiexpr 001+1\relax` did not obtain the expected result ...

  - `\xinttheexpr \xintiexpr 1.23\relax\relax` should have produced `1`,
  but it produced `1.23`

  - the catcode of `;` was not set at package launching time.

  - the `\XINTinFloatPrd:csv` macro name had a typo, hence `prd` was
    non-functional in `\xintfloatexpr`.


`1.09n (2014/04/01)`
----

  * the user manual does not include by default the source code
    anymore: the `\NoSourceCode` toggle in file `xint.tex` has to
    be set to 0 before compilation to get source code inclusion
    (later release `1.1` made source code available as `sourcexint.pdf`).

  * bug fix (**xinttools**) in `\XINT_nthelt_finish` (this bug was
    introduced in `1.09i` of `2013/12/18` and showed up when the index
    `N` was larger than the number of elements of the list).


`1.09m (2014/02/26)`
----

  * new in **xinttools**: `\xintKeep` keeps the first `N` or last
    `N` elements of a list (sequence of braced items); `\xintTrim`
    cuts out either the first `N` or the last `N` elements from a
    list.

  * new in **xintcfrac**: `\xintFGtoC` finds the initial partial
    quotients common to two numbers or fractions `f` and `g`;
    `\xintGGCFrac` is a clone of `\xintGCFrac` which however does not
    assume that the coefficients of the generalized continued
    fraction are numeric quantities. Some other minor changes.


`1.09kb (2014/02/13)`
----

  * bug fix (**xintexpr**): an aloof modification done by `1.09i` to
    `\xintNewExpr` had resulted in a spurious trailing space present
    in the outputs of all macros created by `\xintNewExpr`, making
    nesting of such macros impossible.

  * bug fix (**xinttools**): `\xintBreakFor` and `\xintBreakForAndDo`
    were buggy when used in the last iteration of an `\xintFor` loop.

  * bug fix (**xinttools**): `\xintSeq` from `1.09k` needed a `\chardef`
    which was missing from `xinttools.sty`, it was in `xint.sty`.


`1.09k (2014/01/21)`
----

  * inside `\xintexpr..\relax` (and its variants) tacit multiplication is
    implied when a number or operand is followed directly with an
    opening parenthesis,

  * the `"` for denoting (arbitrarily big) hexadecimal numbers is
    recognized by `\xintexpr` and its variants (package
    **xintbinhex** is required); a fractional hexadecimal part
    introduced by a dot `.` is allowed.

  * re-organization of the first sections of the user manual.

  * bug fix (**xinttools**, **xint**, ...): forgotten catcode check of
    `"` at loading time has been added.


`1.09j (2014/01/09)`
----

  * (**xint**) the core division routines have been re-written for some
    (limited) efficiency gain, more pronounced for small divisors. As a
    result the *computation of one thousand digits of $\pi$* is close
    to three times faster than with earlier releases.

  * some various other small improvements, particularly in the power
    routines.

  * (**xintfrac**) a macro `\xintXTrunc` is designed to produce
    thousands or even tens of thousands of digits of the decimal
    expansion of a fraction. Although completely expandable it has its
    use limited to inside an `\edef`, `\write`, `\message`, \dots. It
    can thus not be nested as argument to another package macro.

  * (**xintexpr**) the tacit multiplication done in `\xintexpr..\relax`
    on encountering a count register or variable, or a `\numexpr`,
    while scanning a (decimal) number, is extended to the case of a sub
    `\xintexpr`-ession.

  * `\xintexpr` can now be used in an `\edef` with no `\xintthe` prefix;
    it will execute completely the computation, and the error message
    about a missing `\xintthe` will be inhibited. Previously, in the
    absence of `\xintthe`, expansion could only be a full one (with
    ``\romannumeral-`0``), not a complete one (with `\edef`). Note
    that this differs from the behavior of the non-expandable
    `\numexpr`: `\the` or `\number` (or `\romannumeral`) are needed
    not only to print but
    also to trigger the computation, whereas `\xintthe` is mandatory
    only for the printing step.

  * the default behavior of `\xintAssign` is changed, it now does not
    do any further expansion beyond the initial full-expansion which
    provided the list of items to be assigned to macros.

  * bug fix (**xintfrac**): `1.09i` did an unexplainable change to
    `\XINT_infloat_zero` which broke the floating point routines for
    vanishing operands =:(((

  * bug fix: the `1.09i` `xint.ins` file produced a buggy `xint.tex` file.


`1.09i (2013/12/18)`
----

  * (**xintexpr**) `\xintiiexpr` is a variant of `\xintexpr` which is
    optimized to deal only with (long) integers, `/` does a euclidean
    quotient.

  * *deprecated*: `\xintnumexpr`, `\xintthenumexpr`, `\xintNewNumExpr` are
    renamed, respectively, `\xintiexpr`, `\xinttheiexpr`, `\xintNewIExpr`. The
    earlier denominations are kept but are to be removed at some point.

  * it is now possible within `\xintexpr...\relax` and its variants to
    use count, dimen, and skip registers or variables without
    explicit `\the/\number`: the parser inserts automatically
    `\number` and a tacit multiplication is implied when a register
    or variable immediately follows a number or fraction. Regarding
    dimensions and `\number`, see the further discussion in
    *Dimensions*.

  * (**xintfrac**) conditional `\xintifOne`; `\xintifTrueFalse`
    renamed to `\xintifTrueAelseB`; macros `\xintTFrac`
    (`fractional part`, mapped to function `frac` in
    `\xintexpr`-essions), `\xintFloatE`.

  * (**xinttools**) `\xintAssign` admits an optional argument to
    specify the expansion type to be used: `[]` (none, default), `[o]`
    (once), `[oo]` (twice), `[f]` (full), `[e]` (`\edef`),... to define
    the macros

  * **xinttools** defines `\odef`, `\oodef`, `\fdef` (if the names have
    already been assigned, it uses `\xintoodef` etc...). These tools are
    provided for the case one uses the package macros in a non-expandable
    context. `\oodef` expands twice the macro replacement text, and `\fdef`
    applies full expansion. They are useful in situations where one does not
    want a full `\edef`. `\fdef` appears to be faster than `\oodef` in almost
    all cases (with less than thousand digits in the result), and even faster
    than `\edef` for expanding the package macros when the result has a few
    dozens of digits. `\oodef` needs that expansion ends up in thousands of
    digits to become competitive with the other two.

  * some across the board slight efficiency improvement as a result of
    modifications of various types to *fork macros* and *branching
    conditionals* which are used internally.

  * bug fix (**xint**): `\xintAND` and `\xintOR` inserted a space token
    in some cases and did not expand as promised in two steps `:-((`
    (bug dating back to `1.09a` I think; this bug was without
    consequences when using `&` and `|` in `\xintexpr-essions`, it
    affected only the macro form).

  * bug fix (**xintcfrac**): `\xintFtoCCv` still ended fractions with
    the `[0]`'s which were supposed to have been removed since release
    `1.09b`.

  * *deprecated*: `\xintifTrueFalse`, `\xintifTrue`; use `\xintifTrueAelseB`.


`1.09h (2013/11/28)`
----

  * parts of the documentation have been re-written or re-organized,
    particularly the discussion of expansion issues and of input and
    output formats.

  * the expansion types of macro arguments are documented in the margin
    of the macro descriptions, with conventions mainly taken over
    from those in the `LaTeX3` documentation.

  * a dependency of **xinttools** on **xint** (inside `\xintSeq`) has
    been removed.

  * (**xintgcd**) `\xintTypesetEuclideAlgorithm` and
    `\xintTypesetBezoutAlgorithm` have been slightly modified
    (regarding indentation).

  * (**xint**) macros `xintiSum` and `xintiPrd` are renamed to
    `\xintiiSum` and `\xintiiPrd`.

  * (**xinttools**) a count register used in `1.09g` in the `\xintFor`
    loops for parsing purposes has been removed and replaced by use of
    a `\numexpr`.

  * the few uses of `\loop` have been replaced by `\xintloop/\xintiloop`.

  * all macros of **xinttools** for which it makes sense are now declared
    `\long`.


`1.09g (2013/11/22)`
----

  * a package **xinttools** is detached from **xint**, to make tools such
    as `\xintFor`, `\xintApplyUnbraced`, and `\xintiloop` available
    without the **xint** overhead.

  * expandable nestable loops `\xintloop` and `\xintiloop`.

  * bugfix: `\xintFor` and `\xintFor*` do not modify anymore the value of
    `\count 255`.


`1.09f (2013/11/04)`
----

  * (**xint**) `\xintZapFirstSpaces`, `\xintZapLastSpaces`,
    `\xintZapSpaces`, `\xintZapSpacesB`, for expandably stripping away
    leading and/or ending spaces.

  * `\xintCSVtoList` by default uses `\xintZapSpacesB` to strip away
    spaces around commas (or at the start and end of the comma
    separated list).

  * also the `\xintFor` loop will strip out all spaces around commas and
    at the start and the end of its list argument; and similarly for
    `\xintForpair`, `\xintForthree`, `\xintForfour`.

  * `\xintFor` *et al.* accept all macro parameters from `#1` to
    `#9`.

  * for reasons of inner coherence some macros previously with one extra
    `i` in their names (e.g. `\xintiMON`) now have a doubled
    `ii` (`\xintiiMON`) to indicate that they skip the overhead of
    parsing their inputs via `\xintNum`. Macros with a *single*
    `i` such as `\xintiAdd` are those which maintain the
    non-**xintfrac** output format for big integers, but do parse
    their inputs via `\xintNum` (since release `1.09a`). They too may
    have doubled-`i` variants for matters of programming optimization
    when working only with (big) integers and not fractions or
    decimal numbers.


`1.09e (2013/10/29)`
----

  * (**xint**) `\xintintegers`, `\xintdimensions`, `\xintrationals`
    for infinite `\xintFor` loops, interrupted with `\xintBreakFor` and
    `\xintBreakForAndDo`.

  * `\xintifForFirst`, `\xintifForLast` for the `\xintFor` and
    `\xintFor*` loops,

  * the `\xintFor` and `xintFor*` loops are now `\long`, the
    replacement text and the items may contain explicit `\par`'s.

  * conditionals `\xintifCmp`, `\xintifInt`, `\xintifOdd`.

  * bug fix (**xint**): the `\xintFor` loop (not `\xintFor*`) did
    not correctly detect an empty list.

  * bug fix (**xint**): `\xintiSqrt {0}` crashed. `:-((`

  * the documentation has been enriched with various additional examples,
    such as the *the quick sort algorithm
    illustrated* or the various ways of *computing prime numbers*.

  * the documentation explains with more details various expansion
    related issues, particularly in relation to conditionals.


`1.09d (2013/10/22)`
----

  * bug fix (**xint**): `\xintFor*` is modified to gracefully
    handle a space token (or more than one) located at the very end of
    its list argument (as the space before `\do` in `\xintFor* #1 in
    {{a}{b}{c}<space>} \do {stuff}`; spaces at other locations were
    already harmless). Furthermore this new version _f-expands_ the
    un-braced list items. After `\def\x{{1}{2}}` and `\def\y{{a}\x
    {b}{c}\x }`, `\y` will appear to `\xintFor*` exactly as if it had
    been defined as `\def\y{{a}{1}{2}{b}{c}{1}{2}}`.

  * same bug fix for `\xintApplyInline`.


`1.09c (2013/10/09)`
----

  * (**xintexpr**) added `bool` and `togl` to the `\xintexpr` syntax;
    also added `\xintboolexpr` and `\xintifboolexpr`.

  * added `\xintNewNumExpr`.

  * the factorial `!` and branching `?`, `:`, operators (in
    `\xintexpr...\relax`) have now less precedence than a function
    name located just before,

  * (**xint**) `\xintFor` is a new type of loop, whose replacement text
    inserts the comma separated values or list items via macro
    parameters, rather than encapsulated in macros; the loops are
    nestable up to four levels (nine levels since `1.09f`) and their
    replacement texts are allowed to close groups as happens with the
    tabulation in alignments,

  * `\xintForpair`, `\xintForthree`, `\xintForfour` are experimental
    variants of `\xintFor`,

  * `\xintApplyInline` has been enhanced in order to be usable for
    generating rows (partially or completely) in an alignment,

  * command `\xintSeq` to generate (expandably) arithmetic sequences
    of (short) integers,

  * again various improvements and changes in the documentation.


`1.09b (2013/10/03)`
----

  * various improvements in the documentation,

  * more economical catcode management and re-loading handling,

  * removal of all those `[0]`'s previously forcefully added at the end
    of fractions by various macros of **xintcfrac**,

  * `\xintNthElt` with a negative index returns from the tail of the
    list,

  * macro `\xintPRaw` to have something like what `\xintFrac` does in
    math mode; i.e. a `\xintRaw` which does not print the denominator
    if it is one.


`1.09a (2013/09/24)`
----

  * (**xintexpr**) `\xintexpr..\relax` and `\xintfloatexpr..\relax`
    admit functions in their syntax, with comma separated values as
    arguments, among them `reduce, sqr, sqrt, abs, sgn, floor, ceil,
    quo, rem, round, trunc, float, gcd, lcm, max, min, sum, prd, add,
    mul, not, all, any, xor`.

  * comparison (`<`, `>`, `=`) and logical (`|`, `&`) operators.

  * the command `\xintthe` which converts `\xintexpr`essions into
    printable format (like `\the` with `\numexpr`) is more efficient,
    for example one can do `\xintthe\x` if `\x` was defined to be an
    `\xintexpr..\relax`:

        \def\x{\xintexpr 3^57\relax}
        \def\y{\xintexpr \x^(-2)\relax}
        \def\z{\xintexpr \y-3^-114\relax}
        \xintthe\z

  * `\xintnumexpr .. \relax` (now renamed `\xintiexpr`) is `\xintexpr
    round( .. ) \relax`.

  * `\xintNewExpr` now works with the standard macro parameter character
    `#`.

  * both regular `\xintexpr`-essions and commands defined by
    `\xintNewExpr` will work with comma separated lists of
    expressions,

  * commands `\xintFloor`, `\xintCeil`, `\xintMaxof`, `\xintMinof`
    (package **xintfrac**), `\xintGCDof`, `\xintLCM`, `\xintLCMof`
    (package **xintgcd**), `\xintifLt`, `\xintifGt`, `\xintifSgn`,
    `\xintANDof`, ...

  * The arithmetic macros from package **xint** now filter their operands
    via `\xintNum` which means that they may use directly count
    registers and `\numexpr`-essions without having to prefix them by
    `\the`. This is thus similar to the situation holding previously
    already when **xintfrac** was loaded.

  * a bug (**xintfrac**) introduced in `1.08b` made `\xintCmp` crash
    when one of its arguments was zero. `:-((`


`1.08b (2013/06/14)`
----

  * (**xintexpr**) Correction of a problem with spaces inside
    `\xintexpr`-essions.

  * (**xintfrac**) Additional improvements to the handling of floating
    point numbers.

  * section *Use of count registers* documenting how count
    registers may be directly used in arguments to the macros of
    **xintfrac**.


`1.08a (2013/06/11)`
----

  * (**xintfrac**) Improved efficiency of the basic conversion from
    exact fractions to floating point numbers, with ensuing speed gains
    especially for the power function macros `\xintFloatPow` and
    `\xintFloatPower`,

  * Better management by `\xintCmp`, `\xintMax`, `\xintMin` and
    `\xintGeq` of inputs having big powers of ten in them.

  * Macros for floating point numbers added to the **xintseries**
    package.


`1.08 (2013/06/07)`
----

  * (**xint** and **xintfrac**) Macros for extraction of square roots,
    for floating point numbers (`\xintFloatSqrt`), and integers
    (`\xintiSqrt`).

  * new package **xintbinhex** providing *conversion routines* to and from
    binary and hexadecimal bases.


`1.07 (2013/05/25)`
----

  * The **xintexpr** package is a new core constituent (which loads
    automatically **xintfrac** and **xint**) and implements the
    expandable expanding parser

        \xintexpr . . . \relax,

    and its variant

        \xintfloatexpr . . . \relax

    allowing on input formulas using the infix operators `+`, `-`, `*`,
    `/`, and `^`, and arbitrary levels of parenthesizing. Within a
    float expression the operations are executed according to the
    current value set by `\xintDigits`. Within an `\xintexpr`-ession the
    binary operators are computed exactly.

    To write the `\xintexpr` parser I benefited from the commented
    source of the `l3fp` parser; the `\xintexpr` parser has its own
    features and peculiarities. *See its documentation*.

  * The floating point precision `D` is set (this is a local assignment
    to a `\mathchar` variable) with `\xintDigits := D;` and queried
    with `\xinttheDigits`. It may be set to anything up to
    `32767`.[^1] The macro incarnations of the binary operations
    admit an optional argument which will replace pointwise `D`; this
    argument may exceed the `32767` bound.

  * The **xintfrac** macros now accept numbers written in scientific
    notation, the `\xintFloat` command serves to output its argument
    with a given number `D` of significant figures. The value of `D`
    is either given as optional argument to `\xintFloat` or set with
    `\xintDigits := D;`. The default value is `16`.

[^1]: but values higher than 100 or 200 will presumably give too slow
evaluations.


`1.06b (2013/05/14)`
----

  * Minor code and documentation improvements. Everywhere in the source
   code, a more modern underscore has replaced the @ sign.


`1.06 (2013/05/07)`
----

  * Some code improvements, particularly for macros of **xint** doing loops.

  * New utilities in **xint** for expandable manipulations of lists:

        \xintNthElt, \xintCSVtoList, \xintRevWithBraces

  * The macros did only a double expansion of their arguments. They now
    fully expand them (using ``\romannumeral-`0``). Furthermore, in the
    case of arguments constrained to obey the TeX bounds they will be
    inserted inside a `\numexpr..\relax`, hence completely expanded, one
    may use count registers, even infix arithmetic operations, etc...


`1.05 (2013/05/01)`
----

Minor changes and additions to **xintfrac** and **xintcfrac**.


`1.04 (2013/04/25)`
----

  * New component **xintcfrac** devoted to continued fractions.

  * **xint**: faster division.

  * **xint**: added expandable macros `\xintListWithSep` and `\xintApply` to
    handle token lists.

  * **xintfrac**: added `\xintRound`.

  * **xintseries** has a new implementation of `\xintPowerSeries` based
    on a Horner scheme, and new macro `\xintRationalSeries`. Both to
    help deal with the *denominator buildup* plague.

  * `tex xint.dtx` extracts style files (no need for a `xint.ins`).

  * Bug fix (**xintfrac**): `\xintIrr {0}` crashed.


`1.03 (2013/04/14)`
----

  * New modules **xintfrac** (expandable operations on fractions) and
    **xintseries** (expandable partial sums with xint package).

  * Slightly improved division and faster multiplication (the best
    ordering of the arguments is chosen automatically).

  * Added illustration of Machin algorithm to the documentation.


`1.0 (2013/03/28)`
----

Initial announcement:

>   The **xint** package implements with expandable TeX macros the basic
    arithmetic operations of addition, subtraction, multiplication
    and division, as applied to arbitrarily long numbers represented
    as chains of digits with an optional minus sign.

>  The **xintgcd** package provides implementations of the Euclidean
    algorithm and of its typesetting.

>  The packages may be used with Plain and with LaTeX.

%</changes>------------------------------------------------------
%<*makefile>------------------------------------------------------
# This file: Makefile.mk (generated from xint.dtx)
# "make --file=Makefile.mk help"

# Starting with xint 1.3c, Latexmk is used for easier compilation of
# sourcexint.pdf as the latter then included indices. These indices
# got removed at 1.3e but usage of Latexmk is maintained for the build.

# Originally tested on Mac OS X Mavericks with GNU Make 3.81,
# TeXLive 2014 and Pandoc 1.13.1.

# Note to myself: I wanted to use .RECIPEPREFIX = > but it is
# supported only with GNU Make 3.82 and later.

# this crazyness is to circumvent a problem with docstrip generation
# of the Makefile; we do not want two empty lines becoming only one
nullstring :=
define newline
$(nullstring)

endef
# will speed-up a little, I think.
newline := $(newline)

define helptext
==== INSTRUCTIONS

It is recommended to work with xint.dtx and Makefile moved to some
otherwise empty temporary repertory.

make help
    displays this help using the more pager.

make helpless
    displays this help using the less pager.

make doc
    produces all documentation, requires Latexmk and Pandoc,
    (as well as grep and file output redirection ">")

make all
    produces all documentation, then creates xint.tds.zip.

make xint.tds.zip
    same as "make all"

make xint.pdf
    extracts files and produces xint.pdf, using latex and dvipdfmx.
    Requires Latexmk.

make sourcexint.pdf
    extracts files and produces sourcexint.pdf, using latex, makeindex
    and dvipdfmx. Requires Latexmk (as well as grep and file ">" redirection).

make CHANGES.html
    requires Pandoc.

make clean
    removes auxiliary files and repertories.

make cleanall
    removes all files, leaving only xint.dtx (and Makefile).

==== INSTALLING

The following has been tested on a TeXLive installation:

make installhome
    creates xint.tds.zip, and unzips it in <TEXMFHOME>

make installlocal
    creates xint.tds.zip, unzips it in <TEXMFLOCAL>
        and then does texhash <TEXMFLOCAL>

    Depending on access rights "sudo make installlocal"
    might be needed. In case of doubt run first "make doc"
    then "make installlocal". If the latter fails, then
    try "sudo make installlocal".

make uninstallhome
    removes all xint files and repertories from <TEXMFHOME>

make uninstalllocal
    removes all xint files and repertories from <TEXMFLOCAL>
        and then does texhash <TEXMFLOCAL>

    Might need "sudo".

endef

.PHONY: help helpless all doc clean cleanall\
        installhome uninstallhome installlocal uninstalllocal

# for printf with subst and \n, got it from
#     http://stackoverflow.com/a/5887751

# I could do the trick with := here, for \n substitution, but this would add
# tiny overhead to all other operations of make

help:
	@printf '$(subst $(newline),\n,$(helptext))' | more

helpless:
	@printf '$(subst $(newline),\n,$(helptext))' | less

# RM         = rm -f
JF_tmpdir  := $(shell mktemp -d TEMP_XINT_XXX)
TEXMF_local = $(shell kpsewhich -var-value TEXMFLOCAL)
TEXMF_home  = $(shell kpsewhich -var-value TEXMFHOME)
packages = xintkernel.sty xintcore.sty xint.sty xintfrac.sty xintexpr.sty\
             xintgcd.sty xintbinhex.sty xintseries.sty xintcfrac.sty\
             xinttools.sty xinttrig.sty xintlog.sty
# Makefile.mk is not included in $(extracted).  Its extraction rule is in
# master Makefile file.  We can not extract Makefile from xint.dtx via
# docstrip, as .tex is always appended if a filename with no extension is
# specified.  Notice that if "make -f Makefile.mk" is executed, this will
# actually extract again Makefile.mk which will be overwritten but this
# does not seem to be problematic.
extracted  = $(packages) xint.tex xint.ins README.md CHANGES.md doHTMLs.sh
filesfortex    = $(packages)
filesforsource = xint.dtx Makefile
filesfordoc    = xint.pdf sourcexint.pdf README.md CHANGES.html
auxiliaryfiles = xint.dvi xint.aux xint.toc xint.log\
     sourcexint.dvi sourcexint.aux sourcexint.toc sourcexint.log
xint_cmd       = latexmk xint
sourcexint_cmd = latexmk -jobname=sourcexint\
     -latex='latex %O "\chardef\dosourcexint=1 \input{%S}"' xint.tex

all: $(extracted) doc xint.tds.zip
	@echo 'make all done.'

$(extracted): xint.dtx
	etex xint.dtx

xint-dates.txt: CHANGES.md
	grep '^`[1-9].* (.*/.*/.*)`$$' CHANGES.md > xint-dates.txt
	rm -f sourcexint.dvi

doc: xint.pdf sourcexint.pdf CHANGES.html
	@echo 'make doc done.'

xint.pdf: xint.dtx xint.tex
	$(xint_cmd)
	dvipdfmx xint.dvi

sourcexint.pdf: xint.dtx xint.tex xint-dates.txt
	$(sourcexint_cmd)
	dvipdfmx sourcexint.dvi

CHANGES.html: CHANGES.md doHTMLs.sh
	chmod u+x doHTMLs.sh && ./doHTMLs.sh

xint.tds.zip: $(filesfordoc) $(filesforsource) $(filesfortex)
	rm -fr $(JF_tmpdir)
	mkdir -p $(JF_tmpdir)/doc/generic/xint
	mkdir -p $(JF_tmpdir)/source/generic/xint
	mkdir -p $(JF_tmpdir)/tex/generic/xint
	chmod -R ugo+rwx $(JF_tmpdir)
	cp -a $(filesfordoc)    $(JF_tmpdir)/doc/generic/xint
	cp -a $(filesforsource) $(JF_tmpdir)/source/generic/xint
	cp -a $(filesfortex)    $(JF_tmpdir)/tex/generic/xint
	cd $(JF_tmpdir); chmod -R ugo+r doc source tex
	umask 0022 && cd $(JF_tmpdir) &&\
                zip -r xint.tds.zip doc source tex &&\
                mv -f xint.tds.zip ../
	rm -fr $(JF_tmpdir)
	@echo 'make xint.tds.zip done.'

xint.zip: $(filesfordoc) $(filesforsource) $(filesfortex) xint.tds.zip
	mkdir -p              $(JF_tmpdir)/xint
	chmod ugo+rwx         $(JF_tmpdir)/xint
	cp -a $(filesfordoc)    $(JF_tmpdir)/xint
	cp -a $(filesforsource) $(JF_tmpdir)/xint
	chmod -R ugo+r        $(JF_tmpdir)/xint
	mv xint.tds.zip       $(JF_tmpdir)/
	umask 0022 && cd $(JF_tmpdir) && zip -r xint.zip xint.tds.zip xint
	mv     $(JF_tmpdir)/xint.tds.zip ./
	mv -f  $(JF_tmpdir)/xint.zip ./
	rm -fr $(JF_tmpdir)
	@echo 'make xint.zip done.'

installhome: xint.tds.zip
	unzip xint.tds.zip -d $(TEXMF_home)

uninstallhome:
	cd $(TEXMF_home) && rm -fr  doc/generic/xint \
	                            source/generic/xint \
	                            tex/generic/xint

# cf http://stackoverflow.com/a/1909390
# as kpsewhich is very slow (.5s) I want to evaluate once only.
installlocal: xint.tds.zip
	$(eval $@_tmp := $(TEXMF_local))
	unzip xint.tds.zip -d $($@_tmp) && texhash $($@_tmp)

uninstalllocal:
	cd $(TEXMF_local) && rm -fr doc/generic/xint \
	                            source/generic/xint \
	                            tex/generic/xint && texhash .
clean:
	rm -fr auto/ TEMP*/
	rm -f $(auxiliaryfiles)\
	  sourcexint.fls sourcexint.fdb_latexmk\
	  xint.fls xint.fdb_latexmk

cleanall: clean
	rm -f $(extracted) CHANGES.html xint-dates.txt \
          xint.pdf sourcexint.pdf xint.tds.zip xint.zip Makefile.mk
%</makefile>$-----------------------------------------------------
%<*dohtmlsh>------------------------------------------------------
#! /bin/sh
# <s>README.html and</s> CHANGES.html from <s>README.md and </s>CHANGES.md
# tested with pandoc 1.13.1

# pandoc -o README.html -s --toc -V highlighting-css='    body{margin-left : 10%; margin-right : 15%; margin-top: 4ex; font-size: 12pt;}
#     pre   {white-space: pre-wrap; }
#     code  {white-space: pre-wrap; }
#     .mono {font-family: monospace;}' README.md

pandoc -o CHANGES.html -s --toc -V highlighting-css='    body{margin-left : 10%; margin-right : 15%; margin-top: 4ex; font-size: 12pt;}
    pre  {white-space: pre-wrap;}
    code {white-space: pre-wrap;}
    #TOC {float: right; position: relative; top: 100px; margin-bottom: 100px;}' CHANGES.md

%</dohtmlsh>------------------------------------------------------
%<*drv>-----------------------------------------------------------
%% Extracting files (but you have done that already probably):
%% - etex xint.dtx
%%
%% To produce manually xint.pdf from extracted xint.tex (and other files):
%% - latex xint.tex (thrice)
%% - dvipdfmx xint.dvi
%% Or xelatex thrice on xint.tex, or pdflatex thrice (produces bigger sized
%% pdf).
%%
%% To produce manually sourcexint.pdf from extracted xint.tex (and other files):
%% - grep '^`[1-9].* (.*/.*/.*)`$' CHANGES.md > xint-dates.txt
%% - latexmk -jobname=sourcexint\
%%           -latex="latex %O \\\\chardef\\\\dosourcexint=1 \\\\input{%S}"\
%%   xint.tex
%% - dvipdfmx sourcexint.dvi
%%
%% (quoting in the above may need shell-dependant changes)
%%
%% To get xint.pdf to include also the source code:
%% - replace 1 by 0 in \chardef\NoSourceCode line below,
%% - make clean then make xint.pdf
%%
%% This will use latexmk.  Without it execute latex thrice on xint.tex then
%% dvipdfmx.  Don't forget producing xint-dates.txt via the grep command.
\NeedsTeXFormat{LaTeX2e}
\ProvidesFile{xint.tex}%
[\xintbndldate\space v\xintbndlversion\space driver file for xint documentation (JFB)]%
\PassOptionsToClass{a4paper,fontsize=10pt}{scrdoc}
\chardef\NoSourceCode 1 % set it to 0 if source code inclusion desired
\input xint.dtx
%%% Local Variables:
%%% mode: latex
%%% TeX-PDF-from-DVI: "Dvipdfmx"
%%% End:
%</drv>-----------------------------------------------------------
%<*dtx>-----------------------------------------------------------
^^Bfi^^Begroup
\chardef\noetex 0
\ifx\numexpr\undefined\chardef\noetex 1 \fi
% changed 2021/05/30: do not extract files at all if not using etex
\ifnum\noetex=1 \chardef\extractfiles 3 % no extract, no typeset
\else
  \ifx\ProvidesFile\undefined
      \chardef\extractfiles 0 % no LaTeX2e: etex, xetex, ... on xint.dtx
  \else
      \ifx\NoSourceCode\undefined
        % latex/pdflatex/xelatex on xint.dtx, we will extract all files
        \chardef\extractfiles 1 % 1 = extract and typeset, 2 = only typeset
        \chardef\NoSourceCode 0 % 0 = include source code, 1 = do not
        \NeedsTeXFormat{LaTeX2e}%
        \PassOptionsToClass{a4paper,fontsize=10pt}{scrdoc}%
      \else
        % latex/pdflatex/xelatex on xint.tex
        \chardef\extractfiles 2 % no extractions, but typeset
        %  \NoSourceCode is set-up in xint.tex
      \fi
    \ProvidesFile{xint.dtx}[bundle source (\xintbndlversion, \xintbndldate) %
                             and documentation (\xintdocdate)]%
  \fi
\fi
\ifnum\extractfiles<2 % extract files
\def\MessageDeFin{\newlinechar10 \let\Msg\message
\Msg{^^J}%
\Msg{********************************************************************^^J}%
\Msg{*^^J}%
\Msg{* To finish the installation you have to move the following^^J}%
\Msg{* files into a directory searched by TeX:^^J}%
\Msg{*^^J}%
\Msg{* \space\space\space\space xintkernel.sty^^J}%
\Msg{* \space\space\space\space xintcore.sty^^J}%
\Msg{* \space\space\space\space xint.sty^^J}%
\Msg{* \space\space\space\space xintbinhex.sty^^J}%
\Msg{* \space\space\space\space xintgcd.sty^^J}%
\Msg{* \space\space\space\space xintfrac.sty^^J}%
\Msg{* \space\space\space\space xintseries.sty^^J}%
\Msg{* \space\space\space\space xintcfrac.sty^^J}%
\Msg{* \space\space\space\space xintexpr.sty^^J}%
\Msg{* \space\space\space\space xinttools.sty^^J}%
\Msg{* \space\space\space\space xinttrig.sty^^J}%
\Msg{* \space\space\space\space xintlog.sty^^J}%
\Msg{*^^J}%
\Msg{* Rename Makefile.mk to Makefile if the latter is absent^^J}%
\Msg{* (or use --file=Makefile.mk option to "make") then "make help"^^J}%
\Msg{* provides information on targets. In particular:^^J}%
\Msg{* - the "doc" target builds all documentation,^^J}%
\Msg{* - the "xint.tds.zip" target additionally prepares a TDS-compliant^^J}%
\Msg{*\space\space\space archive.^^J}%
\Msg{* This requires Latexmk (for xint.pdf and sourcexint.pdf) and Pandoc^^J}%
\Msg{* (for CHANGES.html).^^J}%
\Msg{*^^J}%
\Msg{* Or check instructions in xint.tex for manual compilation.^^J}%
\Msg{*^^J}%
\Msg{* Happy TeXing!^^J}%
\Msg{*^^J}%
\Msg{********************************************************************^^J}%
}%
\ifx\XeTeXinterchartoks\undefined\else
  \expandafter\def\expandafter\MessageDeFin\expandafter{\MessageDeFin
\Msg{* ATTENTION! extraction with xetex must be made with -8bit option.^^J}%
\Msg{* Extracted file Makefile.mk is invalid if this was not the case.^^J}%
\Msg{********************************************************************^^J}%
}%
\fi
\begingroup
    \input docstrip.tex
    \askforoverwritefalse
    \catcode9 11 % do not kill TAB in producing Makefile.mk
    \generate{\nopreamble\nopostamble
    \file{README.md}{\from{xint.dtx}{readme}}
    \file{CHANGES.md}{\from{xint.dtx}{changes}}
    % pure tex will use ^^I notation for TAB character, don't want that.
    % anyway, since 2021/05/30, tex xint.dtx has been made a no-op.
    % there is a problem with xelatex, it generates ^^I if not with -8bit
    \file{Makefile.mk}{\from{xint.dtx}{makefile}}
    \file{doHTMLs.sh}{\from{xint.dtx}{dohtmlsh}}
    \usepreamble\defaultpreamble
    \usepostamble\defaultpostamble
    \file{xint.tex}{\from{xint.dtx}{drv}}
    \file{xintkernel.sty}{\from{xint.dtx}{xintkernel}}
    \file{xinttools.sty}{\from{xint.dtx}{xinttools}}
    \file{xintcore.sty}{\from{xint.dtx}{xintcore}}
    \file{xint.sty}{\from{xint.dtx}{xint}}
    \file{xintbinhex.sty}{\from{xint.dtx}{xintbinhex}}
    \file{xintgcd.sty}{\from{xint.dtx}{xintgcd}}
    \file{xintfrac.sty}{\from{xint.dtx}{xintfrac}}
    \file{xintseries.sty}{\from{xint.dtx}{xintseries}}
    \file{xintcfrac.sty}{\from{xint.dtx}{xintcfrac}}
    \file{xintexpr.sty}{\from{xint.dtx}{xintexpr}}
    \file{xinttrig.sty}{\from{xint.dtx}{xinttrig}}
    \file{xintlog.sty}{\from{xint.dtx}{xintlog}}}
\endgroup
\fi % end of file extraction (from etex/latex/pdflatex/... run on xint.dtx)
\ifnum\extractfiles=0 % no LaTeX, files now extracted. Stop.
  \MessageDeFin\expandafter\end
\fi
\ifnum\extractfiles=3 %
  \newlinechar10
\errhelp{Extraction must be done via an engine with e-TeX extensions.^^J%
         Please try again, using etex rather.^^J%
         (if with xetex, use -8bit option).}%
\errmessage{Extraction aborted (must be done with etex)}%
\expandafter\end
\fi
% From this point on, run is necessarily with e-TeX.
% Check if \MessageDeFin got defined, if yes put it at end of run.
\ifdefined\MessageDeFin\AtEndDocument{\MessageDeFin}\fi
%-----------------------------------------------------------------
% -*- coding: utf-8; mode: latex, fill-column: 78; -*-
%
\ifdefined\dosourcexint % this toggle is set from make sourcexint.pdf rule
    \chardef\NoSourceCode 0
\else
    \chardef\dosourcexint 0
\fi

% default is to assume latex + dvipdfmx
\chardef\Withdvipdfmx 1

\RequirePackage{ifpdf}
\RequirePackage{ifxetex}

\ifpdf  \chardef\Withdvipdfmx 0 \fi
\ifxetex\chardef\Withdvipdfmx 0
% they are used at some locations in this doc
  \let\pdfsetrandomseed\setrandomseed
  \let\pdfuniformdeviate\uniformdeviate
  \let\pdfresettimer\resettimer
  \let\pdfelapsedtime\elapsedtime
\fi

% Get rid of HARASSMENT by KOMA-Script
\makeatletter
\def\class@shoulddisablepackagewarning@list{tocloft.}
\makeatother
\ifnum\Withdvipdfmx=1
\def\pgfsysdriver{pgfsys-dvipdfm.def}
\documentclass [dvipdfm, dvipdfmx, dvipdfmx-outline-open]{scrdoc}
\else
\documentclass {scrdoc}
\fi

% Remove from sectioning commands insertion of marks, because we
% will do it ourself.
\usepackage{etoolbox}
\makeatletter
\patchcmd{\@sect}%
    {\expandafter\csname#1mark\expandafter\endcsname\expandafter{\@currentheadentry}}%
    {}{}{}
\patchcmd{\@sect}%
    {\expandafter\csname#1mark\expandafter\endcsname\expandafter{\@currentheadentry}}%
    {}{}{}
% This one now needed too, Jeudi 30 janvier 2020 
% \expandafter \ifx \csname #1mark\endcsname \@gobble \@mkboth {}{}\else
% \csname #1mark\expandafter \endcsname \expandafter {\@currentheadentry }\fi
% \@gobble est long donc simplement
\def\partmark #1{}%
\makeatother


\makeatletter
  
\PassOptionsToPackage{bookmarks=true}{hyperref}

\ifnum\NoSourceCode=1
  \OnlyDescription
\fi


% counts used in particular in the samples from the documentation of the
% xintseries.sty package
\newcount\cnta
\newcount\cntb
\newcount\cntc

\pagestyle{headings}

\ifxetex
\else
  \usepackage[T1]{fontenc}
  \usepackage[utf8]{inputenc}
  \DeclareUnicodeCharacter{03B4}{\ensuremath{\delta}}%δ
  \DeclareUnicodeCharacter{03BE}{\ensuremath{\xi}}%ξ
  \DeclareUnicodeCharacter{03C0}{\ensuremath{\pi}}%π
  \DeclareUnicodeCharacter{2260}{\ensuremath{\neq}}%≠
\fi

\usepackage{multicol}
\usepackage{geometry}
\AtBeginDocument {\ttzfamily % package newtxtt loaded in preamble
                  \newgeometry{textwidth=\dimexpr92\fontcharwd\font`X\relax,
                               vscale=0.75}}

\unless\ifnum\dosourcexint=1
\usepackage{xintexpr}
\usepackage{xintbinhex}
\usepackage{xintgcd}
\usepackage{xintseries}
\usepackage{xintcfrac}
\usepackage{amsmath}% for \cfrac usage
\DeclareMathOperator{\sinc}{sinc}
\fi
\usepackage{pifont}% for \ding{73} (hollow star)

\usepackage{xinttools}

\usepackage{enumitem}
\usepackage{varioref}
\usepackage{xspace}
\usepackage[para]{footmisc}
\usepackage{picture}
\usepackage{graphicx}

\usepackage[english]{babel}
\usepackage[autolanguage,np]{numprint}
\AtBeginDocument{\npthousandsep{,\hskip .5pt plus .1pt minus .1pt}}

\usepackage[dvipsnames]{xcolor}
\definecolor{joli}{RGB}{225,95,0}
\definecolor{JOLI}{RGB}{225,95,0}
\definecolor{BLUE}{RGB}{0,0,255}
\definecolor{niceone}{RGB}{38,128,192}
\definecolor{sedate}{RGB}{193,132,1}
\definecolor{saddlebrown}{rgb}{.545,.27,.075}
\definecolor{jfbrown}{RGB}{165,100,10}%




\usepackage{eso-pic}% après xcolor sinon Option clash for package xcolor.

\ifnum\dosourcexint=1
\else
% Dependency graph done using TikZ (manually)
  \usepackage{tikz}
  \usetikzlibrary{shapes,arrows.meta}
\fi

\usepackage{framed}
% SNUGFRAMED
% ==========

\makeatletter
\newenvironment{snugframed}{%
  \fboxsep \dimexpr2\fontcharwd\font`X\relax
  \advance\linewidth-2\fboxsep
  \advance\csname @totalleftmargin\endcsname \fboxsep
  \def\FrameCommand##1{\hskip\@totalleftmargin
                       \hskip-\fboxsep
                       \fbox{##1}\hskip-\fboxsep
      % There is no \@totalrightmargin, so:
      \hskip-\linewidth \hskip-\@totalleftmargin \hskip\columnwidth}%
    \MakeFramed {\advance\hsize-\width \@totalleftmargin\z@ \linewidth\hsize
    \@setminipage}%
 }{\par\unskip\@minipagefalse\endMakeFramed}
\makeatother

% HYPERREF
% ========

\usepackage[pdfencoding=unicode]{hyperref}

\hypersetup{%
linktoc=all,%
breaklinks=true,%
colorlinks=true,%
urlcolor=niceone,%
linkcolor=blue,%
pdfauthor={Jean-Fran\c cois Burnol},%
pdftitle={The xintexpr and allied packages},%
pdfsubject={Arithmetic with TeX},%
pdfkeywords={Expansion, arithmetic, TeX},%
pdfstartview=FitH,%
pdfpagemode=UseNone,%
}

\usepackage{hypcap}
\ifnum\dosourcexint=1
\hypersetup{pdftitle={The xint packages source code}}
\fi
\usepackage{bookmark}

% FONTS
% =====

\usepackage[zerostyle=a,straightquotes,scaled=0.95]{newtxtt}
\usepackage{newtxmath}

\makeatletter



\DeclareFontFamily{T1}{newtxttb}{\hyphenchar\font\m@ne}

\DeclareFontShape{T1}{newtxttb}{m}{n}{
      <-> s*[\newtxtt@scale]newtxttbq
}{}
\DeclareFontShape{T1}{newtxttb}{b}{n}{
      <-> s*[\newtxtt@scale]newtxbttbq
}{}
\DeclareFontShape{T1}{newtxttb}{bx}{n}{
      <-> ssub * newtxttb/b/n
}{}
\DeclareFontShape{T1}{newtxttb}{m}{sl}{
     <-> s*[\newtxtt@scale]newtxttslbq
}{}
\DeclareFontShape{T1}{newtxttb}{m}{it}{
     <-> ssub * newtxttb/m/sl
}{}

% Ajouté le 9 mars 2016

\DeclareFontShape{T1}{newtxttb}{m}{sc}{%cap & small cap
     <-> s*[\newtxtt@scale]newtxttscbq
}{}
\DeclareFontShape{T1}{newtxttb}{b}{sc}{%bold cap & small cap
     <-> s*[\newtxtt@scale]newtxbttscbq
}{}
\DeclareFontShape{T1}{newtxttb}{b}{sl}{%bold slanted
     <-> s*[\newtxtt@scale]newtxbttslbq
}{}
\DeclareFontShape{T1}{newtxttb}{b}{it}{%bold italic
     <-> ssub * newtxttb/b/sl%
}{}
\DeclareFontShape{T1}{newtxttb}{bx}{sc}{%bold extended cap & small cap
     <-> ssub * newtxttb/b/sc%
}{}
\DeclareFontShape{T1}{newtxttb}{bx}{sl}{%bold extended slanted
     <-> ssub * newtxttb/b/sl%
}{}
\DeclareFontShape{T1}{newtxttb}{bx}{it}{%bold extended italic
     <-> ssub * newtxttb/b/sl%
}{}

% Ajouté le 9 mars 2016
\DeclareEncodingSubset{TS1}{newtxttb}{0}
\DeclareFontFamily{TS1}{newtxttb}{\hyphenchar\font\m@ne}

\DeclareFontShape{TS1}{newtxttb}{m}{n}{%medium
     <-> s*[\newtxtt@scale]tcxtt%
}{}
\DeclareFontShape{TS1}{newtxttb}{m}{sc}{%cap & small cap
     <->ssub * newtxttb/m/n%
}{}
\DeclareFontShape{TS1}{newtxttb}{m}{sl}{%slanted
     <-> s*[\newtxtt@scale]tcxttsl%
}{}
\DeclareFontShape{TS1}{newtxttb}{m}{it}{%italic
     <->ssub * newtxttb/m/sl%
}{}
\DeclareFontShape{TS1}{newtxttb}{b}{n}{%bold
     <-> s*[\newtxtt@scale]tcxbtt%
}{}
\DeclareFontShape{TS1}{newtxttb}{b}{sc}{%bold cap & small cap
     <->ssub * newtxttb/b/n%
}{}
\DeclareFontShape{TS1}{newtxttb}{b}{sl}{%bold slanted
     <-> s*[\newtxtt@scale]tcxbttsl%
}{}
\DeclareFontShape{TS1}{newtxttb}{b}{it}{%bold italic
     <->ssub * newtxttb/b/sl%
}{}
\DeclareFontShape{TS1}{newtxttb}{bx}{n}{%bold extended
     <->ssub * newtxttb/b/n%
}{}
\DeclareFontShape{TS1}{newtxttb}{bx}{sc}{ %bold extended cap & small cap
     <->ssub * newtxttb/b/sc%
}{}
\DeclareFontShape{TS1}{newtxttb}{bx}{sl}{%bold extended slanted
     <->ssub * newtxttb/b/sl%
}{}
\DeclareFontShape{TS1}{newtxttb}{bx}{it}{%bold extended italic
     <->ssub * newtxttb/b/it%
}{}


\makeatother

% This is with a slashed 0 like the original txtt.
\newcommand\ttbfamily {\fontfamily{newtxttb}\selectfont }

\ifnum\dosourcexint=1
\else
\renewcommand\familydefault\ttdefault
\usepackage[noendash]{mathastext}% pas de endash dans newtxtt
\fi
\frenchspacing
% sans-serif in footnotes, TOC, titles, etc...
\renewcommand\familydefault\sfdefault

% TABLES OF CONTENTS
% ==================

\usepackage{tocloft}
\usepackage{etoc}

\def\gobbletodot #1.{}

\newif\ifinmanualmaintoc
\ifnum\dosourcexint=0
    \inmanualmaintoctrue
\fi
\def\sectioncouleur{{cyan}}

\def\MARGEPAGENO {1.5em}% changera pour la partie implémentation


\def\SKIPSECTIONINTERSPACE{\vskip\bigskipamount}
\etocsetstyle{section}{}
     {\normalfont}
     {\etociffirst{}{\SKIPSECTIONINTERSPACE}%
         \rightskip    \MARGEPAGENO\relax
         \parfillskip  -\MARGEPAGENO\relax
      \bfseries
         \leftskip \leftmarginii
      \noindent\llap            %  \llap
      {\makebox[\leftmarginii][l]%  et \leftmargini le 12/10/2014
              {\expandafter\textcolor\sectioncouleur {\etocnumber}}}%
      \strut\etocname
      \mdseries\nobreak\leaders\etoctoclineleaders\hfill\nobreak\strut
                             \makebox[\MARGEPAGENO][r]{\etocpage}\par
      \let\ETOCsectionnumber\etocthenumber
      }%
     {}%

\newdimen\margegauchetoc
\AtBeginDocument{\margegauchetoc \dimexpr 5\fontcharwd\font`X\relax}
\makeatletter
\etocsetstyle{subsection}
    {\begingroup\normalfont
     \setlength{\premulticols}{0pt}%
     \setlength{\multicolsep}{0pt}%
     \setlength{\columnsep}{\leftmarginii}%
     \setlength{\columnseprule}{.4pt}% n'influence pas séparation colonnes
     \parskip\z@skip
     \raggedcolumns
     \addvspace{\smallskipamount}%
     \begin{multicols}{2}
     \leftskip  \margegauchetoc % 12 octobre 2014
     \ifinmanualmaintoc
        \rightskip \MARGEPAGENO
     \else
        \rightskip \MARGEPAGENO plus 2em minus 1em
     \fi
     \parfillskip -\MARGEPAGENO\relax
    }
    {}
    {\noindent
     \etocifnumbered{\llap{\makebox[\margegauchetoc][l]{\ttzfamily\bfseries\etoclink
             {\ifinmanualmaintoc\expandafter\textcolor\sectioncouleur
               {\normalfont\bfseries\ETOCsectionnumber}\fi
              .\expandafter\gobbletodot\etocthenumber}}}}{\kern-\margegauchetoc}%
     \strut\etocname\nobreak
     \unless\ifinmanualmaintoc\leaders\etoctoclineleaders\fi
     \hfill\nobreak
     \strut\makebox[\MARGEPAGENO][r]{\small\etocpage}\endgraf }
    {\end{multicols}\endgroup
     %\addvspace{\smallskipamount}
    }%

\etocsetstyle{subsubsection}
    {\begingroup\normalfont\small
     \leftskip  \dimexpr\leftmargini+1em\relax }
    {}
    {\noindent
     \llap{\makebox[\dimexpr\leftmargini+1em\relax][l]%
           {\ttzfamily\bfseries\etoclink
                    {\HOOKLOCALTOC.\expandafter\gobbletodot\etocthenumber}}}%
     \strut\etocname\nobreak
     \leaders\etoctoclineleaders
     \hfill\nobreak
     \strut\makebox[\MARGEPAGENO][r]{\small\etocpage}\endgraf }
    {\endgroup }%

\let\HOOKLOCALTOC\empty% quick hack to get style I want in User defined functions
\etocsetlevel{table}{6}

\makeatother

\addtocontents{toc}{\protect\hypersetup{hidelinks}}

% =====================
% MISCELLANEOUS MARK-UP
% =====================


\def\digitstt #1{\begingroup\color[named]{OrangeRed}#1\endgroup}
\let\dtt\digitstt

% \ctexttt is a remnant of 1.09n manual, don't have time to get rid of it now.
\newcommand\ctexttt [1]{\begingroup\color[named]{DarkOrchid}%\bfseries
                        #1\endgroup}

% \fexpan 22 octobre 2013
\newcommand\fexpan {\hyperref[ssec:expansions]{\textit{f}-expan}}
% Septembre 2015
% Address updated to github repo's one, May 2018
\def\liiibigint
   {\href{https://github.com/latex3/latex3/tree/master/l3trial/l3bigint}{l3bigint}}

% \fixmeaning
\makeatletter
% \def\fixmeaning {\expandafter\fix@meaning\meaning}
% \expandafter\edef\expandafter\fix@meaning
%             \expandafter #\expandafter1\string\romannumeral#2#3%
%             {#1\string\romannumeral`\string^\string^@}
% Pour 1.4
\let\fixmeaning\meaning
\makeatother

% Margin Notes
% ============

% Nothing here can be used in vertical mode directly.

\makeatletter
\def\MyMarginNote {\@ifnextchar[\@MyMarginNote{\@MyMarginNote[]}}%
\let\inmarg\MyMarginNote
% \smash needs \hbox here since LaTeX 2018/12/01
%  https://github.com/latex3/latex2e/issues/108
\def\@MyMarginNote [#1]#2{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt
                       {\color[named]{PineGreen}\normalfont\small
                        \hsize 1.6cm\rightskip.5cm minus.5cm
                        \hss\vtop{#2}\ $\to$#1\ }}}%
             \vskip\dp\strutbox
            }\strut\@esphack}
\def\MyMarginNoteWithBrace #1#2{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt
                       {\color[named]{PineGreen}%\normalfont\small
                        \hss #1\ $\bigg\{$#2}}}%
             \vskip\dp\strutbox
            }\strut\@esphack}
\def\IMPORTANT {\MyMarginNoteWithBrace
   {\raisebox{-.5\height}{\resizebox{2\width}{!}{\ding{43}}}}{\ }}
\def\IMPORTANTf {\MyMarginNoteWithBrace
   {\raisebox{-.5\height}{\resizebox{2\width}{!}{\ding{43}}}}%
   {\kern\dimexpr\FrameSep+\FrameRule\relax\ }}
\def\etype #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{PineGreen}%
                    \itshape \xintListWithSep{\,}{#1}\ $\star$\quad }}}%
             \vskip\dp\strutbox
            }\strut\@esphack}
\def\xtype #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{PineGreen}%
                    \itshape \xintListWithSep{\,}{#1}\ \ding{73}\quad }}}%
             \vskip\dp\strutbox }\strut\@esphack}
\def\ntype #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{PineGreen}%
                    \itshape \xintListWithSep{\,}{#1}\quad }}}%
             \vskip\dp\strutbox }\strut\@esphack}
%
\def\Numf {{\vbox{\halign{\hfil##\hfil\cr \footnotesize
    \upshape Num\cr
    \noalign{\hrule height 0pt \vskip1pt\relax}
    \itshape f\cr}}}}
\def\Ff {{\vbox{\halign{\hfil##\hfil\cr \footnotesize
    \upshape Frac\cr
    \noalign{\hrule height 0pt \vskip1pt\relax}
    \itshape f\cr}}}}
\def\numx {{\vbox{\halign{\hfil##\hfil\cr \footnotesize
    \upshape num\cr
    \noalign{\hrule height 0pt \vskip1pt\relax}
    \itshape x\cr}}}}
%
\def\NewWith #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{PineGreen}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent New with #1}\ }}}%
             \vskip\dp\strutbox }\strut\@esphack}
%
\def\CHANGED #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{Red}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent Changed at #1!}\ }}}%
             \vskip\dp\strutbox }\strut\@esphack}

\def\DNU#1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{Red}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent Do not use! #1}\ }}}%
             \vskip\dp\strutbox }\strut\@esphack}

\def\UNSTABLE#1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{Red}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent Unstable! #1}\ }}}%
             \vskip\dp\strutbox }\strut\@esphack}

\def\unstable#1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{Red}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent unstable? #1}\ }}}%
             \vskip\dp\strutbox }\strut\@esphack}

\def\DEPRECATED #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{PineGreen}%
                        \normalfont\small\bfseries
                        \hsize 2cm\rightskip.5cm minus.5cm
                        \vtop{\noindent Deprecated! (#1)}\ }}}%
             \vskip\dp\strutbox }\strut\@esphack}
%
\def\CHANGEDf #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{Red}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent Changed at #1!}\ 
              \kern\dimexpr\FrameSep+\FrameRule\relax}}}%
             \vskip\dp\strutbox }\strut\@esphack}
%
\def\NewWithf #1{\@bsphack
    \vadjust{\vskip-\dp\strutbox
             \hbox{\smash{\hbox to 0pt {\hss\color[named]{PineGreen}%
                        \normalfont\small\bfseries
                        \hsize 1.5cm\rightskip.5cm minus.5cm
                        \vtop{\noindent New with #1}\ 
              \kern\dimexpr\FrameSep+\FrameRule\relax}}}%
             \vskip\dp\strutbox }\strut\@esphack}

\makeatother

% \centeredline: OUR OWN LITTLE MACRO FOR CENTERING LINES
% =======================================================

% 7 mars 2013
% Note (2020): now a package
\usepackage{centeredline}

% \leftedline
% ===========

% 12 octobre 2014

% Note (2020): somewhat obsoleted for many years by my usage of
% everbatim and everbatim* environments.

\makeatletter
\newif\ifinlefted

\newcommand*\leftedline {%
      \ifhmode \\\relax
        \def\leftedline@{\hss\egroup\hskip\z@skip\ignorespaces }%
      \else
        \def\leftedline@{\hss\egroup }%
      \fi
      \afterassignment\@leftedline
      \let\next=}
\def\@leftedline
    {\hbox to \linewidth \bgroup \inleftedtrue
                         \everbatimeverypar
                         \bgroup
                         \aftergroup\leftedline@ }

\makeatother

% verbatim macros and environments
% ================================
%
% June 2013, then October 2014.
% -----------------------------
%
\makeatletter
\catcode`_ 11

% some of my verbatim environments do not make the space active (\lverb e.g.). Then
% \do@noligs must be modified, \char`#1 must be followed by a space token, else,
% the `#1 expansion will swallow one space.
\def\do@noligs #1{%
  \catcode`#1\active
  \begingroup
     \lccode`~`#1\relax
     \lowercase{%
  \endgroup\def~{\leavevmode\kern\z@\char`#1 }}%
}

% \lowast
\def\lowast{\raisebox{-.25\height}{*}}
\catcode`* 13
\def\makestarlowast {\let*\lowast\catcode`\*\active}%
\catcode`* 12



% \MacroFont and \MicroFont
% =========================

\def\restoreMicroFont {\def\MicroFont {\ttbfamily\makestarlowast
%    \ifinlefted\else\ifineverb\else\color[named]{verbatim}\fi\fi
%    \ifinlefted\else\color[named]{saddlebrown}\fi
%    \ifinlefted\else\color[named]{sedate}\fi
    \ifinlefted\else\color[named]{jfbrown}\fi
}}
\restoreMicroFont

% Notice that \macrocode uses \macro@font which stores the \MacroFont meaning
% in force at \begin{document}. But doc.sty's verbatim uses current \MacroFont
% not the meaning at \begin{document}. Comprenne qui pourra...

\def\restoreMacroFont {\def\MacroFont {\ttbfamily
%    \ifinlefted\else\ifineverb\else\color[named]{Blue}\fi\fi
%    \ifinlefted\else\color[named]{Blue}\fi
%    \ifinlefted\else\color[named]{sedate}\fi
    \ifinlefted\else\color[named]{Brown}\fi
}}
\restoreMacroFont

% \verb
% =====

% Initially, June 2013, then Sep 9, 2014, and Oct 9-12 2014
%
% Initial motivation was simply that doc.sty and related classes \verb
% macro is with a hard-coded \ttfamily. There were further issues.
%
% 1. With |stuff with space|, paragraph reformatting in the Emacs/AUCTeX
% buffer caused havoc. Thus I wanted the input to accept linebreaks in
% its contents.
%
% 2. Hence I did not want to have obeyed spaces obeyed, (Emacs reflowing
% of paragraph in certain contexts often adds spaces at beginning of a line)
%
% 3. Also I wanted to allow hyphenated output, at least at some
% locations. I did a first version which treated spaces, \, {, and }
% specially.
%
% 4. At some point I wanted to add some colored background (I have
% dropped that since due to pdf file size increase).
%
% 5. And also I got fed up from the non-compatibility with footnotes due
% to catcode freeze.
%
% Because of 5. I opted for a \scantokens approach, hence for a macro
% with delimited argument. Here is what I do now, this is compatible
% with short verbs.

\def\verb
{%
  \relax \ifmmode\else\leavevmode\null\fi
  \bgroup
  \let\do\@makeother \dospecials
  \@ifstar{\@sverb}% \verb* is used in the index (obsolete: no indices at 1.3e),
                   % leave it using ambient font
          {\MicroFont % used to change font (ttbfamily=slashed 0), color,
                      % will make * active via \makestarlowast
           \catcode 32 10 \endlinechar 32 % allows to fetch across line breaks
           \frenchspacing % done globally in document
           \@@jfverb}%
}%
% Note (Oct 12, 2014): in the improbable situation a newlinechar is
% found in the ##1, \scantokens will convert this to an end of line in
% its "write" phase, which will be then ignored in its "read" phase due
% to \endlinechar-1. This also avoids possible creation of \par which
% would defeat \@@jfverb@@. Thus it is good.
\def\@@jfverb #1{%
   \ifcat\noexpand#1\noexpand~\catcode`#1\active\fi
% No problem with the EOL for the line where the short verb delimiter stands.
   \def\next ##1#1{%
            \@vobeyspaces\everyeof{\relax}\endlinechar\m@ne
            \expandafter\@@jfverb_a\scantokens\expandafter{##1}}%
% hack with \@empty to prevent brace stripping if catcodes have been
% frozen earlier, like in footnotes.
   \next \@empty
}

% We don't want a \discretionary at the very start.
% But then an empty argument is forbidden!
\def\@@jfverb_a #1{#1\@@jfverb_b }

\def\@@jfverb_b #1{\ifx\relax #1%
        \egroup
      \else
% \penalty\z@, or rather (Oct 11, 2014) but I then adjust the textwidth
% precisely:
      \discretionary{\copy\SoftWrapIcon}{}{}%
        #1\expandafter\@@jfverb_b\fi
}

% \SoftWrapIcon box for line-breaking using discretionaries
% =========================================================

\DeclareFontFamily{U}{MdSymbolC}{}
\DeclareFontShape {U}{MdSymbolC}{m}{n}{<-> MdSymbolC-Regular}{}

\newbox\SoftWrapIcon
\colorlet {softwrapicon}{blue}

% Emacs/AUCTeX uses very strange comment-like highlighting for \usefont{U}...
\def\SetSoftWrapIcon{%
    \setbox\SoftWrapIcon\hb@xt@\z@
    {\hb@xt@\fontdimen2\font
        {\hss{\color{softwrapicon}\usefont{U}{MdSymbolC}{m}{n}\char"97}\hss}%
     \hss}%
   }

\AtBeginDocument {\SetSoftWrapIcon }% ttzfamily déjà fait

\catcode`_ 8
\makeatother

% everbatim environment
% =====================

% October 13-14, 2014
% Verbatim with an \everypar hook, mainly to have background color, followed by
% execution of the contents (not limited by a group-scope)

\makeatletter
\catcode`_ 11

% Je modifie Mardi 18 février 2020 \MacroFont
% pour essayer couleur de foreground,
% je supprime donc le \ifineverb de \MacroFont.
% \newif\ifineverb
\def\everbatimtop {\MacroFont \small}
\let\everbatimbottom\relax
\let\everbatimhook\relax

\def\everbatim {\s@everbatim\@everbatim }
\@namedef{everbatim*}{\s@everbatim\expandafter\@everbatimx\expandafter
                      {\the\newlinechar}}

\def\everbatimeverypar{\strut
                   {\color{yellow!5}\vrule\@width\linewidth }%
                   \kern-\linewidth
                   \kern\everbatimindent }
\def\everbatimindent {\z@}
% voir plus loin atbegindocument

\def\endeverbatim  {\if@newlist \leavevmode\fi\endtrivlist }

\@namedef{endeverbatim*}{\endeverbatim\aftergroup\everbatimundoparskip}
\def\everbatimundoparskip{\vbox{}\kern-\baselineskip\kern-\parskip}
% Note 24 juin 2021
%
% Si everbatim[*] suit immédiatement un premier everbatim* dont l'exécution
% n'a fait que des définitions de macros sans créer de paragraphe (ou un
% environnment de liste comme center) le \kern-\parskip émis après le rendu
% verbatim (et qui vient de \everbatimundoparskip) ne sera pas compensé.
% Plus précisément le everbatim* (ou everbatim) qui vient en
% deuxième utilise \trivlist mais en ayant mis les paramètres d'espacements
% verticaux à zéro.  Donc il y aura chevauchement.
%
% Pour régler ce problème, on peut :
% - soit fusionner les deux everbatim*,
% - soit si cela est embêtant, par exemple si le second environnment est
%   un everbatim, pas un everbatim*, ajouter \kern-\parskip entre les deux.

\def\s@everbatim {%
%     \ineverbtrue
     \everbatimtop % put there size changes
       \topsep    \z@skip
       \partopsep \z@skip
       \itemsep   \z@skip
       \parsep    \z@skip
       \parskip   \z@skip
       \lineskip  \z@skip
     \let\do\@makeother \dospecials
     \let\do\do@noligs  \verbatim@nolig@list
     \makestarlowast
     \everbatimhook
     \trivlist
       \@topsepadd \z@skip
       \item\relax
       \leftskip   \@totalleftmargin
       \rightskip  \z@skip
       \parindent  \z@
       \parfillskip\@flushglue
       \parskip    \z@skip
       \@@par
       \def\par{\leavevmode\null\@@par\pagebreak[1]}%
       \everypar\expandafter{\the\everypar \unpenalty
                \everbatimeverypar
                \everypar \expandafter{\the\everypar\everbatimeverypar}%
       }%
       \obeylines \@vobeyspaces
}

\begingroup
\lccode`X 13
\catcode`X \active
\lccode`Y `* % this is because of \makestarlowast.
% I have to think whether this is useful: obviously if I were to provide
% everbatim and everbatim*  in a package I wouldn't do that.
\catcode`Y  \active
\catcode`| 0 \catcode`[ 1 \catcode`] 2 \catcode`* 12
\catcode`{ 12 \catcode`} 12 |catcode`\\ 12
|lowercase[|endgroup% both freezes catcodes and converts X to active ^^M
|def|@everbatim #1X#2\end{everbatim}%
  [#2|end[everbatim]|everbatimbottom ]
|def|@everbatimx #1#2X#3\end{everbatimY}]%
  {#3\end{everbatim*}%
     \everbatimbottom
     \newlinechar 13
     \everbatimxprehook
     \scantokens {#3}%
     \newlinechar #1\relax
     \everbatimxposthook
}%

% L'espace venant du endofline final mis par \scantokens sera inhibé si #3 se
% termine par un % ou un \x, etc...

\def\everbatimxprehook {\colorlet{everbsavedcolor}{.}\color[named]{OrangeRed}}
\def\everbatimxposthook {\color{everbsavedcolor}}
\ifpdf
   \def\everbatimxprehook
      {\pdfcolorstack\@pdfcolorstack push{0 1 0.5 0 k 0 1 0.5 0 K}\relax}
   \def\everbatimxposthook
      {\pdfcolorstack\@pdfcolorstack pop\relax}
\else
% Le 24 juin 2021 je vérifie que ceci est encore nécessaire avex xelatex
% (sinon color leak de OrangeRed à partir de la page 97 dans la doc)
% et aussi pour dvipdfmx (sinon color stack overflow au moment de la page 98
% lors du passage par dvipdfmx)
\ifxetex
   \def\everbatimxprehook  {\special{color push cmyk 0 1 0.5 0}}
   \def\everbatimxposthook {\special{color pop}}
\else
\ifnum\Withdvipdfmx=1
     \def\everbatimxprehook  {\special{pdf:bcolor  OrangeRed}}
     \def\everbatimxposthook {\special{pdf:ecolor}}
\fi\fi\fi



% \everb
% ======
%
% Original was called \dverb and I did it in June 2013.
% Then after doing everbatim, I transformed \dverb, now called \everb
% for itself being as compatible as standard verbatim with list making
% surrounding environments.
% Supposed to be used as
% \everb|@ this will be ignored
% stuff
% escape character: "
% | not necessarily starting a line.
% I chose @ as comment character, mainly for pretty-formatting of the
% source, this can be changed by \everbhook.

% " comme caractère d'échappement. Par exemple pour colorier des parties.
\def\restoreeverbhook{\def\everbhook{%
     \def\"{\begingroup\catcode123 1 \catcode 125 2 \everbescape }%
     \catcode`\" 0 \catcode`\@ 14
}}\restoreeverbhook

\def\everbescape #1;!{#1\endgroup }

\def\everb {%
    \bgroup
    \let\everbatimhook\everbhook
    \s@everbatim
    \@everb
}

\def\@everb #1{\catcode`#1\active
               \lccode`\~`#1%
               \lowercase{\def~{\if@newlist \leavevmode\fi
                                \endtrivlist
                                \egroup
                                \@doendpe
                                \everbatimbottom }}%
              }%

\catcode`_8
\makeatother


% \printnumber
% ============

\catcode`_ 11
\makeatletter
\catcode`& 3
\def\allowsplits_a {\futurelet\printnumber_token\allowsplits_b }%
\def\allowsplits_b{\ifx\printnumber_token\@sptoken\space\fi\allowsplits_c }
\def\allowsplits_c #1{\ifx &#1\xint_dothis\xint_gobble_i\fi
                    \if ,#1\xint_dothis {\discretionary{\rlap,}{}{,}}\fi
                    \xint_orthat{\discretionary
                           {\copy\SoftWrapIcon}%
                           {}%
                           {}#1}\allowsplits_a }%

\def\printnumber #1{\expandafter\allowsplits_a \romannumeral-`0#1&}%
\hyphenpenalty \z@

\catcode`& 4
\makeatother
\catcode`_ 8

% Parameters for lists
% ====================
\AtBeginDocument{%
    \leftmargini \dimexpr4\fontcharwd\font`X\relax
    \leftmarginii\dimexpr3\fontcharwd\font`X\relax
    \leftmarginiii \leftmarginii
    \leftmarginiv  \leftmarginii
    \parindent\dimexpr2\fontcharwd\font`X\relax
    \leftmargin\leftmargini % pourquoi pas 0?
% formerly everbatim indent was set to leftmargingi, reduce it (2017/08/26)
%    \edef\everbatimindent{\the\dimexpr\leftmargini\relax\space }%
% setting it to \parindent does not work with \everb construct
%    \def\everbatimindent{\parindent}%
    \edef\everbatimindent{\the\dimexpr2\fontcharwd\font`X\relax\space}%
    \cftsubsecnumwidth    2\leftmarginii
    \cftsubsubsecnumwidth 2\leftmargini
    \cftsubsecindent 0pt
    \cftsubsubsecindent \cftsubsecnumwidth
}%

% ==========
% Hyperlinks
% ==========

% \csa, \csbxint, \csh etc...
% ===========================

% These definitions in force both in manual and implementation part
\DeclareRobustCommand\csa[1]
    {{\ttzfamily\char92\endlinechar-1 
      \makestarlowast \catcode`_ 12 \catcode`^ 12 
      \scantokens\expandafter{\detokenize{#1}}}}

% csan: n means no backslash
\DeclareRobustCommand\csan[1]
    {{\ttzfamily\endlinechar-1 
      \makestarlowast \catcode`_ 12 \catcode`^ 12 
      \scantokens\expandafter{\detokenize{#1}}}}

\newcommand\csh[1]
    {\texorpdfstring{\csa{#1}}{\textbackslash\detokenize{#1}}}
\newcommand\cshn[1]
    {\texorpdfstring{\csan{#1}}{\detokenize{#1}}}
% \csh and \cshn will be redefined in implementation section
\let\cshnolabel\csh
\let\cshnnolabel\cshn

% These definitions will be re-done for implementation part
% Don't bother about underscore and caret for time being.
\DeclareRobustCommand\csb [1]
        {\hyperref[\detokenize{#1}]%
          {{\char92 \endlinechar-1 \makestarlowast
                 \scantokens\expandafter{\detokenize{#1}}}}}
\DeclareRobustCommand\csbxint [1]
        {\hyperref[\detokenize{xint#1}]%
          {{\char92\mbox{xint}\-\endlinechar-1 \makestarlowast
                 \scantokens\expandafter{\detokenize{#1}}}}}

% \func, \funcdesc, \keyword, \keyworddesc, \prec, \precdesc, \oper, \operdesc
% ============================================================================

\newcommand\func[1]{\hyperlink{\detokenize{func-#1}}{#1}()}
\newcommand\funcdesc[2][x]{\item[#2({#1})]\hypertarget{\detokenize{func-#2}}{}}%

\newcommand\keyword[1]{\hyperlink{\detokenize{kwd-#1}}{#1}}
\newcommand\keyworddesc[1]{\item[#1]\hypertarget{\detokenize{kwd-#1}}{}}%

\let\prec\relax % sinon, c'est \mathchar"321E
\newcommand\prec[1]{\hyperlink{\detokenize{prec-#1}}{#1}}
\newcommand\precdesc[1]{\item[\texttt{#1}]\hypertarget{\detokenize{prec-#1}}{}\leavevmode}%

\newcommand\var[1]{\hyperlink{\detokenize{var-#1}}{#1}}
\newcommand\vardesc[1]{\item[#1]\hypertarget{\detokenize{var-#1}}{}}%

\newcommand\oper[1]{\hyperlink{\detokenize{oper-#1}}{#1}}
\newcommand\operdesc[1]{\item[\texttt{#1}]\hypertarget{\detokenize{oper-#1}}{}}%


% \xintname, \xintnameimp etc...
% ==============================


\xintForpair #1#2 in
{(xint bundle,bundle),
 (xintkernel,kernel),
 (xinttools,tools),
 (xintcore,core),(xint,xint),(xintbinhex,binhex),(xintgcd,gcd),%
 (xintfrac,frac),(xintseries,series),(xintcfrac,cfrac),(xintexpr,expr),%
 (xinttrig, trig), (xintlog, log)}
\do
{%
 \expandafter\def\csname #1name\endcsname
   {\texorpdfstring
                  {\hyperref[sec:#2]%
                     {\relax{\color{joli}\MakeNameUp{#1}}}}%
                  {#1}%
    \xspace }%
 \expandafter\def\csname #1nameimp\endcsname
   {\texorpdfstring
                  {\hyperref[sec:#2imp]%
                    {\relax{\color{blue}\MakeNameUp{#1}}}}%
                  {#1}%
    \xspace }%
}%

 \def\DOCxintfrontpage
   {\texorpdfstring
                  {\hyperref[frontpage]{\relax{\color{joli}TOC}}}%
                  {TOC}%
    \xspace }%

\makeatletter
\protected\def\MakeNameUp#1{%
  \ifcsname #1nameUp\endcsname
    \expandafter\@firstoftwo\else
    \expandafter\@secondoftwo
  \fi
% \ifinheader 2020/01/30
  {\ifinheader\fbox{\textup{#1}}\else#1\fi}%
  {#1}%
}
\makeatother
\newif\ifinheader
% doit être protégé
\protected\def\inheadertrue{\let\ifinheader\iftrue}

% \RaisedLabel
% ============

% Samedi 16 juin 2018 à 15:23:22
% trick to see header of target page
% there is probably better way to use the already in place
% anchor from \section, but no time to go into hyperref source
\newcommand\RaisedLabel[2][6]{%
\vspace*{-#1\baselineskip}%
\begingroup
  \let\leavevmode\relax\phantomsection
  \label{#2}%
\endgroup
\vspace*{#1\baselineskip}%
}

% begin{document}
% ===============
% \ttzfamily done at begin document

\newcommand\ctanpackage[1]{\href{https://ctan.org/pkg/#1}{#1}}

\begin{document}\thispagestyle{empty}
\pdfbookmark[1]{Title page}{TOP}
\def\partname{Part}
\addto\captionsenglish{\def\partname{Part}}

{%
\normalfont\Large\parindent0pt \parfillskip 0pt\relax
 \leftskip 2cm plus 1fil \rightskip 2cm plus 1fil
\ifnum\dosourcexint=1
 The \xintnameimp packages source code\par
 \gdef\DOCxintfrontpage
   {\texorpdfstring
                  {\hyperref[frontpage]{\relax{\color{blue}TOC}}}%
                  {TOC}%
    \xspace }%
\else
 The \xintexprname and allied packages\par
\fi
\RaisedLabel{frontpage}
}

{\centering
  \textsc{Jean-Fran\c cois Burnol}\par
  \footnotesize
  jfbu (at) free (dot) fr\par
  Package version: \xintbndlversion\ (\xintbndldate);
            documentation date: \xintdocdate.\par
  {From source file \texttt{xint.dtx}. \xintdtxtimestamp.}\par
}

\medskip

% 31 janvier 2020

 \def\DOCxintexprintro
   {\texorpdfstring
                  {\hyperref[part:1]{\relax{\color{joli}\MakeNameUp{Start here}}}}%
                  {Start here}%
    \xspace }%

 \def\DOCxintexprmacros
   {\texorpdfstring
                  {\hyperref[sec:oldxintexpr]{\relax{\color{joli}\MakeNameUp{xintexpr}}}}%
                  {xintexpr}%
    \xspace }%

 \def\DOCexamples
   {\texorpdfstring
                  {\hyperref[sec:examples]{\relax{\color{joli}\MakeNameUp{Examples}}}}%
                  {Examples}%
    \xspace }%

% Vendredi 15 juin 2018
% Someone makes the comma active (not me! not sure if doc.sty or KOMA) and
% this derails xspace.sty, in the headers, as it uses \scantokens on a list of
% tokens, so it fails to recognize the commas which of course are of catcode12
% Update Friday, June 25 2021
% I finally take the time to go to the root of this: the culprit is everbatim
% when its output crosses pagebreaks, but the same of course happens with
% verbatim.
% I opened ticket https://github.com/latex3/latex2e/issues/600
\def\xintRunningHeader{{\inheadertrue\catcode`,12\relax
          \DOCxintfrontpage,
          \DOCxintexprintro,
          \xintexprname,
          \xinttrigname,
          \xintlogname,
          \csname xint bundlename\endcsname}}
\markboth{\makebox[0pt]{\xintRunningHeader}}{\makebox[0pt]{\xintRunningHeader}}

% Skips safely.
\ifnum\dosourcexint=1
\catcode`+ 0 \catcode0 9 % n'importe quoi sauf 15 (car ^^@)
\catcode`\\ 12
+expandafter+iffalse+fi
\fi
%

\newcommand\TeXnote{\par\smallskip\textbf{\TeX hackers note: }}

\etocsetlevel{toctobookmark}{6}



\etocsetlevel{table}{2}%  subsection

\renewcommand*{\etocbelowtocskip}{0pt}
\renewcommand*{\etocinnertopsep}{0pt}
\renewcommand*{\etoctoclineleaders}
              {\hbox{\normalfont\normalsize\hbox to 1ex {\hss.\hss}}}
% \etocmulticolstyle [1]{%
%     \phantomsection\section* {Contents}
%     \etoctoccontentsline*{toctobookmark}{Contents}{1}%
% }

\etocsettocstyle{}{}
\etocsettagdepth {part1A}{subsection}
\etocsettagdepth {part1B}{section}
\etocsettagdepth {macros}{none}
\etocsettagdepth {implementation}{none}

\etocsettocdepth{subsection}
\tableofcontents

\renewcommand*\etocabovetocskip{\bigskipamount}
\makeatletter
\etocmulticolstyle [2]{\parskip\z@skip%\raggedcolumns
    \setlength{\columnsep}{\leftmarginii}%
    \setlength{\columnseprule}{0pt}%
}%
\makeatother
   \etocsettagdepth {part1A}{none}
   \etocsettagdepth {part1B}{none}
   \etocsettagdepth {macros}     {section}
\ifnum\NoSourceCode=1
   \etocsettagdepth {implementation}{none}
\else
   \etocsettagdepth {implementation}{section}
\fi

\vspace*{2\baselineskip}
\tableofcontents


\etocignoredepthtags

\etocmulticolstyle [1]{%
    \phantomsection% \section* {Contents}
    \etoctoccontentsline*{toctobookmark}{Contents}{2}%
}

\inmanualmaintocfalse

\clearpage

% ----
% Fibonacci code
% December 7, 2013. Expandably computing a big Fibonacci number
% with the help of TeX+\numexpr+\xintexpr, (c) Jean-François Burnol
\catcode`_ 11
%
% ajouté 7 janvier 2014 au xint.dtx pour 1.07j.
%
% Le 17 janvier je me décide de simplifier l'algorithme car l'original ne tenait
% pas compte de la relation toujours vraie A=B+C dans les matrices symétriques
% utilisées en sous-main [[A,B],[B,C]].
%
% la version ici est celle avec les * omis: car multiplication tacite devant les
% sous-expressions depuis 1.09j, et aussi devant les parenthèses depuis 1.09k.
\def\Fibonacci #1{%
    \expandafter\Fibonacci_a\expandafter
        {\the\numexpr #1\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 1\expandafter\relax\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 1\expandafter\relax\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 1\expandafter\relax\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 0\relax}}
%
\def\Fibonacci_a #1{%
    \ifcase #1
          \expandafter\Fibonacci_end_i
    \or
          \expandafter\Fibonacci_end_ii
    \else
          \ifodd #1
              \expandafter\expandafter\expandafter\Fibonacci_b_ii
          \else
              \expandafter\expandafter\expandafter\Fibonacci_b_i
          \fi
    \fi {#1}%
}%
\def\Fibonacci_b_i #1#2#3{\expandafter\Fibonacci_a\expandafter
  {\the\numexpr #1/2\expandafter}\expandafter
  {\romannumeral0\xintiiexpro sqr(#2)+sqr(#3)\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro (2#2-#3)#3\relax}%
}% end of Fibonacci_b_i
\def\Fibonacci_b_ii #1#2#3#4#5{\expandafter\Fibonacci_a\expandafter
  {\the\numexpr (#1-1)/2\expandafter}\expandafter
  {\romannumeral0\xintiiexpro sqr(#2)+sqr(#3)\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro (2#2-#3)#3\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro #2#4+#3#5\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro #2#5+#3(#4-#5)\relax}%
}% end of Fibonacci_b_ii
\def\Fibonacci_end_i  #1#2#3#4#5{\xintthe#5}
\def\Fibonacci_end_ii #1#2#3#4#5{\xinttheiiexpr #2#5+#3(#4-#5)\relax}
\catcode`_ 8

\def\Fibo #1.{\Fibonacci {#1}}

\def\specialprintone #1%
{%
    \ifx #1\relax \else \makebox[877496sp]{#1}\hskip 0pt plus 2sp\relax
    \expandafter\specialprintone\fi
}%
\def\specialprintnumber #1% first ``fully'' expands its argument.
{\expandafter\specialprintone \romannumeral-`0#1\relax }%

\AddToShipoutPicture*{%
    \put(10.5cm,14.85cm)
    {\makebox(0,0)
      {\resizebox{17cm}{!}{\vbox
       {\hsize 8cm\Huge\baselineskip.8\baselineskip\color{black!5}%
        \specialprintnumber{F(1250)=}%
                  \specialprintnumber{\Fibonacci{1250}}}\par}%
       }%
    }%
}


\pdfbookmark[1]{Dependency graph}{DependencyGraph}




\tikzstyle{block} = [rectangle, draw,
    fill=yellow!10,
%    fill opacity=0.5,
    draw=black!30,
    line width=2pt,
    text width=6em, text centered, rounded corners, minimum height=4em]
\tikzstyle{line} = [draw, line width=1pt, color=black!30]

%\vspace*{\stretch{0.1666}}

\begin{figure}[ht!]
  \phantomsection\label{dependencygraph}
\centeredline{%
\begin{tikzpicture}[node distance = 2.5cm]
    % Place nodes
    \node [block] (kernel) {\xintkernelname};
    \node [left of=kernel] (A) {};
    \node [right of=kernel] (B) {};
    \node [block, below right of=B] (core) {\xintcorename};
    \node [block, below left of=A]  (tools) {\xinttoolsname};
    \node [block, right of=core, xshift=1cm]  (bnumexpr) {\ctanpackage{bnumexpr}};
    \node [block, below of=core] (xint) {\xintname};
    \node [block, left of=xint, xshift=-1cm] (gcd) {\xintgcdname};
    \node [block, left of=gcd] (binhex) {\xintbinhexname};
    \node [block, below of=xint] (frac) {\xintfracname};
    \node [block, below of=frac, yshift=-.5cm] (expr) {\xintexprname};
    \node [block, below right of=expr, yshift=-.5cm, xshift=2.25cm] (polexpr) {\ctanpackage{polexpr}};
    \node [block, below of=polexpr] (session) {rlwrap etex \ctanpackage{xintsession}};
    \node [block, below of=expr, yshift=-.5cm] (trig) {\xinttrigname};
    \node [block, left of=trig] (log) {\xintlogname};
    \node [block, left of=log, xshift=-1cm] (poormanlog) {\ctanpackage{poormanlog}};
    \node [block, below right of=frac, xshift=1cm] (series) {\xintseriesname};
    \node [block, right of=series] (cfrac) {\xintcfracname};
    % Draw edges
    \path [line,-{Stealth[length=5mm]}] (kernel) -- (core);
    \path [line,-{Stealth[length=5mm]}] (kernel) -- (tools);
    \path [line,-{Stealth[length=5mm]}] (core) -- (bnumexpr);
%    \path [line,-{Stealth[length=5mm]}] (core) to [out=180,in=90] (gcd.north);
    \path [line,-{Stealth[length=5mm]}] (kernel) -- (binhex);
    \path [line,-{Stealth[length=5mm]}] (core) -- (xint);
    \path [line,-{Stealth[length=5mm]}] (xint) -- (frac);
    \path [line,-{Stealth[length=5mm]}] (xint) -- (gcd);
    \path [line,-{Stealth[length=5mm]}] (frac) -- (expr);
    \path [line,-{Stealth[length=5mm]}] (expr) -- (polexpr);
    \path [line,{Stealth[length=5mm]}-{Stealth[length=5mm]}] (expr) -- (trig);
    \path [line,{Stealth[length=5mm]}-{Stealth[length=5mm]}] (expr) -- (log);
    \path [line,-{Stealth[length=5mm]}] (poormanlog) -- (log);
    \path [line,-{Stealth[length=5mm]}] (expr) -- (polexpr);
    \path [line,-{Stealth[length=5mm]}] (expr) -- (session);
    \path [line,-{Stealth[length=5mm]}] (polexpr) -- (session);
    \path [line,-{Stealth[length=5mm]}] (trig.south) to [out=-90,in=180] (session.west);
    \path [line,-{Stealth[length=5mm]}] (log.south)  to [out=-90,in=180] (session.west);
    \path [line,-{Stealth[length=5mm]}] (frac) to [out=0,in=90] (series.north);
    \path [line,-{Stealth[length=5mm]}] (frac) to [out=0,in=90] (cfrac.north);
    \path [line,dashed,-{Stealth[length=5mm]}] (binhex.south) -- (expr);
% at 1.3d gcd() and lcm() needs no support from xintgcd
%    \path [line,dashed,-{Stealth[length=5mm]}] (gcd.south) -- (expr);
% at 1.4 xintgcd loads xinttools
    \path [line,-{Stealth[length=5mm]}] (tools) to [out=0, in=90]
    (gcd.north);% je dois positionner mieux mais pas le temps de lire 700 pages
    \path [line,dashed,-{Stealth[length=5mm]}] (tools.south west) to [out=270, in=225]
    (cfrac.south west);% je dois positionner mieux mais pas le temps de lire 700 pages
    \path [line,-{Stealth[length=5mm]}] (tools) to [out=270,in=180] (expr);
  \end{tikzpicture}}\bigskip
\end{figure}

\vspace{\baselineskip}

\begin{addmargin}{2cm}
\normalfont\footnotesize Dependency graph for the
    \xintname bundle components: modules pointed to by arrows \textbf{automatically}
    import the modules originating the continuous line ended by an arrow.
    Dashed lines
    indicate a partial dependency, and to enable the corresponding
    functionalities of the lower module it is thus necessary to use 
    a suitable |\usepackage| (\LaTeX) or |\input| (Plain \TeX.)\par

    \ctanpackage{bnumexpr} is a
    separate (\LaTeX{} only) package by the author which uses (by default)
    \xintcorename as its mathematical engine.

    \ctanpackage{polexpr} handles definitions and algebraic operatione on
    one-variable polynomials, as well as root localization to arbitrary
    precision. It works both with Plain \TeX\ and with \LaTeX.

    \xinttrigname and \xintlogname are loaded automatically by \xintexprname;
    they can not be loaded directly via a separate |\usepackage| (in \LaTeX).

    \ctanpackage{poormanlog} is a \TeX{} and
    \LaTeX{} package by the author which is loaded automatically by \xintlogname.

    \ctanpackage{xintsession} is invoked on the command line as |etex
    xintsession| (or, if available, |rlwrap etex xintsession|). It loads
    \xintbinhexname automatically (but this is not indicated above graphically).
\par
\end{addmargin}

\vfill

\clearpage

\etocdepthtag.toc {part1A}

\csname Start herenameUp\endcsname
\part{The \xintexprname package}
\RaisedLabel[12]{part:1}

\etocsetnexttocdepth{section}
\localtableofcontents

\section {Usage}

\begin{itemize}
\item To use with |etex|, |pdftex|, ..., i.e.\@ with \TeX{} engines activating
  the e\TeX{} extensions:
\begin{everbatim}
\input xintexpr.sty
\xintfloateval{sqrt(13), cos(1), exp(13.3)}% uses 16 digits per default
\xintDigits*:=32;% reload log and trig libraries
\xintfloateval{sqrt(13), cos(1), exp(13.3)}% now with 32 digits
\xinteval{2^1000}% exact computations
\xinteval{reduce(add(1/i, i=1..50))}% dummy variables
\end{everbatim}
\item To use with the \LaTeX{} macro layer (|latex|, |pdflatex|, ...):
\begin{everbatim}
\usepackage{xintexpr}
% and here you have to wait for \begin{document}...
% or rather you can start playing immediately:
\typeout{\xinteval{sqrt(13, 60)}}
\end{everbatim}
\end{itemize}

\xintexprname is a package to do expandable computations, either exactly with
arbitrarily big inputs (fractions, arbitrarily long decimal expansions, ...),
or in the sense of floating point numbers (logarithm, exponential, sine,
cosine, ...).  The math functions are implemented up to \dtt{62} digits of
precision.  The square root (as well of course as the four operations) achieve
correct rounding in arbitrary precision.

The syntax to modify the precision used for floating point evaluations is
\begin{everbatim}
\xintDigits*:= <Number>;
\end{everbatim}
Use the |*| (\csbxint{Digits*}), else the scientific libraries will not be
reloaded.  The current precision is available as \csbxint{theDigits},
but in this documentation I might be using simply |Digits| to refer to it.

The tables of the built-in
\hyperref[tab:precedences]{operators} and  \hyperref[tab:functions]{functions}
will give a quick overview of the available syntax.
The simplest way\footnote{I am assuming here Mac OS or Linux,
adapt to your environment} to test the syntax is to work interactively on the command line
(this feature is available since April 2021, the version of
\ctanpackage{xintsession} used here is |1.3a|):
\begin{everbatim}
rlwrap etex xintsession
[...welcome banner...]
  Magic words: `&pause' (or `;'), `&help', `&bye',
               `&exact', `&fp', `&int', `&pol'.

  \jobname is xintsession
  Transcript will go to log and to xintsession-210609_12h00.tex
Starting in exact mode (floating point evaluations use 16 digits)
>>> 2^100;
@_1     1267650600228229401496703205376
>>> cos(1);
@_2     0.5403023058681397
>>> &fp=32
(/usr/local/texlive/2021/texmf-dist/tex/generic/xint/xintlog.sty)
(/usr/local/texlive/2021/texmf-dist/tex/generic/xint/xinttrig.sty)
fp mode (log and trig reloaded at Digits=32)
>>> cos(1);
@_3     0.54030230586813971740093660744298
>>> 3^1000;
@_4     1.3220708194808066368904552597521e477
>>> &exact
exact mode (floating point evaluations use 32 digits)
>>> 3^1000;

@_5     13220708194808066368904552597521443659654220327521481676649203682268285
9734670489954077831385060806196390977769687258235595095458210061891186534272525
7953674027620225198320803878014774228964841274390400117588618041128947815623094
4380615661730540866744905061781254803444055470543970388958174653682549161362208
3026856377858229022841639830788789691855640408489893760937324217184635993869551
6765018940588109060426089671438864102814350385648747165832010614366132173102768
902855220001
>>> &bye
Did I say something wrong?
Session transcript written on xintsession-210609_12h00.tex
 )
No pages of output.
Transcript written on xintsession.log.
\end{everbatim}

\medskip

Warning: I don't have the time to maintain perfectly such large documentation.
In preparing the |1.4| release I may have missed updating some bits
which got randomly shuffled to new places (at least I did delete large
sections, which was a hard decision to take, almost breaking the palimpsest
quality of the document).  Reports welcome.%
%
\footnote{Thanks to Jürgen Gilg for keeping the author motivated and
  helping proof-read the 1.4 documentation.}

\subsection{Improved support for logarithm, exponential, sine, etc...\@ at
  the \texttt{1.4e} release of \texttt{2021/05/05}}

They are now supported up to \dtt{62} digits and achieve «correct rounding»%
%
\footnote{This means that the produced value is the rounding to |Digits|
  significant digits of the theoretical exact mathematical value.  The
  rounding mode is currently not customizable and is «rounding to nearest,
  ties go to infinity of same sign».}
%
at least
in \dtt{99\%} of cases%
%
\footnote{It is even better than that, but depends a bit on how |Digits| is
  located relative to some thresholds governing Taylor series
  or other approximation means.}
%
for |Digits| being at least \dtt{9}.

For Digits up to \dtt{8}, a special more approximate implementation is used,
and the functions achieve the correct rounding (particularly at |Digits|
equal to \dtt{8} or \dtt{7}) less often, but are significantly faster
(especially logarithm, exponential, powers) than working with \dtt{9} digits
or more.  The achieved precision is largely enough for plots.

See \xintlogname and \xinttrigname for some additional information.


\subsection{Breaking changes at the \texttt{1.4g} release}

\begin{enumerate}[noitemsep]
 \item Power operators |**| and |^| are now parsed
   in a right associative way:%
%
   \footnote{Don't try |2^3^4^5| at home. This is BIG.}
\begin{everbatim*}
$${({(5^4)}^3)}^2=5^{4\cdot3\cdot2}=\xinteval{((5^4)^3)^2}=\xinteval{5^24}$$
$$5^{4^{3^2}} = 5^{262144}\approx \xintTeXfromSci{\xintfloateval{5^4^3^2}}$$
\end{everbatim*}

 \item Single character operators |&|, \verb=|=, and |=| (which were
   deprecated since |1.1|) have been removed.  Use |&&|, \verb=||= and |==|
   respectively.
\end{enumerate}

\subsection{Breaking changes at the \texttt{1.4f} release}

\begin{enumerate}[noitemsep]
\item \csbxint{ieval}|{[-D]...}|\IMPORTANT{}
  which quantizes to a multiple of |1eD| when
  |D| is positive now does \emph{NOT append the |D| trailing zeroes anymore}!
\item \func{float\string_dgt} is the new name of |float_()|, as already
  documented but
  not yet done at |1.4e|.
\item Powers in \csbxint{floateval} with big exponents are computed with more
  accuracy.
\item For |Digits| at most \dtt{8}, powers in \csbxint{floateval} with integer
  or half-integer exponents do not use \func{log10}/\func{pow10} method but
  (as for |Digits| at least \dtt{9}) the \csbxint{FloatPower} based approach,
  allowing very big integer of half-integer exponents while keeping accuracy.
\item Computations with math functions in \csbxint{floateval} at a setting of
  |Digits| higher than \dtt{62} will produce an output limited to the smallest
  of |Digits| and \dtt{64} digits.
\end{enumerate}

\subsection{Breaking changes at the \texttt{1.4e} release}

In principle, I try for breaking changes regarding output to happen only at
major releases.  But it is not as if I had a gigantic user base, and sometimes
it is needed to move forward; and as the \TeX\ world does not have a
``pinning'' mechanism like Python's world, the distinction between minor and
major releases is a bit rhetorical. So |1.4e| behaves like a major release.
It is not one because a complete rework of the foundations of floating point
support is needed...

|1.4e| changes the output format used by \csbxint{eval}, \csbxint{ieval}, and
\csbxint{floateval} (in short all is modified!).

\begin{itemize}
\item \csbxint{floateval} output macro \csbxint{PFloat} has been modified. In
  particular mantissas are trimmed of trailing zeros. Integers are printed
  with a zero after the decimal mark.
\item \csbxint{eval} output macro \csbxint{FracToSci} has been modified,
  regarding the handling of numbers involving a decimal exponent; rather than
  printing out an integer mantissa, it now uses the same conventions as
  \csbxint{PFloat} (of course without pre-rounding to the |Digits| precision).
  The \csa{xintFracToSciE} was removed because \csbxint{PFloatE} is used.

  Notice though that fractions are still not automatically reduced to lowest
  terms even on output. I hesitated about this, but when for example the
  computation is a large power of an already known to be irreducible fraction,
  it would be a very costly operation to apply \csbxint{Irr} or \csbxint{PIrr}
  to it.
\item \csbxint{ieval} was modified to use on output
  \csbxint{DecToString} and not anymore \csbxint{FracToSci}.

  This means than in case of usage of the |[D]| optional argument with a
  negative |D| (i.e.\@ rounding the output to a multiple of a positive power
  of ten) the output does not use scientific notation but is an integer ending
  with explicit zeros.

  Update: at |1.4f| this was changed again.\IMPORTANT{} Now no trailing zeroes
  are added at all!

  Nothing was changed to output for the case of a positive |[D]| (i.e.\@
  rounding to |D| figures after decimal point).  One now only needs to
  configure the new \csbxint{iexprPrintOne} to be \csbxint{DecToStringREZ} in
  order for trailing zeros in decimal expansions to get trimmed rather than
  there being always exactly |D| figures after decimal point.
\item Fractional powers are now allowed in \csbxint{eval}; powers with integer
  exponents are still computed exactly of course, except that if the integer
  exponent is \dtt{10000} or more, or if it is is
  evaluated that the output will contain more than (about) \dtt{10000} digits
  (separately for the numerator and denominator),
  then the floating-point branch is taken forcedly. More details on fractional
  powers are available in \xintlogname.
\item \csa{poormanloghack} is now a no-op; to use the logarithm,
  exponential, and powers based on \ctanpackage{poormanlog}, set Digits to at
  most \dtt{8}. Don't forget the |*| in the |\xintDigits*:=8;| syntax.
  \item The \csbxint{floatexprPrintOne} macro interface has changed, it is now
    to be used with (an expandable) macro either allowing or requiring the
    rounding precision to be present as |[P]|, not |{P}|. Its default 
    is the user level \csbxint{PFloat} whose behaviour has changed.
\end{itemize}

\begin{framed}
  STARTING FROM HERE THE DOCUMENTATION MAY NOT BE UP-TO-DATE AT |1.4e|
  RELEASE, IT MAY CONTAIN OBSOLETE INFORMATION
\end{framed}

\subsection{Features added since the \texttt{1.4} release}

For bugfixes and possibly more details check |CHANGES.html|:
\centeredline{|texdoc --list xint|}
\begin{itemize}
\item The concept of simultaneous assignments is extended:\NewWith{1.4i} in
  case of more variables than values the extraneous variables do not cause an
  error message but are simply set to the |nil| value; in case of more values
  than variables, the last variable is defined to be the ople concatenating
  all the extra values. See \csbxint{defvar}.

\item Built-in\NewWith{1.4i} functions usable with arbitrarily many arguments
  such as \func{max}, \func{gcd}, or \func{len} are now again usable with a
  single numeric argument: since |1.4| a lone argument had to be a |nutple|
  (which was automatically unpacked).  It can now again be a number.

  \item \csbxint{TeXfromSci}\NewWith{1.4g}

  \item The most important feature is at |1.4e| the extended range and accuracy
  of the scientific functions, up to \dtt{62} digits.%\NewWith{1.4e}

  \item The constraints for the replacement macro to be used for
    \csbxint{exprPrintOne} have been much simplified. See the
    documentation of \csbxint{FracToSci} which is the package default.
    %\NewWith{1.4e}

  \item \csbxint{iexprPrintOne} was added, with default
    \csbxint{DecToString}.%\NewWith{1.4e}

  \item \csbxint{DecToStringREZ} was added.%\NewWith{1.4e}

  \item The function \func{zip}.

  \item The function \func{flat}.

  \item Chaining of \hyperlink{\detokenize{prec-10}}{comparison operators} à la
  Python (no short-circuit, though) and |l3fp|.

  \item \csbxint{PFloatE} to specify like \csa{xintFracToSciE} (now defunct at
    |1.4e|) does for
    \csbxint{eval} since |1.4| the separator to use between mantissa and
    exponent in the output of \csbxint{floateval} output.

  \item \csbxint{thespaceseparated} (serves to provide suitable input
    to PS-Tricks |\listplot|).

  \item The optional argument |[D]| to \csbxint{iexpr} (or \csbxint{ieval})
    can be negative, with about the same meaning as the non-negative case, i.e.
    rounding to an integer multiple of |1e-D|.  But attention that at |1.4f|,
    for |D<0|, this is now interpreted as the rounded quotient by |1e-D| (no
    trailing zeros appended, no scientific exponent either).

  \item Also the second argument of \func{trunc} and \func{round} can be
    negative (but then it enacts quantization, e.g |trunc(1234,-2)| produces
    \dtt{\xinteval{trunc(1234,-2)}}).  Matching updates to the support
      macros \csbxint{Trunc}, \csbxint{Round}, \csbxint{iTrunc}, and
      \csbxint{iRound}.

  \item Support by \func{add} and \func{mul} for \keyword{omit}, \keyword{abort}
    and \func{break}.%
%

\end{itemize}

\subsection{The \texttt{1.4} release of \texttt{2020/01/31}}

|1.4| brought some new features (involving significant evolution of the
\xintexprname.sty source code) and a few (but important) breaking changes.
See |CHANGES.html| which contains information which may not yet have been
included into this PDF documentation.

The main new feature was (initial) support for nested structures.  For a quick
idea of already available related abilities check for example \func{ndseq} or
\csbxint{defufunc}.  See also \csbxint{thealign}.  However, please grant the
author a few decades to finish absorbing Python/NumPy.

\begin{framed}
  The main breaking changes were:
  \begin{itemize}
  \item \xintexprname |1.4| requires the |\expanded| primitive, which is
    provided by all major \TeX{} engines since \TeX Live 2019.  The macro
    packages \xintname, \xintfracname, \xinttoolsname et al.\@ do not (yet)
    require |\expanded|.

    It is probable also |\pdfstrcmp| (|\strcmp|) will be required at some point
    but it has been provided by major \TeX{} engines for a long time already.

  \item \csbxint{eval} (and \csbxint{expr}) output does not use anymore the
    \xintfracname ``raw'' format |A/B[N]|, rather it uses scientific notation
    |AeN/B|, dropping the exponent and/or denominator if they are
    respectively \dtt{0} and/or \dtt{1}.  This means that output can now be
    copied pasted directly to competing software on the market, such as Python
    or Maple.  The output format of \csbxint{floatexpr} (which uses macro
    \csbxint{PFloat}) was left un-modified although the prettifying done by it
    is not necessarily the best choice when displaying a nested structure via
    \csbxint{thealign} (perhaps next major release will reconsider that
    choice); and the way the zero value is output by \csbxint{floateval},
    currently \dtt{\xintfloateval{0}} is yet to be chosen definitely.  The
    used (\emph{expandable}) macro for output can be specified by user.

  \item Syntax such as |x*[a, b, c]| or |[a, b, c]+x| for itemwise operation
    on «lists» has been (provisorily) dropped. Indeed, the brackets |[...]|
    are now genuine constructors of nestable structures, and implementing the
    feature (analogous to NumPy's concepts) will require overloading all
    scalar infix operators.  Alternative already exist in the syntax for
    example |seq(x*y, y = a,b,c)|.  Actually in future |x*[a, b, c]| will be
    as |[x*a, x*b, x*c]| i.e.\@ will keep the brackets, which prior to |1.4|
    on their own were no different from parentheses.
  \end{itemize}
\end{framed}

\subsection{Known bugs/features (last updated at \texttt{1.4i})}

\begin{description}
\item[if(100>0,(100,125),(100,128)) breaks my code:]
%
  This is a feature. This is a syntax error, as the comma serves to contatenate
  "oples" (see \autoref{oples}), and parentheses do not create analogs of
  "tuples", so this input is parsed the same as
\begin{everbatim}
    if(100>0,100,125,100,128) 
\end{everbatim}
  which is an error as \func{if} requires exactly three arguments, not
  five. Use:
\begin{everbatim}
    if(100>0,[100,125],[100,128]) 
\end{everbatim}
  which will expand to the "tuple" |[100,125]|.
\item[{\detokenize{\xintdeffunc foo(x):= gcd((x>0)?{[x,125]}{[x,128]});}
    creates a broken function:}]
%
  Bug. Normally \func{gcd} (and other
  multi-arguments functions) work both with open lists of arguments or
  bracketed lists ("nutples") and the above syntax would work perfectly fine
  in numerical context.  But the presence of the \oper{?} breaks in
  \csbxint{deffunc} context the flexibility of \func{gcd}. 

   Currently working alternatives:
\begin{everbatim}
\xintdeffunc foo(x) := gcd(if(x>0, [x,125], [x,128]));
\xintdeffunc foo(x) := if(x>0, gcd(x,125), gcd(x,128));
\xintdeffunc foo(x) := if(x>0, gcd([x,125]), gcd([x,128]));
\xintdeffunc foo(x) := gcd((x>0)?{x,125}{x,128});
\xintdeffunc foo(x) := (x>0)?{gcd(x,125)}{gcd(x,128)};
\xintdeffunc foo(x) := (x>0)?{gcd([x,125])}{gcd([x,128])};
\end{everbatim}
   The same problem will arise with an \oper{??} nested inside \func{gcd} or
   similar functions, in an \csbxint{deffunc}.

 \item[{\detokenize{\xinteval{0^-.5}} says "0 raised to power -1"}]
%
  Feature. Half integer exponents are handled via a square-root extraction, so
  here \xintexprname wanted to first raise \dtt{0} to power \dtt{-1}, as
  reported.

 \item[{Comparison operator \detokenize{==} crashes with nutples}]
%
   Not yet implemented...

 \item[{I liked the ``broadcasting'' \detokenize{[1..10]^10} syntax, but it
     was removed at 1.4}]
%
   Patience...\@ |seq(x^10,x=1..10)| is alternative (add external |[..]| to get
   a nutple).
 \item[{\detokenize{1e\numexpr5+2\relax} crashes}]
%
   Not clear yet if bug or feature.
   The syntax accepted in the scientific part is limited, and failure
   is expected: hitting a |\numexpr|
   when parsing a number triggers insertion of a tacit multiplication and then
   |1e| is missing the scientific exponent. The same happens with |1e(2+3)|.
   Use syntax such as |1e\the\numexpr5+2\relax|, or |1e\xinteval{5+2}|
   (although here this relies on output format of |\xinteval| using integer
   notation with no decoration in this case).

 \item[{seq(1e-i,i=1..5) crashes}]
%
   Not clear if bug or feature. Use |seq(1e\xinteval{-i},i=1..5)| or, as a
   possibly faster way |seq(1e\xintiieval{-i},i=\xintiiexpr1..5\relax)|.

 \item[{omit/abort if nested and not last in the sub-expression cause a crash}]
%
   For example |seq(subs((i)?{i}{abort},t=i)+10, i=-2, -1, 0, 1)| crashes, due
   to the presence of the |+10|.  This is a longstanding limitation, applying
   ever since |omit/abort| were added to the syntax at |1.1|.  Even without
   the |+10| the nested case was broken by a |1.4| regression and got fixed
   only at |1.4h|.  The non-nested case |seq((i)?{i}{abort}+10, i=-2, -1, 0,
   1)| works and the «must be last in expression if nested» limitation is
   currently considered a feature.
\end{description}

The list stops here, but there are certainly other pending bugs in my bug-log,
and many more I am not yet aware of. In particular it is already mentioned in
the \csbxint{deffunc} documentation that it can not parse currently the
entirety of the available purely numerical syntax, some (documented or not,
known or not) limitations apply.

\subsection{License and installation instructions}

\label{ssec:install}

\xintname is made available under the
\href{http://www.latex-project.org/lppl/lppl-1-3c.txt}{LaTeX Project Public
  License 1.3c} and is included in the major \TeX\ distributions, thus there
is probably no need for a custom install: just use the package manager to
update if necessary \xintname to the latest version available.

Else, \href{https://ctan.org/pkg/xint}{CTAN} access provides |xint.tds.zip|
which has all source code and documentation in a TDS-compliant archive, only
waiting to be |unzip -d <DIR>| into some suitable hierarchical structure.

Else, |etex xint.dtx| extracts all source code. A |Makefile| is also provided
with targets such as |xint.pdf| or |sourcexint.pdf|. Even if your system does
not allow executing |make|, the rules it contains can be imitated manually
(if possible using |Latexmk|).

With \TeX\ distributions providing a |"texdoc"| or similar utility,
\centeredline{|texdoc --list xint|}
gives the choice to display one of:
\begin{itemize}[nosep]
\item |xint.pdf| (this file),
\item |sourcexint.pdf| (source code),
\item |README.md|,
\item |CHANGES.html|.
\end{itemize}

\subsection{Printing big numbers on the page}\label{ssec:printnumber}
When producing very long numbers there is the question of printing them on
  the page, without going beyond the page limits. In this document, I have most
  of the time made use of these macros (not provided by the package:)

%
\everb|@
\def\allowsplits #1{\ifx #1\relax \else #1\hskip 0pt plus 1pt\relax
                    \expandafter\allowsplits\fi}%
\def\printnumber #1{\expandafter\allowsplits \romannumeral-`0#1\relax }%
% \printnumber thus first ``fully'' expands its argument.
|

It may be used like this:
%
\begin{everbatim*}
\printnumber{\xintiieval{100!^3}}\newline
\end{everbatim*}%
The rendering here uses extra decoration.

\clearpage
\expandafter\let\csname Start herenameUp\endcsname\undefined
\csname xintexprnameUp\endcsname
\section{\xintexprname syntax reference and user guide}
\RaisedLabel{sec:expr}

\localtableofcontents


\subsection{Oples and nutples: terminology for the \text{1.4} \xintname generation}\label{oples}

\emph{Skip this on first reading, else you will never start using the
  package.} \fbox{SKIP THIS!} (understood?)

In this section I will describe a mathematical terminology which models
how the parser handles the input syntax with numbers, commas, and brackets,
and how it maps internally to \TeX\ specific concept, particularly braces and
macro arguments.

\etocsetnexttocdepth{subsubsection}
\localtableofcontents

\subsubsection{Base terminology}

We start with a set $\mathcal{A}$ of \emph{atoms}, which represent numeric
data. In \TeX{} syntax such \emph{atoms} are always braced, more precisely,
currently they look like
%
\centeredline{\dtt{\{raw format within \TeX{} braces\}}}
%
The \TeX{} braces are not set-theoretical braces here, they are simply used
for \TeX nical reasons (one could imagine using rather some terminator token,
but ultimately support macros for built-in and user defined functions rely on
\TeX\ macros with undelimited parameters, at least so far).

Our category $\mathcal{C}$ of «oples» is the smallest collection of
\emph{totally ordered finite sets} verifying these properties:
\begin{enumerate}
\item The empty set \dtt{$\emptyset$} is an \emph{ople}, i.e.\@ it belongs to
  $\mathcal{C}$.
\item Each singleton set \dtt{$\{O\}$} whose element \dtt{$O$} is either an
  \emph{atom} $a\in\mathcal{A}$ or an \emph{ople} qualifies as an \emph{ople}.
\item $\mathcal{C}$ is stable by concatenation.
\end{enumerate}

Notes:
\begin{itemize}
\item 
We refer to the empty set \dtt{$\emptyset$} via the variable \emph{nil}.%
%
\footnote{There is
actually a built-in variable with this name. At |1.4|, |\xintexpr\relax| is
legal and also generates the \emph{nil}.}

\item It is convenient to accept the empty set as being also an
  \emph{atom}.  If this is done, then we may refer to the original
  \emph{atoms} (elements of $\mathcal{A}$) as \emph{non empty numerical data}.

\item
Concatenation is represented in the syntax by the
comma. Thus repeated commas are like only one and |nil| is a neutral element.

\item A singleton \emph{ople} \dtt{$\{a\}$} whose single element is a
  (non-empty) \emph{atom} is called a \emph{number}.%
%
  \footnote{This has to be taken in a general sense, for example with
    \ctanpackage{polexpr}, polynomials are represented by such «numbers».}
%

\item
The operation of constructing \dtt{$\{O\}$} from the \emph{ople} \dtt{$O$} is
called \emph{bracing} (set theory, \TeX), or \emph{bracketing} (\xintexprname
input syntax, Python |lists|), or \emph{packing} (as a reverse to Python's
unpacking of sequence type objects).  In the expression input syntax it
corresponds to enclosing \dtt{$O$} within square brackets: \dtt{$[O]$}.

\item A braced \emph{ople} is called a \emph{nutple}. Among them \dtt{\{nil\}}
  (aka $\{\emptyset\}$) is a bit special. It is called the \emph{none-ple}.%
%
  \footnote{Prior to version |1.4j| of this documentation it was called the
    \emph{not-ple}.}
%
It is not \dtt{nil}.%
%
\footnote{There is (experimental) a pre-defined «\dtt{None}» variable which
  stands for the \emph{none-ple}. It can also be input as |[]|.}
\end{itemize}

Each \emph{ople} has a \emph{length} which is its cardinality as set. The
singleton |oples| are called \emph{one-ples}. There are thus two types of
\emph{one-ples}:
\begin{itemize}
\item \emph{numbers} \dtt{$\{a\}$}, $a \in \mathcal{A}$,
\item \emph{nutples} \dtt{$\{O\}$}, $O \in \mathcal{C}$.
\end{itemize}

If we consider the empty set |nil| on the same footing as |atoms|, the two
types have only one common object which is the \emph{none-ple}.  As a rule
arithmetic operations will either break or silently convert the \emph{none-ple}
to the zero value:
\begin{everbatim*}
\xinteval{3+[], 5^[], 10*[]}
\end{everbatim*}.
But attention that \csbxint{iieval} in contrast to \csbxint{eval} is broken by
such inputs.

\subsubsection{Items (and sub-items) versus elements}

In order to illustrate these concepts, let us consider how one should
interpret notation such as |3,5,7,9| when it arises in an
\csbxint{expr}|ession|:
\begin{description}
\item[tempting vocabulary:] Each of |3|, |5|, |7|, and |9| is an \emph{item}, or
  \emph{element} of the (comma separated) \emph{list}. In other terms we have
  here a list with 4 items.
\item[rigorous vocabulary:] each one of |3|, |5|, |7|, |9| stands for an
  \emph{ople} (of the \emph{one-ple} type) and |3,5,7,9| stands for their \emph{concatenation}.
\end{description}
It is important to understand that in an \csbxint{expr}|ession|, there is no
difference between |3,5,7| and |3,,,,5,,,,,,,,,7|.  So the view of the comma
as separator is misleading.  In other terms, the comma is NOT a separator but
the (associative) operator of concatenation of totally ordered sets, and the
number |3| for example represents a (singleton) set.

If we want to refer to |3| or |5| or |7| or |9| as «the items of the
(open) list |3,5,7,9|» (and probably this documentation already has such
utterances, due to legacy reasons from the pre-|1.4| internal model), we
\emph{must} realize that this clashes with using the word \emph{item} as
synonymous to \emph{element} in the set-theoretical sense.

To repeat, any ople \dtt{$O$} is a finite totally ordered set: if not the empty
set, it has \emph{elements} \dtt{$a_1$}, \dots, \dtt{$a_k$}, and the above means that
its \emph{items} are the singleton oples (aka one-ples) \dtt{$I_1=\{a_1\}$},
\dots, \dtt{$I_k=\{a_k\}$}.  Each \dtt{$a_j$} may be an |atom|, then
\dtt{$I_j$} is a |number|, or \dtt{$a_j$} is an |ople| (possibly the empty set), then
\dtt{$I_j$} is a |nutple| whose depth is one more than the one of the ople
\dtt{$a_j$}.

Thus we can refer to «items» but must then understand they are not «elements»:
«items» are «singleton sub-sets». The cardinality (aka length) of an ople is
also the number of its
items.  It would be tempting to use the terminology «sub-item» to keep in mind they are «sub-sets»
but this would again create confusion: a |nutple| has only one item which is
itself; and we need some terminology to refer to the individual numbers in the
|nutple| given in input as |[1,2,3]| for example.  It is natural to refer to
|1|, |2|, |3| as «sub-items» of |[1,2,3]| as the latter may be an «item» (it
is in particular an «item» of itself, the unique one at that).

We distinguish the |oples| of length zero (there is only one, the empty set)
or at least two as those which can never be an «item».  Those of length one,
the |one-ples|, are exactly those which can be «items».  Among them some may
have «sub-items», they are the |nutples| with the exception of the |none-ple|.
And the others do not have «sub-items», they are the |numbers| and the |none-ple| (whose input syntax is
either |[]| or the variable |None|).%
%
\footnote{%
A note on the \csbxint{verbosetrue} regime: for a variable defined to be
|3,5,7,9|, it will say that its value is |{3}{5}{7}{9}|, because it does not
keep the external set-theoretical braces.  The braces here are only \TeX{}
braces, and |{3}| is an |atom|.  The |number| would be |{{3}}| with the
external braces being set-theoretical and also used internally as \TeX{}
braces.  From the four numbers |{{3}}|, ..., |{{9}}| concatenation gives
|{{3}{5}{7}{9}}|, which is the |ople| |3,5,7,9|.  But the log view drops
deliberately the external braces.  If the variable is defined to be the
|nutple| |[3,5,7,9]|, then the log view will be |{{3}{5}{7}{9}}| (up to
details on how exactly the numeric quantities are coded) and the actual
internal \TeX{} entity will be |{{{3}{5}{7}{9}}}|, where the two external
layers of braces are both set-theoretical and \TeX nical braces.}


\subsubsection{Oples as trees}

We say that the empty set |nil| and \emph{atoms} are \emph{leaves}.

We associate with any \emph{ople} a tree. The root is the ople. In the case of
the |nil| ople, there is nothing else than the root, which we then consider
also a \emph{leaf}. Else the children at top level are the successive
\emph{elements} (not «items»!) of the ople.%
%
\footnote{\label{fn:alttree}%
  We could also consider a tree for which the children of the root node would
  be its items and recursively; in that case the leaves would be |numbers| and
  possibly the |None|. The tree of the |nil| would be the empty tree, the tree
  of |None| would have a single node and no edges.  Such a tree would match
  the input syntax (of course applying the rule that iterated commas are like
  only one).  The tree which is described in this section matches more
  directly the internal syntax, hence is more useful to the author, who is
  also the sole reader who extracts some benefit from reading this
  documentation once in a while.}
%
Among the elements some are \emph{atoms} giving \emph{leaves} of the tree,
others are \emph{nutples} which in turn have children.  In the special case of
the \emph{none-ple} we consider it has a child, which is the empty set and this
is why we consider the empty set |nil| to be also a potential \emph{leaf}. We
then proceed recursively.  We thus obtain from the root \emph{ople} a tree
whose vertices are either \emph{oples} or \emph{leaves}. Only the empty set
|nil| is both a \emph{leaf} and an \emph{ople}.

Considering the empty set |nil| as an \emph{atom} fits with the \xintexprname
internal implementation based on \TeX: |nil| is an empty pair of braces |{}|,
whereas an \emph{atom} is a braced representation of a numeric value using
digits and other characters. We construct \emph{oples} by putting one after
the other such constituents and bracing them, and then repeating the process
recursively.

It has also an impact on the definition of the \emph{depth} (a.k.a as
\emph{maximal dimension}) of an \emph{ople}. For example the \emph{ople}
$\{\emptyset A_1A_2\}$ with three elements, among them the empty set and two
atoms is said to have depth $1$, or to have maximal dimension $1$. And
$\{\{\emptyset\}A_1A_2\}$ is of depth $2$ because it has a leaf (the empty
set) which is a child of a child of the \emph{ople}. NumPy \emph{ndarrays}
have a more restricted structure for example
$\{\{A_{00}A_{01}\}\{A_{10}A_{11}\}\}$ is a $2$-dimensional array, where all
leaves are at the same depth. When slicing empties the array from its atoms,
NumPy keeps the shape information but prints the array as $[]$. This will not
be the case with \xintexprname, which has no other way to indicate the shape
than display it.
\begin{everbatim*}
\xinteval{[[],[]]}
\end{everbatim*}
\begin{everbatim*}
\xinteval{[[0,1],[10,11]][:,2:]}
\end{everbatim*}

\subsubsection{Ople slicing and indexing}
\label{sssec:opleslicing}

«Set-theoretical» slicing of an \emph{ople} means replacing it with one of its
subsets.  This applies also if it is a \emph{number}.  Then it can be sliced
only to itself or to the empty set (indeed it has only one element, which is
an atom).  Similarly the \emph{none-ple} can only be sliced to give itself or
the empty set.  And more generally a \emph{nutple} is a singleton so also can
only be set-sliced to either the empty set or itself.

\xintexprname extends «Python-like» slicing to act on \emph{oples}:
\begin{itemize}[nosep]
\item if they are not \emph{nutples} set-theoretical slicing applies,
\item if they are \emph{nutples} (only case having a one-to-one
  correspondence in Python) then the slicing happens \emph{within brackets}:
  i.e.\@ the \emph{nutple} is unpacked then the set-theoretical slicing is
  applied, then the result is \emph{repacked} to produce a new \emph{nutple}.
\end{itemize}
With these conventions the \emph{none-ple} for example is invariant under
slicing: unpacking it gives the empty set, which has only the empty set as
subset and repacking gives back the \emph{none-ple}. Slicing a general
\emph{nutple} returns a \emph{nutple} but now of course in general distinct
from the first one.

The input syntax for Python slicing is to postfix a variable or a
parenthesized ople with |[a:b]|.  See \autoref{ssec:lists} for more.  There
are never any out-of-range errors when slicing or indexing.  All operations
are licit and resolved by the |nil|, a.k.a. empty set.

«Set-theoretical» item indexing of an \emph{ople} means reducing it to a
subset which is a singleton. It is thus a special case of set-theoretical
slicing (which is the general process of selecting a subset as replacement of
a set).

\xintexprname extends «Python-like» indexing to act on \emph{oples}:
\begin{itemize}[nosep]
\item if they are not \emph{nutples} set-theoretical item indexing applies,
\item if they are \emph{nutples} (only case having a one-to-one
  correspondence in Python) then the meaning becomes \emph{extracting}: i.e.\@
  the \emph{nutple} is unpacked then the set-theoretical indexing is applied,
  but the result is \emph{not repacked}.
\end{itemize}
For example when applied to the \emph{none-ple} we always obtain
the |nil|. Whereas as we saw slicing the \emph{none-ple} always gives back the
\emph{none-ple}. Indexing is denoted in the syntax by postfixing by |[N]|. Thus
for \emph{nutples} (which are analogous to Python objects), there is genuine
difference between the |[N]| extractor and the |[N:N+1]| slicer. But for
\emph{oples} which are either |nil|, a \emph{number}, or of length at least 2,
there is no difference.

\subsubsection{Nested slicing of oples}

Nested slicing is a concept from NumPy, which is extended by \xintexprname to
trees of varying depths. We have a chain of slicers and extractors. I will
describe only the case of slicers and letting them act on a |nutple|. The
first slicer gives back a new |nutple|.  The second slicer will be applied to
each of one of its remaining elements.  However some of them may be
\emph{atoms} or the empty set.  In the NumPy context all leaves are at the
same depth thus this can happen only when we have reached beyond the last
dimension (axis). This is not permitted by NumPy and generates an error.
\xintexprname does not generate an error. But any attempt to slice an
\emph{atom} or the empty set (as element of its container) removes it. Recall
we call them \emph{leaves}. We can not slice leaves. We can only slice
non-leaf elements: such items are necessarily |nutples|.  The procedure then
applies recursively.

If we handle an extractor rather than a slicer, the procedure is similar: we
can not extract out of an \emph{atom} or the empty set. They are thus
removed. Else we have a |nutple|. It is thus unpacked and replaced by the
selected element. This element may be an atom or the empty set and any further
slicer or extractor will remove them, or it is a |nutple| and the procedure
applies with the next slicer/extractor.

\xintexprname allows to apply such a |[a:b,c:d,N,e:f,...]| chain of
slicing/extracting also to an \emph{ople}, which is not a \emph{nutple}. We
simply apply the first step as has been described previously and successive
steps will only get applied to either \emph{nutples} or \emph{leaves}, the
latter getting silently removed by any attempted operation.

\subsubsection{Function arguments versus variables}
\label{sssec:funcargs}

In a function declaration with \csbxint{deffunc}, the call signature is parsed
as a comma separated list, so here it is not true that repeated commas are
like only one: repeated commas are not allowed and will break the function
declaration.

When \xintexprname parses a function call, it first constructs the ople which
is delimited by the opening and closing parentheses, then it applies the
function body, after having mapped the successive items (not the elements) of
the parsed ople to the variables appearing in the function call
signature.  Hence the arguments in the call signature stand for |one-ples|
(i.e.\@ either |numbers| or |nutples|).

Let me explain why we can not define a function |foo(A,B)| of two oples: the
function call will evaluate as an ople what is enclosed within the
parentheses.  It is then impossible in general to split this uniquely into two
oples |A| and |B|, except if for example we know a priori the length of |A|.
We could imagine defining a declarative interface for a |foo(A,B)| with |A|
preset to have \dtt{37} items or at least a pre-defined number of items but
this is extraneous layer for a functionality no-one will use.

The alternative would be to consider that declaring |foo(A,B)| means |A| will
pick-up always the first item and |B| all the remaining ones, and thus will be
an ople; here, there are some \TeX nical implementation reasons which have
dissuaded the author to do this.

In its place, a special syntax |foo(A,*B)| for the declaration of the function
is available.  It means that |B| stands for the |nutple| which
receives as items all arguments in the function call beyond the first one
already assigned to |A|.

More generally, the last positional argument in a function declaration can
have the form |*|\meta{argname}.  This then means that \meta{argname}
represents a |nutple| which will receive as items all arguments in the
function call remaining after the earlier positional arguments have been
assigned.  The declared function body is free to again use the syntax
|*|\meta{argname} which will unpack it and thus produce the ople concatenating
all such optional arguments.

With \csbxint{defvar} one can define a variable with value an |ople| of
arbitrary cardinality.  Such a variable can be used in a function call, it
will then occupy the place of as many arguments as its cardinality (which is
its number of elements, hence of its associated items).  For example if
function |foo| was declared as a function of 5 arguments |f(a,b,c,d,e)| it is
legitimate to use it as |f(A,B)| if |A| is an ople-valued variable of length
three and |B| of length two.  The actual arguments |a,b,c,d,e| will be made to
match the three items of |A| and the two items of |B|.

\subsubsection{Final words on leaves}

In case things were too clear, let's try to add a bit of confusion with an
extra word on \emph{leaves}. When we discuss informally (particularly to
compare with NumPy) an input such as
\begin{everbatim}
[[1, 2], [3, 4]]
\end{everbatim}
we may well refer to |1|, |2|, |3|, and |4| as being «the leaves of the 2d
array».  But obviously we have here numbers and previously we explained that a
number is not a \emph{leaf}, its \emph{atom} is.  Well, the point here is that
we must make a difference between the input form as above and the actual
constructed \emph{ople} the parser will obtain out of it.  In the input we do
have numbers.  The comma is a \emph{concatenator}, it is not a separator for
enumeration! The \emph{ople} which corresponds to it has a \TeX{}
representation like this:
\begin{everbatim}
{{{1}{2}}{{3}{4}}}
\end{everbatim}
where we don't have the \emph{numbers} anymore (which would look like |{{1}}|,
|{{2}}|, ...)  but numeric \emph{atoms} |{1}|, |{2}|, |{3}|, |{4}| where the
braces are \TeX{} braces and \textbf{not} set-theoretical braces (the other
braces are both). Hence we should see the above as the |ople|
$\{\{A_{00}A_{01}\}\{A_{10}A_{11}\}\}$ with atoms $A_{00}=\{1\}$, ..., being the
\emph{leaves} of the tree associated to (or which is) the \emph{ople}.

Numbers may be called the \emph{leaves} of the \textbf{input}, but once
parsed, the input becomes an \emph{ople} which is
(morally) a tree whose leaves are \emph{atoms} (and the empty set).
This discussion can also be revisited with footnote 
%%\footref{fn:alttree}
\ref{fn:alttree} in mind.

\subsubsection{Farewell, thanks for your visit!}

I hope this is clear to everyone.  If not, maybe time to say this section is
not needed to understand almost all of the manual, but I needed to
write it to be able to maintain in future my own software.

\subsection{The three parsers}

\xintexprname provides three numerical expression parsers and two subsidiary
ones. They are designed to be compatible with expansion only context.  All
computations ultimately rely on (and reduce to) usage of the |\numexpr|
primitive from \eTeX{}%
%
\footnote{It can handle only integers, and they must be at most
$2^{31}-1={}$\dtt{\the\numexpr"7FFFFFFF\relax}. Thus some work has to be done
to handle arbitrarily big integers or arbitrary float precision.}.
%
These \eTeX{} extensions date
back to 1999 and are by default incorporated into the |pdftex|
etc...\@ executables from major modern \TeX{} installations for more than
fifteen years now.

\begin{itemize}
\item \csbxint{eval}\marg{expression} handles integers, decimal numbers,
  numbers in scientific notation and fractions. The algebraic computations are
  done \emph{exactly}, and in particular \oper{/} simply constructs
  fractions. Use \oper{//} for floored division.
\begin{everbatim*}
\xinteval{add(x/(x+1), x = 1000..1014)}\par
\end{everbatim*}
In this example, the fraction obtained by addition is already
irreducible, but this is not always the case:
\begin{snugframed}
  By default, basic operations on fractions do not automatically reduce to
  smallest terms the output: |A/B| multiplied by |C/D|
  returns |AC/BD|, and |A/B| added to |C/D| uses |lcm(B, D)| as denominator.
\end{snugframed}
Arbitrarily long numbers are allowed in the input. The space character
(contrarily to the situation inside |\numexpr|) and also the underscore
character (as allowed in Python too) can serve to separate groups of digits
for better readability.  But the package currently provides no macros to let
the output be formatted with such separators.

Formatting of numeric output is apart from some minimal facilities such as
\csbxint{TeXfromSci}\NewWith{1.4g},
\csbxint{TeXFrac}, \csbxint{DecToString}, \csbxint{PRaw}, \csbxint{FracToSci} or \csbxint{PFloat} left
to user macros or third-party packages%
\begin{everbatim*}
\xinteval{123_456_789_012^5}
\end{everbatim*}

\item \csbxint{iieval}\marg{expression} does exact computations \emph{on (big)
    integers only.} It is (of course) slightly faster than \csbxint{eval} for
  equivalent operations.  The forward slash \oper{/} does the \emph{rounded}
  integer division to match behaviour of |\numexpr|. The \oper{//} operator
  does floored division as in \csbxint{eval}. The \oper{/:} is the associated
  modulo operator (we could easily let the catcode 12 |%|
  character be an alias, but using such an unusual percent character would be
  a bit cumbersome in a \TeX{} workflow, if only for matters of
  syntax highlighting in \TeX-aware text editors).
\begin{everbatim*}
\xintiieval{add((i/:7)?{omit}{i^5}, i=1000..1020)}% only add fifth powers of multiples of 7
\end{everbatim*}

\item \csbxint{floateval}\marg{expression} does floating point computations
  with a given precision \dtt{P}, as specified via a prior assignment
  |\xintDigits:=P\relax |. The \oper{/} will compute the correct rounding of
  the exact fraction. Again \oper{//} is floored division and \oper{/:} its
  associated modulo (see also \func{divmod}).
\begin{everbatim*}
\begingroup
\xintDigits:=64\relax
\xintfloateval{sqrt(3)}
\endgroup 
\end{everbatim*}

  The default is with \dtt{P=16} digits. The four basic
  operations and the square root realize \emph{correct
    rounding.}\footnote{when the inputs are already floating point numbers
    with at most |P|-digits mantissas.}

  It can be used with an optional argument |[Q]| which means to do a final
  float rounding to mantissas of |Q| digits (this makes
  sense only if |Q<P|). ATTENTION: the optional argument |[Q]| is to be
  located \emph{within} the braces at the start of the expression.

  When |Q| is negative it means to round to |P+Q| digits only.

  On output, \csbxint{floateval} uses \csbxint{PFloat} for each number. This
  can be modified (cf.\@ \csbxint{floatexprPrintOne}).
\end{itemize}

The user can define variables and functions. Definition of functions is either
per parser (\csbxint{deffunc}, \csbxint{deffloatfunc}, ...), but there are
some restrictions, or generic (\csbxint{NewFunction}) but the latter is only
syntactic sugar for function-like disguise of a \TeX{} macro having not done
any pre-parsing.

Two derived parsers:
\begin{itemize}
\item \csbxint{ieval}\marg{expression} does all computations like \csbxint{eval}
  but rounds the result to the nearest integer. If there is an optional
  argument |[D]|, the rounding is to:
  \begin{itemize}
  \item if |D>0|: the nearest fixed point number with |D| digits after the
    decimal mark,
  \item if |D=0|: the nearest integer,
  \item if |D<0|: the rounded quotient by |10^(-D)|.\CHANGED{1.4f}
  \end{itemize}
  ATTENTION: the optional argument
  |[D]| is to be located \emph{within} the braces at the start of the expression. 
\item \csbxint{theboolexpr}\meta{expression}|\relax| does all computations like \csbxint{eval}
  then converts all (non-empty) leaves%
%
\footnote{Currently, empty leaves are output using \csbxint{exprEmptyItem},
  i.e.\@ default to \dtt{\xintexprEmptyItem}. This may change.}
%
to |True| or |False|
  (cf.\@ \csbxint{boolexprPrintOne}). There is no |\xintbooleval|.
\end{itemize}

These macros are wrappers for a more core syntax:
\begin{itemize}[nosep]
  \item \csbxint{expr}\meta{expression}|\relax|,
  \item \csbxint{iiexpr}\meta{expression}|\relax|,
  \item \csbxint{floatexpr}\meta{expression}|\relax|,
  \item \csbxint{iexpr}\meta{expression}|\relax|,
  \item \csbxint{boolexpr}\meta{expression}|\relax|.
\end{itemize}
This core syntax can be used directly in typesetting flow.\NewWith{1.4} In an
|\edef| they expand to some braced nested data (all computations having been
done) prefixed with some |\protected| «typesetter» macros. When using
\csbxint{eval} (in contrast to \csbxint{expr}), the protection of the
«typesetter» is by-passed and its action gives (expandably)
explicit digits and other characters such as those of scientific notation or
brackets.%
%
\footnote{\csbxint{eval} and \csbxint{expr} both expand completely in exactly
  two steps. And \csbxint{expr} expands fully under \fexpan sion (of the
  |\romannumeral0| or |-`0| type). As per \csbxint{eval} attention that it may
  expand to nothing, then naturally \fexpan sion propagates to tokens
  following up in the input stream.}

It is possible to use the core syntax\NewWith{1.4}
\csbxint{expr}\meta{expression}|\relax| also in so-called moving arguments,
because when written out to a file the final expansion result uses only
standard catcodes and thus will get retokenized and the typesetter macro
(which being |\protected| is there intact in external file) will expand
as expected.

One needs \csbxint{eval} et al.\@ only if one really wants the final digits (and
other characters), for example in a context where \TeX{} expects a number or a
dimension.

As alternative to \csbxint{eval}\marg{expression}, an equivalent is
\csbxint{the}\csbxint{expr}\meta{expression}|\relax|. Similarly \csbxint{the}
can prefix all other core parsers. And one can also use \csbxint{theexpr} as
shortcut for \csbxint{the}\csbxint{expr}.

Throughout this documentation I will most of the time refer to \csbxint{eval}
and \csbxint{expr}. But beware that doing exact computations with fractions
leads very quickly to very big results (and furthermore one needs to use
explicitly the |reduce()| function to convert the fractions into smallest
terms). Thus most probably what you want is \csbxint{floateval} and
\csbxint{floatexpr}.

\subsection{Expansion}

As mentioned already, the parsers are compatible with expansion-only
context.

Also, they expand the expression piece by piece: the normal mode of operation
of the parsers is to unveil the parsed material token by token.  Unveiling is
a process combining space swallowing, brace removal (one level generally), and
\fexpan sion.

For example a closing parenthesis after some function arguments does not have
to be immediately visible, it and the arguments themselves may arise from
\fexpan sion (applied before grabbing each successive token).  Even the ending
|\relax| may arise from expansion. Even though the \csbxint{eval} user
interface means that the package has at some point the entire expression in
its hands, it immediately re-inserts it into token stream with an additional
postfixed |\relax| and from this point on has lost any ways (a simple-minded
delimited macro won't do because the expression is allowed to contain
sub-\csbxint{expr}essions, even nested) to manipulate formally again the whole
thing; it can only re-discover it one token at a time.

This general behaviour (which allows much more freedom in assembling
expressions than is usually the case with familiar programming languages such
as Python, although admittedly that freedom will prove useful only to
power-\TeX users and possibly does not have that many significant use cases)
has significative exceptions. These exceptions are mostly related to
«pseudo»-functions. A «pseudo»-function will grab some of its arguments via
delimited macros. For example |subs(expr1,x=expr2)| needs to see the comma,
equal sign and closing parenthesis. But it has mechanisms to allow |expr1| and
|expr2| to possess their own commas and parentheses.

Inner semi-colons on the other hand currently always can originate from expansion.
Defining functions or variables requires a visible semi-colon acting as
delimiter of the expression, but inner semi-colons do not need to be
hidden within braces or macros\NewWith{1.4}.

The expansion stops only when the ending |\relax| has been found
(it is then removed from the token stream).

For catcode related matters see \csbxint{exprSafeCatcodes}.

A word of warning on the bracketed optional argument of respectively
\csbxint{floatexpr} and \csbxint{iexpr}. When defining macros which will hand
over some argument to one of these two parsers, the argument may potentially
start with a left square bracket |[| (e.g. argument could be |[1, 2, 3]|) and
this will break the parser. The fix is to use in the macro definition
|\xintfloatexpr\empty|. This extra |\empty| token will prevent the parser
thinking there is an optional argument and it will then disappear during
expansion.

\begin{footnotesize}
  If comparing to other languages able to handle floating point numbers or big
  integers, such as Python, one should take into account that what the \xintname
  packages manipulate are streams of ascii bytes, one per digit. At no time
  (due to expandability) is it possible to store intermediate results in an
  arithmetic CPU register; each elementary operation via |\the\numexpr| will
  output digit tokens (hence as many bytes), not things such as handles to
  memory locations where some numbers are stored as memory words. The process
  can never put aside things but can only possibly permute them with upcoming
  tokens, to use them later, or, via combinations of |\expanded| and
  |\unexpanded| or some other more antiquated means grab some tokens and shift
  the expansion to some distant locations to later come back. The process is a
  never-ending one-dimensional one...\par
\end{footnotesize}

\subsection{\csh{xintthealign} and its customization}
\label{xintthealign}

With \csbxint{thealign} one can get nested data use a \TeX{} alignment in the
output. Attention, this must be followed by \csbxint{expr} et al., never by
\csbxint{theexpr} or \csbxint{eval}.
Here is an example :
\begin{everbatim*}
\xintthealign\xintexpr ndseq(1/(i+j), i = 1..10; j=1..10)\relax
\end{everbatim*}

It is possible to customize the behaviour of |\xintthealign|.\CHANGED{1.4a}
The helper macros, apart from |\xintexpralignbegin| and |\xintexpralignend|
will be subjected to a complete (|\expanded|) expansion (once).%
%
\footnote{\csa{xintexpralignend} is expanded once, after the body has been
  submitted to exhaustive expansion, and prior to the expansion of
  \csa{xintexpralignbegin}.}
%
The package
uses here |\protected| with no strong reason, as the replacement tokens are
not expanding anyhow, but the idea is that this allows to define a macro in an
|\edef| and later change the meaning of the auxiliary macros depending on what
one wants to do with the expansion result. See also further down the \LaTeX{}
example with a matrix environment, where |\noexpand| rather than |\protected|
is used.
\begin{everbatim}
\protected\def\xintexpralignbegin       {\halign\bgroup\tabskip2ex\hfil##&&##\hfil\cr}%
\def\xintexpralignend                   {\crcr\egroup}% removed \protected at 1.4c
\protected\def\xintexpralignlinesep     {,\cr}% separates "lines"
\protected\def\xintexpralignleftsep     {&}% at left of first item in a "line" (after brackets)
\protected\def\xintexpraligninnersep    {,&}% at the left of non-first items
\protected\def\xintexpralignrightsep    {&}% at right of last item in a "line" (before brackets)
\protected\def\xintexpralignleftbracket {[}%
\protected\def\xintexpralignrightbracket{]}%
\end{everbatim}
Although we will try to keep stable the way «regular arrays» are
rendered,\UNSTABLE{} the |\xintthealign| macro (and its associated customizability) is
considered work-in-progress and may experience breaking changes.

Use for example this for outputting to a file or a terminal:
\begin{everbatim}
% Better here without \protected.
% We assume here \newlinechar has the LaTeX setting.
\def\xintexpralignbegin       {}%
\def\xintexpralignend         {}%
\def\xintexpralignlinesep     {,^^J}% separates "lines"
\def\xintexpralignleftsep     { }% at left of first item in a "line" (after brackets)
\def\xintexpraligninnersep    {, }% at the left of non-first items
\def\xintexpralignrightsep    { }% at right of last item in a "line" (before brackets)
\def\xintexpralignleftbracket {[}%
\def\xintexpralignrightbracket{]}%
\end{everbatim}

\medskip

And here is an example using a |pmatrix| environment.  But it will not break
across pages, contrarily to the display produced by the default
\csbxint{thealign} configuration which uses \TeX{}'s |\halign|.

%\kern10\baselineskip
%\hbox{Big empty space here}
%\kern-11\baselineskip
\begin{everbatim*}
\[
\def\xintexpralignbegin       {\begin{pmatrix}}%
\def\xintexpralignend         {\end{pmatrix}}%
\def\xintexpralignlinesep     {\noexpand\\}% needed to counteract an internal \expanded
\def\xintexpraligninnersep    {&}%
\let\xintexpralignleftbracket\empty  \let\xintexpralignleftsep\empty
\let\xintexpralignrightbracket\empty \let\xintexpralignrightsep\empty
% by default amsmath matrices can have 10 columns at most
% (cf amsmath documentation for what to do to allow more)
l.c.m.=\xintthealign\xintiiexpr  ndmap(lcm, 1..12; 1..10)\relax
\]
\end{everbatim*}

\subsection{Customization of typesetting of individual items}
\label{xintexprEmptyItem}
\label{xintexprPrintOne}
\label{xintiexprPrintOne}
\label{xintiiexprPrintOne}
\label{xintfloatexprPrintOne}
\label{xintboolexprPrintOne}

The way individual items are formatted (whether or not using
\csa{xintthealign}) is also customizable. Here are the default package
definitions:
%\kern-2pt
% the \kern is to fix some extra white line from first line being a bit overfull
\begin{everbatim}
\def\xintexprEmptyItem{[]}
\let\xintexprPrintOne\xintFracToSci
\let\xintiexprPrintOne\xintDecToString
\def\xintiiexprPrintOne #1{#1}
\let\xintfloatexprPrintOne\xintPFloat
\def\xintPFloatE{e}
\def\xintboolexprPrintOne#1{\xintiiifNotZero{#1}{True}{False}}
\end{everbatim}
Attention! The above macros convert from \xintexprname internal numeric data
format to «printed» output; they are thus susceptible to require adjustments
if the internal data format changes, which may happen at each release. Of course
the default for |\xintexprPrintOne| etc...\@ will be adjusted accordingly, but
user custom definitions may break.

The interface for \csbxint{floatexprPrintOne} was changed.\CHANGED{1.4e}
It must now be the same as \csbxint{PFloat}, i.e. the target precision is |[P]| not a
braced argument. It will always be used with this |[P|] present so does not
have to consider it to be optional. It still must be expandable.

The \csbxint{PFloatE} is now allowed to a be macro with an argument delimited
by a dot, this argument will be the exponent.\NewWith{1.4e} The output must be produced
\fexpan dably and again be delimited by a dot. The default does not grab the
exponent and simply inserts the letter |e|.

Currently, this means that the macros used in place of \csbxint{FracToSci} and
\csbxint{PFloat} should understand the raw \xintfracname format |A/B[N]|, with
the |/B| and |[N]| parts being optional.%
%
\footnote{The constraints on any replacement to \csbxint{FracToSci} are much
  simplified at |1.4e|. Previously it had to be able to accept also input in
  fixed point notation, and in scientific notation with a catcode 12 |e|.}
%
The typesetter for
\csa{xintiiexpr} simply prints ``as is'', but this may change in future.

The used macros must be compatible with expansion-only context, but do not
have to be \fexpan dable.

Note: when not using \csbxint{thealign}, output of nested structures uses left
and right brackets, and commas and spaces in a non-customizable way, except
via \csa{xintexprEmptyItem}.  Use the \csa{xintthealign} interface for full
customizability.

\subsection{Built-in operators and their precedences}



The parser implements precedence rules based on concepts which are summarized
below (only for binary infix operators):
\begin{itemize}[noitemsep]
\item an infix operator has two associated precedence levels, say |L| for left
  and
  |R| for right,
\item the parser proceeds from left to right, pausing each time it has found a
  new number and an operator following it,
\item the parser compares the left-precedence |L| of the new found operator to
  the right-precedence |R_last| of the last delayed operation (which already
  has one argument and would like to know if it can use the new found one): if
  |L| is at most equal to it, the delayed operation is now executed, else the
  new-found operation is kept around to be executed first, once it will have
  gathered its arguments, of which only one is known at this stage.
\end{itemize}
This means for example in the case of the multiplication \oper{\lowast} and the
division operators \oper{/}, \oper{//}, \oper{/:} that they are parsed in a
left-associative way because they all share the same (left and right)
precedence level. This is the case with the analogous operators from the
Python language, as well.

At |1.4g| the power operators were changed to act in a right associative
way.\CHANGED{1.4g} Again, this matches the behaviour of e.g.\@ Python:%
\begin{everbatim*}
\xinteval{2^-3^4}
\end{everbatim*}


\makeatletter
\def\@floatboxreset{\@setminipage}% faudra contrôler celui-là
\makeatother
\begin{table}[htbp]
\edef\Ampersand{\string&}%
\edef\restorehtdpstrutbox
  {\ht\strutbox\the\ht\strutbox\dp\strutbox\the\dp\strutbox}
\ht\strutbox12pt\dp\strutbox5pt
\capstart
  \centering\begin{tabular}{|c|p{.5\textwidth}|}
    \hline
    \multicolumn{2}{|p{.6\textwidth}|}{\prec{$\infty$}:
    at this top level the syntax elements whose
    execution
    is done prior to operators preceding them:
    \begin{itemize}[nosep]
    \item
      \hyperref[ssec:builtinfunctions]{built-in} or
      \hyperref[ssec:userfunctions]{user-defined} functions, 
    \item \hyperref[ssec:uservariables]{variables}, 
    \item the \oper{\empty\lowast} unpacking operator,
    \item and intrinsic constituents of numbers: decimal mark \oper{\strut.},
      \oper{e} and \oper{E} of scientific notation, hexadecimal prefix
      \oper{"}.
    \end{itemize}\par\kern-\baselineskip\relax}%
  \\\hline\hline
   Precedence&``Operators'' at this level\strut\\
  \hline
  \prec{20}& postfix \oper{!} and branching \oper{?}, \oper{??} operators\strut\\\hline
%
  \prec{-}& minus sign as unary operator inherits the right-\hskip0pt precedence of
  the infix operator it follows, if that precedence is higher than the one of
  binary \oper{+} and \oper{-}, else it inherits the latter\strut\\\hline
%
  \prec{18} (17)& \oper{\string^} and \oper{\lowast\lowast} are synonymous;
  they act in a right-associative way (\textcolor[named]{Red}{\bfseries Changed at 1.4g!})\strut\\\hline
%
  \prec{16} (14)& \hyperref[ssec:tacit multiplication]{Tacit
    multiplication} has an elevated (left) precedence\strut\\\hline
%
  \prec{14}& \oper{\lowast}, \oper{/}, \oper{//} (floored division), 
               and \oper{/:} (associated modulo, alias
               \oper{'mod'})\strut\\\hline
%
  \prec{12}& \oper{+}, \oper{-}\strut\\\hline
%
  \prec{10}& \oper{<}, \oper{>}, \oper{==}, \oper{<=}, \oper{>=},
               \oper{!=} (they can be chained)\strut\\\hline
%
  \prec{8}& Boolean conjunction \oper{\Ampersand\Ampersand} and
              its alias \oper{'and'}\strut\\\hline
%
  \prec{6}& Boolean disjunction \oper{\string|\string|} and
              its alias \oper{'or'}. Also \oper{'xor'} and
              \oper{\strut..}, \oper{..[},
              \oper{{]..}}, and \oper{:} have
              this precedence\strut\\\hline
%
  \prec{4}& the brackets for slicers and extractors \oper{\empty[},
              \oper{\empty]}\strut\\\hline
%
  \prec{3}& the comma \oper{,}\strut\\\hline
%
  \prec{2}& the bracketers \oper{[}, \oper{]} construct nestable «arrays»\strut\\\hline
%
  \prec{1}& the parentheses \oper{(}, \oper{)}, and the semi-colon
              \oper{;} in \func{iter}, \func{rseq}, and further structures\strut\\\hline
%
\hline
%
  \multicolumn{2}{|p{.6\textwidth}|}{%
    \begin{itemize}[nosep]
    \item Binary operators have a left and a right precedence, which for
      most coincide. The right precedence is indicated within parentheses.
    \item 
\hyperref[ssec:tacit multiplication]{Tacit multiplication} has an elevated
left precedence level: |(1+2)/(3+4)5|
is computed as |(1+2)/((3+4)*5)| and |x/2y| is interpreted as |x/(2*y)|
when using variables.
    \end{itemize}
}\\\hline
  \end{tabular}
  \caption{Precedence levels}
  \label{tab:precedences}
\etoctoccontentsline {table}{\protect\emph{Table of precedence levels of operators}}
\restorehtdpstrutbox
\end{table}

The entries of \autoref{tab:precedences} are hyperlinked to the more detailed
discussion at each level. In these entries the number within parentheses
indicates the right-\hskip0pt precedence, if it differs from the left.

\begin{description}
%[parsep=0pt, listparindent=\leftmarginiii]
% [parsep=0pt,align=left,itemindent=0pt,
%   leftmargin=\leftmarginii, labelwidth=\leftmarginii, labelsep=0pt,
%   labelindent=0pt, listparindent=\leftmarginiii]
\edef\Ampersand{\string&}%

\precdesc{$\infty$} At this highest level of precedence, one finds:
\begin{description}
\item[{\hyperref[ssec:builtinfunctions]{functions} and
       \hyperref[ssec:uservariables]{variables}}]
  Functions (even the logic functions
  \func{!} and \func{?} whose names consist of a single non-letter character)
  must be used with parentheses. These parentheses may arise from expansion
  after the function name is parsed (there are exceptions which are documented
  at the relevant locations.)

  \operdesc{\empty\lowast} Python-like «unpacking» prefix operator. Sometimes one
  needs to use it as function |*()| (but I can't find an example right now)
  but most of the time parentheses are unneeded.

  \operdesc{\strut.} is decimal mark; the number scanner treats it as an inherent,
  optional and unique component of a being formed number.
  |\xintexpr 0.^2+2^.0\relax| is interpreted as |0^2+2^0| and
  thus produces \dtt{\xintexpr 0.^2+2^.0\relax}.

  Since release |1.2| an isolated decimal mark is illegal
  input in the \xintexprname parsers (it remains legal as argument to the
  macros of \xintfracname).

  \operdesc{e} scientific notation.

  \operdesc{E} scientific notation. For output, see \csbxint{PFloatE}.

  \operdesc{"} prefix for hexadecimal input. Only uppercase letters, and
  one optional |.| separating integer and fractional hexadecimal parts.
  This functionality
  \centeredline{\fbox{requires to load explicitly package \xintbinhexname.}}%

\begin{everbatim*}
\xintexpr "FEDCBA9876543210\relax\newline
\xintexpr ".FEDCBA9876543210\relax\newline
\xintexpr 16^5-("F75DE.0A8B9+"8A21.F5746+16^-5)\relax
\end{everbatim*}

  It is possible that in future the |"| prefix could be dropped in favour of
  |0x| prefix.  This would free |"| to be used for input of «string»-like
  entities.
\end{description}


\precdesc{20}
  The postfix operators |!| and the branching conditionals |?|, |??|.
\begin{description}
  \operdesc{!} computes the factorial of an integer.

  \operdesc{?} is used as |(stuff)?{yes}{no}|. It
    evaluates |stuff| and chooses the |yes| branch if the result is
    non-zero, else it executes |no|. After evaluation of |stuff| it acts as
    a macro with two mandatory arguments within braces, chooses the
    correct branch \emph{without evaluating the wrong one}. Once the braces
    are removed, the parser scans and expands the uncovered material.
    % so for
    % example
    % %
    % \leftedline{|\xinttheiexpr (3>2)?{5+6}{7-1}2^3\relax|}
    % %
    % is legal and computes
    % |5+62^3=|\dtt{\xinttheiexpr(3>2)?{5+(6}{7-(1}2^3)\relax}. It would be
    % better practice to include here the |2^3| inside the branches. The
    % contents of the branches may be arbitrary as long as once glued to what is
    % next the syntax is respected: {|\xintexpr (3>2)?{5+(6}{7-(1}2^3)\relax|
    %   also works.}

  \operdesc{??} is used as |(stuff)??{<0}{=0}{>0}|,
    where |stuff| is anything, its sign is evaluated and depending on the sign
    the correct branch is un-braced, the two others are discarded with no
    evaluation of their contents.
    % The un-braced branch will then be parsed as
    % usual.
    % %
    % \leftedline{|\def\x{0.33}\def\y{1/3}|}
    % %
    % \leftedline{|\xinttheexpr (\x-\y)??{sqrt}{0}{1/}(\y-\x)\relax|%
    %   \dtt{=\def\x{0.33}\def\y{1/3}%
    %   \xinttheexpr (\x-\y)??{sqrt}{0}{1/}(\y-\x)\relax }}
    % %
  \end{description}

  \precdesc{-} As unary operator, the minus sign inherits as precedence the
  minimum of |12| (which is the precedence for addition and subtraction) and
  of the (right-) precedence of the operators preceding it (if any).
\begin{everbatim*}
\xintexpr -3-4*-5^-7, (-3)-(4*(-(5^(-7))))\relax\newline
\xintexpr -3^-4*-5-7, (-((3^(-4))*(-5)))-7\relax\newline
|2^-10| gives \xintexpr 2^-10\relax\space
\end{everbatim*}and is thus perfectly legal, no need for parentheses.

  The |+| character as prefix unary operator is simply ignored during
  input parsing.

\precdesc{18}
\begin{description}
\operdesc{\string^}
\operdesc{\lowast\lowast} Both compute powers. They act in a right associative
way.\CHANGED{1.4g}
\begin{everbatim*}
\xintiiexpr 2^3^4\relax
\end{everbatim*}
\end{description}
% et:
% *2^-3^-4;
% (@_1)    0.991479137495678
% *2**-3**-4;
% (@_2)    0.991479137495678
% Python:
% >>> 2**-3**-4;
% 0.9914791374956781

\precdesc{16} see \hyperref[ssec:tacit multiplication]{Tacit multiplication}.

\precdesc{14}
\begin{description}
\operdesc{\lowast} multiplication

\operdesc{/} division:
  \begin{itemize}
  \item in \csbxint{eval}: exact division in the field of rational numbers (not
    automatically reduced to lowest terms),
  \item in \csbxint{floateval}: correct rounding of the exact division; the two
    operands are, if necessary, float-rounded before the fraction is
    evaluated and rounded (to obtain the correcty rounded |A/B|
    without prior rounding of |A| and |B| see \func{qfloat}),
  \item in \csbxint{iieval}: for compatibility with the legacy behaviour of
    |/| in |\numexpr|, it rounds the exact fraction \emph{with half-integers
      going towards the infinity of the same sign}.
   \end{itemize}
   The division is left-associative. Example:
\begin{everbatim*}
\xintexpr reduce(100/50/2)\relax
\end{everbatim*}

\operdesc{//} floored division (and thus produces an integer, see
   \func{divmod} for details)

\operdesc{/:} the associated modulo (see \func{divmod} and \func{mod})

   Left-associativity applies to the division operators:
\begin{everbatim*}
\xintexpr 100000/:13, 100000 'mod' 13\relax, \xintexpr 100000/:13/13\relax
\end{everbatim*}
 
   Nothing special needs to be done in contexts such as \LaTeX3
   |\ExplSyntaxOn| where |:| is of catcode letter, but if |:| is an active
   character (for example in \LaTeX\ with babel+french)
   one needs to use input such as |/\string :| (or replace it with usage of the function \func{mod}).

   \operdesc{'mod'} is same as \oper{/:}. 

   \fbox{Attention:}\IMPORTANTf{} with \ctanpackage{polexpr} loaded, which
   allows |'| in variable and function names, |'mod'| can not follow
   a variable name. Add parentheses around the variable, or use |/:|.

\end{description}

\precdesc{12}
\begin{description}
\operdesc{+} addition

\operdesc{-} subtraction. According to the general left-associativity rule in
case of equal precedence, it is
  left associative:
\begin{everbatim*}
\xintiiexpr 100-50-2\relax
\end{everbatim*}
\end{description}

\precdesc{10} Comparison operators are (as in Python) all at the same level of
precedence, use parentheses for disambiguation.
  \begin{description}
  \operdesc{<} |a<b| evaluates to \dtt{1} if the strict inequality holds to \dtt{0}
  if not.

  \operdesc{>} |a>b| evaluates to \dtt{1} if the strict inequality holds to \dtt{0}
  if not.

  \operdesc{==} |a==b| evaluates to \dtt{1} if equality holds to \dtt{0}
  if not.

  \operdesc{<=} |a<=b| evaluates to \dtt{1} if left hand side is at most equal
  to right hand side, to \dtt{0}
  if not.

  \operdesc{>=} |a>=b| evaluates to \dtt{1} if left hand side is at least equal
  to right hand side, to \dtt{0}
  if not.

  \operdesc{!=} |a!=b| evaluates to \dtt{1} if they differ, to \dtt{0}
  if not.
  \end{description}

  Comparisons\NewWith{1.4b} can be chained arbitrarily, e.g., |x < y <= z !=
  t| is equivalent to |x < y 'and' y <= z 'and' z != t| (and also to |all(x<y,
  y<=z, z!=t)|), except that if |y| and |z| involve computations, they
  are evaluated only once.  Currently there is no short-circuit here, i.e.\@
  even if some intermediate comparison turns out false (in fact |0|), all the
  remaining conditionals will still be evaluated.

\begin{everbatim*}
\xintifboolexpr{1<=2!=3<4>1}{true}{\error}, \xintifboolexpr{1<=2>=3<4>1}{\error}{false},
\xintifboolexpr{3 != 3! == 6 != 4! == 24}{true}{\error}
\end{everbatim*}

\precdesc{8}
\begin{description}
  \operdesc{\Ampersand\Ampersand} logical conjunction. Evaluates to \dtt{1} if
  both sides are non-zero, to \dtt{0} if not.

  \operdesc{'and'} same as \verb+&&+.  See
  also the \func{all} multi-arguments function. 

  \fbox{Attention:}\IMPORTANTf{} with \ctanpackage{polexpr} loaded, which
  allows |'| in variable and function names, |'and'| can not follow
  a variable name. Add parentheses around the variable, or use |&&|.
\end{description}

\precdesc{6}
\begin{description}
  \operdesc{\string|\string|} logical (inclusive) disjunction. Evaluates to
  \dtt{1} if one or both sides are non-zero, to \dtt{0} if not.

  \operdesc{'or'} same as as \verb+||+. See also the \func{any} multi-arguments
  function.  

  \fbox{Attention:}\IMPORTANTf{} with \ctanpackage{polexpr} loaded, which
  allows |'| in variable and function names, |'or'| can not follow
  a variable name. Add parentheses around the variable, or use \verb=||=.

  \operdesc{'xor'} logical (exclusive) disjunction. 

  \fbox{Attention:}\IMPORTANTf{} with \ctanpackage{polexpr} loaded, which
  allows |'| in variable and function names, |'xor'| can not follow a variable
  name. Add parentheses around the variable, or use the \func{xor} function syntax.

  \operdesc{\strut..}
  \operdesc{..[}
  \operdesc{{]..}} Syntax for arithmetic
  progressions. See \autoref{ssec:arithseq}.

  \operdesc{:} This is a separator involved in |[a:b]| Python-like slicing syntax.
\end{description}

\precdesc{4} 
\begin{description}
  \operdesc{\empty[}
  \operdesc{\empty]}
  Involved in Python-like slicing |[a:b]| and extracting |[N]| syntax. And its
  extension à la NumPy |[a:b,N,c:d,...,:]|. Ellipsis |...| is not yet implemented.
  The «step» parameter as in |[a:b:step]| is not yet implemented.
\end{description}

\precdesc{3} 
\begin{description}
  \operdesc{,}
  The comma separates expressions (or function arguments).%
%
  \footnote{The comma
    is really like a binary operator, which may be called ``join''. It has
    lowest precedence of all (apart the parentheses) because when it is
    encountered all postponed operations are executed in order to finalize its
    \emph{first} operand; only a new comma or a closing parenthesis or the end
    of the expression will finalize its \emph{second} operand.}
%
\begin{everbatim*}
\xintiiexpr 2^3,3^4,5^6\relax
\end{everbatim*}
\end{description}

\precdesc{2}
\begin{description}
  \operdesc{[}
  \operdesc{]} The bracketers construct nestable «array-like»
  structures. Arbitrary (heterogeneous) nesting is allowed. For output related
  matters see \csbxint{thealign} (its usage is optional, without it rendering
  is «one-dimensional»). Output shape of non-homogeneous arrays is to
  be considered unstable at this time.
\end{description}

\precdesc{1}
\begin{description}
  \operdesc{(}
  \operdesc{)}
  The parentheses serve as mandatory part of the syntax for functions, and to
  disambiguate precedences.%
%
    \footnote{It is not apt to describle the
    opening parenthesis as an operator, but the closing parenthesis is analogous
    to a postfix unary operator. It has lowest precedence which means
    that when it is encountered all postponed operations are executed to finalize
    its operand. The start of this operand was decided by the opening
    parenthesis.}
%
  They do not construct any nested structure.

  \operdesc{;} The semi-colon as involved as part of the syntax of \func{iter},
  \func{rseq}, \func{ndseq}, \func{ndmap} has the same
  precedence as a closing parenthesis.
\end{description}

\item[|\relax|] This is the expression terminator for \csbxint{expr} et al.
  It may arise from expansion during the parsing itself. As alternative to
  \csbxint{expr} (et al.) use
  \csbxint{eval} (et al.) which have the usual macro interface (with one mandatory
  argument).
\end{description}

The |;| also serves as syntax terminator for \csbxint{defvar} and
\csbxint{deffunc}.  It can in this rôle not arise from expansion as the
expression body up to it is fetched by a delimited macro.  But this is done in
a way which does not require any specific hiding for inner semi-colons as
involved in the syntax of \func{iter}, etc...

\subsection{Built-in functions}\label{ssec:builtinfunctions}


See \autoref{tab:functions} whose elements are hyperlinked to the
corresponding definitions.

  Functions are at the same top level of priority. All functions even
  \func{?} and \func{!} require parentheses around their arguments.

% Table of functions

\begin{table}[htbp]
  \capstart
  \centering
\xintAssignArray\xintCSVtoList{!, ?, \textasciigrave\lowast\textasciigrave, \textasciigrave+\textasciigrave,
abs, add, all, any, acos, acosd, Arg, Argd, asin, asind, atan, atand,
atan2, atan2d,
binomial, bool,
ceil, cos, cosd, cot, cotd, cotg, csc, cscd,
divmod, even, exp,
factorial, first, flat, float, float\string_dgt, floor, frac, gcd,
if, ifint, ifone, ifsgn, ilog10, iquo, irem, isint, isone, iter, iterr, inv,
last, lcm, len, log, log10, max, min, mod, mul,
ndmap, ndseq, ndfillraw,
not, num, nuple, odd,
pArg, pArgd, pfactorial, pow, pow10, preduce,
qfloat, qfrac, qint, qrand, qraw,
random, randrange, rbit, reduce, reversed, round, rrseq, rseq,
sec, secd, seq, sgn, sin, sinc, sind, sqr, sqrt, sqrtr,
subs, subsm, subsn,
tan, tand, tg, togl, trunc, unpack,
xor, zip}
\to\Functions
  \cnta\Functions{0} 
  \cntb\xinttheexpr ceil(\cnta/7)\relax\space
\newcommand\builtinfunction[1]{\expandafter\expandafter\expandafter\func
    \expandafter\expandafter\expandafter{\Functions{#1}}}%
\centeredline{\begin{tabular}{|*{7}{p{2cm}|}}
  \hline
  \xintFor* #1 in {\xintSeq{1}{\cntb}}\do
    {\builtinfunction{#1}&
     \builtinfunction{#1+\cntb}&%
     \builtinfunction{#1+2*\cntb}&%
     \builtinfunction{#1+3*\cntb}&%
     \builtinfunction{#1+4*\cntb}&%
     \builtinfunction{#1+5*\cntb}&%
     \ifnumgreater{#1+6*\cntb}{\cnta}
          {}
          {\builtinfunction{#1+6*\cntb}}%
     \\\hline}%
\end{tabular}}
\caption{Functions (click on names)}\label{tab:functions}
\etoctoccontentsline {table}{\protect\emph{Table of functions in expressions}}
\etocsetnexttocdepth{subsubsection}
\localtableofcontents
\end{table}


Miscellaneous notes:
\begin{itemize}[nosep]
    \item since release |1.3d| \func{gcd} and \func{lcm} are extended to apply
      to fractions too, and do NOT require the loading of \xintgcdname,

    \item The randomness related functions \func{random}, \func{qrand} and
      \func{randrange} require that the \TeX\ engine provides the
      \csa{uniformdeviate} or \csa{pdfuniformdeviate} primitive. This is
      currently the case for |pdftex|, |(u)ptex|, |luatex|, and also for
      |xetex| since \TeX Live 2019.\IMPORTANT

    \item \func{togl} is provided for the case |etoolbox| package is loaded,

    \item \func{bool}, \func{togl} use delimited macros to fetch their argument and the
      closing parenthesis must be explicit, it can not arise from
      on the spot expansion. The same holds for \func{qint}, \func{qfrac},
      \func{qfloat}, \func{qraw}, \func{random} and \func{qrand}.

    \item Also \hyperlink{ssec:dummies}{functions with dummy variables} use
      delimited macros for some tasks. See the relevant explanations there.

    \item Functions may be called with \emph{oples} as arguments as long as
       the total length is the number of arguments the function expects.
    \end{itemize}

\subsubsection{Functions with no argument}

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

  \funcdesc[]{random} returns a random float |x| verifying |0 <= x < 1|. It obeys
  the prevailing precision as set by \csbxint{Digits}: i.e.\@ with |P| being the
  precision the random float multiplied by |10^P| is an integer, uniformly
  distributed in the |0..10^P-1| range.

  This description implies that if |x| turns out to be |<0.1| then
  its (normalized) mantissa has |P-1| digits and a trailing zero, if |x<0.01|
  it has |P-2| digits and two trailing zeros, etc... This is what is observed
  also with Python's |random()|, of course with |10| replaced there by radix
  |2|.%
\begin{everbatim*}
  \pdfsetrandomseed 12345
  \xintDigits:=37\relax
  \xintthefloatexpr random()\relax\newline
  \xintthefloatexpr random()\relax\par
\end{everbatim*}

  \funcdesc[]{qrand} returns a random float |0 <= x < 1| using \dtt{16} digits of
  precision (i.e.\@ |10^{16}x| is an integer). This is provided when speed is a
  at premium as it is optimized for precision being precisely \dtt{16}.%
  
\begin{everbatim*}
  % still with 37 digits as prevailing float precision
  \xintthefloatexpr qrand(), random()\relax\newline
  \xintDigits:=16\relax
  \xintthefloatexpr qrand(), random()\relax\par
\end{everbatim*}

  One can use both |qrand()| and |random()| inside the |\xintexpr| parser too.
  But inside the integer only |\xintiiexpr| parser they will cause some
  low-level error as soon as they get involved in any kind of computation as
  they use an internal format not recognized by the integer-only parser.

  See further \func{randrange}, which generates random integers.

  Currently there is no |uniform()| function%
%
\footnote{Because I am not sure how to handle rounding issues: should the
  computation proceed exactly and a rounding be done only at very end?}
%
  but it can be created by user:
\begin{everbatim*}
\xintdeffloatfunc uniform(a, b):= a + (b-a)*random();
\romannumeral\xintreplicate{10}%
{%
    \xintthefloatexpr uniform(123.45678, 123.45679)\relax\newline
}%
\end{everbatim*}

  \funcdesc[]{rbit} returns a random |0| or |1|.\NewWith{1.4}
  
\end{description}

\subsubsection{Functions with one argument}

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

  \funcdesc{num} truncates to the nearest integer (truncation towards zero). It
  has the same sign as |x|, except of course with |-1<x<1| as then |num(x)| is
  zero.
\begin{everbatim*}
\xinttheexpr num(3.1415^20), num(1e20)\relax
\end{everbatim*}
  The output is an explicit integer with as many zeros are as necessary. Even
  in float expressions, there will be an intermediate stage where all needed digits
  are there, but then the integer is immediately reparsed as a float to the target
  precision, either because some operation applies to it, or from the output
  routine of \csbxint{floatexpr} if it stood there alone. Hence,
  inserting something like |num(1e10000)| is costly as it really creates ten
  thousand zeros, even though later the whole thing becomes a float again. On
  the other hand naturally |1e10000| without |num()| would be simply parsed as
  a floating point number and would cause no specific overhead.

  \funcdesc{frac} fractional part.
  For all numbers |x=num(x)+frac(x)|, and |frac(x)| has the same sign as |x|
  except when |x| is an integer, as then |frac(x)| vanishes.
\begin{everbatim*}
\xintthefloatexpr frac(-355/113), frac(-1129.218921791279)\relax
\end{everbatim*}

  \funcdesc{reduce} reduces a fraction to smallest terms
\begin{everbatim*}
\xinttheexpr reduce(50!/20!/20!/10!)\relax
\end{everbatim*}

Recall that this is NOT done automatically, for example when adding fractions.
  \funcdesc{preduce} internally, fractions may have some power of ten part
  (for example when they got input in scientific notation). This function
  ignores the decimal part when doing the reduction. See \csbxint{PIrr}.
\begin{everbatim*}
\xinttheexpr preduce(10e7/2), reduce(10e7/2)\relax
\end{everbatim*}

  \funcdesc{abs} absolute value
  \funcdesc{sgn} sign. See also \csbxint{ifsgnexpr}.
  \funcdesc{inv} inverse.
  \funcdesc{floor} floor function.
  \funcdesc{ceil}  ceil function.
  \funcdesc{sqr}   square.
\item[ilog10(x)]\hypertarget{func:ilog10-ii}
   in |\xintiiexpr| the integer exponent $a$ such that $10^a\leq
  \mathrm{abs}(x)< 10^{a+1}$; returns (this may evolve in future)
  \dtt{\xintiieval{ilog10(0)}} if $x$ vanishes (i.e.\@ \dtt{0x7fff8000}).
\begin{everbatim*}
\xintiieval{ilog10(1), ilog10(-1234567), ilog10(-123456789123456789), ilog10(2**31)}\par
\end{everbatim*}
  See \func{ilog10} for the behaviour in \csbxint{expr}-essions.
  \item[sqrt(x)]\hypertarget{func:sqrt-ii}
    in |\xintiiexpr|, truncated square root; in |\xintexpr| or
    |\xintfloatexpr| this is the floating point square root, and there is an
    optional second argument for the precision. See \func{sqrt}.
  \funcdesc{sqrtr} available \emph{only} in |\xintiiexpr|, rounded square root.
  \item[factorial(x)]\hypertarget{func:factorial-ii} factorial function (like the
    post-fix \oper{!} operator.) When used in |\xintexpr| or
    |\xintfloatexpr| there is an optional second argument. See \func{factorial}.
  \funcdesc{?} is the truth value, $1$ if non zero, $0$ if zero. Must use parentheses.
  \funcdesc{!} is logical not, $0$ if non zero, $1$ if zero. Must use parentheses.
  \funcdesc{not} logical not.
  \funcdesc{even} is the evenness of the truncation |num(x)|.
\begin{everbatim*}
\xintthefloatexpr [3] seq((x,even(x)), x=-5/2..[1/3]..+5/2)\relax
\end{everbatim*}

  \funcdesc{odd} is the oddness of the truncation |num(x)|.
\begin{everbatim*}
\xintthefloatexpr [3] seq((x,odd(x)), x=-5/2..[1/3]..+5/2)\relax
\end{everbatim*}

  \funcdesc{isint} evaluates to 1 if |x| is an integer, to 0 if
  not. See \func{ifint}.
\begin{everbatim*}
$\xinttheexpr -5/3..[1/3]..+5/3\relax
\rightarrow \xinttheexpr seq(isint(x), x=-5/3..[1/3]..+5/3)\relax$
\end{everbatim*}

  \funcdesc{isone} evaluates to 1 if |x| is 1, to 0 if not.
See \func{ifone}.
\begin{everbatim*}
$\xintthefloatexpr subs(((x-1)/x, x/x, (x+1)/x), x=2**30)\relax
\rightarrow
\xintthefloatexpr seq(isone(y), y=subs(((x-1)/x, x/x, (x+1)/x), x=2**30))\relax$
\end{everbatim*}

  \funcdesc{qint} belongs with \func{qfrac}, \func{qfloat}, \func{qraw} to a
  special category:
  \begin{enumerate}[nolistsep]
  \item They require the closing parenthesis of their argument to be
    immediately visible, it can not arise from expansion.
  \item They grab the argument and store it directly; the format must be
    compatible with what is expected at macro level.
  \item And in particular the argument can not be a variable, it has to be
    numerical.
  \end{enumerate}

  \func{qint} achieves the same result as |num|, but the argument is grabbed
  as a whole without expansion and handed over to the
  \csbxint{iNum} macro. The |q| stands for ``quick'', and |qint| is thought
  out for use in \csbxint{iiexpr}|...\relax| with integers having dozens of
  digits.

    Testing showed that using |qint()| starts getting advantageous for inputs
    having more (or \fexpan ding to more) than circa \dtt{20} explicit digits.
    But for hundreds of digits the input gain becomes a negligible
    proportion of (for example) the cost of a multiplication.

  Leading signs and then
  zeroes will be handled appropriately but spaces will not be systematically
  stripped. They should cause no harm and will be removed as soon as the
  number is used with one of the basic operators. This input mode \emph{does
    not accept decimal part or scientific part}.
\begin{everbatim}
\def\x{....many many many ... digits}\def\y{....also many many many digits...}
\xinttheiiexpr qint(\x)*qint(\y)+qint(\y)^2\relax\par
\end{everbatim}

  \funcdesc{qfrac} does the same as \dtt{qint} except that it accepts
    fractions, decimal numbers, scientific numbers as they are understood by
    the macros of package \xintfracname. Thus, it is for use in
    \csbxint{expr}|...\relax|. It is not usable within an
    |\xintiiexpr|-ession, except if hidden inside functions such as
    \dtt{round} or \dtt{trunc} which then produce integers acceptable to the
    integer-only parser. It has nothing to do with |frac| (sigh...).

  \funcdesc{qfloat} does the same as \dtt{qfrac} and then converts to a float
    with the precision given by the setting of |\xintDigits|. This can be used
    in \csbxint{expr} to round a fraction as a float with the same result as
    with the |float()| function (whereas using |\xintfloatexpr A/B\relax|
    inside \csbxint{expr}|...\relax| would first round |A| and |B| to the
    target precision); or it can be used inside
    \csbxint{floatexpr}|...\relax| as a faster alternative to wrapping
    the fraction in a sub-\csbxint{expr}-ession.
    For example, the next two computations done with \dtt{16} digits
    of precision do not give the same result:
\begin{everbatim*}
\xintthefloatexpr qfloat(12345678123456785001/12345678123456784999)-0.5\relax\newline
\xintthefloatexpr 12345678123456785001/12345678123456784999-0.5\relax\newline
\xintthefloatexpr 1234567812345679/1234567812345678-0.5\relax\newline
\xintthefloatexpr \xintexpr12345678123456785001/12345678123456784999\relax-0.5\newline
\end{everbatim*}%
    because the second is equivalent to the third, whereas the
    first one is equivalent to the fourth one. Equivalently one can use
    |qfrac| to the same effect (the subtraction provoking the rounding of its
    two arguments before further processing.)

    Note that if the input needs no special rounding, the internal form of the
    output keeps a short mantissa (it does not add padding zeros to make it of
    length equal to the float precision). For example |qfloat(2[20])| would
    keep internally the input format.

    \funcdesc{float\string_dgt} is like \func{float} and avoids \func{float}'s
    check whether it used with its second optional argument. This is useful
    in the context of converting function definitions done via
    \csbxint{deffunc} (see explanations there) to functions usable in
    \csbxint{floateval}.\NewWith{1.4}

    Breaking change at |1.4e|: formerly the name was |float_()|.\CHANGED{1.4e}

    \funcdesc{nuple} is currently same as |[...]|.\DNU{(1.4)} Reserved for
    possible alternative meaning in future.
\begin{everbatim*}
\xinteval{nuple(1,2,3)}
\end{everbatim*}

   \funcdesc{unpack} is alternative for |*| unpacking operator.\NewWith{1.4}
\begin{everbatim*}
\xinteval{unpack([1,2,3])}
\end{everbatim*}

   \funcdesc[ople]{flat} removes\NewWith{1.4b} all nesting to produce a
   (non-bracketed) ople having the same leaves (some possibly empty) but
   located at depth 1.
\begin{everbatim*}
\xinteval{flat([[[[1,[],3],[4,[[[5,6,[]],[8,9],[[],11]],12],[13,14]]], [[],16]]], [])}
\end{everbatim*}

   I almost\unstable{} delayed indefinitely release because I was hesitating
   on the name: perhaps better with |flattened()|, but long names add
   (negligible, but still) overhead compared to short names. For this reason,
   consider that name may change.


\end{description}

\subsubsection{Functions with an alphanumeric argument}

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

\funcdesc[name]{bool} 
    returns
    $1$ if the \TeX{} conditional |\ifname| would act as |\iftrue| and
    $0$ otherwise. This works with conditionals defined by |\newif| (in
    \TeX{} or \LaTeX{}) or with primitive conditionals such as
    |\ifmmode|. For example:
    %
    \leftedline{|\xintifboolexpr{25*4-if(bool(mmode),100,75)}{YES}{NO}|}
    %
    will return $\xintifboolexpr{25*4-if(bool(mmode),100,75)}{YES}{NO}$
    if executed in math mode (the computation is then $100-100=0$) and
    \xintifboolexpr{25*4-if(bool(mmode),100,75)}{YES}{NO} if not (the
    \func{if} conditional is described below; the
    \csbxint{ifboolexpr} test automatically encapsulates its first
    argument in an |\xintexpr| and follows the first branch if the
    result is non-zero (see \autoref{xintifboolexpr})).

    The alternative syntax |25*4-\ifmmode100\else75\fi| could have been
    used here, the usefulness of |bool(name)| lies in the availability
    in the |\xintexpr| syntax of the logic operators of conjunction
    |&&|, inclusive disjunction \verb+||+, negation |!| (or |not|), of
    the multi-operands functions |all|, |any|, |xor|, of the two
    branching operators |if| and |ifsgn| (see also |?| and |??|), which
    allow arbitrarily complicated combinations of various |bool(name)|.
\funcdesc[name]{togl} 
    returns $1$
    if the \LaTeX{} package \ctanpackage{etoolbox}%
    %
    %
%
\footnote{\url{https://ctan.org/pkg/etoolbox}}
    %
    has been used to define a toggle named |name|, and this toggle is
    currently set to |true|. Using |togl| in an |\xintexpr..\relax|
    without having loaded
    \ctanpackage{etoolbox} will result in an
    error from |\iftoggle| being a non-defined macro. If |etoolbox| is
    loaded but |togl| is used on a name not recognized by |etoolbox|
    the error message will be of the type ``ERROR: Missing |\endcsname|
    inserted.'', with further information saying that |\protect| should
    have not been encountered (this |\protect| comes from the expansion
    of the non-expandable |etoolbox| error message).

    When |bool| or |togl| is encountered by the |\xintexpr| parser, the
    argument enclosed in a parenthesis pair is expanded as usual from
    left to right, token by token, until the closing parenthesis is
    found, but everything is taken literally, no computations are
    performed. For example |togl(2+3)| will test the value of a toggle
    declared to |etoolbox| with name |2+3|, and not |5|. Spaces are
    gobbled in this process. It is impossible to use |togl| on such
    names containing spaces, but |\iftoggle{name with spaces}{1}{0}|
    will work, naturally, as its expansion will pre-empt the
    |\xintexpr| scanner.

    There isn't in |\xintexpr...| a |test| function available analogous
    to the |test{\ifsometest}| construct from the |etoolbox| package;
    but any \emph{expandable} |\ifsometest| can be inserted directly in
    an |\xintexpr|-ession as |\ifsometest10| (or |\ifsometest{1}{0}|),
    for example |if(\ifsometest{1}{0},YES,NO)| (see the |if| operator
    below) works.

    A straight |\ifsometest{YES}{NO}| would do the same more
    efficiently, the point of |\ifsometest10| is to allow arbitrary
    boolean combinations using the (described later) \verb+&&+ and
    \verb+||+ logic operators:
    \verb+\ifsometest10 && \ifsomeothertest10 || \ifsomethirdtest10+,
    etc... |YES| or |NO| above stand for material compatible with the
    |\xintexpr| parser syntax.

    See  also \csbxint{ifboolexpr}, in this context.
\end{description}

\subsubsection{Functions with one mandatory and a second but optional argument}

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

  \funcdesc[{x[, n]}]{round} Rounds its first argument to an integer multiple
  of |10^(-n)| (i.e. it \emph{quantizes}). The case of negative |n| is new with
  |1.4a|. Positive |n| corresponds to conversion to a fixed point number with
  |n| digits after decimal mark.
\begin{everbatim*}
\xinteval{round(-2^30/3^5,12), round(-2^30/3^5,-3)}
\end{everbatim*}
  \funcdesc[{x[, n]}]{trunc} Truncates its first argument to an integer
  multiple of |10^(-n)|. The case of negative |n| is new with
  |1.4a|.
\begin{everbatim*}
\xinteval{trunc(-2^30/3^5,12), trunc(-2^30/3^5,-3)}
\end{everbatim*}

  \funcdesc[{x[, n]}]{float} Rounds its first argument to a floating point number, with a
  precision given by the second argument, which must be positive.
\begin{everbatim*}
\xinteval{float(-2^30/3^5,12), float(-2^30/3^5, 1)}
\end{everbatim*}

  % AUCTeX EXTREMEMENT PENIBLE AVEC L'INDENTATION FORCEE SOUS M-q

    For this example and earlier ones if the parser had been
    \csbxint{floateval}, not \csbxint{eval}, the first argument (here
    |2^30/3^5|) would already have been computed as floating point number with
    numerator and denominator rounded separately first to the prevailing
    precision. To avoid that, use |\xintexpr...\relax| wrapper.
    Then the rounding or truncation will be applied to an exact fraction.

    \funcdesc[{x[, n]}]{sfloat} It is the same as \func{float},
    but in case of a short (non-fractional) input it gets stored internally
    without adding zeros to make the mantissa have the \csbxint{theDigits}
    length. One may wonder then what is the utility of \func{sfloat}? See for
    an example of use the documentation of \csbxint{deffunc}. Notice however
    that this is a bit experimental and may evolve in future when \xintname
    gets a proper internal data structure for floating point numbers. The
    non-normalized format is useful for multiplication or division, but float
    additions and subtractions usually convert their arguments to a normalized
    mantissa.

    \funcdesc[{x[, n]}]{ilog10} If there is an optional argument |n|, returns the (relative) integer $a$ such that $10^a\leq
    \mathrm{abs}(float(x, n)) < 10^{a+1}$. In absence of the
    optional argument:
    \begin{itemize}[nosep]
    \item in \csbxint{expr}, it returns the exponent $a$ such that $10^a\leq
    \mathrm{abs}(x) < 10^{a+1}$.
     \item in \csbxint{floatexpr}, the input is first rounded to
       \csbxint{theDigits} float precision, then the exponent $a$ is evaluated. 
    \end{itemize}
\begin{everbatim*}
\xintfloateval{ilog10(99999999/10000000, 8), ilog10(-999999995/100000000, 8),
                                             ilog10(-999999995/100000000, 9)}\newline
\xinteval{ilog10(-999999995/100000000), ilog10(-999999995/100000000, 8)}
\end{everbatim*}

    If the input vanishes the function outputs
    \dtt{\xinteval{ilog10(0)}} (i.e.\@ |-0x7fff8000| which is near the
    minimal TeX number |-0x7fffffff|). This is also subject to change.

    The \hyperlink{func:ilog10-ii}{integer-only} variant for \csbxint{iiexpr}
    admits no optional argument.

  \funcdesc[{x[, n]}]{sqrt} in \csa{xintexpr}|...\relax| and \csa{xintfloatexpr}|...\relax|
    it achieves the precision given by the optional second argument. For
    legacy reasons the |sqrt| function in \csa{xintiiexpr} \emph{truncates}
    (to an integer), whereas |sqrt| in \csa{xintfloatexpr}|...\relax| (and in
    \csa{xintexpr}|...\relax| which borrows it) \emph{rounds} (in the sense of
    floating numbers). There is |sqrtr| in \csa{xintiiexpr} for
    \emph{rounding} to nearest integer.
\begin{everbatim*}
\xinttheexpr sqrt(2,31)\relax\ and \xinttheiiexpr sqrt(num(2e60))\relax
\end{everbatim*}

    There is an \hyperlink{func:sqrt-ii}{integer only} variant for
    \csbxint{iiexpr}.

  \funcdesc[{x[, n]}]{factorial} when the second optional argument is made
    use of inside \csa{xintexpr}|...\relax|, this switches to the use of the
    float version, rather than the exact one.
\begin{everbatim*}
\xinttheexpr factorial (100,32)\relax, {\xintDigits:=32\relax \xintthefloatexpr
                                           factorial (100)\relax}\newline
\xinttheexpr factorial (50)\relax\newline
\xinttheexpr factorial (50, 32)\relax
\end{everbatim*}

    The \hyperlink{func:factorial-ii}{integer only variant} of course has no
    optional second argument.

  \funcdesc[{A[, B]}]{randrange} when used with a single argument |A| returns a random
    integer |0 <= x < A|, and when used with two arguments |A| and |B| returns
    a random integer |A <= x < B|. As in Python it is an «empty range» error
    in first case if |A| is zero or negative and in second case if |B <= A|.
    
    Attention that the arguments are first converted to integers using
    \csbxint{Num} (i.e.\@ truncated towards zero).

    The function can be used in all three parsers. Of course the size is not
    limited (but in the float parser, the integer will be rounded if involved
    in any operation).
\begin{everbatim*}
  \pdfsetrandomseed 12345
  \xinttheiiexpr randrange(10**20)\relax\newline
  \xinttheiiexpr randrange(1234*10**16, 1235*10**16)\relax\newline
  \printnumber{\xinttheiiexpr randrange(10**199,10**200)\relax}\par
\end{everbatim*}
    For the support macros see \csbxint{RandomDigits}, \csbxint{iiRandRange},
    \csbxint{iiRandRangeAtoB}. For some details regarding how \xintname
    uses the engine provided generator of pseudo-random numbers, see
    \csbxint{UniformDeviate}.

  \end{description}

\subsubsection{Functions with two arguments}

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

  \funcdesc[m, n]{iquo} Only available in |\xintiiexpr/\xintiieval|
  context.  Computes the Euclidean quotient.  Matches with the remainder
  defined in next item. See \csbxint{iiQuo}.

  \funcdesc[m, n]{irem} Only available in |\xintiiexpr/\xintiieval|
  context.  Computes the Euclidean remainder.  Attention that, following
  mathematical definition, it is always non-negative.  See \csbxint{iiRem}.

  \funcdesc[f, g]{mod} computes |f - g*floor(f/g)|. Hence its output is a
  general fraction or floating point number or integer depending on the
  used parser. If non-zero, it has the same sign as |g|.

  Prior to |1.2p| it computed |f - g*trunc(f/g)|.

  The \oper{/:} and \oper{'mod'} infix operators are both mapped to the same underlying
  macro as this |mod(f, g)| function. At |1.3| this macro produces smaller
  denominators when handling fractions than formerly.
\begin{everbatim*}
\xinttheexpr mod(11/7,1/13), reduce(((11/7)//(1/13))*1/13+mod(11/7,1/13)),
mod(11/7,1/13)- (11/7)/:(1/13), (11/7)//(1/13)\relax\newline
\xintthefloatexpr mod(11/7,1/13)\relax\par
\end{everbatim*}

  Attention: the precedence rules mean that |29/5 /: 3/5| is handled like
  |((29/5)/:3)/5|. This is coherent with behaviour of Python language for
  example:
\begin{everbatim}
>>> 29/5 % 3/5, 11/3 % 17/19, 11/57
(0.5599999999999999, 0.19298245614035087, 0.19298245614035087)
>>> (29/5) % (3/5), (11/3) % (17/19), 5/57
(0.4, 0.08771929824561386, 0.08771929824561403)
\end{everbatim}
  For comparison (observe on the last lines how |\xintfloatexpr| is more accurate than
  Python!):
\begin{everbatim*}
\noindent\xinttheexpr 29/5 /: 3/5, 11/3 /: 17/19\relax\newline
         \xinttheexpr (29/5) /: (3/5), (11/3) /: (17/19)\relax\newline
         \xintthefloatexpr 29/5 /: 3/5, 11/3 /: 17/19, 11/57\relax\newline
         \xintthefloatexpr (29/5) /: (3/5), (11/3) /: (17/19), 5/57\relax\newline
         5/57 = \xinttheexpr trunc(5/57, 20)\relax\dots\newline
\end{everbatim*}%
  Regarding some details of behaviour in |\xintfloatexpr|, see discussion of
  |divmod| function next.

  \funcdesc[f, g]{divmod} computes the two mathematical values |floor(f/g)| and
  |mod(f,g)=f - g*floor(f/g)| and produces them as a bracketed
  pair\CHANGEDf{1.4}
  in
  other terms it is analogous to the Python |divmod| function. Its output is
  equivalent to using |f//g, f/:g| but its implementation avoids doing twice
  the needed division.

  In |\xintfloatexpr...\relax| the modulo is rounded to the prevailing
  precision. The quotient is like in the other parsers an exact integer. It
  will be rounded as soon as it is used in further operations, or via the global
  output routine of |\xintfloatexpr|. \emph{Those examples behave as in |1.3f|
  because assignments to multiple variables tacitly unpack if this is
  necessary.}
\begin{everbatim*}
\xintdefvar Q, R := divmod(3.7, 1.2);%
\xinttheexpr Q, R, 1.2Q + R\relax\newline
\xintdefiivar Q, R := divmod(100, 17);%
\xinttheiiexpr Q, R, 17Q + R\relax\newline
\xintdeffloatvar Q, R := divmod(100, 17e-20);%
\xintthefloatexpr Q, R, 17e-20 * Q + R\relax\newline
% show Q exactly, although defined as float it can be used in iiexpr:
\xinttheiiexpr Q\relax\ (we see it has more than 16 digits)\par
\xintunassignvar{Q}\xintunassignvar{R}%
\end{everbatim*}
  Again: |f//g| or the first item output by |divmod(f, g)| is an integer |q|
  which when computed inside |\xintfloatexpr..\relax| is not yet rounded to
  the prevailing float precision; the second item |f-q*g| is the rounding to
  float precision of the exact mathematical value evaluated with this exact
  |q|. \emph{This behaviour may change in future major release;\IMPORTANT{}
    perhaps |q| will be rounded and |f-q*g| will correspond to usage of this
    rounded |q|.}

  As |\xintfloatexpr| rounds its global result, or rounds operands at
  each arithmetic operation, it requires special circumstances to show that
  the |q| is produced unrounded. Either as in the above example or this one
  with comparison operators:
\begin{everbatim*}
\xintDigits := 4\relax
\xintthefloatexpr if(12345678//23==537000, 1, 0), 12345678//23\relax\newline
\xintthefloatexpr if(float(12345678//23)==537000, 1, 0)\relax\par
\xintDigits := 16\relax
\end{everbatim*}
  In the first line, the comparison is done with
  |floor(12350000/23)|\dtt{=\xinttheiiexpr12350000/23\relax} (notice in
  passing that |12345678//23| was evaluated as |12350000//23| because the
  operands are first rounded to \dtt{4} digits of floating point precision), hence the conditional
  takes the "False" branch. In the second line the |float| forces rounding of
  the output to \dtt{4} digits, and the conditional takes the "True" branch.

% pour mémoire, Python :
% >>> divmod(100,17e-20)
% (5.88235294117647e+20, 1.4756182441723705e-19)
% mais faudra voir avec le module Decimal

  This example shows also that comparison operators in
  |\xintfloatexpr..\relax| act on unrounded operands.

  \funcdesc[x, y]{binomial} computes binomial coefficients.
  It returns zero if |y<0| or |x<y| and raises an error if |x<0| (or if
  |x>99999999|.)
\begin{everbatim*}
\xinttheexpr seq(binomial(20, i), i=0..20)\relax
\end{everbatim*}
\begin{everbatim*}
\printnumber{\xintthefloatexpr seq(binomial(100, 50+i), i=-5..+5)\relax}%
\end{everbatim*}

The arguments must be (expand to) short integers.
  \funcdesc[a, b]{pfactorial} computes partial factorials i.e.\@
    |pfactorial(a,b)| evaluates the product |(a+1)...b|.
\begin{everbatim*}
\xinttheexpr seq(pfactorial(20, i), i=20..30)\relax
\end{everbatim*}

The arguments must (expand to) short integers. See \autoref{xintiiPFactorial}
for the behaviour if the arguments are negative.

\funcdesc[\TeX-macro, n-uple]{ndfillraw} The second argument is |[N1, N2, ...,
Nk]|.\NewWith{1.4} The construct fills an |N1xN2x...xNk| hyperrectangular
nested list by evaluating the given |macro| as many times as needed.  The
expansion result goes directly into internal data and must thus comply with
what is expected internally for an individual numeric leaf (at |1.4|,
\xintfracname raw format worked for |\xintexpr| or |\xintfloatexpr|, but not
|\xintiiexpr|, and this may have changed since).\DNU{} This is an experimental
function serving to generate either constant or random arrays.  Attention that
%
% BORDEL \textbf{\sffamily...} NON, {\bfseries\sffamily ...} NON
% LaTeX Font Warning: Font shape `T1/cmss/b/n' undefined
% (Font)              using `T1/cmss/m/n' instead on input line 1826.
%
% ceci est ok {\fontseries{bx}\sffamily\TeX-macro}
% https://github.com/latex3/latex2e/issues/277
{\fontseries{bx}\sffamily \TeX-macro}
stands here for any expandable \TeX{}
\emph{macro}, and an |\xintexpr|-ession at this location thus requires an
explicit |\xinteval| wrapping.
\end{description}

\subsubsection{Functions with 3 or 4 arguments}

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

  \funcdesc[cond,yes,no]{if} (twofold-way conditional)\mbox{}

    checks if |cond| is true or false and takes the corresponding
    branch. Any non zero number or fraction is logical true. The zero
    value is logical false. Both ``branches'' are evaluated (they are
    not really branches but just numbers). See also the \oper{?} operator.

  \funcdesc[x,yes,no]{ifint} (twofold-way conditional)\mbox{}

    checks if |x| is an integer and in that case chooses the ``yes'' branch.%
    
    See also \func{isint}.

  \funcdesc[x,yes,no]{ifone} (twofold-way conditional)\mbox{}

    checks if |x| is equal to one and in that case chooses the ``yes'' branch.%
    
    Slightly more efficient than |if(x==1,..,..)|.  See also \func{isone}.

  \funcdesc[cond,<0,=0,>0]{ifsgn} (threefold-way conditional)\mbox{}

    checks the sign of |cond| and
    proceeds correspondingly. All three are evaluated. See also the \oper{??}
    operator.

  \end{description}

\subsubsection{Functions with an arbitrary number of arguments}

The functions \func{all}, \func{any}, \func{xor},
\func{\textasciigrave+\textasciigrave},
\func{\textasciigrave\lowast\textasciigrave},
\func{max}, \func{min}, \func{gcd}, \func{lcm}, \func{first}, \func{last},
\func{reversed} and \func{len} work both with «open» and «packed» lists (aka |nutples|).

Since |1.4|, when used with a single argument which is a |nutple|, it is automatically unpacked.
But from |1.4| to |1.4h| these functions could not be used with a single numeric argument: either they had at least two arguments, or only one and it had to be a |nutple|.  At |1.4i| it is again possible to use them with a lone numeric argument.\NewWith{1.4i}

  In the specific case of \func{reversed} with a |nutple| argument the output is then repacked so that
  the output is a |nutple| if and only if the input was one (the reversal does
  not propagate to deeper nested |nutple|'s, it applies only at depth one).

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]

  \funcdesc[stuff]{qraw} It\DNU{} injects directly tokens to represent
  internally numerical data.  Will break at any release modifying the internal
  data format specifications (which are not always documented).

\funcdesc[x, y, ...]{all} inserts a logical |AND| in-between its arguments and evaluates the
resulting logical assertion (as with all functions, all arguments are
evaluated).
\begin{everbatim*}
\xinteval{all(1,1,1), all([1,0,1]), all([1,1,1])}
\end{everbatim*}

\funcdesc[x, y, ...]{any} inserts a logical |OR| in-between its arguments and evaluates the
resulting logical assertion,
\begin{everbatim*}
\xinteval{any(0,0,0), any([1,0,1]), any([0,0,0])}
\end{everbatim*}

\funcdesc[x, y, ...]{xor} inserts a logical |XOR| in-between its arguments and evaluates
the resulting logical assertion,
\begin{everbatim*}
\xinteval{xor(1,1,1), xor([1,0,1]), xor([1,1,1])}
\end{everbatim*}

\funcdesc[x, y, ...]{\textasciigrave+\textasciigrave} adds (left ticks mandatory):
\begin{everbatim*}
\xinttheexpr `+`(1,3,19), `+`(1**2,3**2,sqr(19)), `+`([1**2,3**2,sqr(19)])\relax
\end{everbatim*}

\funcdesc[x, y, ...]{\textasciigrave\lowast\textasciigrave} multiplies (left ticks mandatory):
\begin{everbatim*}
\xinttheexpr `*`(1,3,19), `*`(1^2,3^2,19^2), `*`([1^2,3^2,19^2])\relax
\end{everbatim*}

\funcdesc[x, y, ...]{max} maximum of the (arbitrarily many) arguments,
\begin{everbatim*}
\xinttheexpr max(1,3,19), min([1,3,19])\relax
\end{everbatim*}

\funcdesc[x, y, ...]{min} minimum of the (arbitrarily many) arguments,
\begin{everbatim*}
\xinttheexpr min(1,3,19), min([1,3,19])\relax
\end{everbatim*}

\funcdesc[x, y, ...]{gcd} computes the positive generator of the fractional
ideal of rational numbers $x\mathbb Z + y\mathbb Z + ... \subset \mathbb
Q$.  Since |1.4d| the output is always in lowest terms.

This example shows how to reduce an n-uple to its primitive part:
\begin{everbatim*}
\xinteval{gcd(7/300, 11/150, 13/60)}\newline
$(7/300, 11/150, 13/60)\to
(\xinteval{subsn(seq(reduce(x/D), x = L), D=gcd(L); L=7/300, 11/150, 13/60)})$\newline
\xintexpr gcd([7/300, 11/150, 13/60])\relax\par
\end{everbatim*}
MEMO
Perhaps a future release will provide a |primpart()| function as built-in
functionality.

In case of strict integers, using a |\xintiiexpr...\relax| wrapper is
advantageous as the integer-only |gcd()| is more efficient.
%
% ceci semble encore à peu près exact à 1.4d :
% (about |6X|) than the one accepting general fractional inputs. 
%
As \csbxint{iiexpr} accepts only strict integers, doing this may require
wrapping the argument in \func{num}.

\funcdesc[x, y, ...]{lcm} computes the positive generator of the
fractional ideal of rational numbers $x\mathbb Z \cap y\mathbb Z \cap ...
\subset \mathbb Q$.
\begin{everbatim*}
\xinttheexpr lcm([7/300, 11/150, 13/60])\relax
\end{everbatim*}
As for \func{gcd}, since |1.4d| the output is always in lowest terms.
% Memo 1.4d: This
% function got (I did not tests extensively) a |4X| speed gain for inputs being
% only integers
For strict integers it is slightly advantageous to use a sub
\csbxint{iiexpr}-ession.
%
% je disais à 1.4:
%  (about |9X|) than the one accepting general fractional inputs.
% mais à 1.4d c'est seulement 2X : le lcm pour les fractions
% a quadruplé sa vitesse !
%

\funcdesc[x, y, ...]{first} first item of the list or nutple argument:
\begin{everbatim*}
\xintiiexpr first([last(-7..3), [58, 97..105]])\relax
\end{everbatim*}
\funcdesc[x, y, ...]{last} last item of the list or nutple argument:
\begin{everbatim*}
\xintiiexpr last([-7..3, 58, first(97..105)])\relax
\end{everbatim*}
\funcdesc[x, y, ...]{reversed} reverses the order of the comma separated list
or inside a nutple:
\begin{everbatim*}
\xintiieval{reversed(reversed(1..5), reversed([1..5]))}
\end{everbatim*}

The above is correct as \xintexprname functions may produce oples and this is
the case here.
\funcdesc[x, y, ...]{len} computes the number of items in a comma separated
  list or inside a nutple (at first level only: it is not a counter of leaves).
\begin{everbatim*}
\xinttheiiexpr len(37.5), len(1..50, [101..150], 1001..1050), len([1..10])\relax
\end{everbatim*}

   \funcdesc[\lowast nutples]{zip} behaves\NewWith{1.4b} similarly to
   the Python function of the same name: i.e. it produces \emph{an ople of nutples,
   where the i-th nutple contains the i-th element from each of the argument
   nutples. The ople ends when the shortest input nutple is exhausted.
   With a single nutple argument, it returns an ople of 1-nutples.
   With no arguments, it returns the empty ople.}

   As there is no exact match in \xintexprname of the concept of «iterator» object,%
%
\footnote{%
Speaking of iterators, I have some ideas about this: as \csbxint{expr} does not
have the global expression in its hands it is difficult to organize globally
expandably the idea of iterator, but locally via syntax like the one for
\func{seq} this is feasible. When one thinks about it, \func{seq} is closely related
to the iterator idea.}
%
   there is a significant difference here that (for example) the |zip(x,x,x)|
   Python idiom to cluster the iterator |x| into successive chunks of length 3
   does not apply. Consider for this reason even the name of the function as
   work-in-progress, susceptible to change.\unstable{}
\begin{everbatim*}
\xintiieval{zip([1..9], [0, 1, 2], [11..29], [111..139])}
\end{everbatim*}

   See also \csbxint{thespaceseparated} for some possible usage in combination with \func{flat}.

  \end{description}

\subsubsection{Functions requiring dummy variables}
\hypertarget{ssec:dummies}{}

The pseudo-functions \xintFor #1 in {subs, seq, subsm, subsn, iter, add, mul,
  rseq, iterr, rrseq, iterr, ndseq, ndmap, ndfillraw} \do
{\func{#1}\xintifForLast{}{, }} use delimited macros for some tasks:
\begin{itemize}
\item for all of them, whenever a |<varname>=| chunk must be parsed into a
  (non-assigned) variable name, then the equal sign must be visible,
\item and if the syntax is with |,<varname>=| the initial comma also must be
  visible (spaces do not matter),
\item for all of them but \func{ndmap} and \func{ndfillraw} the final closing
  parenthesis must be visible.
\end{itemize}
Although delimited macros involving commas are used to locate |,<varname=|
this is done in a way silently ignoring commas located inside correctly
balanced parentheses. Thus, as the examples will show, nesting works as
expected.

The semi-colons involved in the syntax may arise from expansion alone. For
\func{rseq}, \func{iter}, \func{rrseq} and \func{iterr} the |,<varname>=| part
may also be created from the expansion which will generate the initial comma
separated values delimited by a semi-colon.

Prior to |1.4|, semi-colons needed to be braced or otherwise hidden when
located in an expression parsed by \csbxint{defvar} or
\csbxint{deffunc}, to not be confused with the expression
terminator.

This is not needed anymore.\NewWith{1.4}

\func{seq}, \func{rseq}, \func{iter}, \func{rrseq},
\func{iterr} and also \func{add}, \func{mul}, but not \func{subs} admit the
\keyword{omit}, \keyword{abort}, and \func{break} keywords. This is a
new feature at |1.4| for \func{add} and \func{mul}.

In the case
of a potentially infinite list generated by the |<integer>++| syntax, use of
\keyword{abort} or of \func{break} is mandatory, naturally.

All lowercase and uppercase Latin letters are pre-configured for usage as
dummy variables. In Unicode engines one can use \csbxint{newdummy} to turn any
letter into a usable dummy variable.

And since |1.4|,\NewWith{1.4} \csbxint{newdummy} works (in all engines) to
turn a multi-letter word into a dummy variable.  In the descriptions,
|varname| stands for such a dummy variable, either single-letter or word.

\begin{description}
% [parsep=0pt,align=left,
%    leftmargin=0pt, itemindent=0pt,
%    labelwidth=-\fontdimen2\font, labelsep=\fontdimen2\font, labelindent=0pt,
%    listparindent=\leftmarginiii]
\funcdesc[expr, varname=values]{subs} for variable substitution.
\begin{everbatim*}
\xinttheexpr subs(subs(seq(x*z,x=1..10),z=y^2),y=10)\relax\newline
\end{everbatim*}%
Attention that |xz| generates an error, one must use explicitely |x*z|, else
the parser expects a variable with name |xz|.

\func{subs} is useful when defining macros for which some argument will be
used more than once but may itself be a complicated expression or macro, and
should be evaluated only once, for matters of efficiency. But \func{subs} is
helpless in function definitions: all places where a variable is substituted
will receive the complete recipe to compute the variable, rather than evaluate
only once.

One should rather define auxiliary functions to compute intermediate
results. Or one can use \func{seq}.  See the documentation of
\csbxint{deffunc}.

\funcdesc[expr, varname=values]{add} addition
\begin{everbatim*}
\xintiiexpr add(x^3,x=1..20), add(x(x+1), x=1,3,19)\relax\newline
\xintiiexpr add(x^3, x = 1..[2]..20)\relax\newline         % add only odd cubes
\xintiiexpr add((odd(x))?{x^3}{omit}, x = 1..20)\relax\par % add only odd cubes
\end{everbatim*}

At |1.4| (fixed at |1.4a|),\NewWith {1.4a} the keywords \keyword{omit} (as in
example above), \keyword{abort} and \func{break} are allowed.
The meaning of \func{break} is specific: its argument serves as last
operand for the addition, not as ultimate value.
\begin{everbatim*}
\xintiiexpr add((x>10)?{break(1000)}{x}, x = 1..15)\relax
\end{everbatim*}

The |@| special variable holds the so-far accumulated value. Initially its
value is zero.
\begin{everbatim*}
\xintiiexpr add(1 + @, i=1..10)\relax % iterates x <- 2x+1
\end{everbatim*}

See \func{\textasciigrave+\textasciigrave} for syntax simply adding items of a
list without usage of a dummy variable.

\funcdesc[expr, varname=values]{mul} multiplication
\begin{everbatim*}
\xintiiexpr mul(x^2, x = 1, 3, 19, 37..50)\relax
\end{everbatim*}

The |@| special variable holds the so-far accumulated value. Initially its
value is one.

At |1.4| (fixed at |1.4a|),\NewWith {1.4a} the keywords \keyword{omit},
\keyword{abort} and \func{break} are allowed.  The meaning of \func{break} is
specific: its argument serves as last operand for the multiplication, not as
ultimate value.
\begin{everbatim*}
\xintiieval{mul((i==100)?{break(i^4)}{i}, i = 98, 99, 100)}
\end{everbatim*}
 
See \func{\textasciigrave\lowast\textasciigrave} for syntax without a dummy variable.

\funcdesc[expr, varname=values]{seq} comma separated values generated according to a formula
\begin{everbatim*}
\xintiiexpr seq(x(x+1)(x+2)(x+3),x=1..10), `*`(seq(3x+2,x=1..10))\relax
\end{everbatim*}
\begin{everbatim*}
\smallskip
\leavevmode\vbox{\xintthealign\xintiiexpr [seq([seq(i^2+j^2, i=0..j)], j=0..10)]\relax}
\end{everbatim*}

\funcdesc[initial value; expr, varname=values]{rseq} recursive sequence, |@| for the previous value.
\begin{everbatim*}
\printnumber {\xintthefloatexpr subs(rseq (1; @/2+y/2@, i=1..10),y=1000)\relax }\newline
\end{everbatim*}%
  Attention: in the example above |y/2@| is interpreted as
  |y/(2*@)|.\IMPORTANT{} With versions |1.2c| or earlier it would have been
  interpreted as |(y/2)*@|.

In case the initial stretch is a comma separated list, |@| refers at the first
iteration to the whole list. Use parentheses at each iteration to maintain
this ``nuple''. For example:
\begin{everbatim*}
\printnumber{\xintthefloatexpr rseq(1,10^6;
             (sqrt(@[0]*@[1]),(@[0]+@[1])/2), i=1..7)\relax }
\end{everbatim*}

Prior to |1.4| the above example had to be written with |[@]|. This is still
possible (|@| stands for an ople with two items, bracketing then extracting is
like extracting directly), but it is leaner to drop the extra «packing».

\funcdesc[initial value; expr, varname=values]{iter} is exactly like |rseq|, except that it only prints
  the last iteration.


\hypertarget{BrentSalamin}{}
  |iter()| is convenient to handle compactly higher order iterations.
  We can illustrate its use with an expandable (!)
  implementation of the Brent-Salamin algorithm for the computation of $\pi$:
\begin{everbatim*}
\xintDigits:= 87\relax % we target 84 digits, and use 3 guard digits
\xintdeffloatfunc BS(a, b, t, p):= 0.5*(a+b), sqrt(a*b), t-p*sqr(a-b), \xintiiexpr 2p\relax;
\xinteval
{trunc(% I feel truncation is better than rounding to display decimals of π
  \xintfloatexpr
    iter(1, sqrt(0.5), 1, 1; % initial values
% this 43 is 84/2 + 1
      (@[0]-@[1]<2e-43)?% stopping criteria; takes into account that the
                            % exit computation (break() argument) doubles
                            % number of exact digits (roughly) 
      {break(sqr(@[0]+@[1])/@[2])}      % ... do final computation,
      {BS(@)}, % else do iteration
       i=1++)  % This generates infinite iteration. The i is not used.
  \relax
% this 83 is 84 - 1 (there is a digit known to be 3 actually, before decimal mark)
 , 83)% closing parenthesis of trunc()
}...% some dots following end of \xinteval argument
\xintDigits:=16\relax
\end{everbatim*}\newline
  You can try with |\xintDigits:=1004\relax| and |2e-501| in place of
  |\xintDigits:=87\relax| and |2e-43|, but be patient for some seconds
  for the result. Of course don't truncate the final
  result to only \dtt{83} fractional decimal digits but \dtt{1000}...\@
  and better to wrap the whole thing in |\message| or
  |\immediate\write128| or |\edef| because it will then run in the right margin.

  Prior to |1.4| the above example had to use notation such as |[@][0]|; this
  would still work but |@[0]| is leaner.

\funcdesc[initial values; expr, varname=values]{rrseq} recursive sequence with multiple initial terms. Say, there are
  |K| of them. Then |@1|, ..., |@4| and then |@@(n)| up to |n=K| refer to the
  last |K| values. Notice the difference with |rseq()| for which |@| refers to
  a list of items in case the initial value is a list and not a single item.%
%
\footnote{Prior to |1.4|, one could use |@| in |rrseq()| and |iterr()| as an
  alias to |@1|. This undocumented feature is dropped and |@| will break |rrseq()| and |iterr()|.}
%
  Using |rrseq()| with |@1| etc...\@ accessors may be perhaps a bit
  more efficient than using |rseq()| with a list as staring value and
  constructs such as |@[0]|, |@[1]| (or rather |@[-1]|, |@[-2]| to
  mimick what |@1|, |@2|, |@3|, |@4| and |@@(integer)| do in |rrseq()|.
\begin{everbatim*}
\xinttheiiexpr rrseq(0,1; @1+@2, i=2..30)\relax
\end{everbatim*}
\begin{everbatim*}
\xinttheiiexpr rseq(1; 2@, i=1..10)\relax
\end{everbatim*}
\begin{everbatim*}
\xinttheiiexpr rseq(1; 2@+1, i=1..10)\relax
\end{everbatim*}
\begin{everbatim*}
\xinttheiiexpr rseq(2; @(@+1)/2, i=1..5)\relax
\end{everbatim*}

\begin{everbatim*}
\xinttheiiexpr rrseq(0,1,2,3,4,5; @1+@2+@3+@4+@@(5)+@@(6), i=1..20)\relax
\end{everbatim*}

I implemented an |Rseq| which at all times keeps the memory of \emph{all}
previous items, but decided to drop it as the package was becoming big.

\funcdesc[initial values; expr, varname=values]{iterr} same as |rrseq| but does not print any value until the last |K|.
\begin{everbatim*}
\xinttheiiexpr iterr(0,1; @1+@2, i=2..5, 6..10)\relax
% the iterated over list is allowed to have disjoint defining parts.
\end{everbatim*}

\funcdesc[expr, var1=value1; var2=value2; ....; varN=valueN{[;]}]{subsm}
Simultaneous\NewWith{1.4}
substitutions. The assigned values must not involve the
variables. An optional final
semi-colon is allowed.
\begin{everbatim*}
\xintiieval{subsm(x+2y+3z+4t, x=1; y=10; z=100; t=1000;)}
\end{everbatim*}

\funcdesc[expr, var1=value1; var2=value2; ....; varN=valueN{[;]}]{subsn}
Simultaneous\NewWith{1.4}
substitutions. The assigned values may involve all variables
located further to its right. An optional final
semi-colon is allowed.
\begin{everbatim*}
\xintiieval{subsn(x+y+z+t, x=20y; y=20z; z=20t; t=1)}
\end{everbatim*}

\funcdesc[function, values1; values2; ....; valuesN{[;]}]{ndmap}
%
Construction\NewWith{1.4} of a nested list (a priori having |N| dimensions) from function
values. The function must be an |N|-variable function (or a function accepting
arbitrarily many arguments), but it is not constrained to produce only scalar
values.  Only in the latter case is the output really an |N|-dimensional
«|ndlist|» type object.  An optional final semi-colon in the input before the
closing parenthesis is
allowed.
\begin{everbatim*}
\xintdeffunc foo(a,b,c,d) = a+b+c+d;
\begin{multicols}{2}
\xintthealign\xintexpr ndmap(foo, 1000,2000,3000; 100,200,300; 10,20,30; 1,2,3)\relax
\end{multicols}
\end{everbatim*}

\funcdesc[expr, var1=values1; var2=values2; ....; varN = valuesN{[;]}]{ndseq}
%
Constructs\NewWith{1.4} a nested list (a priori having |N| dimensions) from
substitutions in an expression involving |N| (dummy) variables.  The
expression is not constrained to produce only scalar values.  Only in the
latter case is the output really an |N|-dimensional «|ndlist|» type object.
An optional final semi-colon in the input before the closing parenthesis is
allowed.
\begin{everbatim*}
\begin{multicols}{2}
\xintthealign\xintexpr ndseq(a+b+c+d, a=1000,2000,3000; b=100,200,300; c=10,20,30; d=1,2,3;)\relax
\end{multicols}% in case of page break, this makes amusing zigzag rendering
\end{everbatim*}

\end{description}

Recursions may be nested, with |@@@(n)| giving access to the values of the
outer recursion\dots and there is even |@@@@(n)| to access the outer outer
recursion but I never tried it!

The following keywords are recognized:
\begin{description}
  \keyworddesc{abort} it is a pseudo-variable which indicates to stop here and now.

  \keyworddesc{omit} it is a pseudo-variable which says to omit this value and
  go to next one.

  \funcdesc[stuff]{break} says to abort and insert |stuff| as last value.

  \keyworddesc{<integer>++} serves to generate a potentially infinite list. In
  conjunction with an \keyword{abort} or \func{break} this is often
  more efficient than iterating over a pre-established list of values.
\begin{everbatim*}
\xinttheiiexpr iter(1;(@>10^40)?{break(@)}{2@},i=1++)\relax
\end{everbatim*}
is the smallest power of 2 with at least fourty one digits.

The |i=<integer>++| syntax (any letter is allowed in place of |i|) works only
in the form |<letter>=<integer>++|, something like |x=10,17,30++| is not
legal. The |<integer>| must be a \TeX-allowable integer.
\begin{everbatim*}
First Fibonacci number at least |2^31| and its index
% we use iterr to refer via @1 and @2 to the previous and previous to previous.
\xinttheiiexpr iterr(0,1; (@1>=2^31)?{break(@1, i)}{@2+@1}, i=1++)\relax
\end{everbatim*}%
. If one also wants the previous Fibonacci number one only has to use |break(@2,
@1, i)| in the above example.
\end{description}


\subsection{Generators of arithmetic progressions}
\label{ssec:arithseq}

\begin{itemize}
  \item |a..b| constructs the \textbf{small} integers from the ceil $\lceil
    a\rceil$ to the floor
    $\lfloor b\rfloor$ (possibly a decreasing sequence): one has to be careful
    if using this for algorithms that |1..0| for example is not empty or |1|
    but expands to |1, 0|. Again, |a..b| \emph{can not} be used with |a| and
    |b| greater than $2^{31}-1$. Also, only about at most \dtt{5000} integers
    can be generated (this depends upon some \TeX{} memory settings).

    The |..| has lower precedence than the arithmetic operations.
\begin{everbatim*}
\xintexpr 1.5+0.4..2.3+1.1\relax; \xintexpr 1.9..3.4\relax; \xintexpr 2..3\relax
\end{everbatim*}

    The step of replacing $a$ by its ceil and $b$ by its floor is a kind of
    silly overhead, but $a$ and $b$ are allowed to be themselves the result
    of computations and there is no notion of «int» type in \csbxint{eval}.
    The solution is, when $a$ and $b$ are given explicit integers to
    temporarily switch to the \csbxint{iiexpr} parser:
\begin{everbatim*}
\xintexpr \xintiiexpr 1..10\relax\relax
\end{everbatim*}

    On the other hand integers from |\xintexpr 1..10\relax| are already in
    raw \xintfracname format for example |3/1[0]| which speeds up their usage
    in the macros internally involved in computations...\@ thus perhaps what one
    gains on one side is lost on the other side.

  \item |a..[d]..b| generates «real» numbers along arithmetic progression
    of reason |d|. It does
    \emph{not} replace |a| by its ceil, nor |b| by its floor. The generated
    list is empty if |b-a| and |d| are of opposite signs; if |d=0| or if |a=b|
    the list expands to single element |a|.
\begin{everbatim*}
\xintexpr 1.5..[1.01]..11.23\relax
\end{everbatim*}

     At |1.4|,\CHANGED{1.4} this generator behaves in \csbxint{floatexpr} exactly as in
     \csbxint{expr}, i.e.\@ \emph{exactly}. This is breaking
     change.
\begin{everbatim*}
\xintDigits:=6;
\xintexpr\xintfloatexpr 100..[1.23456]..110\relax\relax
\xintDigits:=16;
\end{everbatim*}

     This demonstration embedded the float expression in the exact parser only
     to avoid the rounding to the prevailing precision on output, thus we can
     see that internally additions are done exactly and not with 
     \dtt{6} digits mantissas (in this example).
\end{itemize}

\subsection{Python slicing and indexing of one-di\-men\-sional sequences}
\label{ssec:lists}

There are some breaking changes in the syntax at |1.4|,\CHANGED{1.4} because previously
\xintexprname had no real notion of a list or sequence type. It now does, and
even allows nesting.

We denote here by \emph{list} or \emph{sequence} a general \emph{ople}, either
given as a variable or explicitly. \emph{In the former case the parentheses are
optional}.%
%
\footnote{Even for an «open list», if it is given as a \emph{variable} then the indexing or slicing will not apply to its last item but to itself as an entity.}

\begin{itemize}
\item |(list)[n]| returns the |n+1|th item if |n>=0|. If |n<0| it enumerates
  items from the tail. Items are numbered as in Python, the
  first element corresponding to |n=0|.
\begin{everbatim*}
\xintexpr (0..10)[6], (0..10)[-1], (0..10)[23*18-22*19]\relax
\end{everbatim*}

This also works for singleton \emph{oples} which are in fact a \emph{number}:
\begin{everbatim*}
\xintexpr (7)[0], (7)[-1], 9, (7)[-2], 9\relax
\end{everbatim*}

In the example above the parentheses serve to disambiguate from the raw
\xintfracname format such as |7[-1]| which, although discouraged, is accepted
on input. And we used a trick to show that |(7)[-2]| returns |nil|.

The behaviour changes for singleton \emph{oples} which are not
\emph{numbers}. They are thus \emph{nutples}, or equivalently they are the
bracketing (bracing, packing) of another \emph{ople}. In this case, the meaning
of the syntax for item indexing is, as in Python, item
\emph{extraction}:

\begin{everbatim*}
\xintexpr [0,1,2,3,4,5][2], [0,1,2,3,4,5][-3]\relax\newline
\xintexpr [0,[1,2,3,4,5],6][1][-1]\relax
\end{everbatim*}

  \item |(list)[:n]| produces the first |n| elements if |n>0|, or suppresses
    the last \verb+|n|+ elements if |n<0|.
\begin{everbatim*}
\xintiiexpr (0..10)[:6]\relax\ and \xintiiexpr (0..10)[:-6]\relax
\end{everbatim*}

As above, the meaning change for \emph{nutples} and fits with expectations
from Python regarding its sequence types:
\begin{everbatim*}
\xintiiexpr [0..10][:6]\relax\ and \xintiiexpr [0..10][:-6]\relax
\end{everbatim*}

 \item |(list)[n:]| suppresses the first |n| elements if |n>0|, or extracts
    the last \verb+|n|+ elements if |n<0|.
\begin{everbatim*}
\xintiiexpr (0..10)[6:]\relax\ and \xintiiexpr (0..10)[-6:]\relax
\end{everbatim*}

As above, the meaning change for \emph{nutples} and fit with expectations
from Python with \emph{tuple} or \emph{list} types:
\begin{everbatim*}
\xintiiexpr [0..10][6:]\relax\ and \xintiiexpr [0..10][-6:]\relax
\end{everbatim*}

\item Finally, |(list)[a:b]| also works according to the Python ``slicing''
  rules (inclusive of negative indices). Notice though that stepping is
  currently not supported.
\begin{everbatim*}
\xinttheiiexpr (1..20)[6:13]\relax\ = \xinttheiiexpr (1..20)[6-20:13-20]\relax\newline
\xinttheiiexpr [1..20][6:13]\relax\ = \xinttheiiexpr [1..20][6-20:13-20]\relax
\end{everbatim*}

\item It is naturally possible to execute such slicing operations one after
  the other (the syntax is simplified compared to before |1.4|):
\begin{everbatim*}
\xintexpr (1..50)[13:37][10:-10]\relax\newline
\xintexpr (1..50)[13:37][10:-10][-1]\relax
\end{everbatim*}
\end{itemize}


\subsection{NumPy like nested slicing and indexing for arbitrary oples and nutples}

This is entirely new with |1.4|.\NewWith{1.4}

I will give one illustrative example and refer to the NumPy documentation for
more.

Notice though that our interpretation of the syntax is more general than
NumPy's concepts (of basic slicing/indexing):
\begin{itemize}
\item slicing and itemizing apply also to non-bracketed objects i.e.\@ \emph{oples},
\item the leaves do not have to be all at the same depth,
\item there are never any out-of-range index errors: out-of-range indices
  are silently ignored.
\end{itemize}

\begin{everbatim*}
\begin{multicols}{3}
\xintdefvar myArray = ndseq(a+b+c, a=100,200,300; b=40,50,60; c=7,8,9);
myArray = \xintthealign\xintexpr myArray\relax
\columnbreak
mySubArray = \xintthealign\xintexpr myArray[0:2,0:2,0:2]\relax
myExtractedSubArray = \xintthealign\xintexpr myArray[0:2,0:2,0:2][0]\relax
\columnbreak
myExtractedSubArray = \xintthealign\xintexpr myArray[0:2,0:2,0:2][0,1]\relax
\noindent
firstExtractedScalar = \xintexpr myArray[0:2,0:2,0:2][0,1,0]\relax\newline
secondExtractedScalar = \xintexpr myArray[0,1,0]\relax\par
\end{multicols}
\end{everbatim*}

As said before, \emph{stepping} is not yet implemented. Also the NumPy
extension to Python for item selection (i.e.\@ via a |tuple| of comma separated
indices) is not yet implemented.

\subsection{Tacit multiplication}
\label{ssec:tacit multiplication}

Tacit multiplication (insertion of a |*|) applies when the parser is currently
either scanning the digits of a number (or its decimal part or scientific
part, or hexadecimal input), or is looking for an infix operator, and:
\begin{enumerate}[nosep, label=(\arabic*.)]
\item \relax\emph{encounters a count or dimen or skip register or variable or an
  \eTeX{} expression,} or
\item \emph{encounters a sub-\csa{xintexpr}ession}, or
\item \emph{encounters an opening parenthesis}, or
\item \emph{encounters a
  letter (which is interpreted as signaling the start of either a variable or
  a function name)}, or
\item (of course, only when in state "looking for an operator") \emph{encounters a digit}.
\end{enumerate}

\begin{framed}
  \centeredline{\textcolor{Red}{\textbf{!!!!ATTENTION!!!!}}}

  Explicit digits prefixing a variable, or a function, whose name starts with
  an |e| or |E| will trap the parser into trying to build a number in
  scientific notation. So the |*| must be explicitly inserted.

\begin{everbatim}
\xintdefiivar e := (2a+4b+6d+N)/:7;%
\xintdefiivar f := (c+11d+22*e)//451;% 22e would raise errors
\end{everbatim}

  I don't think I will fix this anytime soon...
\end{framed}

\begin{framed}
    For example, if |x, y, z| are variables all three of |(x+y)z|, |x(y+z)|,
    |(x+y)(x+z)| will create a tacit multiplication.

    Furthermore starting with release
    |1.2e|, %\MyMarginNote[\kern\dimexpr\FrameSep+\FrameRule\relax]{Changed}
    whenever tacit multiplication is applied, in all cases it \emph{always}
    ``ties'' more\IMPORTANT{} than normal multiplication or division, but
    still less than power. Thus |x/2y| is interpreted as |x/(2y)| and
    similarly for |x/2max(3,5)| but |x^2y| is still interpreted as |(x^2)*y|
    and |2n!| as |2*n!|.

\begin{everbatim*}
\xintdefvar x:=30;\xintdefvar y:=5;%
\xinttheexpr (x+y)x, x/2y, x^2y, x!, 2x!, x/2max(x,y)\relax
\end{everbatim*}

    Since |1.2q| tacit multiplication is triggered also in cases such as
    |(1+2)5| or |10!20!30!|.

\begin{everbatim*}
\xinttheexpr (10+7)5, 4!4!, add(i, i=1..10)10, max(x, y)100\relax
\end{everbatim*}

    The ``tie more'' rule applies to all cases of tacit multiplication. It
    impacts only situations with a division operator as the last seen
    operator, as multiplication is mathematically associative.
\begin{everbatim*}
\xinttheexpr 1/(3)5, (1+2)/(3+4)(5+6), 2/x(10), 2/10x, 3/y\xintiiexpr 5+6\relax, 1/x(y)\relax\
differ from\newline\xinttheexpr 1/3*5, (1+2)/(3+4)*(5+6), 2/x*(10), 2/10*x,
                                3/y*\xintiiexpr 5+6\relax, 1/x*(y)\relax\par
\end{everbatim*}
\end{framed}

    Note that |y|\csbxint{theiiexpr}| 5+6\relax| would have tried to use a variable
    with name |y11| rather than doing |y*11|: tacit multiplication works only
    in front of sub-\csbxint{expr}essions, not in front of
    \csbxint{theexpr}essions which are unlocked into explicit digits.


Here is an expression whose meaning is
    completely modified by the ``tie more'' property of tacit multiplication:


\begin{everbatim}
\xintdeffunc e(z):=1+z(1+z/2(1+z/3(1+z/4)));
\end{everbatim}
will be parsed as
\begin{everbatim}
\xintdeffunc e(z):=1+z*(1+z/(2*(1+z/(3*(1+z/4)))));
\end{everbatim}
which is not at all the presumably hoped for:
\begin{everbatim}
\xintdeffunc e(z):=1+z*(1+(z/2)*(1+(z/3)*(1+(z/4))));
\end{everbatim}
% This case can be handled this way:
% \begin{everbatim}
% \xintdeffunc e(z):=(((z/4+1)z/3+1)z/2+1)z+1;
% \end{everbatim}


\subsection{User defined variables}
\label{ssec:uservariables}
\label{xintdefvar}
\label{xintdefiivar}
\label{xintdeffloatvar}

Since release |1.1| it is possible to make an assignment to a variable name
and let it be known to the parsers of \xintexprname. Since |1.2p| simultaneous
assignments are possible. Since |1.4| simultaneous assignments are possible with
a right-hand-side being a |nutple| which will be automatically unpacked.
\begin{everbatim*}
\xintdefvar myPi:=3.141592653589793238462643;%
$myPi = \xinteval{myPi}$\newline % (there is already built-in Pi variable)
\xintdefvar x_1, x_2, x_3 := 10, 20, 30;%
$x_1 = \xinteval{x_1}, x_2 = \xinteval{x_2}, x_3 = \xinteval{x_3}$\newline
\xintdefvar x_1, x_2, x_3 := [100, 200, 300];%
$x_1 = \xinteval{x_1}, x_2 = \xinteval{x_2}, x_3 = \xinteval{x_3}$\par
\end{everbatim*}


Simultaneous assignments\NewWith{1.4i} with more variables than values do not
raise an error but simply set the extra variables to the \dtt{nil} value.
\begin{everbatim*}
\xintdefiivar a, b, c := [1, 2];% will be automatically unpacked
The value of a is \xinteval{a}, the one of b is \xinteval{b} and the one of c is \xinteval{c}.
\end{everbatim*}
\begin{everbatim*}
\xintdefiivar a, b, c := 314;%
The value of a is \xinteval{a}, the one of b is \xinteval{b} and the one of c is \xinteval{c}.
\end{everbatim*}

Notice that \dtt{nil} variables must be used with caution as they break
arithmetic operations if used as operands to them. And they are not the same
as the \dtt{None} variables, which can also be input as |[]|.

Simultaneous assignments\NewWith{1.4i} with less variables than values do not
raise an error but set the last variable to be the ople concatenating the
remaining values.
\begin{everbatim*}
\xintdefiivar seq := 1..10;%
\xintdefiivar a, seq := seq;%
\xintdefiivar b, seq := seq;%
\xintdefiivar c, d, seq := seq;%
The value of a is \xinteval{a}, the one of b is \xinteval{b}, the one of c is \xinteval{c},
the one of d is \xinteval{d}, the one of seq is \xinteval{seq}.
\end{everbatim*}

In the above we define a variable |seq| but there is a built-in
function \func{seq}.  It is indeed allowed to use the same name for both a
variable and a function.%
%
\footnote{But until a bugfix added at release |1.4i|, some built-in function
  names (those implementing syntax with dummy variables, and the so-called
  «pseudo»-functions) were fragile under such overloading.}
But for safety we will unassign |seq| now:
\begin{everbatim*}
\xintunassignvar{a}\xintunassignvar{b}\xintunassignvar{c}\xintunassignvar{d}%
\xintunassignvar{seq}%
\end{everbatim*}%
Single letter names |a..z| and |A..Z| are pre-declared by the package for use
as a special type of variables called ``dummy variables''.  Unnassigning them
restores this initial meaning.  See further \csbxint{unassignvar} and
\csbxint{newdummy}.  Since |1.4| even assigned variables can be used in the
call signatures of function declarations.

Regarding the manipulation of an «open list» as above, there is no way
to obtain with only one use of the variable both its last item and the
reduction of the variable to its truncated self. One can do rather:
\begin{everbatim*}
\xintdefiivar mylist := 1..10;%
\xintdefiivar z, mylist := last(mylist), mylist[:-1];%
The value of z is \xinteval{z} and mylist is now \xinteval{mylist}.\par
\end{everbatim*}
This uses twice |mylist| and is about the same as doing it in two steps:
\begin{everbatim*}
\xintdefiivar w := last(mylist);%
\xintdefiivar mylist := mylist[:-1];%
The value of w is \xinteval{w} and mylist is now \xinteval{mylist}.%
\xintunassignvar{z}\xintunassignvar{w}\xintunassignvar{mylist}\par
\end{everbatim*}

It is recommended generally speaking to work with «closed (i.e. bracketed)
lists» because only them and numbers can be arguments to functions (but see
\csbxint{deffunc} and the notion of variadic last argument).  For more on the
Python-like slicing used above see \autoref{ssec:lists} and
\autoref{sssec:opleslicing}.  For more information relative to variables
versus arguments see \autoref{sssec:funcargs}.
 
\begin{itemize}[noitemsep]
\item For catcodes issues (particularly, for the semi-colon used to delimit
  the fetched expression), see the discussion of \csbxint{exprSafeCatcodes}.
\item 
  Both syntaxes |\xintdefvar foo := <expr>;| and |\xintdefvar foo = <expr>;|
  are accepted.
\item 
  Spaces in the variable name or around the equal sign are removed and are
  immaterial.
\item 
  The variable names are expanded in an |\edef| (and stripped of spaces).
Example:
\begin{everbatim}
\xintdefvar x\xintListWithSep{, x}{\xintSeq{0}{10}} := seq(2**i, i = 0..10);%
\end{everbatim}
This defines |x0|, |x1|, \dots, |x10| for future usage.
\end{itemize}

Legal variable names are composed of letters, digits, |_| and |@| and
characters.  A variable name must start with a letter. Variable names starting
with a |@| or |_| are reserved for internal usage.%
%
\footnote{The process of variable declaration does not check that these rules
  are met, and breakage will arise on use, if rules are not followed. For
  example, prior to |1.4g|, using a variable (illegally)
  declared with a name starting with a (normal, catcode 8) |_| triggered an
  infinite loop.}

As |x_1x_2| or even |x_1x| are licit variable names, and as the parser does
not trace back its steps, input syntax must be |x_1*x_2| if the aim is to
multiply such variables.

Using \csa{xintdefvar}, \csa{xintdefiivar}, or \csa{xintdeffloatvar} means
that the variable value will be computed using respectively \csa{xintexpr},
\csa{xintiiexpr} or \csa{xintfloatexpr}. It can then be used in all three
parsers, as long as the parser understands the format. Currently this means
that variables using \csa{xintdefvar} or \csa{xintdeffloatvar} can be used
freely either with \csa{xintexpr} or \csa{xintfloatexpr} but not with
\csa{xintiiexpr}, and variables defined via \csa{xintdefiivar} can be used in
all parsers.

When defining a variable with \csa{xintdeffloatvar}\CHANGED{1.4e} it (or
generally speaking its numerical leaves) is rounded to \csbxint{theDigits}
precision.  So the variable holds the same value as would be printed via
\csbxint{floateval} for the same computation.

Prior to |1.4e|, this was the case only if the variable definition actually
involved some computation.

However the \csbxint{floatexpr}|..\relax| wrapper by itself induces no rounding.  If
it is encountered in the typesetting flow, the print-out will be rounded to
\csbxint{theDigits} precision, but this is an effet of behaving like
\csbxint{floateval} in this context.
\begin{everbatim*}
% Since 1.4e, \xintdeffloatvar always rounds (to \xinttheDigits)
\xintdeffloatvar e:=2.7182818284590452353602874713526624977572470936999595749669676;%
1) \xintexpr  e\relax\newline  % shows the recorded value: it is rounded
\xintunassignvar{e}
2) \xintfloatexpr   % when used in typesetting flow, acts like \xintfloateval:
    2.7182818284590452353602874713526624977572470936999595749669676
\relax\newline   % the print-out is rounded.
3) \xintexpr 
  \xintfloatexpr
    2.7182818284590452353602874713526624977572470936999595749669676
  \relax
\relax\newline   % but we can see via the \xintexpr wrapper all the digits were there
% rounding can be forced using an extra 0+, the float() function, or the [D] option.
% tidbit: comparison operators do not pre-round, so 1.2345678 is not same as (1.2345678+0)
% in low precision.
\begingroup\xintDigits:=4;% with 2 guard digits, this makes 4+2=6 digits from "0+" rounding
4) \xintifboolfloatexpr{1.2345 == 1.23456}
   {\error}{Different! Comparisons do not pre-round to Digits precision.}\newline
5) \xintifboolfloatexpr{1.2345 == 1.2345 + 0}
   {\error}{Different! Right hand side rounded from operation, left hand side not rounded.}\par
\endgroup
\end{everbatim*}

% not so exciting example
% In the next examples we examine the effect of cumulated float operations on
% rounding errors:
% \begin{everbatim*}
% \xintdefvar       e_1:=add(1/i!, i=0..10);% exact sum
% \xintdeffloatvar  e_2:=add(1/i!, i=0..10);% float sum
% \xintthefloatexpr e_1, e_2\relax\newline
% \xintdefvar       e_3:=e_1+add(1/i!, i=11..20);% exact sum
% \xintdeffloatvar  e_4:=e_2+add(1/i!, i=11..20);% float sum
% \xintthefloatexpr e_3, e_4\relax\newline
% \xintdeffloatvar e:=2.7182818284590452353602874713526624977572470936999595749669676;%
% \xintDigits:=24\relax 
% \xintthefloatexpr[16] e, e^1000, e^1000000\relax (e rounded to 24 digits first)\newline
% \xintDigits:=16\relax 
% \xintthefloatexpr     e, e^1000, e^1000000\relax (e rounded to 16 digits first)\par
% \end{everbatim*}

After issuing \csbxint{verbosetrue} the values of defined variables are
written out to the log (and terminal). As in this example:
\begin{everbatim}
Package xintexpr Info: (on line 1)
    Variable myPi defined with value {3141592653589793238462643[-24]}.
Package xintexpr Info: (on line 2)
    Variable x_1 defined with value {10}.
Package xintexpr Info: (on line 2)
    Variable x_2 defined with value {20}.
Package xintexpr Info: (on line 2)
    Variable x_3 defined with value {30}.
Package xintexpr Info: (on line 3)
    Variable List defined with value {0}{1}{3}{6}{10}{15}{21}{28}{36}{45}{55}
.
Package xintexpr Info: (on line 4)
    Variable Nuple defined with value {{0}{1}{9}{36}{100}{225}{441}{784}{1296
}{2025}{3025}}.
Package xintexpr Info: (on line 5)
    Variable FourthPowers defined with value {{0}{1}{81}{1296}{10000}{50625}{
194481}{614656}{1679616}{4100625}{9150625}}.
\end{everbatim}
Prior to |1.4e| variable names\CHANGED{1.4e} were printed within straight
double quotes to the log. But this was not the case for function definitions,
so now quotes have been removed.

The braces one can see around the numerical values date back to the |1.4|
release.


\subsubsection{\csh{xintunassignvar}}
\label{xintunassignvar}

Variable declarations obey the current scope. To let a (multi-letter) name be
unknown to (all parsers of) \xintexprname without waiting the
end of the scope one issues \csa{xintunassignvar}\marg{variable}.

In the special case of \csa{xintunassignvar}\marg{letter}, the effect is
different,\IMPORTANT{} as it is synonymous with
\csbxint{newdummy}\marg{letter}: the (catcode 11) \meta{letter} recovers or
acquires meaning as a dummy variable in the current scope.
\begin{everbatim*}
\xintunassignvar{e}%
% overwriting a dummy letter
\xintdefvar i := 3;%
\xinttheiiexpr add(i, i = 1..10)\relax\ ("i" has the fixed value 3)\newline
\xintunassignvar{i}% back to normal
\xinttheiiexpr add(i, i = 1..10)\relax\ ("i" is again a dummy variable)\par
\end{everbatim*}

Under \csbxint{globaldefstrue} regime the effect of \csa{xintunassignvar} is
global.

\subsubsection{\csh{xintnewdummy}}
\label{xintnewdummy}

Any catcode 11 character can serve as a dummy variable, via this declaration:
\begin{everbatim}
\xintnewdummy{<character>}
\end{everbatim}
For example with Xe\TeX\ or Lua\LaTeX\ the following works:
\begin{everbatim}
% use a Unicode engine
\input xintexpr.sty
\xintnewdummy ξ% or any other letter character !
\xinttheexpr add(ξ, ξ=1..10)\relax
\bye
\end{everbatim}
Under \csbxint{globaldefstrue} regime the effect of \csa{xintnewdummy} is
global.

Starting with |1.4|,\NewWith{1.4} it is allowed to use \csa{xintnewdummy} with multi-letter
names (obeying the condition for being a variable name).

\subsubsection{\csh{xintensuredummy}, \csh{xintrestorevariable}}
\label{xintensuredummy}
\label{xintrestorevariable}

Use
\begin{everbatim}
\xintensuredummy{<character>}
...
... code using the (catcode 11) character as a dummy variable
...
\xintrestorevariable{<character>}
\end{everbatim}
if other parts need the letter as an assigned variable name. For example
\xinttrigname being written at high level needs a few genuine dummy variables,
and it uses \csbxint{ensuredummy} to be certain everything is ok.

\csbxint{restorevariable} was formerly called \csa{xintrestorelettervar}.\CHANGED{1.4}

\subsection{User defined functions}
\label{ssec:userfunctions}
\def\HOOKLOCALTOC#1#2#3{}
\etocsetnexttocdepth{subsubsection}\localtableofcontents
\let\HOOKLOCALTOC\empty


\subsubsection{\csh{xintdeffunc}}
\label{xintdeffunc}

Here is an example:
\begin{everbatim*}
\xintdeffunc
    Rump(x,y):=1335 y^6/4 + x^2 (11 x^2 y^2 - y^6 - 121 y^4 - 2) + 11 y^8/2 + x/2y;
\end{everbatim*}(notice the numerous tacit multiplications in this expression;
and that |x/2y| is interpreted as |x/(2y)|.)

\begin{framed}
  \begin{itemize}
  \item The ending semi-colon is allowed to be of active catcode, as
    |\xintdeffunc| temporarily resets catcodes before
    parsing the expression.  But this will fail if the whole thing is inside a
    macro definition. Then the used semi-colon must be the standard one.
  \item Semi-colons used inside the expression need not be hidden inside
    braces. (new with |1.4|)
  \item The colon before the equal sign is optional and its (reasonable)
    catcode does not matter.
  \end{itemize}
\end{framed}

Here are a few important items (bookmark this for reading again later once you
have gained experience in using this interface...):
\begin{itemize}
\item  The function names are composed of letters, digits, underscores or |@|
  signs. A function name must start with a letter. It may be a single letter
  (see \autoref{sssec:overload}).
\item The\NewWith{1.4} variable names used in the function signature may be multi-letter
  words. It is also allowed for them to already be in use for previously
  declared variables. Their meanings will get restored for usage after the
  function declaration.
\item A function can be declared with at most nine arguments. It can be declared as a
  function with no arguments.
\item If\NewWith{1.4} in the function declaration the last argument is
  prefixed by |*|, it stands for a |nutple| which will gather all arguments of
  the function call beyond the first positional ones. See \autoref{sssec:funcargs}
  for additional explanations on such «variadic» arguments.
\item Recursive\CHANGED{1.4}  definitions are possible; for them to not generate error or
  fall in infinite loops, the use of the short-circuit conditionals |?| and
  |??| is \emph{mandatory}.
\item If\CHANGED{1.4} a function is used in another definition it will check if it is
  applied to numerical arguments and if this is the case will expand
  fully. Prior to |1.4| one needed deprecated \csa{xintdefefunc} for this. But
  the latter is now but an alias for \csa{xintdeffunc}, the two have been
  merged.
\item The previous item has an exception for functions with no arguments; they
  never expand immediately in other function definitions (else they would be
  almost like variables). This provides a way to define functions with
  parameters: simply let their definition use some functions with no arguments.
\item A function declared via \csbxint{deffunc} remains \fbox{unknown} to
  \csbxint{floatexpr} (or \csbxint{floateval}). See \csbxint{deffloatfunc},
  \csbxint{defiifunc}. One can use the same formula in a new definition, but
  if one wants the expansion to execute in a parser independent way, one can
  transfer a function with scalar values like this:
\begin{everbatim}
\xintdeffloatfunc foo(x) := float_dgt(\xintexpr foo(x)\relax);
\end{everbatim}
  The \func{float\string_dgt} wrapper\NewWith{1.4} (which was renamed at
  |1.4e|) is in order for the float variant to produce an already-rounded
  value, possibly speeding-up usage if used as input for other
  functions. Using \func{float} here would work the same but the produced
  function would incorporate a routine to check (at time of use, because at
  time of definition it is impossible to tell what will be the case) if it is
  applied to one or two arguments.

\item And in the reverse direction one can do:
\begin{everbatim}
\xintdeffunc bar(x) := \xintfloatexpr bar(float(x))\relax;
\end{everbatim}
  With this the transplanted float-function will expand in \csbxint{expr} as
  it would have in \csbxint{floatexpr}, i.e.\@ using float operations; this is
  different from declaring the function again with the same expression as used
  for the original, as it would have then been parsed with a mapping of infix
  operators to the macros doing the exact operations, not the floating point
  ones.

  The inner \func{float} above is not mandatory but recommended: the macro
  associated to the user float function |bar(x)| may use many times its
  argument |x| and does not worry about rounding it, because its expectation
  is that it is already rounded; but in \csbxint{expr} that value could very
  well be a fraction |19/13| and its float rounding will be done again by each
  float macro receiving it as argument; with a \func{float} used as above this
  will have already been done once and the ulterior roundings are faster: they
  have nothing to do apart from realizing that they have nothing to do.... One
  can also use \func{sfloat}, this would serve to nothing for the |19/13| case
  but would possibly for a short integer input involved in multiplications.

  Here it is not needed to use \func{float\string_dgt}, because it will be
  identified at time of definition that \func{float} is used without optional
  argument.

  An external \func{float\string_dgt} could be added but is not a priori
  necessary, except perhaps if the |bar()| function has been defined at a low
  level using support macros producing output with garbage extra digits, which
  usually would be rounded out in input to other float functions.

\end{itemize}


A function once declared is a first class citizen, its
expression is entirely parsed and converted into a big nested \fexpan dable
macro. 

When used its action is via this defined macro. For example
\begin{everbatim*}
\xintdeffunc
     e(z):=(((((((((z/10+1)z/9+1)z/8+1)z/7+1)z/6+1)z/5+1)z/4+1)z/3+1)z/2+1)z+1;
\end{everbatim*}
creates a macro whose meaning one can find in the log file, after
\csbxint{verbosetrue}. Here it is (it has at |1.4| an extra external brace pair
compared to what happened with earlier releases):
\begin{everbatim}
    Function e for \xintexpr parser associated to \XINT_expr_userfunc_e with me
aning macro:#1->{\xintAdd {\xintMul {\xintAdd {\xintDiv {\xintMul {\xintAdd {\x
intDiv {\xintMul {\xintAdd {\xintDiv {\xintMul {\xintAdd {\xintDiv {\xintMul {\
xintAdd {\xintDiv {\xintMul {\xintAdd {\xintDiv {\xintMul {\xintAdd {\xintDiv {
\xintMul {\xintAdd {\xintDiv {\xintMul {\xintAdd {\xintDiv {#1}{10}}{1}}{#1}}{9
}}{1}}{#1}}{8}}{1}}{#1}}{7}}{1}}{#1}}{6}}{1}}{#1}}{5}}{1}}{#1}}{4}}{1}}{#1}}{3}
}{1}}{#1}}{2}}{1}}{#1}}{1}}
\end{everbatim}

The above is not entirely true.\CHANGED{1.4} At |1.4|, \csbxint{deffunc} is
more powerful and digests more of the syntax but it may have to store it in
such a way that usage will be done via a sub-expression: hence it is not the
case that the original expression has been \emph{entirely} parsed. See
\csbxint{NewFunction} for related discussion.

The main difficulty of \csbxint{deffunc} is with the pseudo-functions
\func{seq}, \func{iter}, etc..., which admit the keywords \keyword{omit},
\keyword{abort}, \func{break}. We have no alternative for them, if the
iterated over values are not entirely numerical than to postpone expansion,
but this means simply storing for later a possibly big sub-expression.

At |1.4| we did some obstinate work to make this working but:
\begin{itemize}
\item this means that the stored function body has not been entirely parsed,
  parsing will happen on the fly at each execution for small or large bits,
\item there remains a main stumbling-block. If the variables used in the
  function declaration are used only in the iterated over values or the
  initial values, then the mechanism may work. If however they are used not
  only in those values iterated over but directly in the expression which the
  generators map to the iterated over values, then it will break
  certainly. Indeed at this stage the variables are simply names, and it is
  impossible to transfer the mechanism which converts these names into
  numerical arguments for delayed usage by the declared function. Except if
  one is ready to basically freeze the entire thing; which then is not any
  different at all than using \csbxint{NewFunction}.
\end{itemize}

Conclusion: if some \csbxint{deffunc} break, check if it does not fit the
above criterion before reporting...\@ and recall \csbxint{NewFunction} is your
friend. It has the big advantage of declaring a function for all parsers
simultaneously!

A special note on \func{subs}: it is and has always been hopeless in \csbxint{deffunc}
context. All it does (if it works at all) after being malaxed by
\csbxint{deffunc} is to copy over at the indicated places the \emph{recipe} to
compute something. Thus at every location where that something is needed it will be
evaluated from scratch again. Yes, this is disappointing. But...\@ on the other
hand the more general \func{seq} does work, or pretends to work. Let me
illustrate to make thinks clear. We start with this:
\begin{everbatim*}
\xintverbosetrue
\xintdeffunc foo(x,y,z) = subs(S + S^2, S = x+y+z);
\xintdeffunc bar(x,y,z) = seq(S + S^2, S = x+y+z);
\xintexpr foo(100,10,1), bar(100,10,1)\relax
\xintverbosefalse
\end{everbatim*}

It produces in the log:
\begin{everbatim}
Package xintexpr Info: (on line 10)
    Function foo for \xintexpr parser associated to \XINT_expr_userfunc_foo wit
h meaning macro:#1#2#3->{\xintAdd {\xintAdd {\xintAdd {#1}{#2}}{#3}}{\xintPow {
\xintAdd {\xintAdd {#1}{#2}}{#3}}{2}}}
Package xintexpr Info: (on line 11)
    Function bar for \xintexpr parser associated to \XINT_expr_userfunc_bar wit
h meaning macro:#1#2#3->\expanded \bgroup \expanded {\unexpanded {\XINT_expr_se
q:_b {\xintbareeval S + S^2\relax !S}}{\xintAdd {\xintAdd {#1}{#2}}{#3}}^}
\end{everbatim}
Even without understanding all details one sees that in the first case the
|\xintAdd {\xintAdd {#1}{#2}}{#3}}| appears twice, and in the second case only
once.  But in the second case we have a yet to evaluate expression. So the
second approach is not much different in its effect than using the more
simple-minded \csbxint{NewFunction}. Besides one gets a feeling why the
function arguments can not appear in the expression but only in the iterated
over values, because there is no way to understand what |x|, |y|, |z| are
supposed to mean without adding extra structure showing they map to |#1|,
|#2|, |#3|.

The above remarks apply to \func{subsm} and \func{subsn}. Even if they do work
in \csbxint{deffunc} context (warning, testing at |1.4| release has remained
minimal), they will not bring added efficiency if the substituted values are
to be used multiple times. They may still be useful to visually simplify the
input of a big expression by expressing it in terms of smaller constituents.

Another workaround if one wants genuine (not «macro»-) functions for some
expression where the same thing is used multiple times is to define helper
functions computing the intermediate data. One can see illustrations of this
in the code source of \xinttrigname (or in the matrix multiplication example
at the end of this chapter).

\subsubsection{\csh{xintdefiifunc}}
\label{xintdefiifunc}

With \csbxint{deffunc} the created function is known by the \csbxint{expr}
parser only. 
For usage in the \csbxint{iiexpr} parser, it is required to use
\csa{xintdefiifunc}.

\subsubsection{\csh{xintdeffloatfunc}}
\label{xintdeffloatfunc}

With \csbxint{deffunc} the created function is known by the \csbxint{expr}
parser only. For usage in the \csbxint{floatexpr} parser, it is required to use
\csa{xintdeffloatfunc}.

Note: the optional argument |[Q]| accepted by \csbxint{floatexpr} does not
work with \csbxint{deffloatfunc}. It is still possible to wrap the expression
in |float(expression,Q)|, if it evaluates to a scalar.


\subsubsection{Deprecated: \csh{xintdefefunc}, \csh{xintdefiiefunc},
               \csh{xintdeffloatefunc}}
\label{xintdefefunc}
\label{xintdeffloatefunc}
\label{xintdefiiefunc}

They\CHANGED{1.4} are deprecated and currently only aliases to \csbxint{deffunc} et
al.. Please update your documents as they may be removed at any
time.

\subsubsection{\csh{xintdefufunc}, \csh{xintdefiiufunc}, \csh{xintdeffloatufunc}}
\label{xintdefufunc}
\label{xintdefiiufunc}
\label{xintdeffloatufunc}

This allows to define so-called «Universal functions». This is terminology
borrowed from |NumPy|.\NewWith{1.4}

Here is an example:
\begin{everbatim*}
\xintdefiivar Array = ndmap(lcm, 1..5; 1..10; 1..10);
Array = \xintthealign\xintiiexpr Array\relax
\xintdefiiufunc foo(x) = x^3;
\begin{figure}[htbp]
\caption{Output of a universal function acting on an array}\label{fig:ufunc}
\centeredline{$\vcenter{\xintthealign\xintiiexpr foo(Array)\relax}$}
\end{figure}
See \autopageref{fig:ufunc} for the output.
\end{everbatim*}

The function can be applied to any nested strucure:
\begin{everbatim*}
\xintiiexpr foo([1, [2, [3, [4, [5, 6, 7, 8, 9, 10]]]]])\relax
\end{everbatim*}

It must be defined as function acting on scalars, but its value type is not constrained.
\begin{everbatim*}
\xintdefiivar Array = [1..10];
\xintdefiiufunc foo(x) = [1..x];
\xintthealign\xintiiexpr foo(Array)\relax
\end{everbatim*}

It is even allowed to produce oples and act on oples:
\begin{everbatim*}
\xintdefiivar Ople = 1..10;
\xintdefiiufunc bar(x) = x, x^2, x^3;
\xintiiexpr bar(Ople)\relax
\end{everbatim*}


\subsubsection{Using the same name for both a variable and a function}
\label{sssec:overload}

It is licit to overload a variable name (all Latin letters are predefined as
dummy variables) with a function name and vice versa. The parsers will decide
from the context if the function or variable interpretation must be used
(dropping various cases of tacit multiplication as normally applied).
\begin{everbatim*}
\xintdefiifunc f(x):=x^3;
\xinttheiiexpr add(f(f),f=100..120)\relax\newline
\xintdeffunc f(x,y):=x^2+y^2;
\xinttheexpr mul(f(f(f,f),f(f,f)),f=1..10)\relax
\xintunassigniiexprfunc{f}\xintunassignexprfunc{f}%
\end{everbatim*}

\subsubsection{\csh{xintunassignexprfunc}, \csh{xintunassigniiexprfunc},
  \csh{xintunassignfloatexprfunc}}
\label{xintunassignexprfunc}
\label{xintunassigniiexprfunc}
\label{xintunassignfloatexprfunc}

Function names can be unassigned via \csa{xintunassignexprfunc}\marg{name},
\csa{xintunassigniiexprfunc}\marg{name}, and
\csa{xintunassignfloatexprfunc}\marg{name}.
\begin{everbatim*}
\xintunassignexprfunc{e}
\xintunassignexprfunc{f}
\end{everbatim*}

Warning: no check is done to avoid undefining built-in functions...

\subsubsection{\csh{ifxintverbose} conditional}
\label{xintverbosetrue}
\label{xintverbosefalse}
\label{ifxintverbose}

With |\xintverbosetrue| the meanings of the
functions (or rather their associated macros) will be written to the log. For
example the |Rump| declaration above generates this in the log file:
\begin{everbatim}
    Function Rump for \xintexpr parser associated to \XINT_expr_userfunc_Rump w
ith meaning macro:#1#2->{\xintAdd {\xintAdd {\xintAdd {\xintDiv {\xintMul {1335
}{\xintPow {#2}{6}}}{4}}{\xintMul {\xintPow {#1}{2}}{\xintSub {\xintSub {\xintS
ub {\xintMul {11}{\xintMul {\xintPow {#1}{2}}{\xintPow {#2}{2}}}}{\xintPow {#2}
{6}}}{\xintMul {121}{\xintPow {#2}{4}}}}{2}}}}{\xintDiv {\xintMul {11}{\xintPow
 {#2}{8}}}{2}}}{\xintDiv {#1}{\xintMul {2}{#2}}}}
\end{everbatim}

\begin{framed}
  The meanings written out to the log for more complicated functions may
  sometimes use the same character at different locations but with different
  catcodes.\IMPORTANTf

  It may thus be impossible to retokenize it (even after having removed the
  extra spaces from the added line breaks).\CHANGEDf{1.4}

  This is in contrast with variable values which are always output in the log
  in the benign way, using digits, braces and some characters of catcode 12.
\end{framed}

\subsubsection{\csh{ifxintglobaldefs} conditional}
\label{xintglobaldefstrue}
\label{xintglobaldefsfalse}
\label{ifxintglobaldefs}

If true user defined variables (\csbxint{defvar}, ...) and functions
(\csbxint{deffunc}, ..., \csbxint{NewFunction}) for the expression parsers, as
well as macros obtained via \csbxint{NewExpr} et al.\@ have global scope. If
false (default) they have local scope.

\subsubsection{\csh{xintNewFunction}}
\label{xintNewFunction}

This is syntactic sugar which allows to use notation of functions for what is
nothing more in disguise than a \TeX{} macro. Here is an example:
\begin{everbatim*}
\xintNewFunction {foo}[3]{add(mul(x+i, i=#1..#2),x=1..#3)}
\end{everbatim*}

We now have a genuine function |foo()| of three variables which can
be used in \emph{all three parsers}.
\begin{everbatim*}
\xintexpr seq(foo(0, 3, j), j= 1..10)\relax
\end{everbatim*}

Each time the created «macro-function» |foo()| will be encountered the
corresponding replacement text will get inserted as a sub-expression (of the
same type as the surrounding one), the macro parameters having been replaced
with the (already evaluated) function arguments, and the parser \emph{will
  then have to parse the expression.} It is very much like a macro
substitution, but with parentheses and comma separated arguments (which can be
arbitrary expressions themselves).

It differs fundamentally from \csbxint{deffunc} as it realizes no pre-parsing
whatsoever of the associated sub-expression; using it shortens the input but
not the parsing time (which however is most of the time negligible compared to
actual numerical computations).  Use it for syntax which \csbxint{deffunc}
does not parse successfully.


\subsection{Examples of user defined functions}

\subsubsection{Example with vectors and matrices}
\label{sssec:csv}

This section\CHANGED{1.4} has changed significantly at |1.4| due to the new extended data
types manipulated by the syntax.

Suppose we want to manipulate 3-dimensional vectors, which will be represented
as |nutples| of length 3. And let's add a bit of matrix algebra.
\begin{everbatim*}
\xintdeffunc dprod(V, W) := V[0]*W[0] + V[1]*W[1] + V[2]*W[2];
\xintdeffunc cprod(V, W) := [V[1]*W[2] - V[2]*W[1],
                             V[2]*W[0] - V[0]*W[2],
                             V[0]*W[1] - V[1]*W[0]];
\xintdeffunc Det3(U, V, W) := dprod(cprod(U, V), W);
\xintdeffunc DetMat(M) = Det3(*M);
\xintdeffunc RowMat(U, V, W) := [U, V, W];
\xintdeffunc ColMat(U, V, W) := [[U[0], V[0], W[0]],
                                 [U[1], V[1], W[1]],
                                 [U[2], V[2], W[2]]];
\xintdeffunc MatMul(A, B) := 
 [[A[0,0]*B[0,0]+A[0,1]*B[1,0]+A[0,2]*B[2,0],
   A[0,0]*B[0,1]+A[0,1]*B[1,1]+A[0,2]*B[2,1],
   A[0,0]*B[0,2]+A[0,1]*B[1,2]+A[0,2]*B[2,2]],
  [A[1,0]*B[0,0]+A[1,1]*B[1,0]+A[1,2]*B[2,0],
   A[1,0]*B[0,1]+A[1,1]*B[1,1]+A[1,2]*B[2,1],
   A[1,0]*B[0,2]+A[1,1]*B[1,2]+A[1,2]*B[2,2]],
  [A[2,0]*B[0,0]+A[2,1]*B[1,0]+A[2,2]*B[2,0],
   A[2,0]*B[0,1]+A[2,1]*B[1,1]+A[2,2]*B[2,1],
   A[2,0]*B[0,2]+A[2,1]*B[1,2]+A[2,2]*B[2,2]]];

\xintdefvar vec1, vec2, vec3 := [1, 1, 1], [1, 1/2, 1/4], [1, 1/3, 1/9];
\xintdefvar mat1 = RowMat(vec1, vec2, vec3);
\xintdefvar mat2 = ColMat(vec1, vec2, vec3);
\xintdefvar mat12 = MatMul(mat1,mat2);
\xintdefvar mat21 = MatMul(mat2,mat1);
Some computations (|align| executes multiple times hence we pre-computed!):
\begin{align*}
   M_1      &= \vcenter{\xintthealign \xintexpr mat1\relax}&&\qquad
   M_2 . M_1 = \vcenter{\xintthealign \xintexpr mat21\relax}\\[3\jot]
   M_2      &= \vcenter{\xintthealign \xintexpr mat2\relax}&&\qquad
   M_1 . M_2 = \vcenter{\xintthealign \xintexpr mat12\relax}
\end{align*}
$$
   \det(M_1)     = \xinteval{DetMat(mat1)},\quad
   \det(M_1.M_2) = \xinteval{reduce(DetMat(mat12))},\quad
   \det(M_2.M_1) = \xinteval{reduce(DetMat(mat21))}
$$
\end{everbatim*}%
For some hair-raising experience check the \csbxint{verbosetrue} output in the
log...\@ here is an alternative with two (three, counting |dprod()|) helper
functions:
\begin{everbatim*}
% annoying that Tr also starts Trace, but Spur is available
% well Sp also starts Spectrum. Big problems.
\xintdeffunc Tr(M) := 
     [[M[0,0], M[1,0], M[2,0]],
      [M[0,1], M[1,1], M[2,1]],
      [M[0,2], M[1,2], M[2,2]]];

\xintdeffunc MatMul_a(r1, r2, r3, c1, c2, c3) :=
     [[dprod(r1, c1), dprod(r1, c2), dprod(r1, c3)],
      [dprod(r2, c1), dprod(r2, c2), dprod(r2, c3)],
      [dprod(r3, c1), dprod(r3, c2), dprod(r3, c3)]];

\xintdeffunc MatMul(A, B) := MatMul_a(*A, *Tr(B));
\end{everbatim*}

And once we have the transpose and the scalar product of vectors, we can
simply use \func{ndmap} for a lean syntax (this would extend to arbitrary
dimension):
\begin{everbatim*}
\xintdeffunc MatMul(A, B) = ndmap(dprod, *A; *Tr(B));
\xintdefvar mat1212 = MatMul(mat12, mat12);
\begingroup
\def\xintexprPrintOne       #1{\xintTeXFrac{#1}}%
\def\xintexpralignbegin       {\begin{pmatrix}}%
\def\xintexpralignend         {\end{pmatrix}}%
\def\xintexpralignlinesep     {\noexpand\\[2\jot]}% needed to counteract an internal \expanded
\def\xintexpraligninnersep    {&}%
\let\xintexpralignleftbracket\empty  \let\xintexpralignleftsep\empty
\let\xintexpralignrightbracket\empty \let\xintexpralignrightsep\empty
$$ \xintthealign \xintexpr mat1\relax \cdot \xintthealign \xintexpr mat2\relax \cdot
   \xintthealign \xintexpr mat1\relax \cdot \xintthealign \xintexpr mat2\relax =
   \xintthealign \xintexpr mat12\relax ^2 = \xintthealign \xintexpr mat1212\relax$$
$$ \det(M_1\cdot M_2 \cdot M_1 \cdot M_2) = \xinteval{reduce(DetMat(mat1212))}$$
\endgroup
\end{everbatim*}

\subsubsection{Example with the \textsc{Rump} test}
\label{sssec:Rump}

Let's try out our |Rump()| function:
\begin{everbatim*}
\xinttheexpr Rump(77617,33096)\relax.
\end{everbatim*}
Nothing problematic for an \emph{exact} evaluation, naturally!

Thus to test the \textsc{Rump} polynomial (it is not quite a polynomial with
its |x/2y| final term) with floats, we \emph{must} also
declare |Rump| as a function to be used there:
\begin{everbatim*}
\xintdeffloatfunc
    Rump(x,y):=333.75 y^6 + x^2 (11 x^2 y^2 - y^6 - 121 y^4 - 2) + 5.5 y^8 + x/2y;
\end{everbatim*}

The numbers are scanned with the current precision, hence as here it is
\dtt{16}, they are scanned exactly in this case. We can then vary the
precision for the evaluation.
\begin{everbatim*}
\def\CR{\cr}
\halign
{\tabskip1ex
\hfil\bfseries#&\xintDigits:=\xintiloopindex\relax \xintthefloatexpr Rump(77617,33096)#\cr
\xintiloop [8+1]
\xintiloopindex &\relax\CR
\ifnum\xintiloopindex<40 \repeat
}
\end{everbatim*}

\subsubsection{Examples of recursive definitions}
\label{sssec:recursive}

Recursive definitions \emph{require} using the short-circuit branching
operators.\CHANGED{1.4} Prior to |1.4|, to the contrary it was explained that
one should use the \func{if} or \func{ifsgn} functions and that they would get
converted into macros doing branching in a short-circuit manner. This was a
bit counter-intuitive.
\begin{everbatim*}
\xintdeffunc GCD(a,b):=(b)?{GCD(b,a/:b)}{a};
\end{everbatim*}
This of course is the Euclide algorithm: it will be here applied to variables
which may be fractions. For example:
\begin{everbatim*}
\xinttheexpr GCD(385/102, 605/238)\relax
\end{everbatim*}

There is already a built-in \func{gcd} (which
accepts arbitrarily many arguments):
\begin{everbatim*}
\xinttheexpr gcd(385/102, 605/238)\relax
\end{everbatim*}

Our second example is modular exponentiation:
\begin{everbatim*}
\xintdefiifunc powmod_a(x, m, n) :=
       isone(m)?
           % m=1, return x modulo n
           {   x /: n  }
           % m > 1 test if odd or even and do recursive call
           {   odd(m)? {  x*sqr(powmod_a(x, m//2, n)) /: n }
                       {    sqr(powmod_a(x, m//2, n)) /: n }
            }
         ;
\xintdefiifunc powmod(x, m, n) := (m)?{powmod_a(x, m, n)}{1};
\end{everbatim*}
I have made the definition here for the |\xintiiexpr| parser; we could do the
same for the |\xintexpr|-parser (but its usage with big powers would quickly
create big denominators, think |powmod(1/2, 1000, 1)| for example.)
\begin{everbatim*}
\xinttheiiexpr seq(powmod(x, 1000, 128), x=9, 11, 13, 15, 17, 19, 21)\relax\par
\end{everbatim*}
The function assumes the exponent is non-negative (the Python |pow| behaved
the same until |3.8| release), but zealous users will add the necessary code for negative
exponents, after having defined another function for modular inverse!

If function |A| needs function |B| which needs function |A| start by giving to
|B| some dummy definition, define |A|, then define |B| properly. TODO: add
some example here...

\subsection {Links to some (old) examples within this document}
\label{sec:awesome}

\begin{itemize}
\item The utilities provided by \xinttoolsname (\autoref{sec:tools}), some
  completely expandable, others not, are of independent interest. Their use
  is illustrated through various examples: among those, it is shown in
  \autoref{ssec:quicksort} how to implement in a completely expandable way
  the \hyperref[ssec:quicksort]{Quick Sort algorithm} and also how to illustrate
  it graphically. Other examples include some dynamically constructed
  alignments with automatically computed prime number cells: one using a
  completely expandable prime test and \csbxint{ApplyUnbraced}
  (\autoref{ssec:primesI}), another one with \csbxint{For*} (\autoref{ssec:primesIII}).

\item  One has also a \hyperref[edefprimes]{computation of primes within an
    \csa{edef}} (\autoref{xintiloop}), with the help of \csbxint{iloop}.
  Also with \csbxint{iloop} an
  \hyperref[ssec:factorizationtable]{automatically generated table of
    factorizations} (\autoref{ssec:factorizationtable}).

\item  The code for the title page fun with Fibonacci numbers is given in
  \autoref{ssec:fibonacci} with \csbxint{For*} joining the game.

\item  The computations of \hyperref[ssec:Machin]{ $\pi$ and $\log 2$}
  (\autoref{ssec:Machin}) using \xintname and the computation of the
  \hyperref[ssec:e-convergents]{convergents of $e$} with the further help of
  the \xintcfracname package are among further examples.

\item Also included,
  an \hyperlink{BrentSalamin}{expandable implementation of the Brent-Salamin
    algorithm} for evaluating $\pi$.

\item The \autoref{ssec:PrimesIV} implements expandably the Miller-Rabin
  pseudo-primality test.


\item The functionalities of \xintexprname are illustrated with various
  other examples, in \autoref{xintdeffunc}, 
  \hyperlink{ssec:dummies}{Functions with dummy variables},
  \autoref{ssec:moredummies} or \hyperref[sssec:recursive]{Recursive definitions}.
\end{itemize}
% ça va sans dire
% Almost all of the computational results interspersed throughout the
% documentation are not hard-coded in the source file of this document but are
% obtained via the expansion of the package macros during the \TeX{}
% run.%


\clearpage
\etocdepthtag.toc {part1B}

\let\xintexprnameUp\undefined
\csname xinttrignameUp\endcsname

\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xinttrigname package}
\RaisedLabel{sec:trig}

\localtableofcontents

This package provides trigonometric functions for use with \xintexprname.
The sole macro is \csbxint{reloadxinttrig}.

This package was first included in release |1.3e| (|2019/04/05|) of
\xintexprname. It is automatically loaded by \xintexprname.

At |1.4e| (|2021/05/05|) the accuracy was significantly
increased:\CHANGED{1.4e} formerly the high-level user interface used to define
the functions had as consequences that intermediate steps of the computations
could not operate with guard digits, and as a result the last two digits
were most of the time off (at least the last one).  Now, computations are done
internally in extended precision, and the accuracy is high up to the last
digits, with faithful rounding and high probability of correct rounding.  And
the maximal number of digits was raised slightly to \dtt{62} digits.

At \dtt{8} digits a special, faster, mode is used, which is less accurate. But faster.

\textbf{Acknowledgements:} I finally decided to release some such functions
under friendly pressure of Jürgen \textsc{Gilg} and Thomas \textsc{Söll}, let
them both be thanked here.

\subsection{\csh{xintreloadxinttrig}}\label{xintreloadxinttrig}

The library is loaded automatically by \xintexprname at start-up.
It is then configured for \dtt{16} digits.

To work for example with \dtt{48} digits, execute \csbxint{SetDigits*}|{48}|
or \csbxint{Digits*}|:=48;| (the ending |;| can be replaced by a |\relax| in
case of problems due to it being active, e.g.\@ with \LaTeX\ and some languages).

With the non-starred variant \csbxint{Digits}|:=48;|\IMPORTANTf{} it is needed
to issue \csbxint{reloadxinttrig} to recalibrate the functions provided by the
library (and the exponential/logarithm functions will only be updated if also
\csbxint{reloadxintlog} is used).

% Absence of guard digits (whether in the used hard-coded constants or in
% passing over values from one auxiliary function to the next) due to high level
% (user) interface used for the programming means that the produced values are
% definitely expected to be wrong in the last digit or last two digits. I should
% actually give some estimate of the actual maximal error in |ulps| unit, but I
% have not done the complete analysis for lack of time.

% Final computation results should thus probably be printed via
% \csbxint{floateval}|{[-2]....}| in order to strip off (with rounding) the last
% two digits, if one does not like seeing those non-meaningful figures in the
% last one or two positions (I don't say those last two figures are
% \emph{systematically} off). For example, to achieve \dtt{16} digits of
% precision one should work with a precision of 18 digits (being careful to have
% issued \csbxint{reloadxinttrig}) and round results using
% \csbxint{floateval}|{[-2]....}|.

% Another approach is to use \csbxint{ieval}|{[D]...}| for conversion to
% a fixed point format.

% In future, lower level coding will probably replace the high-level interface,
% or at least the macros produced by the high-level interface will be hacked
% into to tell the float macros to work at a somewhat elevated precision.

\subsection{Constants}

Their values (with more digits) get incorporated into the trigonometrical
functions at the time of their definitions during loading or reloading of the
package. They are left free to use, or modified, or \csbxint{unassignvar}'d,
as this will have no impact whatsoever on the functions.

\begin{description}
\vardesc{twoPi} what could that be?
\vardesc{threePiover2} 
\vardesc{Pi}
\vardesc{Piover2}
\vardesc{oneRadian} this is one radian in degrees: $180/\pi$
\vardesc{oneDegree} this is one degree in radian: $\pi/180$
\end{description}

Breaking change at |1.4e|:\CHANGED{1.4e} formerly some variables were
defined to hold the inverse factorials. Not anymore.

\subsection{Functions}

\subsubsection{Direct trigonometry}

With the variable in radians:

\begin{description}
\funcdesc{sin} sine
\funcdesc{cos} cosine
\funcdesc{tan} tangent
\funcdesc{cot} cotangent
\funcdesc{sec} secant
\funcdesc{csc} cosecant
\end{description}

With the variable in degrees:

\begin{description}
\funcdesc{sind} sine
\funcdesc{cosd} cosine
\funcdesc{tand} tangent
\funcdesc{cotd} cotangent
\funcdesc{secd} secant
\funcdesc{cscd} cosecant
\end{description}

Only available with the variable in radians:
\begin{description}
\funcdesc{tg} tangent
\funcdesc{cotg} cotangent
\funcdesc{sinc} cardinal sine $\sinc(x) = \sin(x)/x$
\end{description}

\subsubsection{Inverse trigonometry}

With the value in radians:

\begin{description}
\funcdesc{asin} arcsine
\funcdesc{acos} arccosine
\funcdesc{atan} arctangent
\funcdesc[x, y]{Arg} the main branch of the argument of the complex number
|x+iy|, from $-\pi$ (excluded) to $\pi$ (included). As the output is rounded
-\var{Pi} is a possible return value.
\funcdesc[x, y]{pArg} the branch of the argument of the complex number
|x+iy| with values going from $0$ (included) to $2\pi$ (excluded). Inherent
rounding makes \var{twoPi} a possible return value.
\funcdesc[y, x]{atan2} it is |Arg(x, y)|. Note the reversal of the arguments,
this seems to be the most frequently encountered convention across languages.
\end{description}

With the value in degrees:

\begin{description}
\funcdesc{asind} arcsine
\funcdesc{acosd} arccosine
\funcdesc{atand} arctangent
\funcdesc[x, y]{Argd} the main branch of the argument of the complex number
|x+iy|, from $-180$ (excluded) to $180$ (included). Inherent rounding of
output can cause |-180|
to be returned.
\funcdesc[x, y]{pArgd} the branch of the argument of the complex number
|x+iy| with values going from $0$ (included) to $360$ (excluded). Inherent rounding of
output can cause |360| to be returned.
\funcdesc[y, x]{atan2d} it is |Argd(x, y)|. Note the reversal of the arguments,
this seems to be the most frequently encountered convention across languages.
\end{description}

\subsubsection{Conversion functions (optional definitions left to user
  decision)}

Python provides functions |degrees()| and |radians()|. But as most of the
\xinttrigname functions are already defined for the two units, I felt this was
not really needed. It is a oneliner to add them:
\begin{everbatim}
\xintdeffloatfunc radians(x) := x * oneDegree;
\xintdeffloatfunc degrees(x) := x * oneRadian;
\xintdeffunc radians(x) := float_dgt(x * oneDegree);
\xintdeffunc degrees(x) := float_dgt(x * oneRadian);
\end{everbatim}

The \func{float\string_dgt} does a float rounding to \csbxint{theDigits}
precision (recall that |*| is mapped to exact multiplication in
\csbxint{deffunc}).

\subsection{Important implementation notes}
\label{ssec:trignotes}

\begin{itemize}
\item Currently, \xintname is lacking some dedicated internal representation
  of floats which means that most operations re-parse the digit tokens of their
  arguments to count them\dots\ this does not contribute to efficiency (you
  can load the module under |\xintverbosetrue| regime and see how the nested
  macros look like and get an idea of how many times some rather silly
  re-counting of mantissa lengths will get done!)
\item One should not overwrite some function names which are employed as
  auxiliaries; refer to |sourcexint.pdf|.
\item Floats with large exponents are integers and are multiple of \dtt{1000};
  hence modulo \dtt{360} all such ``angles'' are multiple of \dtt{40} degrees.
  Needless to say that considering usage of the |sind()| and |cosd()| functions
  with such large float numbers is meaningless.
\item See |sourcexint.pdf| for some comments on limitations of the range
  reduction implementation.
% \item Regarding |sin()| and |cos()|, \xinttrigname converts their argument to
%   degrees by multiplication by (pre-rounded) $180/\pi$, then does range
%   reduction modulo $360$ and finally goes back to radians in the appropriate
%   octants to use usual Taylor series (roughly said). For large floats, the
%   output value will thus be one of |sind(40n)|, |cosd(40n)|, |n=0..8|. If the
%   unit in the last place of original variable was for example \dtt{1e9} the
%   final result means nothing at all: the unit in the last place interval
%   extends above possibly astronomical numbers of intervals of length $2\pi$.

%   This intrinsic problem is not a by-product of conversion problems to and
%   from degrees, it is an in-built inadequacy of the concept of floating point
%   numbers to provide meaning to evaluating trigonometrical functions. The
%   argument should be treated as a uniformly distributed random variable modulo
%   $2\pi$, and the sine and cosine values should be random variables realizing
%   the value distribution of these mathematical functions. Clearly this adds
%   some (rather severe) implementation complications such as deciding how to
%   make the transition to randomness. Too lazy for that.

%   Opting for a random value also raises the question of how to deal with
%   multiple such evaluations at the same argument in a single expression. I
%   would argue again that as it is evil to consider meaningless quantities, it
%   is not a problem if new compilations give different results, or even single
%   compilation gives different results in various parts of the same formula,
%   that's the whole point of randomness! As said already, I got too lazy to
%   consider seriously implementing such a non-standard philosophy, despite its
%   compelling soundness.
% \item Did I say the implementation was done at very high level (for the most
%   part), hence has ample room for optimization? This is particularly the case
%   for the handling of small inputs by functions such as sine or arcsine.
\end{itemize}

\subsection{Some example evaluations}

\noindent
\begin{everbatim*}
\xintDigits* := 48\relax
$sind(17)\approx\xintfloateval{sind(17)}$\newline
$cosd(17)\approx\xintfloateval{cosd(17)}$\newline
$tand(17)\approx\xintfloateval{tand(17)}$\newline
$sind(43)\approx\xintfloateval{sind(43)}$\newline
$cosd(43)\approx\xintfloateval{cosd(43)}$\newline
$tand(43)\approx\xintfloateval{tand(43)}$\newline
$asind(0.3)\approx\xintfloateval{asind(0.3)}$\newline
$acosd(0.3)\approx\xintfloateval{acosd(0.3)}$\newline
$atand(3)\approx\xintfloateval{atand(3)}$\newline
$tan(atan(7))\approx\xintfloateval{tan(atan(7))}$\newline
$asind(sind(25))\approx\xintfloateval{asind(sind(25))}$\par\medskip
\noindent\xintDigits* := 24\relax
$sind(17)\approx\xintfloateval{sind(17)}$\newline
$cosd(17)\approx\xintfloateval{cosd(17)}$\newline
$tand(17)\approx\xintfloateval{tand(17)}$\newline
$sind(43)\approx\xintfloateval{sind(43)}$\newline
$cosd(43)\approx\xintfloateval{cosd(43)}$\newline
$tand(43)\approx\xintfloateval{tand(43)}$\newline
$asind(0.3)\approx\xintfloateval{asind(0.3)}$\newline
$acosd(0.3)\approx\xintfloateval{acosd(0.3)}$\newline
$atand(3)\approx\xintfloateval{atand(3)}$\newline
$tan(atan(7))\approx\xintfloateval{tan(atan(7))}$\newline
$asind(sind(25))\approx\xintfloateval{asind(sind(25))}$\par
\xintDigits* := 16\relax
\end{everbatim*}

\clearpage
\let\xinttrignameUp\undefined
\csname xintlognameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}

\section{Macros of the \xintlogname package}
\RaisedLabel{sec:log}

This package provides logarithms, exponentials and fractional powers for use
with \xintexprname.

This package was first included in release |1.3e| (|2019/04/05|) of
\xintexprname. It is automatically loaded by \xintexprname.

At release |1.4e| (|2021/05/05|) it was substantially extended.\CHANGED{1.4e}
Formerly, the functions \func{log10}, \func{pow10}, \func{log}, \func{exp},
and \func{pow} used at their core \ctanpackage{poormanlog} which is
automatically imported, this limited the precision to about \dtt{8} accurate
digits, but was fast.

With |1.4e|, the |Digits| range is extended up to \dtt{62}
digits. \ctanpackage{poormanlog} is still used as boot-strap for logarithms,
it is not used at all for exponentials.

At |Digits| set to \dtt{8} or less, the old faster but less accurate macros
based on \ctanpackage{poormanlog} are used.

\localtableofcontents

\subsection{\csh{xintreloadxintlog}}\label{xintreloadxintlog}

The library is loaded automatically by \xintexprname at start-up.
It is then configured for \dtt{16} digits.

To work for example with \dtt{48} digits, execute \csbxint{SetDigits*}|{48}|
or \csbxint{Digits*}|:=48;| (the ending |;| can be replaced by a |\relax| in
case of problems due to it being active, e.g.\@ with \LaTeX\ and some languages).

With the non-starred variant \csbxint{Digits}|:=48;|\IMPORTANTf{} it is needed
to issue \csbxint{reloadxintlog} to recalibrate the functions provided by the
library (and the trigonometric functions will only be updated if also
\csbxint{reloadxinttrig} is used).

\subsection{Functions}

% All those functions achieve only about \dtt{8} or \dtt{9} digits of precision.
% Notice in particular that the digits beyond the ninth printed by \func{log}
% have no significance (here we suppose |1<x<10|), but I did not add the
% rounding overhead as it is expected anyhow that the final result will be
% appropriately rounded. Notice however that \func{log10} should be seen as
% going from floating point to fixed point (in the sense of the number of
% fractional digits) and \func{pow10} from fixed point to floating point.

\begin{description}
\funcdesc{log10} logarithm in base 10

\funcdesc{pow10} fractional powers of 10

\funcdesc{log} natural logarithm

\funcdesc{exp} exponential function

\funcdesc[x, y]{pow} computes $x^y$ either via the formula |pow10(y*log10(x))|
(applied with some internally increased accuracy), for |y| neither an integer
nor an half-integer; or via the legacy \csbxint{FloatPower} and
\csbxint{FloatSqrt} macros if the exponent is integer or half-integer.
Integer exponents trigger an exact evalution in \csbxint{eval} if
the output will not exceed (or will only slightly exceed) \dtt{10000} digits
(separately for numerator and denominator), else the power is computed in the
floating point sense.
\end{description}

\begin{everbatim*}
\xintfloateval{log(2), exp(1), 2^(1/3), 2^10000}
\end{everbatim*}

\pdfstringdefDisableCommands{\let\cs\empty}

\subsection{Some information on how powers are computed}

For powers |a^b| or |a**b| in \csbxint{floateval} the following rules apply:
\begin{enumerate}[noitemsep]
\item a check is made if exponent is integer or half-integer,
\item if this is the case legacy \csbxint{FloatPower} (combined with
  \csbxint{FloatSqrt} for half-integer case)
  are used to evaluate the power (and |a| can be negative if exponent is
  integer),
\item else the power is computed as |pow10(b*log10(a))| (but keeping some
  extra digits in intermediate evaluations; in particular |b| is not
  float-rounded, but |a| is).
\end{enumerate}
The reason is that the log/exp approach loses accuracy for very big
exponents (say for exponents of the order of \dtt{100000000} or more). Here is
an example of a precise computation with a very large exponent (\dtt{\xintiieval{12^16}}):
\begin{everbatim*}
$\xintTeXfromSci{\xintfloateval{1.00000001^\xintiiexpr 12^16\relax}}$\newline
\xintDigits:=48;%\xintreloadxintlog is not done as log10/pow10 will not be used
$\xintTeXfromSci{\xintfloateval{1.00000001^12^16}}$\newline
\xintDigits:=64;%\xintreloadxintlog is not done as log10/pow10 will not be used
$\xintTeXfromSci{\xintfloateval{1.00000001^12^16}}$\newline
\xintDigits:=80;%\xintreloadxintlog is not done as log10/pow10 will not be used
$\xintTeXfromSci{\xintfloateval{1.00000001^12^16}}$
\xintDigits:=16;%
\end{everbatim*}

Notes:
\begin{itemize}[noitemsep]
\item in the case with \dtt{16} digits precision, we ensured |12^16| got
  computed exactly with all its \dtt{18} digits and was not rounded to only
  \dtt{16} digits (and confirmation is that the result matches the second one
  at \dtt{48} digits),
\item the |1.4g| right associativity
  of powers is taken into account to drop parentheses.
\end{itemize}

As the legacy \csbxint{FloatPower} and \csbxint{FloatSqrt} work in arbitrary
precision, the result for integer or half-integer exponents is produced with a
full-size mantissa, even if |Digits| is more than \dtt{62} (as is examplified
above).

In the |10^(b*log10(a))| branch the mantissa size is limited to the minimum of
|Digits| and of \dtt{64}. Its last digits will start being wrong if |b|
becomes about (in absolute value) \dtt{100000000}. If you really need to
compute powers with exponents that large or larger, it is recommended to
decompose the exponent as a sum of the nearest integer or half-integer and a
fractional part and express the power as a product. This is not done
automatically as it would add some overhead in general for some a priori very
rare use cases.

% And recall that the decimal exponents of final and
% intermediate results should obey the \TeX\ bound for integers anyhow, i.e. be
% at most (in absolute value) \dtt{\number"7FFFFFFF}, add some safety margin...

In \csbxint{eval}, this is as in \csbxint{floateval} but for one difference: integer
exponents will trigger an exact evaluation, as long as:
\begin{itemize}
\item the exponent absolute value is at most \dtt{9999},
\item it is evaluated a priori, based on the length of the input, that the
  output will have at most \dtt{10000} digits (or only a bit more), separately
  for numerator and denominator.
\end{itemize}
The check for integralness of exponent is not on its mathematical value
but on its internal representation, for speed. So |6/3| is not recognized as
being an integer exponent in \csbxint{eval}; but in \csbxint{floateval}, the
|6/3| will have been computed and recognized as |2|. Also |2.00| or |200e-2|
is recognized as an integer in both parsers. Similar remarks apply to
half-integer case.

To compute exactly higher powers than |2^9999| or |9^9999| or |99^5000| or
|999^3333|, etc..., use \csbxint{iieval}.  See \csbxint{iiPow} for related
comments if you don't want to melt your CPU.

If |Digits| is at most \dtt{8}, logarithms are computed faster but
with less accuracy; internally, using \dtt{9} \emph{fixed
  point fractional digits}.  And powers |a^b| lose accuracy in
last digits quickly as |b| rises.  Here is what was observed
with some random tests:
\begin{itemize}[noitemsep]
\item for |b| neither integer nor half-integer and |1<b<10|, 
  roughly \dtt{8} correct digits for between \dtt{80}\% and \dtt{90}\% of
  cases and in the remaining cases only a \dtt{1ulp} error.
\item for |b| neither integer nor half-integer and |10^e<b<10^(e+1)|,
  roughly \dtt{8-e} digits are correct for about
  \dtt{90}\% of cases and there is a one unit error in the last of
  those digits in the remaining cases.
\end{itemize}
To maintain higher accuracy, split the input as |a^n a^h| with |n|
integer or half-integer nearest to |b|. After having considered (and
implemented) the method, decision was made to not incorporate it as it
would induce serious overhead generally speaking. The |a^b| with
fractional exponent |b| such that |abs(b)<10| are currently computed
with at most \dtt{1ulp} error in the vast majority of cases it seems,
which is largely precise enough for plots, and then speed matters
most. Larger exponents can be handled (since |1.4f|) via manually
implementing the splitting trick, as described above.

The documentation of the legacy macro \csbxint{FloatPower} (which is used for
powers with integer and half-integer exponents) explains it has a guaranteed
error bound of |0.52ulp|, in arbitrary precision.  Generally speaking, the
math functions added at |1.4e| target even smaller errors (but only up to
\dtt{62} digits), something of the order of |0.505ulp|, and in practice they
seem to achieve even better than \dtt{99\%} of correct rounding probability
(at least in their natural ranges, and it varies according to the value of
|Digits|).  Perhaps in future I will re-examine whether it is worthwile to
increase a bit the theoretical accuracy of \csbxint{FloatPower}, as I have not
had the time to really measure systematically its pratical accuracy, all
anecdotical evidence showing it is good.

\subsection{Deprecated: \csh{poormanloghack}}
\label{ssec:poormanloghack}

|\poormanloghack| is now a no-op, and will be removed at next major release.


\clearpage
\let\xintlognameUp\undefined
\csname xint bundlenameUp\endcsname

\def\xintRunningHeader{{\inheadertrue\catcode`,12\relax
          \DOCxintfrontpage,
          \csname xint bundlename\endcsname,
          \xintkernelname,
          \xintcorename,
          \xintname,
          \xintfracname,
          \xintbinhexname,
          \xintgcdname,
          \xintseriesname,
          \xintcfracname, 
          \xinttoolsname,
          \DOCxintexprmacros,
          \DOCexamples}}
\markboth{\makebox[0pt]{\xintRunningHeader}}{\makebox[0pt]{\xintRunningHeader}}

\etocdepthtag.toc {macros}
\addtocontents{toc}{\gdef\string\sectioncouleur{{joli}}}
\addtocontents{toc}{\gdef\string\SKIPSECTIONINTERSPACE{\kern\smallskipamount}}
\renewcommand{\etocaftertochook}{\addvspace{\bigskipamount}}

\part{The macro layer for expandable computations: \xintcorename, \xintname,
  \xintfracname,...\@}
\RaisedLabel[15]{sec:bundle}


\begin{framed}
  WARNING !

  The documentation is getting old, and is in need of rewrites for many
  sections, particularly for examples.

  We do try to keep updated the description of macros
  provided by the packages.
\end{framed}

\etocsetnexttocdepth{section}
\localtableofcontents

\section{The \xintname bundle}

\localtableofcontents

\subsection{Characteristics}

\begin{framed}
  The main characteristics are:
  \begin{enumerate}
  \item exact algebra on ``big numbers'', integers as well as
    fractions,
  \item floating point variants with user-chosen precision,
  \item the computational macros are compatible with expansion-only context,
  \item the bundle comes with parsers (integer-only, or handling fractions, or
    doing floating point computations) of infix operations implementing
    beyond infix operations extra features such as dummy variables.
  \end{enumerate}


  Since |1.2| ``big numbers'' must have less than about \dtt{19950} digits:
  the maximal number of digits for addition is at \dtt{19968} digits, and it
  is \dtt{19959} for multiplication. The reasonable range of use of the
  package is with numbers of up to a few hundred digits.\footnotemark

  \TeX\ does not know off-hand how to print on the page such very long
  numbers, see \autoref{ssec:printnumber}.
\end{framed}
\footnotetext{For example multiplication of integers having from \dtt{50} to
  \dtt{100} digits takes roughly of the order of the millisecond on a 2012
  desktop computer. I compared this to using Python3: using timeit module on a
  wrapper defined as |return w*z| with random integers of \dtt{100} digits, I
  observe on the same computer a computation time of roughly $4.10^{-7}$s per
  call. And with |return str(w*z)| then this becomes more like $16.10^{-7}$s
  per call. And with |return str(int(W)*int(Z))| where |W| and |Z| are
  strings, this becomes about $26.10^{-7}$s (I am deliberately ignoring
  Python's Decimal module here...) Anyway, my sentence from earlier version of
  this documentation: \emph{this is, I guess, at least about 1000 times slower
    than what can be expected with any reasonable programming language,} is
  about right. I then added: \emph{nevertheless as compilation of a typical
    \LaTeX\ document already takes of the order of seconds and even dozens of
    seconds for long ones, this leaves room for reasonably many computations
    via \xintexprname or via direct use of the macros of
    \xintname/\xintfracname.}}

Integers with only $10$ digits and starting with a $3$ already exceed the
\TeX{} bound; and \TeX{} does not have a native processing of floating point
numbers (multiplication by a decimal number of a dimension register is allowed
--- this is used for example by the
\href{http://mirrors.ctan.org/graphics/pgf/base}{pgf} basic math engine.)

\TeX{} elementary operations on numbers are done via the non-expandable
\emph{\char92advance, \char92multiply, \emph{and} \char92divide} assignments.
This was changed with \eTeX{}'s |\numexpr| which does expandable computations
using standard infix notations with \TeX{} integers. But \eTeX{} did not
modify the \TeX{} bound on acceptable integers, and did not add floating point
support.

The \ctanpackage{bigintcalc} package by
\textsc{Heiko Oberdiek} provided expandable macros (using some of |\numexpr|
possibilities, when available) on arbitrarily big integers, beyond the \TeX{}
bound. It does not provide an expression parser.%
%
\footnote{One can currently use package
  \href{http://ctan.org/pkg/bnumexpr}{bnumexpr} to associate the |bigintcalc|
  macros with an expression parser. This may be unavailable in future if
  |bnumexpr| becomes more tightly associated with future evolutions or
  variants of \xintcorename.}
%
\xintname did it again using more of |\numexpr| for higher speed, and in a
later evolution added handling of exact fractions, of scientific numbers, and
an expression parser. Arbitrary precision floating points operations were
added as a derivative, and not part of the initial design goal.

The concept of signed infinities, signed zeroes, |NaN|'s, error
traps\dots,\footnote{The latter exist as work-in-progress for some time in the
  source code.} have not been implemented, only the notion of `scientific
notation with a given number of significant figures'.%
%
\footnote{Multiplication of two floats with |P=\xinttheDigits| digits is
  first done exactly then rounded to |P| digits, rather than using a
  specially tailored multiplication for floating point numbers which
  would be more efficient (it is a waste to evaluate fully the
  multiplication result with |2P| or |2P-1| digits.)}

The \LaTeX3 project has implemented expandably floating-point computations with
\dtt{16} significant figures
(\href{https://ctan.org/pkg/l3kernel}{l3fp}), including
functions such as exp, log, sine and cosine.\footnote{at the time of writing (2014/10/28) the
  \href{https://ctan.org/pkg/l3kernel}{l3fp} (exactly represented) floating
  point numbers have their exponents limited to $\pm$\dtt{9999}.}
%

More directly related to the \xintname bundle there is the \liiibigint{}
package, also devoted to big integers and in development a.t.t.o.w (2015/10/09,
no division yet). It is part of the experimental trunk of the
\href{http://latex-project.org}{\LaTeX3 Project} and provides an expression
parser for expandable arithmetic with big integers. Its author Bruno
\textsc{Le Floch} succeeded brilliantly into implementing expandably the
Karatsuba multiplication algorithm and he achieves \emph{sub-quadratic growth
  for the computation time}. This shows up very clearly with numbers having
thousands of digits, up to the maximum which a.t.t.o.w is at $8192$ digits.


The \liiibigint{} multiplication from late |2015| is observed to be roughly
|3x--4x| faster than the one from \csbxint{iiexpr} in the range of \dtt{4000}
to \dtt{5000} digits integers, and isn't far from being |9x| faster at
\dtt{8000} digits. On the other hand \csbxint{iiexpr}'s multiplication is
found to be on average roughly |2.5x| faster than \liiibigint's for numbers up
to \dtt{100} digits and the two packages achieve about the same speed at
\dtt{900} digits: but each such multiplication of numbers of \dtt{900} digits
costs about one or two tenths of a second on a 2012 desktop computer, whereas
the order of magnitude is rather the |ms| for numbers with \dtt{50--100}
digits.\footnote{I have tested this again on |2016/12/19|, but the macros have
  not changed on the \liiibigint{} side and barely on the \xintcorename side,
  hence I got again the same results\dots}

Even with the superior \liiibigint{} Karatsuba multiplication it takes about
|3.5s| on this 2012 desktop computer for a single multiplication of two
\dtt{5000}-digits numbers. Hence it is not possible to do routinely such
computations in a document. I have long been thinking that without the
expandability constraint much higher speeds could be achieved, but perhaps I
have not given enough thought to sustain that optimistic stance.\footnote{The
  \ctanpackage{apnum} package implements
  (non-expandably) arbitrary precision fixed point algebra and (v1.6)
  functions exp, log, sqrt, the trigonometrical direct and inverse functions.}

I remain of the opinion that if one really wants to do computations with
\emph{thousands} of digits, one should drop the expandability requirement.
Indeed, as clearly demonstrated long ago by the
\href{https://ctan.org/pkg/pi}{pi computing file} by \textsc{D. Roegel} one
can program \TeX{} to compute with many digits at a much higher speed than
what \xintname achieves: but, direct access to memory storage in one form or
another seems a necessity for this kind of speed and one has to renounce at
the complete expandability.%
%
\footnote{The Lua\TeX{} project possibly makes endeavours such as \xintname
  appear even more insane that they are, in truth: \xintname is able to handle
  fast enough computations involving numbers with less than one hundred digits
  and brings this to all engines.}

\subsection{Floating point evaluations}
\label{ssec:floatingpoint}

Floating point macros are provided by package \xintfracname to work with a
given arbitrary precision |P|. The default value is $P=16$ meaning that the
significands of the produced (non-zero) numbers have \dtt{16} decimal digits.
The syntax to set the precision to |P| is
%
\centeredline{|\xintDigits:=P\relax|}
%
The value is local to the group or environment (if using \LaTeX). To query the
current value use \csbxint{theDigits}.

Most floating point macros accept an optional first argument |[P]| which then
sets the target precision and replaces the |\xintDigits| assigned value (the
|[P]| must be repeated if the arguments are themselves \xintfracname macros
with arguments of their own.) In this section |P| refers to the prevailing
|\xinttheDigits| float precision or to the target precision set in this way as
an optional argument.

\csbxint{floatexpr}|[Q]...\relax| also admits an optional argument |[Q]| but
it has an altogether different meaning: the computations are always done with
the prevailing |\xinttheDigits| precision and the optional argument |Q| is
used for the final rounding. This makes sense only if |Q<\xinttheDigits| and
is intended to clean up the result from dubious last digits (when |Q<0| it
indicates rather by how many digits one should reduce the mantissa lengths via a
final rounding).


 




\begin{framed}
  The |IEEE 754|\footnotemark\ requirement of \emph{correct rounding} for
  addition, subtraction, multiplication, division and square root is achieved
  (in arbitrary precision) by the macros of \xintfracname hence also by the
  infix operators |+|, |-|, |*|, |/|.

    This means that for operands given with at most |P| significant digits
    (and arbitrary exponents) the output coincides exactly with the rounding
    of the exact theoretical result (barring overflow or underflow).


{\footnotesize Due to a typographical oversight, this documentation
    (up to |1.2j|) adjoined |^| and |**| to the above list of
    infix operators. But as
    is explained in \autoref{xintFloatPower}, what is guaranteed regarding
    integer powers is an error of at most |0.52ulp|, not the correct rounding.
    Half-integer powers are computed as square roots of integer powers.\par }%

  The rounding mode is ``round to nearest, ties away from zero''.
  It is not customizable.

  Currently \xintfracname has no notion of |NaN|s or signed infinities or signed
  zeroes, but this is intended for the future.
\end{framed}
%
\footnotetext{The |IEEE 754-1985| standard was for hardware implementations of
  binary floating-point arithmetic with a specific value for the precision
  ($24$ bits for single precision, $53$ bits for double precision). The newer
  {\texttt{IEEE 754-2008}}
  (\url{https://en.wikipedia.org/wiki/IEEE_floating_point}) normalizes five
  basic formats, three binaries and two decimals ($16$ and $34$ decimal
  digits) and discusses extended formats with higher precision. These
  standards are only indirectly relevant to libraries like \xintname dealing
  with arbitrary precision.%
}


Since release
|1.2f|, square root extraction achieves correct rounding in arbitrary
precision.

See \xintlogname for fractional powers and \xinttrigname for trigonometrical
functions.


The maximal floating point decimal exponent is currently
\dtt{\number"7FFFFFFF} which is the maximal number handled by \TeX. The
minimal exponent is its opposite. But this means that overflow or underflow
are detected only via low-level |\numexpr| arithmetic overflows which are
basically un-recoverable. Besides there are some border effects as the
routines need to add or subtract lengths of numbers from exponents, possibly
triggering the low-level overflows. In the future not only the Precision but
also the maximal and minimal exponents |Emin| and |Emax| will be specifiable
by the user.

Since |1.2f|, the float macros round their inputs to the target precision |P|
before further processing. Formerly, the initial rounding was done to |P+2|
digits (and at least |P+3| for the power operation.)

The more ambitious model would be for the computing macros to obey the
intrinsic precision of their inputs, i.e.\@ to compute the correct rounding to
|P| digits of the exact mathematical result corresponding to inputs allowed to
have their own higher precision.%
%
\footnote{The |MPFR| library
  \url{http://www.mpfr.org/} implements this but it does not know fractions!}
%
This would be feasible by \xintfracname which after all knows how to compute
exactly, but I have for the time being decided that for reasons of efficiency,
the chosen model is the one of rounding inputs to the target precision first.

The float macros of \xintfracname have to handle inputs which
not only may have much more digits than the target float precision, but may
even be fractions: in a way this means infinite precision.

From releases |1.08a| to |1.2j| a fraction input $AeM/BeN$ had its numerator
and denominator $A$ and $B$ truncated to |Q+2| digits of precision, then the
substituted fraction was correctly rounded to |Q| digits of precision (usually
with |Q| set to |P+2|) and then the operation was implemented on such rounded
inputs. But this meant that two fractions representing the same rational
number could end up being rounded differently (with a difference of one unit
in the last place), if it had numerators and denominators with at least |Q+3|
digits.

Starting with release |1.2k| a fractional input $AeM/BeN$ is handled
intrinsically: the fraction, independently of its representation $AeM/BeN$, is
\emph{correctly rounded} to |P| digits during the input parsing. Hence the
output depends only on its arguments as mathematical fractions and not on
their representatives as quotients.

Notice that in float expressions, the |/| is treated as operator, and is
applied to arguments which are generally already |P|-floats, hence the above
discussion becomes relevant in this context only for the special input form
|qfloat(A/B)| or when using a sub-expression |\xintexpr A/B\relax| embedded in
the float expression with |A| or |B| having more digits than the prevailing
float precision |P|.






\subsection{Expansion matters}

\subsubsection{Full expansion of the first token}
\label{ssec:expansions}

The whole business of \xintname is to build upon |\numexpr| and handle
arbitrarily large numbers. Each basic operation is thus done via a macro:
\csbxint{iiAdd}, \csbxint{iiSub}, \csbxint{iiMul}, \csbxint{iiDivision}. In
order to handle more complex operations, it must be possible to nest these
macros.
%
An expandable macro can not execute a |\def| or an |\edef|. But the macro must
expand its arguments to find the digits it is supposed to manipulate. \TeX{}
provides a tool to do the job of (expandable !) repeated expansion of the
first token found until hitting something non expandable, such as a digit, a
|\def| token, a brace, a |\count| token, etc...\@ is found. A space token also
will stop the expansion (and be swallowed, contrarily to the non-expandable
tokens).

By convention in this manual \fexpan sion (``full expansion'' or ``full first
expansion'') will be this \TeX{} process of expanding repeatedly the first
token seen. For those familiar with \LaTeX3 (which is not used by \xintname)
this is what is called in its documentation full expansion (whereas expansion
inside |\edef| would be described I think as ``exhaustive'' expansion).

Most of the package macros, and all those dealing with computations%
%
\footnote{except \csbxint{XTrunc}.},
%
are expandable in the strong sense that they expand to their final result via
this \fexpan sion. This will be signaled in their descriptions via a
\etype{}star in the margin.

These macros not only have this property of \fexpan dability, they all begin
by first applying \fexpan sion to their arguments. Again from \LaTeX3's
conventions this will be signaled by a%
%
\ntype{{\setbox0 \hbox{\Ff}\hbox to \wd0 {\hss f\hss}}}
%
margin annotation next to the description of the arguments.

\subsubsection{Summary of important expandability aspects}

\begin{enumerate}
\item the macros \fexpan d their arguments, this means that they expand the
  first token seen (for each argument), then expand, etc..., until something
  un-expandable such as a\strut{} digit or a brace is hit against. This
  example
%
  \leftedline{|\def\x{98765}\def\y{43210}| |\xintiiAdd {\x}{\x\y}|}
%
  is \emph{not} a legal construct, as the |\y| will remain untouched by
  expansion and not get converted into the digits which are expected by the
  sub-routines of |\xintiiAdd|. It is a |\numexpr| which will expand it and an
  arithmetic overflow will arise as |9876543210| exceeds the \TeX{} bounds.
  The same would hold for |\xintAdd|.

  \begingroup\slshape
  To the contrary \csbxint{theiiexpr} and others have no issues with
  things such as |\xinttheiiexpr \x+\x\y\relax|.\hfill
  \endgroup

\item\label{fn:expansions} using |\if...\fi| constructs \emph{inside} the
  package macro arguments requires suitably mastering \TeX niques
  (|\expandafter|'s and/or swapping techniques) to ensure that the \fexpan sion
  will indeed absorb the \csa{else} or closing \csa{fi}, else some error will
  arise in further processing. Therefore it is highly recommended to use the
  package provided conditionals such as \csbxint{ifEq}, \csbxint{ifGt},
  \csbxint{ifSgn},\dots\ or, for \LaTeX{} users and when dealing
  with short integers the
  \ctanpackage{etoolbox}%
%
\footnote{\url{https://ctan.org/pkg/etoolbox}}
  expandable conditionals (for small integers only) such as \texttt{\char92
    ifnumequal}, \texttt{\char92 ifnumgreater}, \dots . Use of
  \emph{non-expandable} things such as \csa{ifthenelse} is impossible inside the
  arguments of \xintname macros.

  \begingroup\slshape
  One can use naive |\if..\fi| things inside an \csbxint{theexpr}-ession
  and cousins,  as long as the test is
  expandable, for example\upshape
%
\leftedline{|\xinttheiexpr\ifnum3>2 143\else 33\fi
  0^2\relax|$\to$\dtt{\xinttheiexpr \ifnum3>2 143\else 33\fi 0^2\relax
    =1430\char`\^2}}
%
  \endgroup

\item after the definition |\def\x {12}|, one can not use
  {\color{blue}|-\x|} as input to one of the package macros: the \fexpan sion
  will act only on the minus sign, hence do nothing. The only way is to use the
  \csbxint{Opp} macro (or \csbxint{iiOpp} which is integer only)
  which obtains the opposite of a given number.

  \begingroup\slshape
  Again, this is otherwise inside an \csbxint{theexpr}-ession or
  \csbxint{thefloatexpr}-ession. There, the
  minus sign may prefix macros which will expand to numbers (or parentheses
  etc...)
  \endgroup

\def\x {12}%
\def\AplusBC #1#2#3{\xintAdd {#1}{\xintMul {#2}{#3}}}%

\item \label{item:xpxp} With the definition
%
\leftedline{|\def\AplusBC #1#2#3{\xintAdd {#1}{\xintMul {#2}{#3}}}|}
%
one obtains an
  expandable macro producing the expected result, not in two, but rather in
  three steps: a first expansion is consumed by the macro expanding to its
  definition. As the package macros expand their arguments until no more is
  possible (regarding what comes first), this |\AplusBC| may be used inside
  them: {|\xintAdd {\AplusBC {1}{2}{3}}{4}|} does work and returns
  \dtt{\xintAdd {\AplusBC {1}{2}{3}}{4}}.

  If, for some reason, it is important to create a macro expanding in two steps
  to its final value, one may either do:
%
\smallskip
%
\leftedline {|\def\AplusBC #1#2#3{\romannumeral-`0\xintAdd {#1}{\xintMul
      {#2}{#3}}}|}
%
or use the \emph{lowercase} form of \csa{xintAdd}:
%
\smallskip
%
\leftedline {|\def\AplusBC #1#2#3{\romannumeral0\xintadd {#1}{\xintMul
      {#2}{#3}}}|}

  and then \csa{AplusBC} will share the same properties as do the
  other \xintname `primitive' macros.

\item
The |\romannumeral0| and |\romannumeral-`0| things above look like an invitation
to hacker's territory; if it is not important that the macro expands in two
steps only, there is no reason to follow these guidelines. Just chain
arbitrarily the package macros, and the new ones will be completely expandable
and usable one within the other.

Since release |1.07| the \csbxint{NewExpr} macro automatizes the creation of
such expandable macros:
%
\leftedline{|\xintNewExpr\AplusBC[3]{#1+#2*#3}|}
%
creates the |\AplusBC| macro doing the above and expanding in two expansion
steps.

\item In the expression parsers of \xintexprname such as
  \csbxint{expr}|..\relax|, \csbxint{floatexpr}|..\relax| the contents are
  expanded completely from left to right until the ending |\relax| is found
  and swallowed, and spaces and even (to some extent) catcodes do not matter.

\item For all variants, prefixing with \csbxint{the} allows to print the
  result or use it in other contexts. Shortcuts \csbxint{theexpr},
  \csbxint{thefloatexpr}, \csbxint{theiiexpr}, \dots\ are available.

\end{enumerate}

\subsection {Input formats for macros}\label{ssec:inputs}

Macros can have different types of arguments (we do not consider here the
\csbxint{expr}-parsers but only the macros of
\xintcorename/\xintname/\xintfracname). In a macro description, a
margin annotation signals what is the argument type.
\begin{enumerate}
\item \TeX\ integers\ntype{\numx} are handled inside a |\numexpr..\relax|
  hence may be count registers or variables. Beware that |-(1+1)| is not legal
  and raises an error, but |0-(1+1)| is. Also |2\cnta| with |\cnta| a |\count|
  isn't legal. Integers must be kept less than \dtt{\number "7FFFFFFF} in
  absolute value, although the \emph{scaling} operation |(a*b)/c| computes the
  intermediate product with twice as many bits.

  The slash |/| does a \fbox{rounded} division which is a fact of life of
  |\numexpr| which I have found very annoying in at least nine cases out of
  ten, not to say ninety-nine cases out of one hundred. Besides, it is at odds
  with \TeX's |\divide| which does a truncated division (non-expandably).

  But to follow-suit |/| also does rounded integer division in
  \csbxint{iiexpr}|..\relax|, and the operator |//| does there the truncated
  division.

\item the strict format\ntype{f} applies to macros handling big integers but
  only \fexpan ding their arguments. After this \fexpan sion the input should
  be a string of digits, optionally preceded by a unique minus sign. The first
  digit can be zero only if it is the only digit. A plus sign is not accepted.
  |-0| is not legal in the strict format. Macros of \xintname with a double
  |ii| require this `strict' format for the inputs.

\item the extended integer format\ntype{\Numf} applies when the macro parses
  its arguments via \csbxint{Num}. The input may then have arbitrarily many
  leading minus and plus signs, followed by leading zeroes, and further
  digits. With \xintfracname loaded, \csbxint{Num} is extended to
  accept fractions and its action is to truncate them to integers.

\item the fraction input format\ntype{\Ff} applies to the arguments of
  \xintfracname macros handling genuine fractions. It allows two types
  of inputs: general and restricted. The restricted type is parsed faster,
  but...\@ is restricted.
  \begin{description}
  \item[general:] inputs of the shape |A.BeC/D.EeF|. Example:
\begin{everbatim*}
\noindent\xintRaw{+--0367.8920280e17/-++278.289287e-15}\newline
\xintRaw{+--+1253.2782e++--3/---0087.123e---5}\par
\end{everbatim*}
    The input parser does not reduce fractions to smallest terms.
    Here are the rules of this general fraction format:
    \begin{itemize}
    \item everything is optional, absent numbers are treated as zero, here are
      some extreme cases:
\begin{everbatim*}
\xintRaw{}, \xintRaw{.}, \xintRaw{./1.e}, \xintRaw{-.e}, \xintRaw{e/-1}
\end{everbatim*}
    \item |AB| and |DE| may start with pluses and minuses, then leading
      zeroes, then digits.
    \item |C| and |F| will be given to |\numexpr| and can be anything
      recognized as such and not provoking arithmetic overflow (the lengths of
      |B| and |E| will also intervene to build the final exponent naturally
      which must obey the \TeX{} bound).
    \item the |/|, |.| (numerator and/or denominator) and |e|
      (numerator and/or denominator) are all optional components.
    \item each of |A|, |B|, |C|, |D|, |E| and |F| may arise from \fexpan sion
      of a macro.
    \item the whole thing may arise from \fexpan sion, however the |/|, |.|,
      and |e| should all come from this initial expansion. The |e| of
      scientific notation is mandatorily lowercased.
    \end{itemize}
  \item[restricted:] inputs either of the shape |A[N]| or |A/B[N]|, which
    represents the fraction |A/B| times |10^N|. The whole thing or
    each of |A|, |B|, |N| (but then not |/| or |[|) may arise from \fexpan
    sion, |A| (after expansion) \emph{must} have a unique optional minus sign
    and no leading zeroes, |B| (after expansion) if present \emph{must} be a
    positive integer with no signs and no leading zeroes, |[N]| if present
     will be given to |\numexpr|. Any deviation from the rules above will
     result in errors.
  \end{description}
  Notice that |*|, |+| and |-| contrarily to the |/| (which is treated simply
  as a kind of delimiter) are not acceptable within arguments of this
  type\ntype{\Ff} (see \autoref{sec:useofcount}
   for some exceptions to this.)
\end{enumerate}

Generally speaking, there should be no spaces among the digits in the inputs
(in arguments to the package macros). Although most would be harmless in most
macros, there are some cases where spaces could break havoc.%
\footnote{The \csbxint{Num} macro does not remove spaces between digits beyond
  the first non zero ones; however this should not really alter the subsequent
  functioning of the arithmetic macros, and besides, since \xintcorename 1.2
  there is an initial parsing of the entire number, during which spaces will
  be gobbled. However I have not done a complete review of the legacy code to
  be certain of all possibilities after |1.2| release. One thing to be aware
  of is that \csa{numexpr} stops on spaces between digits (although it
  provokes an expansion to see if an infix operator follows); the exponent for
  \csbxint{iiPow} or the argument of the factorial \csbxint{iiFac} are only
  subjected to such a \csa{numexpr} (there are a few other macros with such
  input types in \xintname). If the input is given as, say |1 2\x| where
  \csa{x} is a macro, the macro \csa{x} will not be expanded by the
  \csa{numexpr}, and this will surely cause problems afterwards. Perhaps a
  later \xintname will force \csa{numexpr} to expand beyond spaces, but I
  decided that was not really worth the effort. Another immediate cause of
  problems is an input of the type |\xintiiAdd{<space>\x}{\y}|, because the
  space will stop the initial expansion; this will most certainly cause an
  arithmetic overflow later when the \csa{x} will be expanded in a
  \csa{numexpr}. Thus in conclusion, damages due to spaces are unlikely if
  only explicit digits are involved in the inputs, or arguments are single
  macros with no preceding space.}
So the best is to avoid them entirely.

This is entirely otherwise inside an |\xintexpr|-ession, where spaces are
ignored (except when they occur inside arguments to some macros, thus
escaping the |\xintexpr| parser). See the \autoref{sec:expr}.

There are also some slighly more obscure expansion types: in particular, the
\csbxint{ApplyInline} and \csbxint{For*} macros from \xinttoolsname apply a
special iterated \fexpan sion, which gobbles spaces, to the non-braced items
(braced items are submitted to no expansion because the opening brace stops
it) coming from their list argument; this is denoted by a special
symbol\ntype{{\lowast f}} in the margin. Some other macros such as
\csbxint{Sum} from \xintfracname first do an \fexpan sion, then treat each
found (braced or not) item (skipping spaces between such items) via the
general fraction input parsing, this is signaled as
here\ntype{f{$\to$}{\lowast\Ff}} in the margin where the signification of the
\lowast{} is thus a bit different from the previous case.

A few macros from \xinttoolsname do not expand, or expand only once their
argument\ntype{n{{\color{black}\upshape, resp.}} o}. This is also
signaled in the margin with notations \`a la \LaTeX3.


\subsection{Output formats of macros}
\label{ssec:outputs}

We do not consider here the \csbxint{expr}-parsers but only the macros from \xintcorename, \xintname and \xintfracname. Macros of other
components of the bundle may have their own output formats, for example for
continuous fractions with \xintcfracname.
There are mainly three types of outputs:%

\begin{itemize}[nosep,listparindent=\leftmarginiii]
\item arithmetic macros from \xintcorename/\xintname deliver integers
 in the strict format as described in the previous section.
\item arithmetic macros from \xintfracname produce on output the strict
fraction format |A/B[N]|, which stands for |(A/B)|$\times$|10^N|, where |A|
and |B| are integers, |B| is positive, and |N| is a ``short'' integer. The
output is not reduced to smallest terms. The |A| and |B| may end with zeroes
(\emph{i.e}, |N| does not represent all powers of ten). The denominator |B| is
always strictly positive. There is no |+| sign. The |-| is always first if
present (i.e.\@ the denominator on output is always positive.) The output will
be expressed as such a fraction even if the inputs are both integers and the
mathematical result is an integer. The |B=1| is not removed.%
%
\footnote{refer to the documentation of \csbxint{PRaw} for an alternative.}
\item macros with |Float| in their names produce on output scientific
format with |P=|\nobreak\csbxint{theDigits} digits, a lowercase |e| and an
exponent |N|. The first digit is not zero, it is preceded by an optional minus
sign and is followed by a dot and |P-1| digits. Trailing zeroes are not
trimmed. There is one exceptional case:
\begin{itemize}[nosep]
\item if the value is mathematically zero, it is output as |0.e0|,
  i.e.\@ zeros after the decimal mark are removed and the exponent is always |0|.
\end{itemize}
Future versions of the package may modify this.
\end{itemize}


\subsection{Count registers and variables}\label{sec:useofcount}

Inside |\xintexpr..\relax| and its variants, a count register or count control
sequence is automatically unpacked using |\number|, with tacit multiplication:
|1.23\counta| is like |1.23*\number\counta|. There
is a subtle difference between count \emph{registers} and count
\emph{variables}. In |1.23*\counta| the unpacked |\counta| variable defines a
complete operand thus |1.23*\counta 7| is a syntax error. But |1.23*\count0|
just replaces |\count0| by |\number\count0| hence |1.23*\count0 7| is like
|1.23*57| if |\count0| contains the integer value |5|.

Regarding now the package macros, there is first the case of arguments having to
be short integers: this means that they are fed to a |\numexpr...\relax|, hence
submitted to a \emph{complete expansion} which must deliver an integer, and
count registers and even algebraic expressions with them like
|\mycountA+\mycountB*17-\mycountC/12+\mycountD| are admissible arguments (the
slash stands here for the rounded integer division done by |\numexpr|). This
applies in particular to the number of digits to truncate or round with, to the
indices of a series partial sum, \dots

The macros allowing the extended format for long numbers or dealing with
fractions will \emph{to some extent} allow the direct use of count
registers and even infix algebra inside their arguments: a count
register |\mycountA| or |\count 255| is admissible as numerator or also as
denominator, with no need to be prefixed by |\the| or |\number|. It is possible
to have as argument an algebraic expression as would be acceptable by a
|\numexpr...\relax|, under this condition: \emph{each of the numerator and
  denominator is expressed with at most \emph{nine}
  tokens}.%
%
\footnote{The |1.2k| and earlier versions manual claimed up to 8
  tokens, but low-level TeX error arose if the |\numexpr...\relax| occupied
  exactly 8 tokens \emph{and} evaluated to zero. With |1.2l| and later, up to
  9 tokens are always safe and one may even drop the ending |\relax|. But
  well, all these explanations are somewhat silly because prefixing by |\the|
  or |\number| is always working with arbitrarily many tokens.}
%
%
\footnote{Attention! in the \LaTeX{} context a
  \csa{value}\texttt{\{countername\}} will behave ok only if it is first in
  the input, if not it will not get expanded, and braces around the name will
  be removed and chaos\IMPORTANT{} will ensue inside a \csa{numexpr}. One
  should enclose the whole input in \csa{the}\csa{numexpr}|...|\csa{relax} in
  such cases.}
%
Important: a slash for rounded division in a |\numexpr| should be written with
braces |{/}| to not be confused with the \xintfracname delimiter between
numerator and denominator (braces will be removed internally and the slash
will count for one token). Example:
|\mycountA+\mycountB{/}17/1+\mycountA*\mycountB|, or |\count 0+\count
2{/}17/1+\count 0*\count 2|.
%
\leftedline{|\cnta 10 \cntb 35 \xintRaw
  {\cnta+\cntb{/}17/1+\cnta*\cntb}|\dtt{->\cnta 10 \cntb 35 \xintRaw
    {\cnta+\cntb{/}17/1+\cnta*\cntb}}}
%
For longer algebraic expressions using
count registers, there are two possibilities:
\begin{enumerate}[nosep]
\item let the numerator and the denominator be presented as |\the\numexpr...\relax|,
\item or as |\numexpr {...}\relax| (the braces are removed during processing;
  they are not legal for |\numexpr...\relax| syntax.)
\end{enumerate}
\everb|@
\cnta 100 \cntb 10 \cntc 1
\xintPRaw {\numexpr {\cnta*\cnta+\cntb*\cntb+\cntc*\cntc+
                    2*\cnta*\cntb+2*\cnta*\cntc+2*\cntb*\cntc}\relax/%
          \numexpr {\cnta*\cnta+\cntb*\cntb+\cntc*\cntc}\relax }
|
\cnta 100 \cntb 10 \cntc 1
%
\leftedline{\dtt{\xintPRaw {\numexpr
      {\cnta*\cnta+\cntb*\cntb+\cntc*\cntc+
                    2*\cnta*\cntb+2*\cnta*\cntc+2*\cntb*\cntc}\relax/%
          \numexpr {\cnta*\cnta+\cntb*\cntb+\cntc*\cntc}\relax }}}

\subsection{Dimension registers and variables}
\label{sec:Dimensions}

\meta{dimen} variables can be converted into (short) integers suitable for the
\xintname macros by prefixing them with |\number|. This transforms a dimension
into an explicit short integer which is its value in terms of the |sp| unit
($1/65536$\,|pt|).
When |\number| is applied to a \meta{glue} variable, the stretch and shrink
components are lost.

For \LaTeX{} users: a length is a \meta{glue} variable, prefixing a
length macro defined by \csa{newlength} with \csa{number} will thus discard
the |plus| and |minus| glue components and return the dimension component as
described above, and usable in the \xintname bundle macros.

This conversion is done automatically inside an
|\xintexpr|-essions, with tacit multiplication implied if prefixed by some
(integral or decimal) number.

One may thus compute areas or volumes with no limitations, in units of |sp^2|
respectively |sp^3|, do arithmetic with them, compare them, etc..., and possibly
express some final result back in another unit, with the suitable conversion
factor and a rounding to a given number of decimal places.

A \hyperref[tableofdimensions]{table of dimensions} illustrates that the
internal values used by \TeX{} do not correspond always to the closest
rounding. For example a millimeter exact value in terms of |sp| units is
\dtt{72.27/10/2.54*65536=\xinttheexpr trunc(72.27/10/2.54*65536,3)\relax ...}
and \TeX{} uses internally \dtt{\number\dimexpr 1mm\relax}|sp| (\TeX{}
truncates to get an integral multiple of the |sp| unit; see at the end of this
section the exact rules applied internally by \TeX).

\begin{figure*}[ht!]
\phantomsection\label{tableofdimensions}
\begingroup\let\ignorespaces\empty
           \let\unskip\empty
           \def\T{\expandafter\TT\number\dimexpr}
           \def\TT#1!{\gdef\tempT{#1}}
           \def\E{\expandafter\expandafter\expandafter
                  \EE\xintexpr reduce(}
           \def\EE#1!{\gdef\tempE{#1}}
\centeredline{\begin{tabular}{%
                >{\bfseries\strut}c%
                c%
                >{\E}c<{)\relax!}@{}%
                >{\xintthe\tempE}r@{${}={}$}%
                >{\xinttheexpr trunc(\tempE,3)\relax...}l%
                >{\T}c<{!}@{}%
                >{\tempT}r%
                >{\xinttheexpr round(100*(\tempT-\tempE)/\tempE,4)\relax\%}c}
   \hline
   Unit&%
   definition&%
   \omit &%
   \multicolumn{2}{c}{Exact value in \texttt{sp} units\strut}&%
   \omit &%
   \omit\parbox{2cm}{\centering\strut\TeX's value in \texttt{sp} units\strut}&%
   \omit\parbox{2cm}{\centering\strut Relative error\strut}\\\hline
  cm&0.01 m&72.27/2.54*65536&&&1cm&&\\
  mm&0.001 m&72.27/10/2.54*65536&&&1mm&&\\
  in&2.54 cm&72.27*65536&&&1in&&\\
  pc&12 pt&12*65536&&&1pc&&\\
  pt&1/72.27 in&65536&&&1pt&&\\
  bp&1/72 in&72.27*65536/72&&&1bp&&\\
  \omit\hfil\llap{3}bp\strut\hfil&1/24 in&72.27*65536/24&&&3bp&&\\
  \omit\hfil\llap{12}bp\strut\hfil&1/6 in&72.27*65536/6&&&12bp&&\\
  \omit\hfil\llap{72}bp\strut\hfil&1 in&72.27*65536&&&72bp&&\\
  dd&1238/1157 pt&1238/1157*65536&&&1dd&&\\
  \omit\hfil\llap{11}dd\strut\hfil&11*1238/1157 pt&11*1238/1157*65536&&&11dd&&\\
  \omit\hfil\llap{12}dd\strut\hfil&12*1238/1157 pt&12*1238/1157*65536&&&12dd&&\\
  sp&1/65536 pt&1&&&1sp&&\\\hline
  \multicolumn{8}{c}{\bfseries\large\TeX{} \strut dimensions}\\\hline
\end{tabular}}
\endgroup
\end{figure*}

There is something quite amusing with the Didot point. According to the \TeX
Book, $1157$\,|dd|=$1238$\,|pt|. The actual internal value of $1$\,|dd| in \TeX{} is $70124$\,|sp|. We can use \xintcfracname to display the list of
centered convergents of the fraction $70124/65536$:
%
\leftedline{|\xintListWithSep{, }{\xintFtoCCv{70124/65536}}|}
%
\xintFor* #1 in {\xintFtoCCv{70124/65536}}\do {$\printnumber{#1}$, }%
and we don't find
$1238/1157$ therein, but another approximant $1452/1357$!

And indeed multiplying $70124/65536$ by $1157$, and respectively $1357$, we find
the approximations (wait for more, later):
%
\leftedline{``$1157$\,|dd|''\dtt{=\xinttheexpr trunc(1157\dimexpr
    1dd\relax/\dimexpr 1pt\relax,12)\relax}\dots|pt|}
%
\leftedline{``$1357$\,|dd|''\dtt{=\xinttheexpr trunc(1357\dimexpr
    1dd\relax/\dimexpr 1pt\relax,12)\relax}\dots|pt|}
%
and we seemingly discover that $1357$\,|dd|=$1452$\,|pt| is \emph{far more
  accurate} than
the \TeX Book formula $1157$\,|dd|=$1238$\,|pt|~!
The formula to compute $N$\,|dd| was
%
\leftedline{|\xinttheexpr trunc(N\dimexpr 1dd\relax/\dimexpr
  1pt\relax,12)\relax}|}
%

What's the catch? The catch is that \TeX{} \emph{does not} compute $1157$\,|dd|
like we just did:%
%
\leftedline{$1157$\,|dd|=|\number\dimexpr 1157dd\relax/65536|%
      \dtt{=\xintTrunc{12}{\number\dimexpr 1157dd\relax/65536}}\dots|pt|}
%
\leftedline{$1357$\,|dd|=|\number\dimexpr 1357dd\relax/65536|%
      \dtt{=\xintTrunc{12}{\number\dimexpr 1357dd\relax/65536}}\dots|pt|}
%
We thus discover that \TeX{} (or rather here, e-\TeX{}, but one can check that
this works the same in \TeX82), uses  $1238/1157$ as a conversion
factor (and necessarily intermediate computations simulate higher precision
than a priori available with  integers less than $2^{31}$ or rather $2^{30}$ for
dimensions). Hence the $1452/1357$ ratio is irrelevant, an artefact
of the rounding (or rather, as we see, truncating) for one |dd| to be
expressed as an integral number of |sp|'s.

Let us now
use |\xintexpr| to compute the value of the Didot point in millimeters, if
the above rule is exactly verified:
%
\leftedline{|\xinttheexpr
 trunc(1238/1157*25.4/72.27,12)\relax|%
  \dtt{=\xinttheexpr trunc(1238/1157*25.4/72.27,12)\relax}|...mm|}
%
This fits very well with the possible values of the Didot point as listed in
the
\href{http://en.wikipedia.org/wiki/Point_%28typography%29#Didot}{Wikipedia Article}.
%
The value $0.376065$\,|mm| is said to be \emph{the traditional value in
  European printers' offices}. So the $1157$\,|dd|=$1238$\,|pt| rule refers to
this Didot point, or more precisely to the \emph{conversion factor} to be used
between this Didot and \TeX{} points.

The actual value in millimeters of exactly one Didot point as implemented in
\TeX{} is
%
\leftedline {|\xinttheexpr trunc(\dimexpr
  1dd\relax/65536/72.27*25.4,12)\relax|}
%
\leftedline{\dtt{=\xinttheexpr trunc(\dimexpr
    1dd\relax/65536/72.27*25.4,12)\relax}|...mm|}
%
The difference of circa $5$\AA\ is arguably tiny!

% 543564351/508000000

By the way the \emph{European printers' offices \emph{(dixit Wikipedia)}
  Didot} is thus exactly
%
\leftedline{|\xinttheexpr reduce(.376065/(25.4/72.27))\relax|%
   \dtt{=\xinttheexpr reduce(.376065/(25.4/72.27))\relax}\,|pt|}
%
and the centered convergents of this fraction are \xintFor* #1 in
{\xintFtoCCv{543564351/508000000}}\do {\dtt{\printnumber{#1}}\xintifForLast{.}{, }} We do
recover the $1238/1157$ therein!

\begin{framed}
  Here is how \TeX\ converts |abc.xyz...<unit>|. First the decimal is
  \emph{rounded} to the nearest integral multiple of |1/65536|, say |X/65536|.
  The |<unit>| is associated to a ratio |N/D|, which represents |<unit>/pt|.
  For the Didot point the ratio is indeed |1238/1157|. \TeX\ \emph{truncates}
  the fraction |XN/D| to an integer |M|. The dimension is represented by |M
  sp|.

\end{framed}


\subsection{\csh{ifcase}, \csh{ifnum}, ...\@ constructs}\label{sec:ifcase}

When using things such as |\ifcase \xintSgn{\A}| one has to make sure to leave
a space after the closing brace for \TeX{} to
stop its scanning for a number: once \TeX{} has finished expanding
|\xintSgn{\A}| and has so far obtained either |1|, |0|, or |-1|, a
space (or something `unexpandable') must stop it looking for more
digits. Using |\ifcase\xintSgn\A| without the braces is very dangerous,
because the blanks (including the end of line) following |\A| will be
skipped and not serve to stop the number which |\ifcase| is looking for.
%
\begin{everbatim*}
\begin{enumerate}[nosep]\def\A{1}
\item \ifcase \xintSgn\A 0\or OK\else ERROR\fi
\item \ifcase \xintSgn\A\space 0\or OK\else ERROR\fi
\item \ifcase \xintSgn{\A} 0\or OK\else ERROR\fi
\end{enumerate}
\end{everbatim*}

In order to use successfully |\if...\fi| constructions either as arguments to
the \xintname bundle expandable macros, or when building up a completely
expandable macro of one's own, one needs some \TeX nical expertise (see also
\autoref{fn:expansions} on page~\pageref{fn:expansions}).

It is thus much to be recommended to use the expandable branching macros,
provided by \xintfracname succh as \csbxint{ifSgn}, \csbxint{ifZero},
\csbxint{ifOne}, \csbxint{ifNotZero}, \csbxint{ifTrueAelseB}, \csbxint{ifCmp},
\csbxint{ifGt}, \csbxint{ifLt}, \csbxint{ifEq},
\csbxint{ifInt}... See their respective documentations. All these conditionals
always have either two or three branches, and empty brace pairs |{}| for
unused branches should not be forgotten.

If these tests are to be applied to standard \TeX{} short integers, it is more
efficient to use (under \LaTeX{}) the equivalent conditional tests from the
\ctanpackage{etoolbox}%
%
\footnote{\url{https://ctan.org/pkg/etoolbox}}
package.

\subsection{No variable declarations are needed}

  There is no notion of a \emph{declaration of a variable}.

  To do a computation and assign its result to some macro |\z|, the user will employ the |\def|, |\edef|, or |\newcommand| (in \LaTeX)
  as usual, keeping in mind that two expansion steps are needed, thus |\edef|
  is initially the main tool:
%
\begin{everbatim*}
\def\x{1729728} \def\y{352827927} \edef\z{\xintiiMul {\x}{\y}}
\meaning\z
\end{everbatim*}

As an alternative to |\edef| the package provides |\oodef| which expands
exactly twice the replacement text, and |\fdef| which applies \fexpan sion to
the replacement text during the definition.
\begin{everbatim*}
\def\x{1729728} \def\y{352827927} \oodef\w {\xintiiMul\x\y} \fdef\z{\xintiiMul {\x}{\y}}
\meaning\w, \meaning\z
\end{everbatim*}

In practice |\oodef| is slower than |\edef|, except for computations ending in
very big final replacement texts (thousands of digits). On the other hand
|\fdef|\IMPORTANT{} appears to be slightly faster than |\edef| already in the
case of expansions leading to only a few dozen digits.

\xintexprname does provide an interface to declare and assign values to
identifiers which can then be used in expressions: \autoref{xintdefvar}.


\subsection{Possible syntax errors to avoid}

\edef\x{\xintMul {3}{5}/\xintMul{7}{9}}

Here is a list of imaginable input errors. Some will cause compilation errors,
others are more annoying as they may pass through unsignaled.
\begin{itemize}
\item using |-| to prefix some macro: |-\xintiiSqr{35}/271|.%
%
\footnote{to the
    contrary, this \emph{is}
    allowed inside an |\xintexpr|-ession.}
\item using one pair of braces too many |\xintIrr{{\xintiiPow {3}{13}}/243}| (the
  computation goes through with no error signaled, but the result is completely
  wrong).
\item things like |\xintiiAdd { \x}{\y}| as the space will cause \csa{x} to be
  expanded later, most probably within a |\numexpr| thus provoking possibly an
  arithmetic overflow.
\item using |[]| and decimal points at the same time |1.5/3.5[2]|, or with a
  sign in the denominator |3/-5[7]|. The scientific notation has no such
  restriction, the two inputs |1.5/-3.5e-2| and |-1.5e2/3.5| are equivalent:
  |\xintRaw{1.5/-3.5e-2}|\dtt{=\xintRaw{1.5/-3.5e-2}},
  |\xintRaw{-1.5e2/3.5}|\dtt{=\xintRaw{-1.5e2/3.5}}.
\item generally speaking, using in a context expecting an integer (possibly
  restricted to the \TeX{} bound) a macro or expression which returns a
  fraction: |\xinttheexpr 4/2\relax| outputs \dtt{\xinttheexpr 4/2\relax},
  not $2$. Use |\xintNum {\xinttheexpr 4/2\relax}| or |\xinttheiexpr 4/2\relax|
  (which rounds the result to the nearest integer, here, the result is already
  an integer) or |\xinttheiiexpr 4/2\relax|. Or, divide in your head |4| by
  |2| and insert the result directly in the \TeX{} source.
\end{itemize}

\subsection{Error messages}

In situations such as division by zero, the \TeX{} run will be interrupted
with some error message.  It conveys some short information on
the cause of the problem, then an optimistic statement about a possible
recovery if the user (in interactive mode) simply hits the |<return>| key.%
\CHANGED{1.4g} 
In non-interactive (|nonstopmode|) the \TeX{} run goes on uninterrupted and
the error data will be found in the compilation log.  Often, \xintname will
fall-back to using a zero value.  This is still an experimental feature.%
%
\footnote{Customizable handlers, error traps, error flags are implemented
  in embryonic form but without user interface since |1.2l| release.  This is
  not ready yet.}
%
\footnote{The |1.4g| new formatting implementation benefited from a May 2021
  thread at the \LaTeX3 site where expandable error messages were discussed,
  with in particular contributions of |@blefloch| and |@Skillmon|.}


The encouragements will be slightly better formatted if the run is with
\LaTeX\ compared to Plain \eTeX: Plain by default does not set the
|\newlinechar| which allows to issue linebreaks in messages at chosen
locations.  In the examples here, \ctanpackage{xintsession} is used, and it
loads \xintname in a way activating the nicer |\newlinechar| formatted
messages, even though it runs (a priori, but not necessarily) under Plain
\eTeX.
\begin{everbatim}
>>> 1/0;
Runaway argument?
! xint error: Division by zero: 1/0.
! Paragraph ended before \romannumeral (or \expanded,
  \numexpr, ...) expansion could produce its final output.
  See above the exception specifics.
  xint will try to recover (if in interactive mode, hit <return>
  at the ? prompt) and will go ahead hoping repair was complete.
<to be read again> 
                   \par 
...
l.491 \xintsession
                  \endinput%^^M
? 
@_1     0
>>> (-1)^3.2;
Runaway argument?
! xint error: Fractional power 32/1[-1] of negative -1[0].
! Paragraph ended before \romannumeral (or \expanded,
  \numexpr, ...) expansion could produce its final output.
  See above the exception specifics.
  xint will try to recover (if in interactive mode, hit <return>
  at the ? prompt) and will go ahead hoping repair was complete.
<to be read again> 
                   \par 
...
l.491 \xintsession
                  \endinput%^^M
? 
@_2     0
>>> cos 1);
Runaway argument?
! xint error: `cos1' unknown, say `Isome_var' or I use 0.
! Paragraph ended before \romannumeral (or \expanded,
  \numexpr, ...) expansion could produce its final output.
  See above the exception specifics.
  xint will try to recover (if in interactive mode, hit <return>
  at the ? prompt) and will go ahead hoping repair was complete.
<to be read again> 
                   \par 
...
l.491 \xintsession
                  \endinput%^^M
? 
Runaway argument?
! xint error: Extra ) removed. Hit <return>, fingers crossed.
! Paragraph ended before \romannumeral (or \expanded,
  \numexpr, ...) expansion could produce its final output.
  See above the exception specifics.
  xint will try to recover (if in interactive mode, hit <return>
  at the ? prompt) and will go ahead hoping repair was complete.
<to be read again> 
                   \par 
...
l.491 \xintsession
                  \endinput%^^M
? 
@_3     0
>>> 3=4;
Runaway argument?
! xint error: Expected an operator but got `='. Ignoring.
! Paragraph ended before \romannumeral (or \expanded,
  \numexpr, ...) expansion could produce its final output.
  See above the exception specifics.
  xint will try to recover (if in interactive mode, hit <return>
  at the ? prompt) and will go ahead hoping repair was complete.
<to be read again> 
                   \par 
...
l.491 \xintsession
                  \endinput%^^M
? 
@_4     12
>>> 
\end{everbatim}

In the last example, tacit multiplication was applied as \xintexprname was
looking for an operator, got some invalid input and then a number.

Some constructs in \xintexprname-essions use delimited macros and there is
thus possibility in case of an ill-formed expression to end up beyond the
|\relax| end-marker. Such a situation can also occur from |\relax| being
swallowed by a non-terminated
|\numexpr|:
\begin{everbatim}
\xintexpr 3 + \numexpr 5+4\relax followed by some LaTeX code...
\end{everbatim}
The correct input is
\begin{everbatim}
\xintexpr 3 + \numexpr 5+4\relax\relax
\end{everbatim}
But people in their right mind will have done
\begin{everbatim}
\xintexpr 3 + 5 + 4\relax
\end{everbatim}
A few will have done the computation in their heads.

In such cases low-level errors will arise and may
lead to very cryptic messages; but nothing unusual or especially traumatizing
for the daring experienced \TeX/\LaTeX\ user, whose has seen zillions of
un-helpful error messages already in her daily practice of
\TeX/\LaTeX.


\subsection{Package namespace, catcodes}


The bundle packages needs that the \csa{space} and \csa{empty} control
sequences are pre-defined with the identical meanings as in Plain \TeX{} (or
\LaTeX2e which has the same macros).

Private macros of \xintkernelname, \xintcorename, \xinttoolsname,
\xintname, \xintfracname, \xintexprname, \xintbinhexname, \xintgcdname,
\xintseriesname, and \xintcfracname{} use one or more underscores |_| as
private letter, to reduce the risk of getting overwritten. They almost
all begin either with |\XINT_| or with |\xint_|, a handful of these
private macros such as \csa{XINTsetupcatcodes}, \csa{XINTdigits} and
those with names such as |\XINTinFloat...| or |\XINTinfloat...| do not
have any underscore in their names (for obscure legacy reasons).

\xintkernelname provides \hyperref[odef]{|\odef|}, \hyperref[oodef]{|\oodef|},
\hyperref[fdef]{|\fdef|}: if macros with these names already exist
\xinttoolsname will not overwrite them. The same meanings are independently
available under the names |\xintodef|, |\xintoodef|, etc...

Apart from |\thexintexpr|, |\thexintiexpr|, ...\@
all other public macros from the \xintname bundle packages start with |\xint|.

For the good functioning of the macros, standard catcodes are assumed for the
minus sign, the forward slash, the square brackets, the letter `e'. These
requirements are dropped inside an |\xintexpr|-ession: spaces are gobbled,
catcodes mostly do not matter, the |e| of scientific notation may be |E| (on
input) \dots{}

If a character used in the |\xintexpr| syntax is made active,
this will surely cause problems; prefixing it with |\string| is one option.
There is \csbxint{exprSafeCatcodes} and \csbxint{exprRestoreCatcodes} to
temporarily turn off potentially active characters.

\begin{framed}
  For advanced \TeX\ users. At loading time of the packages the
  catcode configuration may be arbitrary as long as it satisfies the following
  requirements: the percent is of category code comment character, the
  backslash is of category code escape character, digits have category code
  other and letters have category code letter. Nothing else is assumed.
\end{framed}

\subsection{Origins of the package}
\label{ssec:origins}

|2013/03/28.| Package |bigintcalc| by \textsc{Heiko Oberdiek} already
provides expandable arithmetic operations on «big integers», i.e. integers
beyond the \TeX\ bound $2^{31}-1$, so why another%
%
\footnote{this section was written before the \xintfracname package; the
  author is not aware of another package allowing expandable
  computations with arbitrarily big fractions.}
%
one?

I got started on this in early March 2013, via a thread on the
|c.t.tex| usenet group, where \textsc{Ulrich D\,i\,e\,z} used the
previously cited package together with a macro (|\ReverseOrder|)
which I had contributed to another thread.%
%
\footnote{the \csa{ReverseOrder} could be avoided in that circumstance,
  but it does play a crucial r\^ole here.}
%
What I had learned in this
other thread thanks to interaction with \textsc{Ulrich D\,i\,e\,z} and
\textsc{GL} on expandable manipulations of tokens motivated me to
try my hands at addition and multiplication.

I wrote macros \csa{bigMul} and \csa{bigAdd} which I posted to the
newsgroup; they appeared to work comparatively fast. These first
versions did not use the \eTeX{} \csa{numexpr} primitive, they worked
one digit at a time, having previously stored carry-arithmetic in
1200 macros.

I noticed that the |bigintcalc| package used \csa{numexpr}
if available, but (as far as I could tell) not
to do computations many digits at a time. Using \csa{numexpr} for
one digit at a time for \csa{bigAdd} and \csa{bigMul} slowed them
a tiny bit but avoided cluttering \TeX{} memory with the 1200
macros storing pre-computed digit arithmetic. I wondered if some speed
could be gained by using \csa{numexpr} to do four digits at a time
for elementary multiplications (as the maximal admissible number
for \csa{numexpr} has ten digits).

|2013/04/14|. This initial \xintname was followed by \xintfracname which
handled exactly fractions and decimal numbers.

|2013/05/25|. Later came \xintexprname and at the same time \xintfracname got
extended to handle floating point numbers.

|2013/11/22|. Later, \xinttoolsname was detached.

|2014/10/28|. Release |1.1| significantly extended the \xintexprname parsers.

|2015/10/10|. Release |1.2| rewrote the core integer routines which had
remained essentially unmodified, apart from a slight improvement of division
early 2014.

This |1.2| release also got its impulse from a fast
``reversing'' macro, which I wrote after my interest got awakened again as a
result of correspondence with Bruno \textsc{Le Floch} during September 2015:
this new reverse uses a \TeX nique which \emph{requires} the tokens to be
digits. I wrote a routine which works (expandably) in quasi-linear time, but a
less fancy |O(N^2)| variant which I developed concurrently proved to be faster
all the way up to perhaps $7000$ digits, thus I dropped the quasi-linear one.
The less fancy variant has the advantage that \xintname can handle numbers
with more than $19900$ digits (but not much more than $19950$). This is with
the current common values of the input save stack and maximal expansion depth:
$5000$ and $10000$ respectively.

\clearpage
\expandafter\let\csname xint bundlenameUp\endcsname\undefined
\csname xintkernelnameUp\endcsname
\section{Macros of the \xintkernelname package}
\RaisedLabel{sec:kernel}

The \xintkernelname package contains mainly the common code base for handling
the load-order of the bundle packages, the management of catcodes at loading
time, definition of common constants and macro utilities which are used
throughout the code etc ...\@ it is automatically loaded by all packages of the
bundle.

It provides a few macros possibly useful in other contexts.

\localtableofcontents

\subsection{\csh{odef}, \csh{oodef}, \csh{fdef}}
\label{odef}
\label{oodef}
\label{fdef}

\csa{oodef}|\controlsequence {<stuff>}| does
\everb|@
    \expandafter\expandafter\expandafter\def
    \expandafter\expandafter\expandafter\controlsequence
    \expandafter\expandafter\expandafter{<stuff>}
|

This works only for a single
|\controlsequence|, with no parameter text, even without parameters. An
alternative would be:
\everb|@
\def\oodef #1#{\def\oodefparametertext{#1}%
               \expandafter\expandafter\expandafter\expandafter
               \expandafter\expandafter\expandafter\def
               \expandafter\expandafter\expandafter\oodefparametertext
               \expandafter\expandafter\expandafter }
|

\noindent
but it does not allow |\global| as prefix, and, besides, would have anyhow its
use (almost) limited to parameter texts without macro parameter tokens
(except if the expanded thing does not see them, or is designed to deal with
them).

There is a similar macro |\odef| with only one expansion of the replacement text
|<stuff>|, and |\fdef| which expands fully |<stuff>| using |\romannumeral-`0|.

They can be prefixed with |\global|. It appears than |\fdef| is generally a bit
faster than |\edef| when expanding macros from the \xintname bundle, when the
result has a few dozens of digits. |\oodef| needs thousands of digits it seems
to become competitive.


\subsection{\csh{xintReverseOrder}}\label{xintReverseOrder}

\csa{xintReverseOrder}\marg{list}\etype{n} does not do any expansion of its
argument and just reverses the order of the tokens in the \meta{list}. Braces
are removed once and the enclosed material, now unbraced, does not get
reversed. Unprotected spaces (of any character code) are gobbled.
%
\leftedline{|\xintReverseOrder{\xintDigitsOf\xintiiPow {2}{100}\to\Stuff}|}
%
\leftedline{gives:
  \ttfamily{\string\Stuff\string\to1002\string\xintiiPow\string\xintDigitsOf}}

\xinttoolsname provides a variant \csbxint{RevWithBraces} which keeps brace
pairs in the output, and \fexpan ds its input first.

For inputs consisting only digit tokens, see \csbxint{ReverseDigits} from
\xintname.

\subsection{\csh{xintLength}}
\label{xintLength}

\csa{xintLength}\marg{list}\etype{n} counts how many tokens (or braced items)
there are (possibly none). It does no expansion of its argument, so to use it
to count things in the replacement text of a macro |\x| one should do
|\expandafter\xintLength\expandafter{\x}|. Blanks between items are not
counted. See also \csbxint{NthElt}|{0}| (from \xinttoolsname) 
which first \fexpan ds its argument and then applies the same code.
%
\leftedline{|\xintLength {\xintiiPow {2}{100}}|\dtt{=\xintLength
    {\xintiiPow{2}{100}}}}
%
\leftedline{${}\neq{}$|\xintLen {\xintiiPow {2}{100}}|\dtt{=\xintLen
    {\xintiiPow{2}{100}}}}

\subsection{\csh{xintFirstItem}}
\label{xintFirstItem}

\csa{xintFirstItem}\marg{list}\etype{n} returns the first item of its
argument, one pair of braces removed. If the list has no items the output is
empty.\NewWith{1.4}

It does no expansion. For this and the next similar ones, see
|sourcexint.pdf| for comments on limitations.

\subsection{\csh{xintLastItem}}
\label{xintLastItem}

Added at |1.2i|.

\csa{xintLastItem}\marg{list}\etype{n} returns the last item
of its argument, one pair of braces removed. If the list has no items the
output is empty.

It does no expansion, which should be obtained via suitable |\expandafter|'s.
See also \csbxint{NthElt}|{-1}| from \xinttoolsname which obtains the same
result (but with another code) after having however \fexpan ded its
argument first.

\subsection{\csh{xintFirstOne}}
\label{xintFirstOne}

\csa{xintFirstOne}\marg{list}\etype{n} returns the first item as a braced
item. i.e.\@ if it was braced the braces are kept, else the braces are added.
It looks like using \csbxint{FirstItem} within braces, but the difference is
when the input was empty. Then the output is empty.\NewWith{1.4}

It does no expansion, which should be obtained via suitable |\expandafter|'s.

\subsection{\csh{xintLastOne}}
\label{xintLastOne}

\csa{xintLastOne}\marg{list}\etype{n} returns the last item as a braced
item. i.e.\@ if it was braced the braces are kept, else the braces are added.
It looks like using \csbxint{LastItem} within braces, but the
difference is when the input was empty. Then the output is empty.\NewWith{1.4}

It does no expansion, which should be obtained via suitable |\expandafter|'s.

\subsection{\csh{xintReplicate}, \csh{xintreplicate}}
\label{xintreplicate}
\label{xintReplicate}

\csa{romannumeral}\csa{xintreplicate}|{x}|\marg{stuff}\etype{\numx n} is simply
copied over from \LaTeX3's |\prg_replicate:nn| with some minor changes.%
%
\footnote{I started with the code from Joseph \textsc{Wright}
available on an online site.}

And \csa{xintReplicate}|{x}| integrates the
\csa{romannumeral} prefix.\NewWith{1.4}

It
does not do any expansion of its second argument but inserts it in the upcoming
token stream precisely |x| times. Using it with a negative |x| raises no error
and does nothing.%
%
\footnote{This behaviour may change in future.}


 
\subsection{\csh{xintGobble}, \csh{xintgobble}}
\label{xintgobble}
\label{xintGobble}

\csa{romannumeral}\csa{xintgobble}|{x}|\etype{\numx} is a Gobbling macro
written in the spirit of \LaTeX3's |\prg_replicate:nn| (which I cloned as
\csbxint{replicate}.) It gobbles |x| tokens upstream, with |x| allowed to be
as large as \dtt{531440}. Don't use it with |x<0|.

And \csa{xintGobble}|{x}| integrates the \csa{romannumeral}.\NewWith{1.4}


\csbxint{gobble} looks as if it must be related to \csbxint{Trim} from
\xinttoolsname, but the latter uses different code (using directly
\csbxint{gobble} is not possible because one must make sure not to gobble more
than the number of available items; and counting available items first is an
overhead which \csbxint{Trim} avoids.) It is rather\csbxint{Keep} with a
negative first argument which hands over to \csbxint{gobble} (because in that
case it is needed to count anyhow beforehand the number of items, hence
\csbxint{gobble} can then be used safely.)

I wrote an \csa{xintcount} in the same spirit as \csa{xintreplicate} and
\csa{xintgobble}. But it needs to be counting hundreds of tokens to be worth
its salt compared to \csbxint{Length}.

\subsection{(WIP) \csh{xintUniformDeviate}}
\label{xintUniformDeviate}

\csa{xintUniformDeviate}|{x}|\etype{\numx} is a wrapper of engine
|\pdfuniformdeviate| (or |\uniformdeviate|).%
%
\footnote{The |\uniformdeviate| primitive has been added to Xe\TeX\
  and will be available with \TeX Live 2019 release.}
The implementation is to be
considered experimental for the time being.%

The argument is expanded in |\numexpr| and the macro itself needs two
expansion steps. It produces like the engine primitive an integer (digit
tokens) with minimal value \dtt{0} and maximal one \dtt{x-1} if |x| is
positive, or minimal value \dtt{x+1} and maximal value \dtt{0} if |x| is
negative. For the discussion next, |x| is supposed positive as this
avoids having to insert absolute values in formulas.

The underlying engine Random Number Generator works with an array of 55 28bits
integers. To produce a « uniform » random integer in a given range
\dtt{0..x-1} it produces next pseudo-random |y| (supposedly uniformly
distributed, i.e.\@ non-uniformity can be neglected) such that \dtt{$0\leq y <
  2^{28}$} and the output is the rounding of \dtt{$x*(y/2^{28})$}, with upper
bound |x| remapped to |0|. This has following corollaries:
\begin{enumerate}
\item with |x=2^{29}| or |x=2^{30}| the engine primitive produces only even
  numbers,
\item with |x=3*2^{26}| the integers produced by the RNG when taken modulo
  three obey the proportion |1:1:2|, not |1:1:1|,
\item with |x=3*2^{14}| there is analogous although weaker non-uniformity of
  the random integers when taken modulo 3,
\item generally speaking pure powers of two should generate uniform random
  integers, but when the range is divisible by large powers of
  two, the non-uniformity may be amplified in surprising ways by modulo
  operations.
\end{enumerate}
These observations are not to be construed as criticism of the engine
primitive itself, which comes from MetaPost, as the code comments and more
generally the whole of \emph{The Art of Computer Programming, Vol. 2} stresses
that it should rather be seen as producing random fractions (the unit fraction
being $2^{28}$). Using it as a generator for \emph{integers} is a bit of an
abuse.

The first goal of \csa{xintUniformDeviate} is to guarantee a better uniformity
for the distribution of random integers in any given range |x|.

\emph{If the probability to obtain a given |y| in |0..x-1| is
  \verb$(1+e(y))/x$, the ``{relative non-uniformity}'' for that value |y| is
  \verb$|e(y)|$.}

The engine primitive guarantees only \dtt{$x/2^{28}$} relative non-uniformity, and
\csa{xintUniformDeviate} (in its current implementation) improves this by
a factor \dtt{|2^{28}=|\number"10000000}: the non-uniformity is guaranteed to
be bounded by \dtt{$x/2^{56}$}.%
%
\expandafter\footnote\expandafter{\ifnum\value{footnote}=55 This «56» is proof
  of existence of devil, no? \fi These estimates assume that the engine RNG underlying stream of
  28-bits integers can be considered uniform; it is known that the
  parity bits of these 28-bits integers have a period of |55(2^{55}-1)| and
  that after that many draws the count of 1s has only an excess of 55 compared
  to the count of 0s, so the scale seems to be an intrinsic non-uniformity of
  |2^{-55}| but it is not obvious if it applies to much shorter ranges. At any
  rate we assumed that the non-uniformity for |x| a power of two less than
  |2^{28}| is negligible in comparison to |2^{-28}|. Bigger powers of 2
  produce only even integers because the output is rescaled by
  factor |x/2^{28}|!}
%
With such a small non-uniformity, modulo phenomena as mentioned earlier are
not observable in reasonable computing time.%
%
\begin{everbatim*}
%\xintdefiifunc mod3(x):= x 'mod' 3;
\xintNewIIExpr\ModThree[1]{#1 'mod' 3}

\pdfsetrandomseed 87654321
\xintdefiivar BadDigits:=qraw(%
   \romannumeral\xintreplicate{504}{{\ModThree{\pdfuniformdeviate "C000000}}}%
);%

\pdfsetrandomseed 87654321
\xintdefiivar GoodDigits:=qraw(%
   \romannumeral\xintreplicate{504}{{\ModThree{\xintUniformDeviate{"C000000}}}}%
);%

These 504 digits generated from \string\pdfuniformdeviate:
\xinttheiiexpr BadDigits\relax\hfill\break
contain these respective amounts of 0, 1, and 2:
% (this is definitely not the fastest way to count, but it is fun - and expandable)
% !(i) (logical not) is short-cut for the vanishing test i==0, but it would be
% simpler to use (i)?{i is not zero}{i is zero} with permuted branches
\xinttheiiexpr iter(0,0,0;(!(i))?{[@][0]+1,[@][1],[@][2]}
                                 {(isone(i))?{[@][0],[@][1]+1,[@][2]}
                                             {[@][0],[@][1],[@][2]+1}}, 
                    i=BadDigits)\relax\par

These 504 digits generated from \string\xintUniformDeviate:
\xinttheiiexpr GoodDigits\relax\hfill\break
contain these respective amounts of 0, 1, and 2:
\xinttheiiexpr iter(0,0,0;(!(i))?{[@][0]+1,[@][1],[@][2]}
                                 {(isone(i))?{[@][0],[@][1]+1,[@][2]}
                                             {[@][0],[@][1],[@][2]+1}}, 
                    i=GoodDigits)\relax\par
% % output to data file for double-check with python
% \newwrite\out
% \immediate\openout\out=\jobname.data
% \immediate\write\out{Lbad=[\xinttheiiexpr BadDigits\relax]}
% \immediate\write\out{Lgood=[\xinttheiiexpr GoodDigits\relax]}
% \immediate\closeout\out
\end{everbatim*}

There is a second peculiarity of the engine RNG: two seeds sharing the same
low |k| bits generate sequences of 28-bits integers which are identical modulo
|2^k|! In particular after setting the seed, there are only 2 distinct
sequences of parity bits for the integers generated by |\pdfuniformdeviate (2
to the power 28)|...

In order to mitigate, \csa{xintUniformDeviate} currently only uses the
seven high bits from the underlying random stream, using multiple calls to
|\pdfuniformdeviate 128|. From the Birthday Effect, after about |2^{11}| seeds
one will likely pick a new one sharing its 22 low bits with an earlier one.

\begin{enumerate}
\item but as the final random integer is obtained by additional operations
  involving the range |x| (currently a modulo operation), for odd ranges it is
  more difficult for bit correlations to be seen,
\item anyway as they are only
|2^{28}| seeds in total, after only |2^{14}| seeds it is likely to encounter
one already explored, and then random integers are identical, however
complicated the RNG's raw output is malaxed, and whatever the target range
|x|. And |2^{14}| is only eight times as large as |2^{11}|.
\end{enumerate}

It would be nice if the engine provided some user interface for
  letting its RNG execute a given number of iterations without the overhead
  of replicated executions of |\pdfuniformdeviate|. This could help gain
  entropy and would reduce correlations across series from distinct seeds.

\smallskip
\emph{The description above summarizes parts of discussions held with Bruno Le
  Floch in May 2018 on occasion of his LaTeX3 contributions related to this.}
\par
\smallskip

\TeXnote 
currently the implementation of \csbxint{UniformDeviate} consumes exactly 5
calls to the engine primitive at each execution; the improved |x/2^{56}|
non-uniformity could be obtained with only 2 calls, but paranoïa about the
phenonemon of seeds with common bits has led me to accept the overhead of
using the 7 high bits of 4 random 28bits integers, rather than one single
28bits integer, or two, or three.

Timings indicate that one \csbxint{UniformDeviate} has a time cost about 13
times the one for one call to the engine primitive (and not only 5, as the
extra arithmetic expressions add overhead which is more costly than the
primitive itself). Except if the code using the pseudo-random number is very
short, this time penalty will prove in practice much less severe (and this is
one important reason why we opted for obtaining 28bits via the 7 high bits of
4 successive pseudo random numbers from the engine primitive). For example
let's raise 100 times a random integer to the tenth power:%
%
\InputIfFileExists{JFBUORDI}{}{}%
\ifdefined\JFBUORDI
\footnote{%
          \if0\pdfstrcmp{IMAC}{\JFBUORDI}Timings done during dvi build on a computer with a |2.4GHz| cpu.\else
          \if0\pdfstrcmp{MBA}{\JFBUORDI}Timings done during dvi build on a computer with a |2GHz| cpu.\else
          %\error
          DOCUMENTATION SEEMINGLY BUILT NOT ON AN AUTHOR-OWNED SYSTEM, BUT \string\JFBUORDI\space IS DEFINED, WEIRD!\fi\fi
}
\else
\footnote{These timings are done during a build on a computer not belonging to the author.}
\fi
%
\begin{everbatim*}
\pdfsetrandomseed 12345678
\pdfresettimer\romannumeral\xintreplicate
        {100}{\fdef\foo{\xintiiPow{\xintUniformDeviate{100000000}}{10}}}%
\the\dimexpr\pdfelapsedtime sp\relax\space (with \string\xintUniformDeviate)\newline
(last result: \foo)\newline
\pdfsetrandomseed 12345678
\pdfresettimer\romannumeral\xintreplicate
        {100}{\fdef\foo{\xintiiPow{\pdfuniformdeviate 100000000}{10}}}%
\the\dimexpr\pdfelapsedtime sp\relax\space (with \string\pdfuniformdeviate)\newline
(last result: \foo)\par
\end{everbatim*}

\TeXnote 
the macros \csbxint{RandomDigits} or \csbxint{iiRandRange}, and their
variants, as well as the supporting macros for \func{random} generate random
decimal digits eight by eight as if using
\csa{xintUniformDeviate}|{100000000}|, but via a direct optimized call made
possibly by the range being a power of 10.

\clearpage
\let\xintkernelnameUp\undefined
\csname xintcorenameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintcorename package}
\RaisedLabel{sec:core}

Package \xintcorename is automatically loaded by \xintname.

\xintcorename provides for big integers the four basic arithmetic operations
(addition, subtraction, multiplication, division), as well as powers and
factorials.

In the descriptions of the macros \texttt{\n} and \texttt{\m} stand
for (big) integers or macros \hyperref[ssec:expansions]{\fexpan ding} to
such big integers in strict format as described in \autoref{ssec:inputs}.

All macros require strict integer format on input and produce
strict integer format on output, except:\IMPORTANT
\begin{itemize}[nosep]
\item \csbxint{iNum} which converts to strict integer format an input in
  \emph{extended} integer format, i.e.\@ admitting multiple leading plus or
  minus signs, then possibly leading zeroes, then digits,
\item and \csbxint{Num} which is an alias for the former, which gets redefined by
  \xintfracname to accept more generally also decimal numbers or fractions as
  input and which truncates them to integers.
\end{itemize}

% Most removed macros listed in \autoref{ssec:coredeprecated} were by design
% applying \csbxint{Num} to their inputs. Typically these macros had a single
% |i| in their names, for example \csa{xintiAdd} was such a companion to
% \csa{xintiiAdd}. \xintfracname redefined \csbxint{Num} to be the macro
% accepting general fractional input and truncating it to an integer. Hence a
% macro such as \csa{xintiAdd} was compatible with the output format of
% \xintfracname macros, contrarily to \csbxint{iiAdd} which handles only strict
% integer format for its inputs. Of course, \xintfracname defined also its own
% \csbxint{Add} which did the addition of its arguments without truncating them
% to integers... but whose output format is the |A/B[N]| format explained in
% \autoref{ssec:outputs}, hence even if representing a small integer it can not
% be used directly in a \TeX\ context such as |\ifnum|, contrarily to
% \csa{xintiAdd} or to \csbxint{iiAdd}.

% \begin{framed}
%   This situation was the result of some early-on design
%   decisions which now appear misguided and impede further development. Hence,
%   at |1.2o| it has been decided to deprecate \emph{all} such |i|-macros. And
%   they got removed from the package at |1.3|.
% \end{framed}

The |ii| in the names of the macros such as \csbxint{iiAdd} serves to stress
that they accept only strict integers as input (this is signaled by the margin
annotation \textcolor[named]{PineGreen}{\emph{f}}), or macros \fexpan ding to
such strict format (big) integers and that they produce strict integers as
output.

Other macros, such as \csbxint{Double}, lack the |ii|, but this is only a
legacy of the history of the package and they have the same requirements for
input and format of output as the |ii|-macros.

%
% %
% \footnote{Regarding \csbxint{FDg} and \csbxint{LDg}, this is a breaking change
%   because formerly they used \csbxint{Num}.}

The letter \texttt{x} (with margin annotation
\smash{\textcolor[named]{PineGreen}{\numx}}) stands for an argument which will
be handled embedded in |\numexpr..\relax|. It will thus be completely expanded
and must give an integer obeying the \TeX{} bounds. See also
\autoref{sec:useofcount}. This is the case for the argument of \csbxint{iiFac}
or the exponent argument of \csbxint{iiPow}.

The {\color[named]{PineGreen}$\star$}'s in the margin are there to remind of
the complete expandability, even \fexpan dability of the macros, as discussed
in \autoref{ssec:expansions}.

\localtableofcontents

\subsection{\csh{xintiNum}}\label{xintiNum}

|\xintiNum|\n\etype{f} removes chains of plus or minus signs, followed by
zeroes.
\begin{everbatim*}
\xintiNum{+---++----+--000000000367941789479}
\end{everbatim*}

\subsection{\csh{xintDouble}}\label{xintDouble}

|\xintDouble|\n\etype{f} computes |2N|.

\subsection{\csh{xintHalf}}\label{xintHalf}

|\xintHalf|\n\etype{f} computes |N/2|
truncated towards zero.

\subsection{\csh{xintInc}}\label{xintInc}

|\xintInc|\n\etype{f} evaluates |N+1|.

\subsection{\csh{xintDec}}\label{xintDec}

|\xintDec|\n\etype{f} evaluates |N-1|.

\subsection{\csh{xintDSL}}\label{xintDSL}

|\xintDSL|\n\etype{f} is decimal shift left, \emph{i.e.\@} multiplication by
ten.

\subsection{\csh{xintDSR}}\label{xintDSR}

|\xintDSR|\n\etype{f} is truncated decimal shift right, \emph{i.e.\@} it is the
truncation of |N/10| towards zero.

\subsection{\csh{xintDSRr}}\label{xintDSRr}

|\xintDSRr|\n\etype{f} is rounded decimal shift right, \emph{i.e.\@} it is the
rounding of |N/10| away from zero. It is needed in \xintcorename for use by
\csbxint{iiDivRound}.

\subsection{\csh{xintFDg}}\label{xintFDg}

|\xintFDg|\n\etype{f} outputs the first digit (most significant) of the
number.

\subsection{\csh{xintLDg}}\label{xintLDg}

|\xintLDg|\n\etype{f} outputs the least significant digit. When the number
is positive, this is the same as the remainder in the Euclidean division by
ten.

\subsection{\csh{xintiiSgn}}\label{xintiiSgn}

|\xintiiSgn|\n\etype{f} returns 1 if the number is positive, 0 if it is zero
and -1 if it is negative.

\subsection{\csh{xintiiOpp}}\label{xintiiOpp}

|\xintiiOpp|\n\etype{f} outputs the opposite |-N| of the number |N|.

Important note: an input such as |-\foo| is not legal, generally speaking, as
argument to the macros of the \xintname bundle (except, naturally in
\csbxint{expr}-essions). The reason is that the minus sign stops the \fexpan
sion done during parsing of the inputs. One must use the syntax
|\xintiiOpp{\foo}| if one wants to pass |-\foo| as
argument to other macros.

\subsection{\csh{xintiiAbs}}\label{xintiiAbs}

|\xintiiAbs|\n\etype{f} outputs the absolute value of the number.

\subsection{\csh{xintiiAdd}}\label{xintiiAdd}

|\xintiiAdd|\n\m\etype{ff} computes the sum of the two (big) integers.

\subsection{\csh{xintiiCmp}}\label{xintiiCmp}

|\xintiiCmp|\n\m\etype{ff} produces \dtt{1} if |N>M|, \dtt{0} if |N=M|,
and \dtt{-1} if |N<M|.

At |1.2l| this macro was moved from package \xintname to \xintcorename.

\subsection{\csh{xintiiSub}}\label{xintiiSub}

|\xintiiSub|\n\m\etype{ff} computes the difference |N-M|.

\subsection{\csh{xintiiMul}}\label{xintiiMul}

|\xintiiMul|\n\m\etype{ff} computes the product of two (big) integers.

\subsection{\csh{xintiiSqr}}\label{xintiiSqr}

|\xintiiSqr|\n\etype{f} produces the square.

\subsection{\csh{xintiiPow}}\label{xintiiPow}

|\xintiiPow|\n\x\etype{f\numx} computes |N^x|. For |x=0|, this is 1. For |N=0|
and |x<0|, or if \verb+|N|>1+ and |x<0|, an error is raised. There will also
be an error if |x| exceeds the maximal \eTeX{} number \dtt{\number"7FFFFFFF},
but the real limit for exponents comes from either the computation time or the
settings of some \TeX\ memory parameters.

\begin{framed}
  Indeed, the maximal power of $2$ which \xintname is able to compute
  explicitely is |2^2^17=2^131072| which has \dtt{39457} digits. This
  exceeds the maximal size on input for the \xintcorename multiplication, hence
  any |2^N| with a higher |N| will fail. On the other hand |2^2^16| has
  \dtt{19729} digits, thus it can be squared once to obtain |2^2^17| or
  multiplied by anything smaller, thus all exponents up to and including |2^17|
  are allowed (because the power operation works by squaring things and making
  products).
\end{framed}

% Side remark: after all it does pay to think! I almost melted my CPU trying by
% dichotomy to pin-point the exact maximal allowable |N| for |\xintiiPow 2{N}|
% before finally making the reasoning above. Indeed, each such computation with
% |N>130000| activates the fan of my laptop and results in so warm a keyboard
% that I can hardly go on working on it! And it takes about 12 minutes for each
% |\xintiiPow2{N}| with such |N|'s of the order of $130000$ (a.t.t.o.w.).

\subsection{\csh{xintiiFac}}\label{xintiiFac}

|\xintiiFac|\x\etype{\numx} computes the factorial.

\begin{framed}
  The (theoretically) allowable range is $0\leqslant x\leqslant10000$.

  However the maximal possible computation depends on the values of some memory
  parameters of the |tex| executable: with the current default settings of
  TeXLive 2015, the maximal computable factorial (a.t.t.o.w. 2015/10/06) turns
  out to be $5971!$ which has $19956$ digits.%\footnotemark
\end{framed}



The |factorial| function, or equivalently |!| as post-fix operator is
available in \csbxint{iiexpr}, \csbxint{expr}:
\begin{everbatim*}
\printnumber{\xinttheiiexpr 200!\relax}\par
\end{everbatim*}
See also \csbxint{FloatFac} from package \xintfracname for the float variant,
used in \csbxint{floatexpr}.



\subsection{\csh{xintiiDivision}}\label{xintiiDivision}


|\xintiiDivision|\m\n\etype{ff} produces |{quotient}{remainder}|, in the sense
of (mathematical) Euclidean division: |M = QN + R|,
|0|${}\leq{}$\verb+R < |N|+. So the remainder is always non-negative and the
formula |M = QN + R| always holds independently of the signs of |N| or |M|.
Division by zero is an error (even if |M| vanishes) and returns |{0}{0}|.

\subsection{\csh{xintiiQuo}}\label{xintiiQuo}

|\xintiiQuo|\m\n\etype{ff} computes the quotient from the Euclidean division.

\subsection{\csh{xintiiRem}}\label{xintiiRem}

|\xintiiRem|\m\n\etype{ff} computes the remainder from the Euclidean
division.

\subsection{\csh{xintiiDivRound}}\label{xintiiDivRound}

|\xintiiDivRound|\m\n\etype{ff} returns the rounded value of the algebraic
quotient $M/N$ of two big integers. The rounding is ``away from zero.''
\begin{everbatim*}
\xintiiDivRound {100}{3}, \xintiiDivRound {101}{3}
\end{everbatim*}

\subsection{\csh{xintiiDivTrunc}}\label{xintiiDivTrunc}

|\xintiiDivTrunc|\m\n\etype{ff} computes $trunc(M/N)$. For positive arguments
$M,N>0$ it is the same as the Euclidean quotient \csbxint{iiQuo}.
\begin{everbatim*}
\xintiiQuo{1000}{57} (Euclidean), \xintiiDivTrunc{1000}{57} (truncated),
\xintiiDivRound{1000}{57} (rounded)\newline
\xintiiQuo{-1000}{57}, \xintiiDivTrunc{-1000}{57} (t), \xintiiDivRound{-1000}{57} (r)\newline
\xintiiQuo{1000}{-57}, \xintiiDivTrunc{1000}{-57} (t), \xintiiDivRound{1000}{-57} (r)\newline
\xintiiQuo{-1000}{-57}, \xintiiDivTrunc{-1000}{-57} (t), \xintiiDivRound{-1000}{-57} (r)\par
\end{everbatim*}

\subsection{\csh{xintiiDivFloor}}\label{xintiiDivFloor}

|\xintiiDivFloor|\m\n\etype{ff} computes $floor(M/N)$. For positive divisor
$N>0$ and arbitrary dividend $M$ it is the same as the Euclidean quotient
\csbxint{iiQuo}.
\begin{everbatim*}
\xintiiQuo{1000}{57} (Euclidean), \xintiiDivFloor{1000}{57} (floored)\newline
\xintiiQuo{-1000}{57}, \xintiiDivFloor{-1000}{57}\newline
\xintiiQuo{1000}{-57}, \xintiiDivFloor{1000}{-57}\newline
\xintiiQuo{-1000}{-57}, \xintiiDivFloor{-1000}{-57}\par
\end{everbatim*}

\subsection{\csh{xintiiMod}}\label{xintiiMod}

|\xintiiMod|\m\n\etype{ff} computes $M - N*floor(M/N)$. For positive divisor
$N>0$ and arbitrary dividend $M$ it is the same as the Euclidean remainder
\csbxint{iiRem}.

Formerly, this macro computed $M - N*trunc(M/N)$. The former meaning is
retained as \csa{xintiiModTrunc}.
\begin{everbatim*}
\xintiiRem {1000}{57} (Euclidean), \xintiiMod {1000}{57} (floored),
\xintiiModTrunc {1000}{57} (truncated)\newline
\xintiiRem {-1000}{57}, \xintiiMod {-1000}{57}, \xintiiModTrunc {-1000}{57}\newline
\xintiiRem {1000}{-57}, \xintiiMod {1000}{-57}, \xintiiModTrunc {1000}{-57}\newline
\xintiiRem {-1000}{-57}, \xintiiMod {-1000}{-57}, \xintiiModTrunc {-1000}{-57}\par
\end{everbatim*}

\subsection{\csh{xintNum}}\label{xintNum}

|\xintNum|\etype{f} is originally an alias for \csbxint{iNum}. But with
\xintfracname loaded its meaning is \hyperref[xintNumFrac]{modified} to accept
more general inputs. It then becomes an alias to \csbxint{TTrunc} which
truncates the general input to an integer in strict format.

% \subsection{Removed macros}\label{ssec:coredeprecated}

% These macros were deprecated at |1.2o| and removed at |1.3|.
% |\xintiiFDg| (renamed to \csbxint{FDg}), 
% |\xintiiLDg| (renamed to \csbxint{LDg}), 
% |\xintiOpp|, 
% |\xintiAbs|, 
% |\xintiAdd|, 
% |\xintCmp| (it gets defined by \xintfracname, so deprecation will usually not be
% seen; the macro with this name from former \xintcorename should have been
% called |\xintiCmp| actually),
% |\xintSgn| (it also gets its proper definition from \xintfracname),
% |\xintiSub|, 
% |\xintiMul|, 
% |\xintiDivision|, 
% |\xintiQuo|, 
% |\xintiRem|, 
% |\xintiDivRound|, 
% |\xintiDivTrunc|, 
% |\xintiMod|, 
% |\xintiSqr|, 
% |\xintiPow|, 
% |\xintiFac|.


\clearpage
\let\xintcorenameUp\undefined
\csname xintnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintname package}
\RaisedLabel{sec:xint}

This package loads automatically \xintcorename (and \xintkernelname) hence
all macros described in \autoref{sec:core} are still available.

This is \texttt{\xintbndlversion} of
\texttt{\xintbndldate}.

Version |1.0| was released |2013/03/28|.
Since |1.1 2014/10/28| the core arithmetic macros have been moved to a separate
package \xintcorename, which is automatically loaded by \xintname.
Only the \csbxint{iiSum}, \csbxint{iiPrd}, \csbxint{iiSquareRoot},
\csbxint{iiSqrt}, \csbxint{iiSqrtR},
\csbxint{iiPFactorial}, \csbxint{iiBinomial} genuinely add to the arithmetic
macros from \xintcorename. (\csbxint{iiFac} which computes factorials is
already in \xintcorename.)

With the exception of \csbxint{Len}, of the «Boolean logic macros» (see
next paragraphs) all macros require inputs being integers in strict format, see \autoref{ssec:inputs}.%
%
\footnote{of
course for conditionals such as \csbxint{iiifCmp} this constraint applies only
to the first two arguments.}
%
The |ii| in the macro names is here as a reminder of that fact. The output is
an integer in strict format, or a pair of two braced such integers for
\csbxint{iiSquareRoot}, with the exception of \csbxint{iiE} which may produce
strings of zero's if its first argument is zero.

Macros \csbxint{DecSplit} and \csbxint{ReverseDigits} are non-arithmetic and
have their own specific rules.

For all macros described here for which it makes sense, package \xintfracname
defines a similar one without |ii| in its name. This will handle more general
inputs: decimal, scientific numbers, fractions. The |ii| macros provided here
by \xintname can be nested inside macros of \xintfracname but the opposite
does not apply, because the output format of the \xintfracname macros, even
for representing integers, is not understood by the |ii| macros. The «Boolean
macros» \csbxint{AND} etc...\@ are exceptions though, they work fine if served
as inputs some \xintfracname output, despite doing only \fexpan
sion. Prior to |1.2o|, these macros did apply the \csbxint{Num}
or the more general \xintfracname general parsing, but this overhead was
deemed superfluous as it serves only to handle hand-written input and is not
needed if the input is obtained as a nested chain of \xintfracname macros for
example.

Prior to release |1.2o|, \xintname defined additional macros which applied
\csbxint{Num} to their input arguments. All these macros were deprecated at
|1.2o| and have been removed at |1.3|.

At |1.3d| macros \csbxint{iiGCD} and \csbxint{iiLCM} from package \xintgcdname
are also available from loading \xintname only. They are support macros for
the (multi-arguments) functions \func{gcd} and \func{lcm} in \csbxint{iiexpr}.

See \autoref{ssec:expansions} for the significance of the
\textcolor[named]{PineGreen}{\Numf}, \textcolor[named]{PineGreen}{\emph{f}},
\textcolor[named]{PineGreen}{\numx} and \textcolor[named]{PineGreen}{$\star$}
margin annotations.





\etocsetnexttocdepth{subsubsection}
\localtableofcontents

\subsection{\csh{xintiLen}}\label{xintiLen}

|\xintiLen|\n\etype{\Numf} returns the length of the number, after its parsing
via \csbxint{iNum}. The count does not include the sign.
\begin{everbatim*}
\xintiLen{-12345678901234567890123456789}
\end{everbatim*}

Prior to |1.2o|, the package defined only \csbxint{Len}, which is extended by
\xintfracname to fractions or decimal numbers, hence acquires a bit more
overhead then.

\subsection{\csh{xintReverseDigits}} \label{xintReverseDigits}

\the\dp\strutbox, \the\ht\strutbox, \the\baselineskip

|\xintReverseDigits|\n\etype{f} will reverse the order of the digits of the
number. \csa{xintRev} is the former denomination and is kept as an alias.
Leading zeroes resulting from the operation are not removed. Contrarily to
\csbxint{ReverseOrder} this macro \fexpan ds its argument; it is only usable
with digit tokens. It does \emph{not} apply \csbxint{Num} to its argument (so
this must be done explicitely if the argument is an integer produced from some
\xintfracname macros). It does accept a leading minus sign which will be left
upfront in the output.

\begingroup
\begin{everbatim*}
\oodef\x{\xintReverseDigits
  {98765432109876543210987654321098765432109876543210}}\meaning\x\par
\noindent\oodef\x{\xintReverseDigits {\xintReverseDigits
  {98765432109876543210987654321098765432109876543210}}}\meaning\x\par
\end{everbatim*}
\endgroup

\subsection{\csh{xintDecSplit}}
\label{xintDecSplit}

|\xintDecSplit|\x\n\etype{\numx f} cuts the |N| (a list of digits) into two
pieces |L| and |R|: it outputs |{L}{R}| where the original |N|
is the concatenation |LR|. These two pieces are decided according to |x|:
\begin{itemize}[nosep]
\item for |x>0|, |R| coincides with the |x| least significant digits. If |x|
  equals or exceeds the length of |N| the first piece |L| will thus be
  \emph{empty},
\item for |x=0|, |R| is empty, and |L| is all of |N|,
\item for |x<0|, the first piece |L| consists of the \verb+|x|+ most
  significant digits and the second piece |R| gets the remaining ones. If |x|
  equals or exceeds the length of |N| the second piece |R| will thus be
  \emph{empty}.
\end{itemize}

This macro provides public interface to some functionality which is primarily
of internal interest. It operates only (after \fexpan sion) on ``strings'' of
digits tokens: leading zeroes are allowed but a leading sign (even a minus
sign) will provoke an error.

Breaking change with |1.2i|: formerly |N<0| was replaced by its
  absolute value. Now, a sign (positive or negative) will create an error.


\subsection{\csh{xintDecSplitL}, \csh{xintDecSplitR}}
\label{xintDecSplitL}
\label{xintDecSplitR}

|\xintDecSplitL|\x\n\etype{\numx f} returns the first piece (unbraced) from
the \csa{xintDecSplit} output.

\noindent|\xintDecSplitR|\x\n\etype{\numx f} returns the second piece
(unbraced) from the \csa{xintDecSplit} output.

\subsection{\csh{xintiiE}}\label{xintiiE}

|\xintiiE|\n\x\etype{f\numx } serves to extend |N| with |x| zeroes. The
parameter |x| must be non-negative. The same output would be obtained via
\csbxint{DSH}|{-x}{N}|, except for |N=0|, as |\xintDSH{-x}{N}| multiplies |N|
by |10^x| hence produces |0| if |N=0| whereas
|\xintiiE{0}{x}| produces |x+1| zeros.
\begin{everbatim*}
\xintiiE {0}{91}\par
\end{everbatim*}

\subsection{\csh{xintDSH}}\label{xintDSH}

|\xintDSH|\x\n\etype{\numx f} is parametrized decimal shift. When |x| is
negative, it is like iterating \csbxint{DSL} \verb+|x|+ times (\emph{i.e.\@}
multiplication by $10^{-x}$). When |x| positive, it is like iterating
\csbxint{DSR} |x| times (and is more efficient), and for a non-negative |N|
this is thus the same as the quotient from the Euclidean division by |10^x|.

\subsection{\csh{xintDSHr}, \csh{xintDSx}}\label{xintDSHr}\label{xintDSx}

|\xintDSHr|\x\n\etype{\numx f} expects |x| to be zero or positive and it
returns then a value |R| which is correlated to the value |Q| returned by
\csbxint{DSH}\x\n{} in the following manner:
\begin{itemize}
\item if |N| is
  positive or zero, |Q| and |R| are the quotient and remainder in
  the Euclidean division by |10^x| (obtained in a more efficient
  manner than using \csa{xintiiDivision}),
\item if |N| is negative let
  |Q1| and |R1| be the quotient and remainder in the Euclidean
  division by |10^x| of the absolute value of |N|. If |Q1|
  does not vanish, then |Q=-Q1| and |R=R1|. If |Q1| vanishes, then
  |Q=0| and |R=-R1|.
\item for |x=0|, |Q=N| and |R=0|.
\end{itemize}
So one has |N = 10^x Q + R| if |Q| turns out to be zero or
positive, and |N = 10^x Q - R| if |Q| turns out to be negative,
which is exactly the case when |N| is at most |-10^x|.

|\xintDSx|\x\n\etype{\numx f} for |x| negative is exactly as
|\xintDSH|\x\n, \emph{i.e.\@} multiplication by $10^{-|x|}$. For |x| zero or
positive it returns the two numbers |{Q}{R}| described above, each one within
braces. So |Q| is |\xintDSH|\x\n, and |R| is |\xintDSHr|\x\n, but computed
simultaneously.

\subsection{\csh{xintiiEq}}\label{xintiiEq}

|\xintiiEq|\n\m\etype{ff} returns 1 if |N=M|, 0 otherwise.

\subsection{\csh{xintiiNotEq}}\label{xintiiNotEq}

|\xintiiNotEq|\n\m\etype{ff} returns 0 if |N=M|, 1 otherwise.


\subsection{\csh{xintiiGeq}}\label{xintiiGeq}

|\xintiiGeq|\n\m\etype{ff} returns 1 if the \emph{absolute value}
of the first number is at least equal to the absolute value of the second
number. If \verb+|N|<|M|+ it returns 0.

Important: the macro compares \emph{absolute values}.

\subsection{\csh{xintiiGt}}\label{xintiiGt}

|\xintiiGt|\n\m\etype{ff} returns 1 if |N|$>$|M|, 0 otherwise.

\subsection{\csh{xintiiLt}}\label{xintiiLt}

|\xintiiLt|\n\m\etype{ff} returns 1 if |N|$<$|M|, 0 otherwise.

\subsection{\csh{xintiiGtorEq}}\label{xintiiGxstorEq}

|\xintiiGtorEq|\n\m\etype{ff} returns 1 if |N|$\geqslant$|M|, 0 otherwise.
Extended by \xintfracname to fractions.

\subsection{\csh{xintiiLtorEq}}\label{xintiiLtorEq}

|\xintiiLtorEq|\n\m\etype{ff} returns 1 if |N|$\leqslant$|M|, 0 otherwise.

\subsection{\csh{xintiiIsZero}}\label{xintiiIsZero}

|\xintiiIsZero|\n\etype{f} returns 1 if |N=0|, 0 otherwise.

\subsection{\csh{xintiiIsNotZero}}\label{xintiiIsNotZero}

|\xintiiIsNotZero|\n\etype{f} returns 1 if |N!=0|, 0 otherwise.

\subsection{\csh{xintiiIsOne}}\label{xintiiIsOne}

|\xintiiIsOne|\n\etype{f} returns 1 if |N=1|, 0 otherwise.

\subsection{\csh{xintiiOdd}}\label{xintiiOdd}

|\xintiiOdd|\n\etype{f} is 1 if the number is odd and 0 otherwise.

\subsection{\csh{xintiiEven}}\label{xintiiEven}

|\xintiiEven|\n\etype{f} is 1 if the number is even and 0 otherwise.

\subsection{\csh{xintiiMON}}\label{xintiiMON}

|\xintiiMON|\n\etype{f} computes |(-1)^N|.
\begin{everbatim*}
\xintiiMON {-280914019374101929}
\end{everbatim*}

\subsection{\csh{xintiiMMON}}\label{xintiiMMON}

|\xintiiMMON|\n\etype{f} computes |(-1)^{N-1}|.
\begin{everbatim*}
\xintiiMMON {280914019374101929}
\end{everbatim*}

\subsection{\csh{xintiiifSgn}}\label{xintiiifSgn}

\csh{xintiiifSgn}\marg{N}\marg{A}\marg{B}\marg{C}\etype{fnnn} executes either
the \meta{A}, \meta{B} or \meta{C} code, depending on its first argument being
respectively negative, zero, or positive.

\subsection{\csh{xintiiifZero}}\label{xintiiifZero}

\csa{xintiiifZero}\marg{N}\marg{IsZero}\marg{IsNotZero}\etype{fnn} expandably
checks if the first mandatory argument |N| (a number, possibly a fraction if
\xintfracname is loaded, or a macro expanding to one such) is zero or not. It
then either executes the first or the second branch.

Beware that both branches must be present.

\subsection{\csh{xintiiifNotZero}}\label{xintiiifNotZero}

\csa{xintiiifNotZero}\marg{N}\marg{IsNotZero}\marg{IsZero}\etype{fnn}
expandably checks if the first mandatory argument |N| is not
zero or is zero. It then either executes the first or the second branch.

Beware that both branches must be present.

\subsection{\csh{xintiiifOne}}\label{xintiiifOne}

\csa{xintiiifOne}\marg{N}\marg{IsOne}\marg{IsNotOne}\etype{fnn} expandably
checks if the first mandatory argument |N| is one or not one. It
then either executes the first or the second branch. Beware that both branches
must be present.

\subsection{\csh{xintiiifCmp}}\label{xintiiifCmp}

\csa{xintiiifCmp}\marg{A}\marg{B}\marg{A<B}\marg{A=B}\marg{A>B}\etype{ffnnn}
compares its first two arguments and chooses accordingly the correct branch.

\subsection{\csh{xintiiifEq}}\label{xintiiifEq}

\csa{xintiiifEq}\marg{A}\marg{B}\marg{A=B}\marg{not(A=B)}\etype{ffnn} checks
equality of its two first arguments and executes the corresponding branch.

\subsection{\csh{xintiiifGt}}\label{xintiiifGt}

\csa{xintiiifGt}\marg{A}\marg{B}\marg{A>B}\marg{not(A>B)}\etype{ffnn}
checks if $A>B$ and executes the corresponding branch.

\subsection{\csh{xintiiifLt}}\label{xintiiifLt}

\csa{xintiiifLt}\marg{A}\marg{B}\marg{A<B}\marg{not(A<B)}\etype{ffnn}
checks if $A<B$ and executes the corresponding branch.

\subsection{\csh{xintiiifOdd}}\label{xintiiifOdd}

\csa{xintiiifOdd}\marg{A}\marg{A odd}\marg{A even}\etype{fnn} checks if $A$ is
and odd integer and executes the corresponding branch.

\subsection{\csh{xintiiSum}}\label{xintiiSum}

\csa{xintiiSum}\marg{braced things}\etype{{\lowast f}} after expanding its
argument expects to find a sequence of tokens (or braced material). Each is
\fexpan ded, and the sum of all these numbers is returned.
\begin{everbatim*}
\xintiiSum{{123}{-98763450}{\xintiiFac{7}}{\xintiiMul{3347}{591}}}\newline
\xintiiSum{1234567890}\newline
\xintiiSum{1234}\newline
\xintiiSum{}
\end{everbatim*}

A sum with only one term returns that
number: |\xintiiSum {{-1234}}|\dtt{=\xintiiSum {{-1234}}}.
Attention that |\xintiiSum {-1234}| is not legal input and would make the
\TeX{} run fail.

\subsection{\csh{xintiiPrd}}\label{xintiiPrd}

\csa{xintiiPrd}\marg{braced things}\etype{{\lowast f}} after expanding its
argument expects to find a sequence of (of braced items or unbraced
single tokens). Each is
expanded (with the usual meaning), and the product of all these numbers is
returned.
\begin{everbatim*}
\xintiiPrd{{-9876}{\xintiiFac{7}}{\xintiiMul{3347}{591}}}\newline
\xintiiPrd{123456789123456789}\newline
\xintiiPrd {1234}\newline
\xintiiPrd{}
\end{everbatim*}

Attention that |\xintiiPrd {-1234}| is not legal input and would make the \TeX{}
compilation fail.
\begin{everbatim*}
$2^{200}3^{100}7^{100}=\printnumber
       {\xintiiPrd {{\xintiiPow {2}{200}}{\xintiiPow {3}{100}}{\xintiiPow {7}{100}}}}$
\end{everbatim*}

With \xintexprname, the syntax is the natural one:
\begin{everbatim*}
$2^{200}3^{100}7^{100}=\printnumber{\xinttheiiexpr 2^200 * 3^100 * 7^100\relax}$
\end{everbatim*}

\subsection{\csh{xintiiSquareRoot}}
\label{xintiiSquareRoot}

|\xintiiSquareRoot|\n\etype{f} returns two braced integers |{M}{d}| which
satisfy |d>0| and |M^2-d=N| with
|M| the smallest (hence if |N=k^2| is a perfect square then |M=k+1|, |d=2k+1|).

\begin{everbatim*}
\xintAssign\xintiiSquareRoot {17000000000000000000000000}\to\A\B
\xintiiSub{\xintiiSqr\A}\B=\A\string^2-\B
\end{everbatim*}

A rational approximation to $\sqrt{|N|}$ is $|M|-\frac{|d|}{|2M|}$ which is a
majorant and the error is at most |1/2M| (if |N| is a perfect square |k^2|
this gives |k+1/(2k+2)|, not |k|.)

Package \xintfracname has \csbxint{FloatSqrt} for square roots of floating
point numbers.

\subsection{\csh{xintiiSqrt}, \csh{xintiiSqrtR}}
\label{xintiiSqrt}\label{xintiiSqrtR}

\noindent|\xintiiSqrt|\n\ computes the largest integer whose square
is at most equal to |N|.\etype{f} |\xintiiSqrtR| 
produces the rounded, not truncated, square root.\etype{f}
\begin{everbatim*}
\begin{itemize}[nosep]
\item \xintiiSqrt  {3000000000000000000000000000000000000}
\item \xintiiSqrtR {3000000000000000000000000000000000000}
\item \xintiiSqrt  {\xintiiE {3}{100}}
\end{itemize}
\end{everbatim*}

\subsection{\csh{xintiiBinomial}}\label{xintiiBinomial}

|\xintiiBinomial{x}{y}|\etype{\numx\numx} computes binomial coefficients.

If |x<0| an out-of-range error is raised. Else, if |y<0| or if |x<y| the macro
evaluates to \dtt{\xintiiBinomial{1}{-1}}.


%\begin{framed}
  The allowable range is $0\leqslant x\leqslant99999999$.
%\end{framed}
  % Thus the maximal computable value is ${9999 \choose 5000}$ which turns out
  % to have \dtt{3008} digits.
  But this theoretical range includes binomial coefficients with more than the
  roughly 19950 digits that the arithmetics of \xintname can handle. In such
  cases, the computation will end up in a low-level \TeX{} error after a
  long time.

%
It turns out that ${65000 \choose 32500}$ has \dtt{19565} digits and
${64000 \choose 32000}$ has \dtt{19264} digits. The latter can be evaluated
(this takes a long long time) but presumably not the former (I didn't try).
Reasonable feasible evaluations are with binomial coefficients not exceeding
about one thousand digits.


%
The |binomial| function is available in the \xintexprname parsers.
\begin{everbatim*}
\xinttheiiexpr seq(binomial(100,i), i=47..53)\relax
\end{everbatim*}

See \csbxint{FloatBinomial} from package \xintfracname for the float variant,
used in \csbxint{floatexpr}.


In order to
evaluate binomial coefficients ${x \choose y}$ with $x>99999999$, or even
$x\geqslant 2^{31}$, but $y$ is not too large, one may use an ad hoc function
definition such as:
\begin{everbatim*}
\xintdeffunc mybigbinomial(x,y):=`*`(x-y+1..[1]..x)//y!;%
%                            without [1], x would have been limited to < 2^31
\printnumber{\xinttheexpr mybigbinomial(98765432109876543210,10)\relax}
\end{everbatim*}


To get this functionality in macro form, one can do:
\begin{everbatim*}
\xintNewIIExpr\MyBigBinomial [2]{`*`(#1-#2+1..[1]..#1)//#2!}
\printnumber{\MyBigBinomial {98765432109876543210}{10}}
\end{everbatim*}

As we used \csa{xintNewIIExpr}, this macro will only accept strict integers.
Had we used \csa{xintNewExpr} the |\MyBigBinomial| would have accepted general
fractions or decimal numbers, and computed the product at the numerator
without truncating them to integers; but the factorial at the denominator
would truncate its argument.

\subsection{\csh{xintiiPFactorial}}\label{xintiiPFactorial}

|\xintiiPFactorial{a}{b}|\etype{\numx\numx} computes the partial factorial
|(a+1)(a+2)...b|. For |a=b| the product is considered empty hence returns |1|.

%\begin{framed}
  The allowed range
%
%
%
  is $-100000000\leqslant a, b\leqslant99999999$.
  The
  rule is to interpret the formula as the product of the
  $j$'s such that $a<j\leqslant b$, hence in particular if $a\geqslant b$ the
  product is empty and the macro evaluates to |1|.

  Only for $0\leqslant a\leqslant b$ is the behaviour to be considered
  stable. For $a>b$ or negative arguments, the definitive rules have not yet
  been fixed.

\begin{everbatim*}
\xintiiPFactorial {100}{130}
\end{everbatim*}
%\end{framed}

This theoretical range allows computations whose result values would have more
than the roughly 19950 digits that the arithmetics of \xintname can handle. In
such cases, the computation will end up in a low-level \TeX{} error after a
long time.

The |pfactorial| function is available in the \xintexprname parsers.
\begin{everbatim*}
\xinttheiiexpr pfactorial(100,130)\relax
\end{everbatim*}

See \csbxint{FloatPFactorial} from package \xintfracname for the float
variant, used in \csbxint{floatexpr}.


In case values are needed with $b>99999999$, or even $b\geqslant 2^{31}$, but
$b-a$ is not too large, one may use an ad hoc function definition such as:
\begin{everbatim*}
\xintdeffunc mybigpfac(a,b):=`*`(a+1..[1]..b);%
%                            without [1], b would have been limited to < 2^31
\printnumber{\xinttheexpr mybigpfac(98765432100,98765432120)\relax}
\end{everbatim*}

\subsection{\csh{xintiiMax}}\label{xintiiMax}

|\xintiiMax|\n\m\etype{ff} returns the largest of the two in the sense
of the order structure on the relative integers (\emph{i.e.\@} the right-most
number if they are put on a line with positive numbers on the right):
|\xintiiMax {-5}{-6}|\dtt{=\xintiiMax{-5}{-6}}.

\subsection{\csh{xintiiMin}}\label{xintiiMin}

|\xintiiMin|\n\m\etype{ff} returns the smallest of the two in the sense of the
order structure on the relative integers (\emph{i.e.\@} the left-most number if
they are put on a line with positive numbers on the right): |\xintiiMin
{-5}{-6}|\dtt{=\xintiiMin{-5}{-6}}.

\subsection{\csh{xintiiMaxof}}\label{xintiiMaxof}

\csa{xintiiMaxof}|{{a}{b}{c}...}|\etype{f{$\to$}\lowast f} returns the
maximum. The list argument may be a macro, it is \fexpan ded first.

\subsection{\csh{xintiiMinof}}\label{xintiiMinof}

\csa{xintiiMinof}|{{a}{b}{c}...}|\etype{f{$\to$}\lowast f} returns the
minimum. The list argument may be a macro, it is \fexpan ded first.

\subsection{\csh{xintifTrueAelseB}}
\label{xintifTrueAelseB}

\csa{xintifTrueAelseB}\marg{f}\marg{true branch}\marg{false branch}\etype{fnn}
is a synonym for \csbxint{iiifNotZero}.

{\small
  \noindent |\xintiiifnotzero| is lowercase companion macro.\par }

Note 1: as it does only \fexpan sion on its argument it fails with inputs such
as |--0|. But with \xintfracname loaded, it does work fine if nested with
other \xintfracname macros, because the output format of such macros is fine
as input to \csbxint{iiifNotZero}. This remark applies to all other «Boolean
logic» macros next.

Note 2: prior to |1.2o| this macro was using \csbxint{ifNotZero} which applies
\csbxint{Num} to its argument (or gets redefined by \xintfracname to handle
general decimal numbers or fractions). Hence it would have
worked with input such as |--0|. But it was decided at |1.2o| that the
overhead was not worth it. The same remark applies to the other «Boolean
logic» type macros next.

\subsection{\csh{xintifFalseAelseB}}
\label{xintifFalseAelseB}

\csa{xintifFalseAelseB}\marg{f}\marg{false branch}\marg{true
  branch}\etype{fnn} is a synonym for \csbxint{iiifZero}.

{\small
  \noindent |\xintiiifzero| is lowercase companion macro.\par }

\subsection{\csh{xintNOT}}\label{xintNOT}

\csa{xintNOT}\etype{f} is a synonym for \csa{xintiiIsZero}.

{\small |\xintiiiszero| serves as lowercase companion macro.\par}


\subsection{\csh{xintAND}}\label{xintAND}

|\xintAND{f}{g}|\etype{ff} returns \dtt{1} if |f!=0| and |g!=0| and \dtt{0}
otherwise.

\subsection{\csh{xintOR}}\label{xintOR}

|\xintOR{f}{g}|\etype{ff} returns \dtt{1} if |f!=0| or |g!=0| and \dtt{0}
otherwise.

\subsection{\csh{xintXOR}}\label{xintXOR}

|\xintXOR{f}{g}|\etype{ff} returns \dtt{1} if exactly one of |f| or |g|
is true (i.e.\@ non-zero), else \dtt{0}.

\subsection{\csh{xintANDof}}\label{xintANDof}

\csa{xintANDof}|{{a}{b}{c}...}|\etype{f{$\to$}\lowast f} returns \dtt{1} if
all are true (i.e.\@ non zero) and \dtt{0} otherwise. The list argument may be a
macro, it (or rather its first token) is \fexpan ded first to deliver its
items.

\subsection{\csh{xintORof}}\label{xintORof}

\csa{xintORof}|{{a}{b}{c}...}|\etype{f{$\to$}\lowast f} returns \dtt{1} if at
least one is true (i.e.\@ does not vanish), else it produces \dtt{0}. The list
argument may be a macro, it is \fexpan ded first.

\subsection{\csh{xintXORof}}\label{xintXORof}

\csa{xintXORof}|{{a}{b}{c}...}|\etype{f{$\to$}\lowast f} returns \dtt{1} if an
odd number of them are true (i.e.\@ do not vanish), else it produces \dtt{0}.
The list argument may be a macro, it is \fexpan ded first.

\subsection{\csh{xintiiGCD}}
\label{xintiiGCD}

|\xintiiGCD|\n\m\etype{ff} computes the greatest common divisor. It is
positive, except when both |N| and |M| vanish, in which case the macro returns
zero.
%
\leftedline{\csa{xintiiGCD}|{10000}{1113}|\dtt{=\xintiiGCD{10000}{1113}}}
%
\leftedline{|\xintiiGCD{123456789012345}{9876543210321}=|\dtt
  {\xintiiGCD{123456789012345}{9876543210321}}}

At |1.3d|, this macro (which is used by the \func{gcd} function in
\csbxint{iiexpr}) was copied over to \xintname, thus removing a partial
dependency of \xintexprname on \xintgcdname.

At |1.4| \xintgcdname requires \xintname and the latter is thus the one
providing the macro.

\subsection{\csh{xintiiLCM}}
\label{xintiiLCM}

|\xintiiLCM|\n\m\etype{ff} computes the least common multiple. It is positive,
except if one of |N| or |M| vanish, in which case the macro returns zero.
%
\leftedline{\csa{xintiiLCM}|{10000}{1113}|\dtt{=\xintiiLCM{10000}{1113}}}
%
\leftedline{|\xintiiLCM{123456789012345}{9876543210321}=|\dtt
  {\xintiiLCM{123456789012345}{9876543210321}}}

At |1.3d|, this macro (which is used by the \func{lcm} function in
\csbxint{iiexpr}) was copied over to \xintname, thus removing a partial
dependency of \xintexprname on \xintgcdname.

At |1.4| \xintgcdname requires \xintname and the latter is thus the one
providing the macro.

\subsection{\csh{xintiiGCDof}}\label{xintiiGCDof}

\csa{xintiiGCDof}|{{a}{b}{c}...}|\etype{f{$\to$}{\lowast f}} computes the
greatest common divisor of the integers |a|, |b|, \dots{}. It is a support
macro for the |gcd()| function of the \csbxint{iiexpr} parser.

It replaces the \csbxint{GCDof} which was formerly provided by \xintgcdname
and is now available via \xintfracname in a version handling also
fractions.


\subsection{\csh{xintiiLCMof}}\label{xintiiLCMof}

\csa{xintiiLCMof}|{{a}{b}{c}...}|\etype{f{$\to$}{\lowast f}} computes the
least common multiple of the integers |a|, |b|, \dots{}. It is a support
macro for the |lcm()| function of the \csbxint{iiexpr} parser.

It replaces the \csbxint{LCMof} which was formerly provided by \xintgcdname
and is now available via \xintfracname in a version handling also
fractions.

\subsection{\csh{xintLen}}\label{xintLen}

|\xintLen|\etype{\Numf} is originally an alias for \csbxint{iLen}. But with
\xintfracname loaded its meaning is \hyperref[xintLenFrac]{modified} to accept
more general inputs.

% \subsection{Removed macros (they require \xintfracname)}\label{ssec:xintdeprecated}

% These macros now require \xintfracname. They have been removed from \xintname
% at |1.3|.
% |\xintEq|,
% |\xintNeq|,
% |\xintGeq|,
% |\xintGt|,
% |\xintLt|,
% |\xintGtorEq|,
% |\xintLtorEq|,
% |\xintIsZero|,
% |\xintIsNotZero|,
% |\xintIsOne|,
% |\xintOdd|,
% |\xintEven|,
% |\xintifSgn|,
% |\xintifCmp|,
% |\xintifEq|,
% |\xintifGt|,
% |\xintifLt|,
% |\xintifZero|,
% |\xintifNotZero|,
% |\xintifOne|,
% |\xintifOdd|.

% With the exception of |\xintNeq| which was renamed to |\xintNotEq|, the above
% listed macros all belong to \xintfracname.

% At |1.4|, these macros formerly available via \xintgcdname have been moved to
% \xintfracname as well.
% |\xintGCD|,
% |\xintLCM|,
% |\xintGCDof|,
% |\xintLCMof|.

% \subsection{Removed macros (they used \csh{xintNum})}\label{ssec:xintdeprecatedNum}

% These macros filtered their arguments via \csbxint{Num}. They got deprecated
% at |1.2o| and removed at |1.3|:
% |\xintMON|,
% |\xintMMON|,
% |\xintiMax|,
% |\xintiMin|,
% |\xintiMaxof|,
% |\xintiMinof|,
% |\xintiSquareRoot|,
% |\xintiSqrt|,
% |\xintiSqrtR|,
% |\xintiBinomial|,
% |\xintiPFactorial|.

\subsection{(WIP) \csh{xintRandomDigits}}\label{xintRandomDigits}

\begin{framed}
  All randomness related macros are Work-In-Progress: implementation and user
  interface may change. They work only if the \TeX\ engine provides the
  \csa{uniformdeviate} or \csa{pdfuniformdeviate} primitive. See
  \csbxint{UniformDeviate} for additional information.
\end{framed}

|\xintRandomDigits{N}|\etype{\numx} expands in two steps to |N| random decimal
digits. The argument must be non-negative and is limited by \TeX\ memory
parameters.
On \TeX Live 2018 with input save stack size at \dtt{5000} the
maximal allowed |N| is at most \dtt{19984} (tested within a |\write| to an
auxiliary file, the macro context may cause a reduced maximum).
\begin{everbatim*}
\pdfsetrandomseed 271828182
\xintRandomDigits{92}
\end{everbatim*}

\TeXnote the digits are produced eight by eight by the same method which would
result from \csbxint{UniformDeviate}|{100000000}| but with less overhead.

% \subsection{\csh{\xintOneRandomDigit}}\label{xintOneRandomDigit}

\subsection{(WIP) \csh{xintXRandomDigits}}\label{xintXRandomDigits}

|\xintXRandomDigits{N}|\xtype{\numx} expands under exhaustive expansion
(|\edef|, |\write|, |\csname| ...) to |N| random decimal
digits. The argument must be non-negative.
For example:
\begin{everbatim}
\newwrite\out
\immediate\openout\out=\jobname-out.txt
\immediate\write\out{\xintXRandomDigits{4500000}}
\immediate\closeout\out
\end{everbatim}
creates a \dtt{4500001} bytes file (it ends with a line feed character).
Trying with \dtt{5000000} raises this error:
\begin{everbatim}
Runaway text?
588875947168511582764514135070217555354479805240439407753451354223283\ETC.
! TeX capacity exceeded, sorry [main memory size=5000000].
<inserted text> 666515098
                         
l.15 ...ate\write\out{\xintXRandomDigits{5000000}}
                                                  
No pages of output.
Transcript written on temp.log.
\end{everbatim}
This can be lifted by increasing the \TeX\ memory settings (installation
dependent).

\TeXnote the digits are produced eight by eight by the same method which would
result from \csbxint{UniformDeviate}|{100000000}| but with less overhead.

\subsection{(WIP) \csh{xintiiRandRange}}\label{xintiiRandRange}

|\xintiiRandRange{A}|\etype{f} expands to a random (big) integer |N|
such that |0<=N<A|. It is a supporting macro for \func{randrange}. As with
Python's function of the same name, it is an error if |A<=0|.
\begin{everbatim*}
\pdfsetrandomseed 271828314
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx\newline
\xintiiRandRange{\xintNum{1e40}}\newline
\pdfsetrandomseed 271828314
\xinttheiiexpr randrange(num(1e40))\relax\newline % bare 1e40 not understood by \xintiiexpr
\pdfsetrandomseed 271828314
\xinttheexpr randrange(1e40)\relax
\end{everbatim*}

Of course, keeping in mind that the set of seeds is of cardinality |2^{28}|,
randomness is a bit illusory here say with |A=10^N|, |N>8|, if we proceed
immediately after having set the seed. If we add some entropy in any way, then
it is slightly more credible; but I think that for each seed the period is
something like |2^{27}(2^{55}-1)55|,%
%
\footnote{Compare the result of exercise 3.2.2-30 in TAOCP, vol II.}
%
so we expect at most about |2^{110}55|
``points in time'', and this is already small compared to the |10^40|
from example above. Thus already we are very far from being intrinsically
able to generate all numbers with fourty digits as random numbers, and this
makes the previous section about usage of \csbxint{XRandomDigits} to generate
millions of digits a bit comical...

\TeXnote the digits are produced eight by eight by the same method which would
result from \csbxint{UniformDeviate}|{100000000}| but with less overhead.

\subsection{(WIP) \csh{xintiiRandRangeAtoB}}\label{xintiiRandRangeAtoB}

|\xintiiRandRangeAtoB{A}{B}|\etype{ff} expands to a random (big) integer |N|
such that |A<=N<B|. It is a supporting macro for \func{randrange}. As with
Python's function of the same name, it is an error if |B<=A|.
\begin{everbatim*}
\pdfsetrandomseed 271828314
12345678911111111111111111111\newline
\xintiiRandRangeAtoB{12345678911111111111111111111}{12345678922222222222222222222}\newline
\pdfsetrandomseed 271828314
\def\test{%
\xinttheiiexpr randrange(12345678911111111111111111111,12345678922222222222222222222)\relax}%
\romannumeral\xintreplicate{10}{\test\newline}%
12345678922222222222222222222
\end{everbatim*}

\TeXnote the digits are produced eight by eight by the same method which would
result from \csbxint{UniformDeviate}|{100000000}| but with less overhead.

\clearpage
\let\xintnameUp\undefined
\csname xintfracnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintfracname package}
\RaisedLabel{sec:frac}

First version of this package was in release |1.03| (|2013/04/14|) of the
\xintname bundle.

At release |1.3| (|2018/02/28|) the behaviour of \csbxint{Add} (and of
\csbxint{Sub}) was modified: when adding |a/b| and |c/d| they
will use always the least common multiple of the denominators. This helps
limit the build-up of denominators, but the author still hesitates if the
fraction should be reduced to smallest terms. The current method allows (for
example when multiplying two polynomials) to keep a well-predictable
denominator among various terms, even though some may be reducible.

\xintfracname loads automatically \xintcorename and \xintname and inherits
their macro definitions. Only these two are redefined:
\hyperref[xintNumFrac]{\string\xintNum} and
\hyperref[xintLenFrac]{\string\xintLen}. As explained in \autoref{ssec:inputs}
and \autoref{ssec:outputs} the interchange format for the \xintfracname
macros, i.e.\@ |A/B[N]|, is not understood by the |ii|-named macros of
\xintcorename/\xintname which expect the so-called strict integer format.
Hence, to use such an |ii|-macro with an output from an \xintfracname macro,
an extra \csbxint{Num} wrapper is required. But macros already defined by
\xintfracname cover most use cases hence this should be a rarely needed.

In the macro descriptions, the variable |f|\ntype{\Ff} and the margin
indicator stand for the \xintfracname input format for integers, scientific
numbers, and fractions as described in \autoref{ssec:inputs}.

As in the \hyperref[sec:xint]{xint.sty} documentation, |x|\ntype{\numx} stands
for something which internally will be handled in a \csa{numexpr}. It may thus
be an expression as understood by \csa{numexpr} but its evaluation and
intermediate steps must obey the \TeX\ bound.

The output format for most macros is the |A/B[N]| format but naturally the
float macros use the scientific notation on output. And some macros are
special, for example \csbxint{Trunc} produces decimal numbers, \csbxint{Irr}
produces an |A/B| with no |[N]|, \csbxint{iTrunc} and \csbxint{iRound} produce
integers without trailing |[N]| either, etc\dots

At |1.4g|,\CHANGED{1.4g} old legacy typesetting macros |\xintFrac|, |\xintSignedFrac|,
|\xintFwOver| and |\xintSignedFwOver| were renamed into \csbxint{TeXFrac},
\csbxint{TeXsignedFrac}, \csbxint{TeXOver}, \csbxint{TeXsignedOver}.  The old
names will raise warnings and will be removed at next major release.

\localtableofcontents

\subsection{\csh{xintTeXfromSci}}\label{xintTeXfromSci}

Experimental.\etype{} This typesetting math-mode-only macro expects an input
which is already in, or will expand to, decimal or scientific
notation.\NewWith{1.4g} A trailing |/B| is accepted and will be handled
differently according to whether it follows some scientific exponent |eN|
part or not.

This macro is designed to be used as a typesetting wrapper for \csbxint{eval}
or \csbxint{floateval} output.  It does no «reflowing» of the mantissa as the
input is expected to have been already «prettified» (for example via the
removal of trailing zeros, of a scientific part if the exponent is small,
etc...) and simply transforms the |e<exponent>| part, if actually present,
into |\cdot 10^{exponent}|.  A fractional part |/B| if found in the expansion
of the input must be last and will be tranformed into |\cdot B^{-1}| if there
was a scientific part, else the output will be |\frac{A}{B}| (or the \TeX\
equivalent) with |A| the numerator.%
%
\footnote{I am hesitating whether the |\frac{A}{B}| branch choice should
  require |A| to be an integer, or will also, as currently, be done with |A|
  being a number in decimal notation.}
%
\begin{everbatim*}
$\xintTeXfromSci{\xintfloateval{1.1^10000/5}}$, $\xintTeXfromSci{\xinteval{1.1^10000/5}}$\par
\end{everbatim*}
The macro applies complete expansion to its argument, hence its input can be
expressed using \csbxint{FracToSci} but it is simpler to use \csbxint{eval}
which applies and expands completely \csbxint{FracToSci}, as per default
setting of \csbxint{exprPrintOne}.

Note though that the macro can handle only one value.  If one intends to
achieve the same effect but with comma separated or bracketed values, the way
to proceed is via a redefinition of \csbxint{floatexprPrintOne} and/or
\csbxint{exprPrintOne}.
\begin{everbatim*}
\[\def\xintfloatexprPrintOne[#1]#2{\xintTeXfromSci{\xintPFloat[#1]{#2}}}
  \xintfloateval{[10] 2^100, 3^100, 13^100}\] 
\end{everbatim*}
\par\vskip-\belowdisplayskip
\begin{everbatim*}
\[\def\xintexprPrintOne#1{\xintTeXfromSci{\xintFracToSci{#1}}}
  \xinteval{sqrt(2^101,60), 355/113, 6.02e23/1000}\]
\end{everbatim*}% attention à l'espace TeX est en mode horizontal après display
This will however make impossible, due to the added \TeX\ mark-up in the
output, the nesting of \csbxint{floateval}/\csbxint{eval} inside one another.
But the \csbxint{expr}|...\relax| syntax remains usable and is indeed the
recommended way as it is more efficient, only in certain very special
situations does one need the «eval» which produces explicit digits, and then
usually the output will be an integer, so will remain invariant and usable as
before.  Anyway, in the above examples we localized the redefinitions to the
equation construct which is scope-limiting (as is already the underlying \TeX\
math display using |$$|).

Some similar effect is also possible without \csbxint{TeXfromSci},
simply by a customization of \csbxint{PFloatE} like this:
\begin{everbatim*}
\begingroup
\def\xintPFloatE#1.{\cdot10^{#1}.}%
$\xintfloateval{1.1^10000/5}$, $\xinteval{1.1^10000/5}$
\endgroup\newline
\end{everbatim*}%
This method is simpler-minded and will leave trailing |/B|'s «as is», though. 

And of course, with this poor man method too the presence of \TeX\ mark-up
makes nesting dangerous, so we limited by security the definition to a group.

\subsection{\csh{xintTeXFrac}}\label{xintTeXFrac}

This is a typesetting \LaTeX{} only macro\etype{\Ff}, math mode only as it
expands to |\frac{A}{B}10^n| for an input which ends up parsed into raw
format |A/B[n]|.

If the denominator |B| is \dtt{1}, the output is |A\cdot 10^n|. If the
exponent |n| is \dtt{0}, the |[\cdot]10^n| part is omitted.
\begin{everbatim*}
$\xintTeXFrac {178.000/25600000}$, $\xintTeXFrac {178.000/1}$, $\xintTeXFrac {3.5/5.7}$\newline
\end{everbatim*}%
The input, if in a fraction form, is not simplified in any way, except for
transforming numerator and denominator into integers, separating a power of
ten part.  Macros such as \csbxint{Irr}, \csbxint{PIrr}, \csbxint{REZ} can be
inserted to wrap the input and help simplify it.  The minus sign ends up in
the numerator.

It is the new name of |\xintFrac|.\CHANGED{1.4g} The old name is
deprecated and will emit a warning.

\subsection{\csh{xintTeXsignedFrac}}\label{xintTeXsignedFrac}


This is as \csbxint{TeXFrac}\etype{\Ff} except that a negative fraction has the
sign ending up in front, not in the numerator.
\begin{everbatim*}
$\xintTeXFrac {-355/113}=\xintTeXsignedFrac {-355/113}$\newline
\end{everbatim*}%
It is the new name of |\xintSignedFrac|.\CHANGED{1.4g}

\subsection{\csh{xintTeXOver}}\label{xintTeXOver}

This does the same as \csa{xintTeXFrac}\etype{\Ff} except that the \csa{over}
primitive is used for the fraction (in case the denominator is not one; and a
pair of braces contains the |A\over B| part).
\begin{everbatim*}
$\xintTeXOver {178.000/25600000}$, $\xintTeXOver {178.000/1}$, $\xintTeXOver {3.5/5.7}$\newline
\end{everbatim*}%
It is the new name of |\xintFwOver|.\CHANGED{1.4g}

\subsection{\csh{xintTeXsignedOver}}\label{xintTeXsignedOver}

This is as \csbxint{TeXOver}\etype{\Ff} except that a negative fraction has the
sign put in front, not in the numerator.
\begin{everbatim*}
$\xintTeXOver{-355/113}=\xintTeXsignedOver{-355/113}$\newline
\end{everbatim*}%
It is the new name of |\xintSignedFwOver|.\CHANGED{1.4g}

\subsection{\csh{xintLen}}\label{xintLenFrac}

The \csbxint{Len} macro\etype{\Ff} from \xintname is extended to accept a fraction
on input: the length of |A/B[n]| is the length of |A| plus the length of |B|
plus the absolute value of |n| and minus one (an integer input as |N| is
internally represented in a form equivalent to |N/1[0]| so the minus one means
that the extended \csa{xintLen} behaves the same as the original for
integers).
\begin{everbatim*}
\xintLen{201710/298219}=\xintLen{201710}+\xintLen{298219}-1\newline
\xintLen{1234/1}=\xintLen{1234}=\xintLen{1234[0]}=\xintiLen{1234}\newline
\xintLen{-1e3/5.425} (\xintRaw {-1e3/5.425})\par
\end{everbatim*}
The length is computed on the |A/B[n]| which would have been returned by
\csbxint{Raw}, as illustrated by the last example above.

|\xintLen| is only for use with such (scientific) numbers or fractions. See
also \csbxint{NthElt} from \xinttoolsname. See also \csbxint{Length} (which
however does not expand its argument) from \xintkernelname for counting more
general tokens (or rather braced items).

\subsection{\csh{xintNum}}\label{xintNumFrac}

The \csbxint{Num} \etype{\Ff} from \xintname is transformed into a synonym to
\csbxint{TTrunc}.

Attention that for example |\xintNum{1e100000}| expands to the needed
\dtt{100001} digits...

The original \hyperref[xintiNum]{\string\xintNum} from \xintcorename which
does not understand the fraction slash or the scientific notation is still
available under the name \csbxint{iNum}.

\subsection{\csh{xintRaw}}\label{xintRaw}

This macro `prints' the\etype{\Ff}
fraction |f| as it is received by the package after its parsing and
expansion, in a form |A/B[N]| equivalent to the internal
representation: the denominator |B| is always strictly positive and is
printed even if it has value |1|.
\begin{everbatim*}
\xintRaw{\the\numexpr 571*987\relax.123e-10/\the\numexpr-201+59\relax e-7}
\end{everbatim*}

No simplification is done, not even of common zeroes between numerator and
denominator:
\begin{everbatim*}
\xintRaw {178000/25600000}
\end{everbatim*}

\subsection{\csh{xintNumerator}}\label{xintNumerator}

The input data\etype{\Ff} is parsed as if by \csbxint{Raw} into |A/B[N]|
format and
the macro outputs |A| if |N<=0|, or |A| extended by |N| zeroes if |N>0|.
\begin{everbatim*}
\xintNumerator {178000/25600000[17]}\newline
\xintNumerator {312.289001/20198.27}\newline
\xintNumerator {178000e-3/256e5}\newline
\xintNumerator {178.000/25600000}
\end{everbatim*}

\subsection{\csh{xintDenominator}}\label{xintDenominator}

The input data\etype{\Ff} is parsed as if by \csbxint{Raw} into |A/B[N]|
format and
the macro outputs |B| if |N>0|, or |B| extended by \verb+|N|+ zeroes if |N<=0|.
\begin{everbatim*}
\xintDenominator {178000/25600000[17]}\newline
\xintDenominator {312.289001/20198.27}\newline
\xintDenominator {178000e-3/256e5}\newline
\xintDenominator {178.000/25600000}
\end{everbatim*}

\subsection{\csh{xintRawWithZeros}}\label{xintRawWithZeros}

This macro parses the input\etype{\Ff} and outputs |A/B|, with |A|
as would be returned by \csa{xintNumerator}|{f}| and |B| as would be returned by
\csa{xintDenominator}|{f}|.
\begin{everbatim*}
\xintRawWithZeros{178000/25600000[17]}\newline
\xintRawWithZeros{312.289001/20198.27}\newline
\xintRawWithZeros{178000e-3/256e5}\newline
\xintRawWithZeros{178.000/25600000}\newline
\xintRawWithZeros{\the\numexpr 571*987\relax.123e-10/\the\numexpr-201+59\relax e-7}
\end{everbatim*}

\subsection{\csh{xintREZ}}\label{xintREZ}

The input\etype{\Ff} is first parsed into |A/B[N]| as by \csbxint{Raw}, then
trailing zeroes of |A| and |B| are suppressed and |N| is accordingly adjusted.
\begin{everbatim*}
\xintREZ {178000/25600000[17]}
\end{everbatim*}

This macro is used internally by various other constructs; its implementation
was redone entirely at |1.3a|, and it got faster on long inputs.

\subsection{\csh{xintIrr}}\label{xintIrr}

This puts the fraction\etype{\Ff} into its unique irreducible form:
\begin{everbatim*}
\xintIrr {178.256/256.1780}, \xintIrr {178000/25600000[17]}
\end{everbatim*}

The current implementation does not cleverly first factor powers of
2 and 5, and |\xintIrr {2/3[100]}| will execute the
Euclidean division of |2|\raisebox{.5ex}{|.|}|10^{100}| by |3|, which is a bit
stupid as it could have known that the \dtt{100} trailing zeros can not bring
any divisibility by \dtt{3}.

Starting with release |1.08|, \csa{xintIrr} does not remove the trailing |/1|
when the output is an integer. This was deemed better for various (questionable?)
reasons, anyway the output format is since \emph{always} |A/B| with |B>0|, even
in cases where it turns out that |B=1|.
Use \csbxint{PRaw} on top of \csa{xintIrr} if it is needed to get rid of such a
trailing |/1|.

\subsection{\csh{xintPIrr}}\label{xintPIrr}

This puts the fraction\etype{\Ff} into irreducible form,
\emph{keeping as is the
  decimal part} |[N]| from raw internal |A/B[N]| format.
(|P| stands here for \emph{Partial})
\begin{everbatim*}
\xintPIrr {178.256/256.1780}, \xintPIrr {178000/25600000[17]}
\end{everbatim*}

Notice that the output always has the ending |[N]|, which is exactly the
opposite of \csbxint{Irr}'s behaviour. The interest of this macro is mainly in
handling fractions which somehow acquired a big |[N]| (perhaps from input in
scientific notation) and for which the reduced fraction would have a very
large number of digits. This large number of digits can considerably slow-down
computations done afterwards.

For example package \href{http://ctan.org/pkg/polexpr}{polexpr} uses
\csa{xintPIrr} when differentiating a polynomial, or in setting up a Sturm
chain for localization of the real roots of a polynomial. This is relevant to
polynomials whose coefficients were input in decimal notation, as this
automatically creates internally some |[N]|. Keeping and combining those
|[N]|'s during computations significantly increases their speed.

\subsection{\csh{xintJrr}}\label{xintJrr}

This also puts the fraction\etype{\Ff} into its unique irreducible form:
\begin{everbatim*}
\xintJrr {178.256/256.178}
\end{everbatim*}

This is (supposedly, not tested for ages) faster than \csa{xintIrr} for
fractions having some big common factor in the numerator and the denominator.
\begin{everbatim*}
\xintJrr {\xintiiPow{\xintiiFac {15}}{3}/%
              \xintiiPrd{{\xintiiFac{10}}{\xintiiFac{30}}{\xintiiFac{5}}}}
\end{everbatim*}

But to notice the difference one would need computations with much bigger
numbers than in this example. As \csbxint{Irr}, \csa{xintJrr} does not remove
the trailing |/1| from a fraction reduced to an integer.

\subsection{\csh{xintPRaw}}\label{xintPRaw}

|PRaw|\etype{\Ff} stands for ``pretty raw''. It does like \csbxint{Raw} apart
from removing the |[N]| part if |N=0| and removing the |B| if |B=1|.
\begin{everbatim*}
\xintPRaw {123e10/321e10}, \xintPRaw {123e9/321e10}, \xintPRaw {\xintIrr{861/123}}
\end{everbatim*}

\subsection{\csh{xintFracToSci}}\label{xintFracToSci}


\csa{xintFracToSci}\NewWith{1.4} is for usage by \csbxint{eval} for formatting
the output of numbers: the output routine of \csbxint{eval} uses
\csbxint{exprPrintOne} whose current default definition is:
\begin{everbatim}
\let\xintexprPrintOne\xintFracToSci
\end{everbatim}
Any replacement should obey the following blueprint:
\begin{itemize}[noitemsep]
\item to\xtype{} be expandable, but not necessarily \fexpan dable,
\item to accept on input |A|, |A/B|, |A[N]|, or |A/B[N]|, i.e. the ``raw''
      \xintfracname format, but with optional |/B| and |[N]| parts,
      which can be called the ``relaxed raw format''.
\end{itemize}
These constraints\CHANGED{1.4e} are much simplified at |1.4e| (and \csa{xintFracToSci} has
been internally simplified to only have to obey the reduced constraints, which
is a breaking change).

At |1.4e| the handling\CHANGED{1.4e} by this macro of input with a scientific
exponent part has changed. Rather than producing an integer mantissa it now
does as \csbxint{PFloat} (apart from the float-rounding of course) in
particular it trims out trailing zeros.

Attention, \csa{xintFracToSci} does not behave as the other public macros from
\xintfracname:
\begin{itemize}[noitemsep]
\item it is expandable, but not \fexpan dable, so it can't appear as argument
  to other \xintfracname macros without an explicit |\expanded{...}| wrapper
  (as they only \fexpan d their arguments).
\item it expects input already (after \fexpan sion) in ``relaxed raw''
  \xintfracname format.\IMPORTANT{}
\end{itemize}

\noindent\csa{xintFracToSciE} has been removed at |1.4e|, see \csbxint{PFloatE}.

\subsection{\csh{xintDecToStringREZ}}\label{xintDecToStringREZ}

\csa{xintDecToStringREZ}\etype{\Ff} uses fixed point notation
for the output.  The argument is first parsed in the same way as for any other
\xintfracname macros,\NewWith{1.4e} which means that it is first transformed into an
internal format having a numerator |A|, a denominator |B| and a power of ten
exponent |N|.  The following recipe applies:
\begin{itemize}[noitemsep]
\item the zero value is printed as \dtt{\xintDecToStringREZ{0}} (no decimal point).
\item trailing zeros of |A| and |B| are removed and |N| is adjusted,
\item if the new |B| is not \dtt{1}, it will appear in the output as |/B|,
\item fixed point notation is used for |AeN|:
  \begin{itemize}[noitemsep]
  \item if |N| is non-negative, the output is an integer with |N| trailing
    zeros,
  \item if |N| is negative a decimal point is used, and if |AeN| is less than
    one in absolute value, output will start with \dtt{0.} (with a decimal point).
  \end{itemize}
\end{itemize}
Please note the following:
\begin{enumerate}[noitemsep]
\item the fraction |AeN/B| or even |A/B| is not pre-reduced into lowest terms,
\item the macro does not check if |B| contains only powers of \dtt{2} and
  \dtt{5}, so |1/2| is printed as \dtt{\xintDecToString{1/2}}, not as \dtt{0.5}.
\end{enumerate}
The definitive behaviour remains to be decided regarding these two points.

\begin{everbatim*}
\xintDecToStringREZ{0}, \xintDecToStringREZ{1/2}, \xintDecToStringREZ{0.5000}\newline
\xintDecToStringREZ{1.23456789e5}, \xintDecToStringREZ {1.23456789e-3}\newline
\xintDecToStringREZ{12345e-1}, \xintDecToStringREZ {12345e-2}, \xintDecToStringREZ{12345e-3}\newline
\xintDecToStringREZ{12345e-4}, \xintDecToStringREZ {12345e-5}, \xintDecToStringREZ{12345e-6}\newline
\xintDecToStringREZ{1.234567890000e12}, \xintDecToStringREZ{1.23456000e-5/10}\newline
\xintDecToStringREZ{70/14} % is not reduced to lowest terms
\end{everbatim*}

\subsection{\csh{xintDecToString}}\label{xintDecToString}

\csa{xintDecToString}\etype{\Ff} uses fixed point notation for the output.  It
was introduced at |1.3| as experimental backport from a
\href{http://ctan.org/pkg/polexpr}{polexpr} macro, and its behaviour remains
somewhat undecided in particular regarding whether it should identify inputs
which correspond to decimal numbers, \emph{after reduction to lowest terms}.

It follows the same rules as \csbxint{DecToStringREZ} except that it does not
remove trailing zeros, in fact \csbxint{DecToStringREZ}|{f}| is defined as
\csbxint{DecToString}|{|\csbxint{REZ}|{f}}|.

\begin{everbatim*}
\xintDecToString{0}, \xintDecToString{1/2}, \xintDecToString{0.5000}\newline
\xintDecToString{1.23456789e5}, \xintDecToString {1.23456789e-3}\newline
\xintDecToString{12345e-1}, \xintDecToString {12345e-2}, \xintDecToString{12345e-3}\newline
\xintDecToString{12345e-4}, \xintDecToString {12345e-5}, \xintDecToString{12345e-6}\newline
\xintDecToString{1.234567890000e12}, \xintDecToString{1.23456000e-5/10}\newline
\xintDecToString{70/14}
\end{everbatim*}

Since |1.4e|\CHANGED{1.4e}, \csbxint{DecToString} is the default for
\csbxint{iexprPrintOne}, which governs the \csbxint{ieval} output format (in
this use case there is never a |/B| fractional part).
\begin{everbatim}
\let\xintiexprPrintOne\xintDecToString
\end{everbatim}

Any replacement of \csbxint{iexprPrintOne} should obey the following blueprint:
\begin{itemize}[noitemsep]
\item to\xtype{} be expandable, but not necessarily \fexpan dable,
\item to accept on input |A| or |A[N]|.
\end{itemize}

\subsection{\csh{xintTrunc}}\label{xintTrunc}

\csa{xintTrunc}|{x}{f}|\etype{\numx\Ff} returns the start of the decimal
expansion of the fraction |f|, truncated to:
\begin{itemize}
\item if |x>0|, |x| digits after the decimal mark,
\item if |x=0|, an integer,
\item if |x<0|, an integer multiple of |10^{-x}| (in scientific notation).\NewWith{1.4a}
\end{itemize}
The output is the sole digit token \dtt{0} if and only if the input was exactly
zero; else it contains always either a decimal mark (even if |x=0|) or a
scientific part and it conserves the sign of |f| (even if the truncated value
represents the zero value).

Truncation is done towards zero.
\begin{everbatim*}
\begin{multicols}{2}
  \noindent\xintFor* #1 in {\xintSeq[-1]{7}{-14}}:{\xintTrunc{#1}{-11e12/7}\newline}%
  \xintTrunc{10}{1e-11}\newline \xintTrunc{10}{1/65536}\par
\end{multicols}
\end{everbatim*}

\begin{framed}
  \textbf{Warning:} \emph{it is not yet decided is the current behaviour is
    definitive.}

  Currently \xintfracname has no notion of a positive zero or a negative zero.
  Hence transitivity of \csbxint{Trunc} is broken for the case where the first
  truncation gives on output \dtt{0.00...0} or \dtt{-0.00...0}: a second
  truncation to less digits will then output \dtt{0}, whereas if it had been
  applied directly to the initial input it would have produced \dtt{0.00...0}
  or respectively \dtt{-0.00...0} (with less zeros after decimal mark).

  If \xintfracname distinguished zero, positive zero, and
  negative zero then it would be possible to maintain transitivity.

  The problem would also be fixed, even without distinguishing a negative zero
  on input, if \csbxint{Trunc} always produced \dtt{0.00...0} (with no sign)
  when the mathematical result is zero, discarding the information on original
  input being positive, zero, or negative.

  I have multiple times hesitated about what to do and must postpone again
  final decision.
\end{framed}

\subsection{\csh{xintXTrunc}}\label{xintXTrunc}


\csa{xintXTrunc}|{x}{f}|\xtype{\numx\Ff} is similar to \csbxint{Trunc} with
the following important differences:
\begin{itemize}[nosep]
\item it is completely expandable but not
\fexpan dable, as is indicated by the hollow star in the margin,
\item hence it can not be used as argument to the other package macros, but as
  it \fexpan ds its |{f}| argument, it accepts arguments expressed with other
  \xintfracname macros,
\item it requires |x>0|,
\item contrarily to \csbxint{Trunc} the number of digits on output is not
  limited to about \dtt{19950} and may go well beyond \dtt{100000} (this is
  mainly useful for outputting a decimal expansion to a file),
\item when the mathematical result is zero, it always prints it as
  \dtt{0.00...0} or \dtt{-0.00...0} with |x| zeros after the decimal mark.
\end{itemize}

\textbf{Warning:} 
transitivity is broken too (see discussion of \csbxint{Trunc}), due to the
sign in the last item. Hence \emph{the definitive policy is yet to be fixed.}

Transitivity is here in the sense of using a first |\edef| and then a second
one, because it is not possible to nest \csb{xintXTrunc} directly as argument
to itself. Besides, although the number of digits on output isn't limited,
nevertheless |x| should be less than about |19970| when the number of digits
of the input (assuming it is expressed as a decimal number) is even bigger:
|\xintXTrunc{30000}{\Z}| after |\edef\Z{\xintXTrunc{60000}{1/66049}| raises an
error in contrast with a direct |\xintXTrunc{30000}{1/66049}|. But
|\xintXTrunc{30000}{123.456789}| works, because here the number of digits
originally present is smaller than what is asked for, thus the routine only
has to add trailing zeros, and this has no limitation (apart from \TeX\ main
memory).

\csbxint{XTrunc} will expand fully in an |\edef| or a |\write| (|\message|,
|\wlog|, \dots) or in an \csbxint{expr}-ession, or as list argument to
\csbxint{For*}.

Here is an example session where the
user checks that the decimal expansion of $1/66049=1/257^2$ has the maximal
period length $257*256=65792$ (this period length must be a divisor of
$\phi(66049)$ and to check it is the maximal one it is enough to show that
neither $32896$ nor $256$ are periods.)

\begingroup\small
\everb|@
$ rlwrap etex -jobname worksheet-66049
This is pdfTeX, Version 3.14159265-2.6-1.40.17 (TeX Live 2016) (preloaded format=etex)
 restricted \write18 enabled.
**xintfrac.sty
entering extended mode
(/usr/local/texlive/2016/texmf-dist/tex/generic/xint/xintfrac.sty
(/usr/local/texlive/2016/texmf-dist/tex/generic/xint/xint.sty
(/usr/local/texlive/2016/texmf-dist/tex/generic/xint/xintcore.sty
(/usr/local/texlive/2016/texmf-dist/tex/generic/xint/xintkernel.sty))))
*% we load xinttools for \xintKeep, etc... \xintXTrunc itself has no more

*% any dependency on xinttools.sty since 1.2i

*\input xinttools.sty
(/usr/local/texlive/2016/texmf-dist/tex/generic/xint/xinttools.sty)
*\def\m#1;{\message{#1}}

*\m \the\numexpr 257*257\relax;
66049
*\m \the\numexpr 257*256\relax;
65792
*% Thus 1/66049 will have a period length dividing 65792.

*% Let us first check it is indeed periodical.

*\edef\Z{\xintXTrunc{66000}{1/66049}}

*% Let's display the first decimal digits.

*\m \xintXTrunc{208}{\Z};

0.00001514027464458205271843631243470756559523989765174340262532362337052794137
6856576178291874214598252812306015231116292449545034746930309315810988811337037
6538630410755651107511090251177156353616254598858423
*% let's now fetch the trailing digits

*\m \xintKeep{65792-66000}{\Z};% 208 trailing digits

0000151402746445820527184363124347075655952398976517434026253236233705279413768
5657617829187421459825281230601523111629244954503474693030931581098881133703765
38630410755651107511090251177156353616254598858423
*% yes they match! we now check that 65792/2 and 65792/257=256 aren't periods.

*\m \xintXTrunc{256}{\Z};

0.00001514027464458205271843631243470756559523989765174340262532362337052794137
6856576178291874214598252812306015231116292449545034746930309315810988811337037
6538630410755651107511090251177156353616254598858423291798513225029902042423049
554118911717058547442
*\m \xintXTrunc{256+256}{\Z};

0.00001514027464458205271843631243470756559523989765174340262532362337052794137
6856576178291874214598252812306015231116292449545034746930309315810988811337037
6538630410755651107511090251177156353616254598858423291798513225029902042423049
5541189117170585474420505987978621932201850141561567926842192917379521264515738
3154930430438008145467758785144362518736089872670290239064936637950612424109373
3440324607488379839210283274538600130206361943405653378552286938485064119063119
8049932625777831609865402958409665551333
*% now with 65792/2=32896. Problem: we can't do \xintXTrunc{32896+100}{\Z}

*% but only direct \xintXTrunc{32896+100}{1/66049}. Anyway we want to nest it

*% hence let's do it all with (slower) \xintKeep, \xintKeepUnbraced.

*\m \xintKeep {-100}{\xintKeepUnbraced{2+65792/2+100}{\Z}};

9999848597253554179472815636875652924344047601023482565973746763766294720586231
434238217081257854017
*% This confirms 32896 isn't a period length.

*% To conclude let's write the 66000 digits to the log.

*\wlog{\Z}

*% We want always more digits:

*\wlog{\xintXTrunc{150000}{1/66049}}

*\bye
|
\endgroup % $ à cause de fontification de AUCTeX.

The acute observer will have noticed that there is something funny when one
compares the first digits with those after the middle-period:
\begin{everbatim}
0000151402746445820527184363124347075655952398976517434026253236233705279413768...
9999848597253554179472815636875652924344047601023482565973746763766294720586231...
\end{everbatim}
Mathematical exercise: can you explain why the two indeed add to |9999...9999|?

You can try your hands at this simpler one:
\begin{everbatim*}
1/49=\xintTrunc{42+5}{1/49}...\newline
\xintTrim{2}{\xintTrunc{21}{1/49}}\newline
\xintKeep{-21}{\xintTrunc{42}{1/49}}
\end{everbatim*}

This was again an example of the type |1/N| with |N| the square of a prime.
One can also find counter-examples within this class: |1/31^2| and |1/37^2|
have an odd period length (|465| and respectively |111|) hence they can not
exhibit the symmetry.

\begin{framed}
  Mathematical challenge: prove generally that if the period length of the
  decimal expansion of |1/p^r| (with |p| a prime distinct from |2| and |5| and
  |r| a positive exponent) is even, then the previously observed symmetry
  about the two halves of the period adding to a string of nine's applies.
\end{framed}



\subsection{\csh{xintTFrac}}\label{xintTFrac}

\csa{xintTFrac}|{f}|\etype{\Ff} returns the fractional part,
|f=trunc(f)+frac(f)|. Thus if |f<0|, then |-1<frac(f)<=0| and if |f>0| one has
|0<= frac(f)<1|. The |T| stands for `Trunc', and there should exist also
similar macros associated respectively with `Round', `Floor', and `Ceil', each
type of rounding to an integer deserving arguably to be associated with a
fractional ``modulo''. By sheer laziness, the package currently implements
only the ``modulo'' associated with `Truncation'. Other types of modulo may be
obtained more cumbersomely via a combination of the rounding with a subsequent
subtraction from |f|.

Notice that the result is filtered through \csbxint{REZ}, and will thus be of
the form |A/B[N]|, where neither |A| nor |B| has trailing zeros. But the
output fraction is not reduced to smallest terms.

The function call in expressions (\csbxint{expr}, \csbxint{floatexpr}) is
|frac|. Inside |\xintexpr..\relax|, the function |frac| is mapped to
\csa{xintTFrac}. Inside |\xintfloatexpr..\relax|, |frac| first applies
\csa{xintTFrac} to its argument (which may be an exact fraction with more
digits than the floating point precision) and only in a second stage makes the
conversion to a floating point number with the precision as set by |\xintDigits|
(default is \dtt{16}).
\begin{everbatim*}
\xintTFrac {1235/97}, \xintTFrac {-1235/97}\newline
\xintTFrac {1235.973}, \xintTFrac {-1235.973}\newline
\xintTFrac {1.122435727e5}\par
\end{everbatim*}

\subsection{\csh{xintRound}}\label{xintRound}

\csa{xintRound}|{x}{f}|\etype{\numx\Ff} returns the start of the decimal
expansion of the fraction |f|, rounded to:
\begin{itemize}
\item if |x>0|, |x| digits after the decimal mark,
\item if |x=0|, an integer,
\item if |x<0|, an integer multiple of |10^{-x}| (in scientific notation).\NewWith{1.4a}
\end{itemize}
The output is the sole digit token \dtt{0} if and only if the input was
exactly zero; else it contains always either a decimal mark (even if |x=0|) or
a scientific part and it conserves the sign of |f| (even if the rounded value
represents the zero value).
\begin{everbatim*}
\begin{multicols}{2}
  \noindent\xintFor* #1 in {\xintSeq[-1]{7}{-14}}:{\xintRound{#1}{-11e12/7}\newline}%
  \xintRound{10}{1e-11}\newline \xintRound{10}{1/65536}\newline
\end{multicols}
\end{everbatim*}%
Rounding is done with half-way numbers going towards infinity of
the same sign.

\subsection{\csh{xintFloor}}\label{xintFloor}

|\xintFloor {f}|\etype{\Ff} returns the largest relative integer |N| with
|N|${}\leqslant{}$|f|.
\begin{everbatim*}
\xintFloor {-2.13}, \xintFloor {-2}, \xintFloor {2.13}
\end{everbatim*}
Note the trailing |[0]|, see \csbxint{iFloor} if it is not desired.

\subsection{\csh{xintCeil}}\label{xintCeil}

|\xintCeil {f}|\etype{\Ff} returns the smallest relative integer |N| with
|N|${}>{}$|f|.
\begin{everbatim*}
\xintCeil {-2.13}, \xintCeil {-2}, \xintCeil {2.13}
\end{everbatim*}

\subsection{\csh{xintiTrunc}}\label{xintiTrunc}

\csa{xintiTrunc}|{x}{f}|\etype{\numx\Ff} returns the integer equal to |10^x|
times what \csa{xintTrunc}|{x}{f}| would produce. Attention that leading zeros
are automatically removed: the output is in strict integer format.
\begin{everbatim*}
\begin{multicols}{2}
  \noindent\xintFor* #1 in {\xintSeq[-1]{7}{-14}}:{\xintiTrunc{#1}{-11e12/7}\newline}%
  \xintiTrunc{10}{1e-11}\newline \xintiTrunc{10}{1/65536}\par
\end{multicols}
\end{everbatim*}

\subsection{\csh{xintTTrunc}}\label{xintTTrunc}

\csa{xintTTrunc}|{f}|\etype{\Ff} truncates to an integer (truncation towards
zero). This is the same as |\xintiTrunc {0}{f}| and also the same as
\csbxint{Num}.

\subsection{\csh{xintiRound}}\label{xintiRound}

\csa{xintiRound}|{x}{f}|\etype{\numx\Ff} returns the integer equal to |10^x|
times what \csa{xintRound}|{x}{f}| would return. The output has no leading
zeroes, it is always in strict integer format.
\begin{everbatim*}
\begin{multicols}{2}
  \noindent\xintFor* #1 in {\xintSeq[-1]{7}{-14}}:{\xintiRound{#1}{-11e12/7}\newline}%
  \xintiRound{10}{1e-11}\newline \xintiRound{10}{1/65536}\par
\end{multicols}
\end{everbatim*}

\subsection{\csh{xintiFloor}}\label{xintiFloor}

|\xintiFloor {f}|\etype{\Ff} does the same as \csbxint{Floor} but without the
trailing |/1[0]|.
\begin{everbatim*}
\xintiFloor {-2.13}, \xintiFloor {-2}, \xintiFloor {2.13}
\end{everbatim*}

\subsection{\csh{xintiCeil}}\label{xintiCeil}

|\xintiCeil {f}|\etype{\Ff} does the same as \csbxint{Ceil} but its output is
without the |/1[0]|.
\begin{everbatim*}
\xintiCeil {-2.13}, \xintiCeil {-2}, \xintiCeil {2.13}
\end{everbatim*}

\subsection{\csh{xintE}}\label{xintE}

|\xintE {f}{x}|\etype{\Ff\numx} multiplies the fraction |f| by $10^x$. The
\emph{second} argument |x| must obey the \TeX{} bounds. Example:
\begin{everbatim*}
\count 255 123456789 \xintE {10}{\count 255}
\end{everbatim*}
Don't feed this example to \csbxint{Num}!

\subsection{\csh{xintCmp}}\label{xintCmp}

This\etype{\Ff\Ff} compares two fractions |F| and |G| and produces
|-1|, |0|, or |1| according to |F<G|, |F=G|, |F>G|.

For choosing branches according to the result of comparing |f| and |g|, see
\csbxint{ifCmp}.

\subsection{\csh{xintEq}}\label{xintEq}

|\xintEq{f}{g}|\etype{\Ff\Ff} returns 1 if |f=g|, 0 otherwise.

\subsection{\csh{xintNotEq}}\label{xintNotEq}

|\xintNotEq{f}{g}|\etype{\Ff\Ff} returns 0 if |f=g|, 1 otherwise.


\subsection{\csh{xintGeq}}\label{xintGeq}

This\etype{\Ff\Ff} compares the \emph{absolute values} of two
fractions.
|\xintGeq{f}{g}| outputs |1| if {\catcode`| 12 $|f|\geqslant|g|$} and |0|
if not.

Important: the macro compares \emph{absolute values}.

\subsection{\csh{xintGt}}\label{xintGt}

|\xintGt{f}{g}|\etype{\Ff\Ff} returns \dtt{1} if |f|$>$|g|, \dtt{0} otherwise.

\subsection{\csh{xintLt}}\label{xintLt}

|\xintLt{f}{g}|\etype{\Ff\Ff} returns \dtt{1} if |f|$<$|g|, \dtt{0} otherwise.

\subsection{\csh{xintGtorEq}}\label{xintGxstorEq}

|\xintGtorEq{f}{g}|\etype{\Ff\Ff} returns \dtt{1} if |f|$\geqslant$|g|, \dtt{0} otherwise.
Extended by \xintfracname to fractions.

\subsection{\csh{xintLtorEq}}\label{xintLtorEq}

|\xintLtorEq{f}{g}|\etype{\Ff\Ff} returns \dtt{1} if |f|$\leqslant$|g|, \dtt{0} otherwise.

\subsection{\csh{xintIsZero}}\label{xintIsZero}

|\xintIsZero{f}|\etype{f} returns \dtt{1} if |f=0|, \dtt{0} otherwise.

\subsection{\csh{xintIsNotZero}}\label{xintIsNotZero}

|\xintIsNotZero{f}|\etype{f} returns \dtt{1} if |f!=0|, \dtt{0} otherwise.

\subsection{\csh{xintIsOne}}\label{xintIsOne}

|\xintIsOne{f}|\etype{f} returns \dtt{1} if |f=1|, \dtt{0} otherwise.

\subsection{\csh{xintOdd}}\label{xintOdd}

|\xintOdd{f}|\etype{f} returns \dtt{1} if the integer obtained by truncation is
odd, and \dtt{0} otherwise.

\subsection{\csh{xintEven}}\label{xintEven}

|\xintEven{f}|\etype{f} returns \dtt{1} if the integer obtained by truncation is
even, and \dtt{0} otherwise.

\subsection{\csh{xintifSgn}}\label{xintifSgn}

\csh{xintifSgn}\marg{f}\marg{A}\marg{B}\marg{C}\etype{\Ff nnn} executes either the
\meta{A}, \meta{B} or \meta{C} code, depending on its first argument being
respectively negative, zero, or positive.

\subsection{\csh{xintifZero}}\label{xintifZero}

\csa{xintifZero}\marg{f}\marg{IsZero}\marg{IsNotZero}\etype{\Ff nn} expandably
checks if the first mandatory argument |N| (a number, possibly a fraction if
\xintfracname is loaded, or a macro expanding to one such) is zero or not. It
then either executes the first or the second branch.

Beware that both branches must be present.

\subsection{\csh{xintifNotZero}}\label{xintifNotZero}

\csa{xintifNotZero}\marg{N}\marg{IsNotZero}\marg{IsZero}\etype{\Ff nn}
expandably checks if the first mandatory argument |f| is not
zero or is zero. It then either executes the first or the second branch.

Beware that both branches must be present.

\subsection{\csh{xintifOne}}\label{xintifOne}

\csa{xintifOne}\marg{N}\marg{IsOne}\marg{IsNotOne}\etype{\Ff nn} expandably
checks if the first mandatory argument |f| is one or not one. It
then either executes the first or the second branch. Beware that both branches
must be present.

\subsection{\csh{xintifOdd}}\label{xintifOdd}

\csa{xintifOdd}\marg{N}\marg{odd}\marg{not odd}\etype{\Ff nn} expandably
checks if the first mandatory argument |f|, after truncation to an integer, is
odd or even. It then executes accordingly the first or the second branch.
Beware that both branches must be present.

\subsection{\csh{xintifCmp}}\label{xintifCmp}

\csa{xintifCmp}\marg{f}\marg{g}\marg{if f<g}\marg{if f=g}\marg{if
  f>g}\etype{\Ff\Ff nnn} compares its first two arguments and chooses accordingly
the correct branch.

\subsection{\csh{xintifEq}}\label{xintifEq}

\csa{xintifEq}\marg{f}\marg{g}\marg{YES}\marg{NO}\etype{\Ff\Ff nn} checks
equality of its two first arguments and executes accordingly the |YES| or the
|NO| branch.

\subsection{\csh{xintifGt}}\label{xintifGt}

\csa{xintifGt}\marg{f}\marg{g}\marg{YES}\marg{NO}\etype{\Ff\Ff nn}
checks if $f>g$ and in that case executes the |YES| branch.

\subsection{\csh{xintifLt}}\label{xintifLt}

\csa{xintifLt}\marg{f}\marg{g}\marg{YES}\marg{NO}\etype{\Ff\Ff nn}
checks if $f<g$ and in that case executes the |YES| branch.

\subsection{\csh{xintifInt}}\label{xintifInt}

\csa{xintifInt}|{f}{YES branch}{NO branch}|\etype{\Ff nn} expandably chooses
the |YES| branch if |f| reveals itself after expansion and simplification to
be an integer.

\subsection{\csh{xintSgn}}\label{xintSgn}

The sign of a fraction.\etype{\Ff}

\subsection{\csh{xintOpp}}\label{xintOpp}

The opposite of a fraction.\etype{\Ff}
Note that |\xintOpp {3}| produces \dtt{\xintOpp
  {3}} whereas |\xintiiOpp {3}| produces \dtt{\xintiiOpp {3}}.

\subsection{\csh{xintAbs}}\label{xintAbs}

The absolute value\etype{\Ff}. Note that |\xintAbs {-2}|\dtt{=\xintAbs {-2}}
where |\xintiiAbs {-2}| outputs \dtt{=\xintiiAbs {-2}}.

\subsection{\csh{xintAdd}}\label{xintAdd}

Computes the addition\etype{\Ff\Ff} of two fractions.

Since |1.3| always uses the least common multiple of the
denominators.

\subsection{\csh{xintSub}}\label{xintSub}

Computes the difference\etype{\Ff\Ff} of two fractions (|\xintSub{F}{G}|
computes |F-G|).

Since |1.3| always uses the least common multiple of the
denominators.

\subsection{\csh{xintMul}}\label{xintMul}

Computes the product\etype{\Ff\Ff} of two fractions.

Output is not reduced to smallest terms.

\subsection{\csh{xintDiv}}\label{xintDiv}

Computes the quotient \etype{\Ff\Ff} of two fractions.
(|\xintDiv{F}{G}| computes |F/G|).

Output is not reduced to smallest terms.

\subsection{\csh{xintDivFloor}}
\label{xintDivFloor}

Computes the quotient \etype{\Ff\Ff} of two arguments then apply floor
function to get an integer (in strict format). This macro was defined at |1.1|
(but was left not documented until |1.3a|...) and changed at |1.2p|, formerly
it appended |/1[0]| to output.
\begin{everbatim*}
\xintDivFloor{-170/3}{23/2}
\end{everbatim*}

\subsection{\csh{xintMod}}
\label{xintMod}

Computes the remainder associated to the floored division\etype{\Ff\Ff}
\csbxint{DivFloor}. Prior to |1.2p| the meaning was the one of
\csbxint{ModTrunc}. Was left undocumented until |1.3a|.
\begin{everbatim*}
\xintMod{-170/3}{23/2}
\end{everbatim*}

Modified at |1.3| to use a l.c.m. for the denominator of the result.


\subsection{\csh{xintDivMod}}
\label{xintDivMod}

Computes both the floored division and the remainder\etype{\Ff\Ff}
\csbxint{DivFloor}. New at |1.2p| and documented at |1.3a|.
\begin{everbatim*}
\oodef\foo{\xintDivMod{-170/3}{23/2}}\meaning\foo
\end{everbatim*}

\subsection{\csh{xintDivTrunc}}
\label{xintDivTrunc}

Computes the quotient \etype{\Ff\Ff} of two arguments then 
truncates to an integer  (in strict format).
\begin{everbatim*}
\xintDivTrunc{-170/3}{23/2}
\end{everbatim*}

\subsection{\csh{xintModTrunc}}
\label{xintModTrunc}

Computes the remainder\etype{\Ff\Ff} associated with the truncated division of
two arguments. Prior to |1.2p| it was named \csbxint{Mod}, but the latter then
got associated with floored division.
\begin{everbatim*}
\xintModTrunc{-170/3}{23/2}
\end{everbatim*}

Modified at |1.3| to use a l.c.m. for the denominator of the result.


\subsection{\csh{xintDivRound}}
\label{xintDivRound}

Computes the quotient \etype{\Ff\Ff} of the two arguments then rounds to an
integer (in strict format).
\begin{everbatim*}
\xintDivRound{-170/3}{23/2}
\end{everbatim*}

\subsection{\csh{xintSqr}}\label{xintSqr}

Computes the square\etype{\Ff} of one fraction.

\subsection{\csh{xintPow}}\label{xintPow}

\csa{xintPow}{|{f}{x}|}:\etype{\Ff\Numf} computes |f^x| with |f| a fraction
and the exponent |x| possibly also, but if only \xintfracname is loaded it
will be truncated to an integer.

At |1.4e|\CHANGED{1.4e} the behaviour of the macro is enhanced if
\xintexprname is loaded, at it then becomes the support macro for powers
|a^b|, |a**b| (and the \func{pow} function) in \csbxint{eval}: it now
handles also non-integer exponents.  Also, if the exponent is an integer, it
checks a priori if an exact evaluation
would produce more than about \dtt{10000} digits and then does in its place a
floating point evaluation.

The check whether the exponent is integer is not on the mathematical value but
on the format (for reasons of efficiency). So |4/2| will not be recognized as
integer and it will thus trigger usage of the floating point evaluations;
however |2.0| will be recognized as an integer, as of course |2|.

If the exponent is considered an integer it is then checked if it is less than
\dtt{10000} (in absolute value) and if the output would contain less than
\dtt{10000} digits (separately for numerator and denominator) and only
then\CHANGED{1.4e} is the power computed exactly.  Else it is computed as by
\csbxint{FloatPower} (but the output uses raw |A[N]| format not scientific
notation). Use \csbxint{iiPow} (on integers only, not fractions) for exact
powers with larger exponents.

Also, a check is done whether the exponent is half-integer. Again this check
is not on the value but on the format, so |2.5| is an half integer, as is
|25e-1|, or |2.50| but |5/2| is not considered an half-integer (for reasons of
internal efficiency). If the exponent is half-integer\CHANGED{1.4e} the power
is computed by combining suitably \csbxint{FloatPower} with \csbxint{FloatSqrt}
(but the output uses raw |A[N]| format not scientific notation).

If the exponent is neither an integer nor an half-integer, the power is
computed using\CHANGED{1.4e} logarithm and exponential based approach (and uses raw |A[N]|
output format). If |Digits| is at most \dtt{8} (which triggers
\ctanpackage{poormanlog} usage, for very fast logarithms but only with about
\dtt{8} or \dtt{9} accurate fractional digits) this will start being
inaccurate in the last digit already with fractional exponents |x > 10|. It
is recommended to split then the exponent into an integer or half-integer part
and a fractional part. Powers with integer or half-integer exponents, even
very big, are always computed accurately, for any value of |Digits|.

% Note: in all cases where the macro has been
% extended at |1.4e| to proceed via a floating point evaluation, the output is
% in raw |A[N]| format.




Within an \csbxint{iiexpr}|..\relax| the infix operators |^| and |**| are
mapped to \csbxint{iiPow} and powers are always computed exactly even if they
would produce more than \dtt{10000} digits and melt your CPU; within an
\csbxint{expr}-ession |^| and |**| are mapped to \csbxint{Pow} as described
here.

\subsection{\csh{xintFac}}\label{xintFac}

This is a convenience variant of \csbxint{iiFac} which applies \csbxint{Num}
to its argument\etype{\Numf}. Notice however that the output will have a trailing
|[0]| according to the \xintfracname format for integers.


\subsection{\csh{xintBinomial}}\label{xintBinomial}

This is a convenience variant of \csbxint{iiBinomial} which applies
\csbxint{Num} to its arguments\etype{\Numf\Numf}. Notice however that the
output will have a trailing |[0]| according to the \xintfracname format for
integers.


\subsection{\csh{xintPFactorial}}\label{xintPFactorial}

This is a convenience variant of \csbxint{iiPFactorial} which applies
\csbxint{Num} to its arguments\etype{\Numf\Numf}. Notice however that the
output will have a trailing |[0]| according to the \xintfracname format for
integers.


\subsection{\csh{xintMax}}\label{xintMax}

The maximum of two fractions.\etype{\Ff\Ff} Beware that |\xintMax {2}{3}|
produces \dtt{\xintMax {2}{3}}. The original, for use with
integers only with no need of normalization, is available as \csbxint{iiMax}:
|\xintiiMax {2}{3}=|\dtt{\xintiiMax {2}{3}}.\etype{ff}

\begin{everbatim*}
\xintMax {2.5}{7.2}
\end{everbatim*}

\subsection{\csh{xintMin}}\label{xintMin}

The minimum of two fractions.\etype{\Ff\Ff} Beware that |\xintMin {2}{3}|
produces \dtt{\xintMin {2}{3}}. The original, for use with 
integers only with no need of normalization, is available as \csbxint{iiMin}:
|\xintiiMin {2}{3}=|\dtt{\xintiiMin {2}{3}}.\etype{ff}

\begin{everbatim*}
\xintMin {2.5}{7.2}
\end{everbatim*}

\subsection{\csh{xintMaxof}}\label{xintMaxof}

The maximum of any number of fractions, each within braces, and the whole
thing within braces. \etype{f{$\to$}{\lowast\Ff}}

\begin{everbatim*}
\xintMaxof {{1.23}{1.2299}{1.2301}} and \xintMaxof {{-1.23}{-1.2299}{-1.2301}}
\end{everbatim*}

\subsection{\csh{xintMinof}}\label{xintMinof}

The minimum of any number of fractions, each within braces, and the whole
thing within braces. \etype{f{$\to$}{\lowast\Ff}}

\begin{everbatim*}
\xintMinof {{1.23}{1.2299}{1.2301}} and \xintMinof {{-1.23}{-1.2299}{-1.2301}}
\end{everbatim*}

\subsection{\csh{xintSum}}\label{xintSum}

This\etype{f{$\to$}{\lowast\Ff}} computes the sum of fractions. The output
will now always be in the form |A/B[n]|. The original, for big integers only
(in strict format), is available as \csa{xintiiSum}.

\begin{everbatim*}
\xintSum {{1282/2196921}{-281710/291927}{4028/28612}}
\end{everbatim*}

No simplification attempted.

\subsection{\csh{xintPrd}}\label{xintPrd}

TThis\etype{f{$\to$}{\lowast\Ff}} computes the product of fractions. The output
will now always be in the form |A/B[n]|. The original, for big integers only
(in strict format), is available as \csa{xintiiPrd}.

\begin{everbatim*}
\xintPrd {{1282/2196921}{-281710/291927}{4028/28612}}
\end{everbatim*}

No simplification attempted.

\begin{everbatim*}
$\xintIsOne {21921379213/21921379213}\neq\xintIsOne {1.00000000000000000000000000000001}$
\end{everbatim*}

\subsection{\csh{xintGCD}}\label{xintGCD}

The greatest common divisor of its two arguments, which are possibly
\emph{fractions}.\etype{\Ff\Ff}

Prior to |1.4| a macro of the same name existed in \xintgcdname. But
it truncated its two arguments to integers via \csbxint{Num}.

See \csbxint{iiGCD} for the integer only variant.

\subsection{\csh{xintLCM}}\label{xintLCM}

The least common multiple of its two arguments, which are possibly
\emph{fractions}.\etype{\Ff\Ff}

Prior to |1.4| a macro of the same name existed in \xintgcdname. But
it truncated its two arguments to integers via \csbxint{Num}.

See \csbxint{iiLCM} for the integer only variant.

\subsection{\csh{xintGCDof}}\label{xintGCDof}

\csa{xintGCDof}|{{a}{b}{c}...}|\etype{f{$\to$}{\lowast\Ff}} computes the
greatest common divisor of |a|, |b|, \dots{}. The arguments are allowed to be
\emph{fractions}: the macro produces the non-negative generator of the
fractional ideal they generate. The list argument may be a macro as it is
\fexpan ded first. If all arguments vanish, then also the output.

Prior to |1.4| a macro of the same name existed in \xintgcdname. But
it truncated all its arguments to integers via \csbxint{Num} and then
proceeded with integer only computations.

See \csbxint{iiGCDof} for the integer only variant.

% Semble encore vrai à 1.4d
% Mais je n'ai testé que sur un exemple...
% (which is about |6X| faster than this one for integer arguments).

\subsection{\csh{xintLCMof}}\label{xintLCMof}

\csa{xintLCMof}|{{a}{b}{c}...}|\etype{f{$\to$}{\lowast\Ff}} computes the least
common multiple of |a|, |b|, \dots{}.  The arguments are allowed to be
\emph{fractions}: the macro produces the non-negative generator of the
intersection of the corresponding fractional ideals. The list argument may be
a macro, it is \fexpan ded first. If one of the item vanishes, then also the
output.

Prior to |1.4| a macro of the same name existed in \xintgcdname. But
it truncated all its arguments to integers via \csbxint{Num}.

See \csbxint{iiLCMof} for the integer only variant.

% Avant 1.4d on avait ceci :
% (which is about |9X| faster han this one for integer arguments).
% mais à 1.4d le lcm des fractions est environ 4X fois plus efficace,
% en ce qui concerne son emploi avec des entiers (testé sur un seul exemple)
% donc le gain de faire \xintiiexpr n'est plus que 2X !

\subsection{\csh{xintDigits}, \csh{xinttheDigits}}
\label{xintDigits}
\label{xinttheDigits}

The syntax |\xintDigits := D;| assigns the value of |D| to the number of
digits to be used by floating point operations (this uses internally a
|\mathchardef| assignement, and |D| stands for (or expands to) a legal \TeX\
number). The default is |16|. The maximal value is |32767|.
\begin{framed}
  \xintexprname adds\IMPORTANTf{} the variant \csbxint{Digits*} which executes
  \csbxint{reloadxinttrig}.
\end{framed}

The expandable macro |\xinttheDigits|\etype{} serves to retrieve (internally
it uses |\number|) the current value.

Spaces do not matter as long as they do not occur in-between digits:
\begin{everbatim*}
\xintDigits := 24;\xinttheDigits, %
\xintDigits:=36  ;\xinttheDigits, %
\xintDigits:= 16 ;and \xinttheDigits.
\end{everbatim*}

% It has always been the case that an active colon |:| was allowed.

Also |\xintDigits = D;| (i.e.\@ without a colon) is accepted syntax.

An ending active semi-colon |;| is \emph{not} compatible: it can and will
cause low-level \TeX{} errors. But this has a trivial workaround: any
non-expanding token can be used in place of the ending semi-colon. For
example a full stop or a |\relax| token. This
non-expanding ending token will get removed from the token stream.

%  This
% restriction always applied, but never got properly documented prior to
% |1.3f| release... It is possible to use |\string;| but then there can not be
% any space separating it from the digits.

The recommended syntax is thus now |\xintDigits := D\relax| (with or without the colon).
This is the syntax in use in most examples from the documentation.
\begin{everbatim*}
\xintDigits  = 24\def\xinttheDigits, % only for showing it works! don't do that!
\xintDigits := 36.\xinttheDigits,    % one can use a dot in place of semi-colon
\xintDigits  = 16\relax and \xinttheDigits.\par % with \relax, even better
\end{everbatim*}

\subsection{\csh{xintSetDigits}}
\label{xintSetDigits}

To be used as |\xintSetDigits|\marg{expression}\ntype{\numx} where the
expression will be fed to |\numexpr|. It is a shortcut for doing
|\xintDigits := \numexpr|\meta{expression}|\relax \relax|.
\begin{everbatim*}
\xintSetDigits{1+2+3+4+5}The value is now \xinttheDigits.
\xintSetDigits{2*8}The value is now \xinttheDigits.\par
\end{everbatim*}
\begin{framed}
  See also the \xintexprname-added variant \csbxint{SetDigits*}.
\end{framed}

\subsection{\csh{xintFloat}}\label{xintFloat}


The macro |\xintFloat [P]{f}|\etype{{\upshape[\numx]}\Ff} has an optional
argument |P| which replaces the current value of |\xinttheDigits|. The
fraction |f| is then printed in scientific notation with a rounding to |P| digits.

That is, on output: the first digit is from |1| to |9|, it is possibly
prefixed by a minus sign and is followed by a dot and |P-1| digits, then a
lower case |e| and an exponent |N|. The trailing zeroes are not trimmed.

\begin{framed}
  There is currently one exceptional case: the zero value, which gets output
  as \dtt{\xintFloat{0}}. It is yet to be decided what the final policy will be.
\end{framed}

Starting with |1.2k|, when the input is a fraction |AeN/BeM|
the output always is the \emph{correct rounding} to |P| digits. Formerly, this
was guaranteed only when |A| and |B| had at most |P+2| digits, or when |B| was
|1| and |A| was arbitrary, but in other cases it was only guaranteed that the
difference between the original fraction and the rounding was at most
\dtt{0.6} unit in the last place (of the output), hence the output could
differ in the last digit (and earlier ones in case of chains of zeros or
nines) from the correct rounding.

Also: for releases |1.2j| and earlier, in the special case when
|A/B| ended up being rounded up to the next power of ten, the output was with
a mantissa of the shape |10.0...0eN|. However, this worked only for |B=1| or
when both |A| and |B| had at most |P+2| digits, because the detection of the
rounding-up to next power of ten was done not on original |A/B| but on an
approximation |A'/B'|, and it could happen that |A'/B'| was itself being
rounded \emph{down} to a power of ten which however was a rounding \emph{up}
of original |A/B|. With the |1.2j| refactoring which achieves correct rounding
in all cases, it was decided not to add to the code the extra overhead of
detecting with 100\% fiability the rounding up to next power of ten (such
overhead would necessitate alterations of the algorithm and as a result we
would end up with a slightly less efficient one; it would make sense in a
model where inputs have their intrinsic precisions which is obeyed by the
implementation of the basic operations, but currently the design decision for
the floating point macros is that when the target precision is |P| the inputs
are rounded first to |P| digits before further processing.)
\begin{everbatim*}
{\def\x{99999999999999994999999999999999/99999999999999999999999999999999}%
\xintFor #1 in {13, 14, 15, 16, 17, 18, 19, 47, 48, 49, 50, 79, 80, 81}
\do{#1: \xintFloat[#1]{\x}\xintifForLast{\par}{\newline}}}%
\end{everbatim*}
As an aside, which is illustrated by the above, rounding is not
transitive in the number of kept digits.
\begin{everbatim*}
{\def\x{137893789173289739179317/13890138013801398}%
\xintFor* #1 in {\xintSeq{4}{20}}
\do{#1: \xintFloat[#1]{\x}\newline}}%
\xintFloat{5/9999999999999999}\newline
\xintFloat[32]{5/9999999999999999}\newline
\xintFloat[48]{5/9999999999999999}\par
\end{everbatim*}



\subsection{\csh{xintPFloat}}
\label{xintPFloat}
\label{xintPFloatE}

|\xintPFloat [P]{f}|\etype{{\upshape[\numx]}\Ff} is like \csbxint{Float} but
``pretty-prints'' the output.

This macro was added at |1.1| as a (very primitive) "prettifying printer" for
floating point number, and was basically influenced by Maple.

The old rules were:
\begin{enumerate}[nosep]
\item The input is float-rounded to either |Digits| or the optional argument,
\item zero is printed as \dtt{0.},
\item \dtt{x.yz...eN} is printed ``as is'' if the exponent |N| is at least
  \dtt{6} or at most \dtt{-6},
\item else fixed point decimal notation is used,
\item and there is no trimming of trailing zeroes.
\end{enumerate}

At |1.4e|, there is breaking change\CHANGED{1.4e}. The new rules are:
\begin{enumerate}[nolistsep]
\item The input is float-rounded to either |Digits| or the optional argument,
\item zero is printed as \dtt{0.0},
\item \dtt{x.yz...eN} is printed in decimal fixed point if |-4<=N<=+5|
  else it is printed in scientific notation,
\item Trailing zeros of the mantissa are trimmed always,
\item In case of decimal fixed point output format, and the value is an integer, there
  is a trailing |.0|,
\item In case of scientific notation with a one-digit trimmed mantissa
  there is an added |.0| too.
\end{enumerate}

The |1.4e| changes will affect all usages of \csbxint{floateval} as the latter
applies per default (cf.\@ \csbxint{floatexprPrintOne}) \csbxint{PFloat} to
each numerical leaf of the computed expression.

\csa{xintPFloatE}\NewWith{1.4b} was added to allow customizing the
symbol used on output for separating the significand from the exponent, if
output uses scientific notation. The separator defaults to |e|, according to
this definition:
\begin{everbatim}
\def\xintPFloatE{e}
\end{everbatim}
It is now possible\NewWith{1.4e} to let it grab the exponent as an argument (delimited by a
dot) and format it (output must be delimited by a dot, which will be removed
later on).

\begin{everbatim*}
\begin{multicols}2
\def\test #1{#1${}\to{}$\xintPFloat{#1}}\string\xintDigits\ at \xinttheDigits
\begin{itemize}[nosep]
\item \test {0}
\item \test {1.2340000e-7}
\item \test {1.2340000e-6}
\item \test {1.2340000e-5}
\item \test {1.2340000e-4}
\item \test {1.2340000e-3}
\item \test {1.2340000e-2}
\item \test {1.2340000e-1}
\end{itemize}
Change of scientific separator to |E|.\def\xintPFloatE{E}%
\begin{itemize}[nosep]
\item \test {1.2340000e0}
\item \test {1.2340000e1}
\item \test {1.2340000e2}
\item \test {1.2340000e3}
\item \test {1.2340000e4}
\item \test {1.2340000e5}
\item \test {1.2340000e6}
\item \test {1.2340000e7}
\end{itemize}
\end{multicols}
\end{everbatim*}


% \subsection{\csh{xintFloatE}}\label{xintFloatE}
% %! {\small New with |1.097|.}

% |\xintFloatE [P]{f}{x}|\etype{{\upshape[\numx]}\Ff\numx} multiplies the input
% |f| by $10^x$, and
% converts it to float format according to the optional first argument or current
% value of |\xinttheDigits|.
% \begin{everbatim*}
% \xintFloatE {1.23e37}{53}
% \end{everbatim*}

% There is since |1.4b| an unfortunate proximity in name with \csbxint{PFloatE}
% despite the two things having absolutely nothing in common.

\subsection{\csh{xintFloatAdd}}\label{xintFloatAdd}


|\xintFloatAdd [P]{f}{g}|\etype{{\upshape[\numx]}\Ff\Ff} first replaces |f|
and |g| with their float approximations |f'| and |g'| to |P| significant
places or to the precision from |\xintDigits|. It then produces
the sum |f'+g'|, correctly rounded to nearest with the same number of
significant places.


\subsection{\csh{xintFloatSub}}\label{xintFloatSub}


|\xintFloatSub [P]{f}{g}|\etype{{\upshape[\numx]}\Ff\Ff} first replaces |f|
and |g| with their float approximations |f'| and |g'| to |P| significant
places or to the precision from |\xintDigits|. It then produces
the difference |f'-g'| correctly rounded to nearest |P|-float.


\subsection{\csh{xintFloatMul}}\label{xintFloatMul}


|\xintFloatMul [P]{f}{g}|\etype{{\upshape[\numx]}\Ff\Ff} first replaces |f|
and |g| with their float approximations |f'| and |g'| to |P| (or
|\xinttheDigits|) significant places. It then correctly rounds
the product |f'*g'| to nearest |P|-float.

See \autoref{ssec:floatingpoint} for more.

\begin{framed}
  It is obviously much needed that the author improves its algorithms to avoid
  going through the exact |2P| or |2P-1| digits before
  throwing to the waste-bin half of those digits !

  % \xintname initially was purely an \emph{exact} arbitrary precision
  % arithmetic machine, and the introduction of floating point numbers was an
  % after-thought. I got it working in release |1.07 (2013/05/25)| and never had
  % time to come back to it.
\end{framed}

\subsection{\csh{xintFloatDiv}}\label{xintFloatDiv}


|\xintFloatDiv [P]{f}{g}|\etype{{\upshape[\numx]}\Ff\Ff} first replaces |f|
and |g| with their float approximations |f'| and |g'| to |P| (or
|\xinttheDigits|) significant places. It then correctly rounds
the fraction |f'/g'| to nearest |P|-float.

See \autoref{ssec:floatingpoint} for more.

Notice in the special situation with |f| and |g| integers that |\xintFloatDiv
[P]{f}{g}| will \emph{not necessarily} give the correct rounding of the
exact fraction |f/g|. Indeed the macro arguments are each first individually
rounded to |P| digits of precision. The correct syntax to get the correctly
rounded integer fraction |f/g| is \csbxint{Float}|[P]{f/g}|.

\subsection{\csh{xintFloatPow}}\label{xintFloatPow}

|\xintFloatPow [P]{f}{x}|\etype{{\upshape[\numx]}\Ff\numx} uses either the
optional argument |P| or in its absence the value of |\xinttheDigits|. It
computes a floating approximation to |f^x|.

The exponent |x| will be handed over to a |\numexpr|, hence count registers are
accepted on input for this |x|. And the absolute value \verb+|x|+ must obey the
\TeX{} bound.

The argument |f| is first rounded to |P| significant places to give
|f'|. The output |Z| is such that the exact |f'^x| differs from
|Z| by an absolute error less than |0.52 ulp(Z)|.
\begin{everbatim*}
\xintFloatPow [8]{3.1415}{1234567890}
\end{everbatim*}

\subsection{\csh{xintFloatPower}}\label{xintFloatPower}

\csa{xintFloatPower}|[P]{f}{g}|\etype{{\upshape[\numx]}\Ff\Numf} computes a
floating point value |f^g| where the exponent |g| is not constrained to be at
most the \TeX{} bound \dtt{\number "7FFFFFFF}. It may even be a fraction
|A/B| but will be truncated to an integer. The exponent of the
\emph{output} however \emph{must} at any rate obey the \TeX{} bound.

The argument |f| is first rounded to |P| significant places to give
|f'|. The output |Z| is then such that the exact |f'^g| differs from
|Z| by an absolute error less than |0.52 ulp(Z)|.


\emph{For integer exponents} this is the support macro which
is used for the |^| (or |**|) infix operators in \csbxint{floateval}, or also
in \csbxint{eval} for very big integer exponents. It is also used in
\csbxint{floateval} and \csbxint{eval} for half-integer exponents, via a
combination with the \csbxint{FloatSqrt} square-root extraction.

The macro itself was \emph{NOT} modified at |1.4e|: when used directly it
still starts by truncating the exponent to an integer... As for other
user-level floating-point macros, its output is handled by \csbxint{Float},
i.e. it uses scientific notation.

The |0.52 ulp(Z)| guaranteed error bound applies also to the
\csbxint{floateval} evaluations for the half-integer exponent case. It is
valid only when |f| already had a mantissa of at most |P| digits and was
not modified by the initial rounding done by the macro to reduce |f| to |P|
digits.

The integer exponent |g| may have more than |P| (or |Digits|) digits, it is
handled exactly. And as said above its absolute value may exceed the \TeX\ bound.

% Notice that this is a bound on the distance from
% |f'^g| to |Z|, where |f'| is the rounded value of the original input |f|; 
% the distance from |f^g| to |Z| can be much worse if |g| is very
% large. Roughly, when |g| is negligible compared to |10^P|, we get an extra
% difference of up to about |50g ulp(Z)| which completely dwarfs the |0.52
% ulp(Z)|. Thus, if |f| has strictly more than |P| digits, then the computation
% must be done with an elevated working precision |P'|. For example with
% |g=1000| we should use |P'=P+6| to achieve a total error at worst slightly
% bigger than |0.55 ulp(Z)| after the final rounding from |P'| to |P| digits to
% get |Z|.

% Examples:%
% %
% \footnote{|\np| is formatting macro from the \url{http://ctan.org/pkg/numprint}
%   package.}
% %
% \begin{everbatim*}
% \np{\xintFloatPower [8]{3.1415}{3e9}}\newline% Notice that 3e9>2^31
% \np{\xintFloatPower [48]{1.1547}{\xintiiPow {2}{35}}}\newline
% \end{everbatim*}%
% $2^{35}=\xintiiPow {2}{35}$ exceeds \TeX's bound, but what
% counts is the exponent of the result which, while dangerously close to
% $2^{31}$ is not quite there yet.

% With expressions:
% \begin{everbatim*}
% {\xintDigits:=48\relax \np{\xintthefloatexpr 1.1547^2^35\relax}}
% \end{everbatim*}

% There is a subtlety here that the |2^35| will be evaluated as a floating point
% number but fortunately it only has \dtt{11} digits, hence the final evaluation
% is done with a correct exponent. It would have been safer, and also more
% efficient to code the above rather as:
% \begin{everbatim}
% \xintthefloatexpr 1.1547^\xintiiexpr 2^35\relax\relax
% \end{everbatim}

% Here is an example with
% |12^16| as exponent, which has $18$ digits (\dtt{={\xintiiPow{12}{16}}}).
% \begin{everbatim*}
% {\xintDigits:=12\relax \np{\xintthefloatexpr (1+1e-8)^\xintiiexpr 12^16\relax\relax}}\newline
% \np{\xintthefloatexpr (1+1e-8)^\xintiiexpr 12^16\relax\relax}\newline
% {\xintDigits:=27\relax \np{\xintthefloatexpr (1+1e-8)^12^16\relax}}\newline
% {\xintDigits:=48\relax \np{\xintthefloatexpr (1+1e-8)^12^16\relax}}
% \end{everbatim*}

% There is an important difference between |\xintFloatPower[Q]{X}{Y}| and
% |\xintthefloatexpr[Q] X^Y\relax|: in the former case the computation is done
% with |Q| digits or precision,%
% %
% \footnote{if |X| and |Y| themselves stand for some
% floating point macros with arguments, their respective evaluations obey the
% precision |\xinttheDigits| or as set optionally in the macro calls
% themselves.}
% %
% whereas with \csbxint{thefloatexpr}|[Q]| the evaluation of the
% expression proceeds with |\xinttheDigits| digits of precision, and the final
% result is then rounded to |Q| digits: thus this makes real sense only if used
% with |Q<\xinttheDigits|.

\subsection{\csh{xintFloatSqrt}}\label{xintFloatSqrt}

\csa{xintFloatSqrt}|[P]{f}|\etype{{\upshape[\numx]}\Ff} computes a floating
point approximation of $\sqrt{|f|}$, either using the optional precision |P| or
the value of |\xinttheDigits|.

More precisely since |1.2f| the macro achieves so-called \emph{correct
  rounding}:\IMPORTANT{} the produced value is the rounding to |P| significant
places of the abstract exact value, \emph{if the input has itself at most |P|
  digits} (and an arbitrary exponent).
\begin{everbatim*}
\xintFloatSqrt [89]{10}\newline
\xintFloatSqrt [89]{100}\newline
\xintFloatSqrt [89]{123456789}\par
\end{everbatim*}

And now some tests to check that correct rounding applies correctly (sic):
\begin{everbatim*}
The argument has 16 digits, hence escapes initial rounding:\newline
\xintFloatSqrt {5625000075000001}\newline
This one gets rounded hence same value is computed:\newline
\xintFloatSqrt {5625000075000001.4}\newline
but actual value is more like:\newline
\xintFloatSqrt [24]{5625000075000001.4}\newline
\xintFloatSqrt [32]{5625000075000001.4}\newline
The argument has 48 digits, hence escapes initial rounding:\newline
\xintFloatSqrt [48]{562500000000000000000000750000000000000000000001}\newline
\xintFloatSqrt [64]{562500000000000000000000750000000000000000000001}\newline
\xintFloatSqrt [80]{562500000000000000000000750000000000000000000001}\newline
\end{everbatim*}
(we observe in passing illustrations that rounding to nearest is not
transitive.)\par






\subsection{\csh{xintFloatFac}}\label{xintFloatFac}

\csa{xintFloatFac}|[P]{f}|\etype{{\upshape[\numx]}\Numf} returns the
factorial with either \csa{xinttheDigits} or |P| digits of precision.



The exact theoretical value differs from the calculated one |Y| by an absolute
error strictly less than |0.6 ulp(Y)|.

\begin{everbatim*}
$1000!\approx{}$\xintFloatFac [30]{1000}
\end{everbatim*}
The computation proceeds via doing explicitely the product, as
the Stirling formula cannot be used for lack so far of |exp/log|.

The maximal allowed argument is $99999999$, but already $100000!$ currently
takes, for \dtt{16} digits of precision, a few seconds on my laptop (it
returns \dtt{2.824229407960348e456573}).

The |factorial| function is available in \csbxint{floatexpr}:
\begin{everbatim*}
\xintthefloatexpr factorial(1000)\relax % same as 1000!
\end{everbatim*}

\subsection{\csh{xintFloatBinomial}}\label{xintFloatBinomial}

\csa{xintFloatBinomial}|[P]{x}{y}|\etype{{\upshape[\numx]}\Numf\Numf} computes
binomial coefficients with either \csa{xinttheDigits} or |P| digits of
precision.

When |x<0| an out-of-range error is raised. Else if |y<0| or if |x<y| the
macro evaluates to \dtt{\xintFloatBinomial{1}{-1}}.
The exact theoretical value differs from the calculated one |Y| by an absolute
error strictly less than |0.6 ulp(Y)|.

\begin{everbatim*}
${3000\choose 1500}\approx{}$\xintFloatBinomial [24]{3000}{1500}
\end{everbatim*}

% \begin{everbatim*}
% ${9999\choose 5000}\approx{}$\xintFloatBinomial [24]{9999}{5000}
% \end{everbatim*}

% 2015/11/28
% 7.95895131766219474168799e3007
% aparté: (testé avec Maple 16, 2015/11/28)
% > binomial (9999.,5000.);
%                                               3008
%                              0.795895131768 10
%
% > Digits:=32;
%                                  Digits := 32
%
% > binomial (9999.,5000.);
%                                               3008
%                              0.795895131768 10
% apparemment le binomial de Maple ne sait pas calculer avec plus de
% précision!
% et son dernier chiffre est faux! Pourtant GAMMA(9999.) fonctionne. Sauf si
% je n'ai pas compris quelque chose il me semble donc que le binomial de Maple
% est bogué...binomial(100.,50.); marche lui et binomial(4999.,2000.); aussi,
% bon clairement on a un bug de Maple ! oui binomial(8999.,5000.); ainsi que
% binomial(10999.,5000.); fonctionnent avec Digits:=32 mais **pas**
% binomial(9999.,5000.)... binomial(10000.,5000.); et binomial(9998.,5000.);
% sont OK. Est-ce qu'on gagne quelque chose pour un bug report ?
% > binomial(9999.,5000.);
%                                               3008
%                              0.795895131768 10
% > binomial(10000.,5000.);
%                                                          3009
%                   0.1591790263532438948337597273641521 10
% > binomial(9998.,5000.);
%                                                          3008
%                   0.3979077671466477799149739359402922 10
% en plus je lui demande 32 chiffres et il m'en sort 34.

The associated function in \csbxint{floatexpr} is \func{binomial}:
\begin{everbatim*}
\xintthefloatexpr binomial(3000,1500)\relax
\end{everbatim*}

The computation is based on the formula |(x-y+1)...x/y!| (here one arranges
|y<=x-y| naturally).


\subsection{\csh{xintFloatPFactorial}}\label{xintFloatPFactorial}

\csa{xintFloatPFactorial}|[P]{x}{y}|\etype{{\upshape[\numx]}\Numf\Numf}
computes the product |(x+1)...y|.




The arguments must be integers (they are expanded inside |\numexpr|)
and the allowed range is $-100000000\leqslant x, y\leqslant99999999$. If
$x\geqslant y$ the product is considered empty hence returns one (as a
floating point value).
See also \csbxint{iiPFactorial}.


The exact theoretical value differs from the calculated one |Y| by an absolute
error strictly less than |0.6 ulp(Y)|.

The associated function in \csbxint{floatexpr} is \func{pfactorial}:
\begin{everbatim*}
\xintthefloatexpr pfactorial(2500,5000)\relax
\end{everbatim*}

\xintDigits:=16\relax 


\clearpage
\let\xintfracnameUp\undefined
\csname xintbinhexnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintbinhexname package}
\RaisedLabel{sec:binhex}

This package provides expandable conversions of (big) integers to
and from binary and hexadecimal.

First version of this package was in the |1.08| (|2013/06/07|) release of
\xintname. Its routines remained un-modified until their complete rewrite at
release |1.2m| (|2017/07/31|). The new macros are faster, using techniques
from the |1.2| (|2015/10/10|) release of \xintcorename. But the inputs are now
limited to a few thousand digits, whereas the |1.08| could handle (slowly...)
tens of thousands of digits.

\autoref{tab:binhexsizes} recapitulates the maximal allowed sizes (they got
increased at |1.2n|):
for macro |\xintFooToBar| in the first column, the value in the second column
is the maximal |N| such that |\edef\X{\xintFooToBar{<N digits>}}| does not
raise an error with standard \TeX\ memory parameters (input stack
size=\dtt{5000}, expansion depth=\dtt{10000}, parameter stack
size=\dtt{10000}). The tests were done with TL2017 and |etex|. Nested calls
will allow slightly lesser values only. The third column gives the
corresponding maximal size of output. The fourth column gives the \TeX\
parameter cited in the error message when trying with |N+1| digits.

\begin{table}[htbp]
\capstart
  \centering
\def\E#1#2!{\edef\F{\the\numexpr(#1-\xintLength{#2})/2}%
          \relax\romannumeral\xintreplicate{\F}{ }#2%
          \romannumeral\xintreplicate{#1-\F-\xintLength{#2}}{ }\relax}%
% non satisfactory because depends on #1 oddness, but well. Temporary destined
% to stay...
\begin{tabular}{r>{\E{19}}c<{!}>{\E{19}}c<{!}r}
  \hline
  &Max\ length\ of\ input&->\ length\ of\ output&Limiting factor\\
  \csbxint{DecToHex}&6014&4995&input stack size=5000\\
  \csbxint{DecToBin}&6014&19979&input stack size=5000\\
  \csbxint{HexToDec}&8298&9992&input stack size=5000\\
  \csbxint{BinToDec}&19988&6017&input stack size=5000\\
  \csbxint{BinToHex}&19988&4997&input stack size=5000\\
  \csbxint{HexToBin}&4996&19984&input stack size=5000\\
  \csbxint{CHexToBin}&4997&19988&input stack size=5000\\
  \hline
\end{tabular}
\caption{Maximal sizes of inputs (at \texttt{1.2n}) for \xintbinhexname macros}\label{tab:binhexsizes}
\end{table}

Roughly, base |10| numbers are limited to \dtt{6000} digits, hexadecimal
numbers to (almost) \dtt{5000} digits, and binary numbers to (almost)
\dtt{20000} digits. With the surprising exception of \csbxint{HexToDec} which
allows almost \dtt{8300} hexadecimal digits on input.

The argument is first \fexpan ded.
It may optionally have a unique leading minus sign (a plus sign is not
allowed), and leading zeroes.

An input (possibly signed) with no leading zeroes is guaranteed to give an
output without leading zero, with the sole, deliberate, exception of
\csbxint{CHexToBin}: from |N| hexadecimal digits it produces |4N| binary
digits, hence possibly with up to three leading zeroes (if the
input had none.)

Inputs with leading zeroes usually produce outputs with an unspecified,
case-dependent, number of leading zeroes (\csbxint{BinToHex} always uses the
minimal number of hexadecimal digits needed to represent the binary digits,
inclusive of leading zeroes if present.)

The macros converting from binary or decimal are robust against
non terminated inputs like |\the\numexpr 2+3| or |\the\mathcode`\-|. The macro
\csbxint{HexToDec} also but not \csbxint{HexToBin} and \csbxint{CHexToBin}
(anyway there are no primitive in (e)-\TeX\ to my knowledge which will
generate hexadecimal digits and may force expansion of next token).

Hexadecimal digits |A..F| must be in uppercase. Category code for them on
input may be \emph{letter} or \emph{other}. On output they are of category
code \emph{letter}, and in uppercase.

Low-level unrecoverable errors will happen if for example a supposedly binary
input contains other digits than |0| and |1|. Inputs can not start with a
|0b|, |0x|, |#x|, |"| or similar prefix: only digits/letters according to the
binary, decimal, or hexadecimal notation.


With this package loaded additionally to \xintexprname, hexadecimal input is
possible in expressions: simply by using the prefix |"|. Such hexadecimal
numbers may have a fractional part. Lowercase hexadecimal letters are
currently \emph{not} recognized as such in expressions.
Currently the |p| postfix notation from standard programming languages
standing for an extra
power of two multiplicand is not implemented.

\localtableofcontents

\subsection{\csh{xintDecToHex}}\label{xintDecToHex}

Converts from decimal to hexadecimal.\etype{f}

\texttt{\string\xintDecToHex \string{\printnumber{2718281828459045235360287471352662497757247093699959574966967627724076630353547594571382178525166427427466391932003}\string}}\endgraf\noindent\dtt{->\printnumber{\xintDecToHex{2718281828459045235360287471352662497757247093699959574966967627724076630353547594571382178525166427427466391932003}}}

\subsection{\csh{xintDecToBin}}\label{xintDecToBin}

Converts from decimal to binary.\etype{f}

\texttt{\string\xintDecToBin \string{\printnumber{2718281828459045235360287471352662497757247093699959574966967627724076630353547594571382178525166427427466391932003}\string}}\endgraf\noindent\dtt{->\printnumber{\xintDecToBin{2718281828459045235360287471352662497757247093699959574966967627724076630353547594571382178525166427427466391932003}}}

\subsection{\csh{xintHexToDec}}\label{xintHexToDec}

Converts from hexadecimal to decimal.\etype{f}

\texttt{\string\xintHexToDec
  \string{\printnumber{11A9397C66949A97051F7D0A817914E3E0B17C41B11C48BAEF2B5760BB38D272F46DCE46C6032936BF37DAC918814C63}\string}}\endgraf\noindent
\dtt{->\printnumber{\xintHexToDec{11A9397C66949A97051F7D0A817914E3E0B17C41B11C48BAEF2B5760BB38D272F46DCE46C6032936BF37DAC918814C63}}}

\subsection{\csh{xintBinToDec}}\label{xintBinToDec}

Converts from binary to decimal.\etype{f}

\texttt{\string\xintBinToDec
  \string{\printnumber{100011010100100111001011111000110011010010100100110101001011100000101000111110111110100001010100000010111100100010100111000111110000010110001011111000100000110110001000111000100100010111010111011110010101101010111011000001011101100111000110100100111001011110100011011011100111001000110110001100000001100101001001101101011111100110111110110101100100100011000100000010100110001100011}\string}}\endgraf\noindent
\dtt{->\printnumber{\xintBinToDec{100011010100100111001011111000110011010010100100110101001011100000101000111110111110100001010100000010111100100010100111000111110000010110001011111000100000110110001000111000100100010111010111011110010101101010111011000001011101100111000110100100111001011110100011011011100111001000110110001100000001100101001001101101011111100110111110110101100100100011000100000010100110001100011}}}

\subsection{\csh{xintBinToHex}}\label{xintBinToHex}

Converts from binary to hexadecimal.\etype{f} The input is first zero-filled
to |4N| binary digits, hence the output will have |N| hexadecimal digits
(thus, if the input did not have a leading zero, the output will not either).

\texttt{\string\xintBinToHex
  \string{\printnumber{100011010100100111001011111000110011010010100100110101001011100000101000111110111110100001010100000010111100100010100111000111110000010110001011111000100000110110001000111000100100010111010111011110010101101010111011000001011101100111000110100100111001011110100011011011100111001000110110001100000001100101001001101101011111100110111110110101100100100011000100000010100110001100011}\string}}\endgraf\noindent
\dtt{->\printnumber{\xintBinToHex{100011010100100111001011111000110011010010100100110101001011100000101000111110111110100001010100000010111100100010100111000111110000010110001011111000100000110110001000111000100100010111010111011110010101101010111011000001011101100111000110100100111001011110100011011011100111001000110110001100000001100101001001101101011111100110111110110101100100100011000100000010100110001100011}}}

\subsection{\csh{xintHexToBin}}\label{xintHexToBin}

Converts from hexadecimal to binary. Up to three leading zeroes of the output
are trimmed.\etype{f}

\texttt{\string\xintHexToBin
  \string{\printnumber{11A9397C66949A97051F7D0A817914E3E0B17C41B11C48BAEF2B5760BB38D272F46DCE46C6032936BF37DAC918814C63}\string}}\endgraf\noindent
\dtt{->\printnumber{\xintHexToBin{11A9397C66949A97051F7D0A817914E3E0B17C41B11C48BAEF2B5760BB38D272F46DCE46C6032936BF37DAC918814C63}}}

\subsection{\csh{xintCHexToBin}}\label{xintCHexToBin}

Converts from hexadecimal to binary.\etype{f} Same as \csbxint{HexToBin}, but
an input with |N| hexadecimal digits will give an output with exactly |4N|
binary digits, leading zeroes are not trimmed.

\texttt{\string\xintCHexToBin
  \string{\printnumber{11A9397C66949A97051F7D0A817914E3E0B17C41B11C48BAEF2B5760BB38D272F46DCE46C6032936BF37DAC918814C63}\string}}\endgraf\noindent
\dtt{->\printnumber{\xintCHexToBin{11A9397C66949A97051F7D0A817914E3E0B17C41B11C48BAEF2B5760BB38D272F46DCE46C6032936BF37DAC918814C63}}}

This can be combined with \csbxint{BinToHex} for round-trips preserving
leading zeroes for |4N| binary digits numbers, whereas using
\csbxint{HexToBin} gives reproducing round-trips only for |4N| binary numbers
numbers not starting with |0000|.
\begin{everbatim*}
This zero-fills to 4N digits the input, hence gives here a leading zero in output:
\xintBinToHex{0001111}\newline
Chaining, we end up with 4N-3 digits, as three binary zeroes are trimmed:
\xintHexToBin{\xintBinToHex{0001111}}\newline
But this will always reproduce the initial input zero-filled to length 4N:
\xintCHexToBin{\xintBinToHex{0001111}}\par
Another example (visible space characters manually inserted):\newline
$000000001111101001010001\xrightarrow{\text{\string\xintBinToHex}}
\xintBinToHex{000000001111101001010001}\xrightarrow{\text{\string\xintHexToBin\hphantom{X}}}
\text{\textvisiblespace\textvisiblespace\textvisiblespace}
\xintHexToBin{\xintBinToHex{000000001111101001010001}}$\newline
$000000001111101001010001\xrightarrow{\text{\string\xintBinToHex}}
\xintBinToHex{000000001111101001010001}\xrightarrow{\text{\string\xintCHexToBin}}
\xintCHexToBin{\xintBinToHex{000000001111101001010001}}$
\par
\end{everbatim*}

\clearpage
\let\xintbinhexnameUp\undefined
\csname xintgcdnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintgcdname package}
\RaisedLabel{sec:gcd}

This package was included in the original release |1.0| (|2013/03/28|) of the
\xintname bundle.



At |1.3d| macros \csbxint{iiGCD} and \csbxint{iiLCM} are copied over to
\xintname, hence \func{gcd} and \func{lcm} functions in \csbxint{iiexpr} were
available simply from loading only \xintexprname, and the \xintgcdname dependency
got removed.

\begin{framed}
From |1.1| to |1.3f| the package loaded only \xintcorename,
not \xintname and neither \xinttoolsname.

But at |1.4| it loads automatically both \xintname and \xinttoolsname (the
latter being a requirement since |1.09h| of the \csbxint{TypesetEuclideAlgorithm} and
\csbxint{TypesetBezoutAlgorithm} macros).

The macros \csbxint{iiGCD} and \csbxint{iiLCM} got relocated into
\xintname. \IMPORTANTf The macros \csbxint{GCD}, \csbxint{LCM},
\csbxint{GCDof}, and \csbxint{LCMof} are \emph{removed}:
\xintfracname provides under these names more powerful macros handling
general fractions and not only integers.
\end{framed}

\localtableofcontents

\subsection{\csh{xintBezout}}\label{xintBezout}

|\xintBezout|\n\m\etype{\Numf\Numf} returns three numbers |U|, |V|,
|D| within braces where |D| is the (non-negative) GCD, and \dtt{UN + VM = D}.
\begin{everbatim*}
\oodef\X{\xintBezout {10000}{1113}}\meaning\X\par
\xintAssign {\xintBezout {10000}{1113}}\to\U\V\D
U: \meaning\U, V: \meaning\V, D: \meaning\D\par
AU+BV: \xinttheiiexpr 10000*\U+1113*\V\relax\par
\noindent\oodef\X{\xintBezout {123456789012345}{9876543210321}}\meaning\X\par
\xintAssign \X\to\U\V\D
U: \meaning\U, V: \meaning\V, D: \meaning\D\par
AU+BV: \xinttheiiexpr 123456789012345*\U+9876543210321*\V\relax
\end{everbatim*}

\subsection{\csh{xintEuclideAlgorithm}}\label{xintEuclideAlgorithm}

|\xintEuclideAlgorithm|\n\m\etype{\Numf\Numf} applies the Euclide algorithm
and keeps a copy of all quotients and remainders.
\begin{everbatim*}
\edef\X{\xintEuclideAlgorithm {10000}{1113}}\meaning\X
\end{everbatim*}

The first item is the number of steps, the second is |N|, the
third is the GCD, the fourth is |M| then the first quotient and
remainder, the second quotient and remainder, \dots until the
final quotient and last (zero) remainder.

\subsection{\csh{xintBezoutAlgorithm}}\label{xintBezoutAlgorithm}

|\xintBezoutAlgorithm|\n\m\etype{\Numf\Numf} applies the Euclide algorithm
and keeps a copy of all quotients and remainders. Furthermore it computes the
entries of the successive products of the 2 by 2 matrices
$\left(\vcenter{\halign {\,#&\,#\cr q & 1 \cr 1 & 0 \cr}}\right)$ formed from
the quotients arising in the algorithm.
\begin{everbatim*}
\edef\X{\xintBezoutAlgorithm {10000}{1113}}\printnumber{\meaning\X}
\end{everbatim*}

The first item is the number of steps, the second is |N|, then
|0|, |1|, the GCD, |M|, |1|, |0|, the first quotient, the first
remainder, the top left entry of the first matrix, the bottom left
entry, and then these four things at each step until the end.

\subsection{\csh{xintTypesetEuclideAlgorithm}}\label{xintTypesetEuclideAlgorithm}

This macro is just an example of how to organize the data returned by
\csa{xintEuclideAlgorithm}.\ntype{\Numf\Numf} Copy the source code to a new
macro and modify it to what is needed.

\leftedline{|\xintTypesetEuclideAlgorithm {123456789012345}{9876543210321}|}
\xintTypesetEuclideAlgorithm {123456789012345}{9876543210321}

\subsection{\csh{xintTypesetBezoutAlgorithm}}%
\label{xintTypesetBezoutAlgorithm}

This macro is just an example of how to organize the data returned by
\csa{xintBezoutAlgorithm}.\ntype{\Numf\Numf} Copy the source code to a new
macro and modify it to what is needed.

\leftedline{|\xintTypesetBezoutAlgorithm {10000}{1113}|}
\xintTypesetBezoutAlgorithm {10000}{1113}

\clearpage
\let\xintgcdnameUp\undefined
\csname xintseriesnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintseriesname package}
\RaisedLabel{sec:series}

This package was first released with version |1.03| (|2013/04/14|) of the
\xintname bundle.

The \Ff{} expansion type of various macro arguments is only a \Numf{} if only
\xintname but not \xintfracname is loaded. The macro \csbxint{iSeries} is
special and expects summing big integers obeying the strict format, even if
\xintfracname is loaded.

The arguments serving as indices are of the \numx{} expansion type.

In some cases one or two of the macro arguments are only expanded at a later
stage not immediately.

\begin{framed}
  Since |1.3|, \csbxint{Add} and \csbxint{Sub} use systematically the least
  common multiple of the denominators. Some of the comments in this chapter
  refer to the earlier situation where often the denominators were simply
  multiplied together. \emph{They have yet to be updated to reflect the new
    situation brought by the |1.3| release.} Some of these comments may now be
  off-synced from the actual computation results and thus may be wrong.
\end{framed}
%% \clearpage

\localtableofcontents

\subsection{\csh{xintSeries}}\label{xintSeries}

\csa{xintSeries}|{A}{B}{\coeff}|\etype{\numx\numx\Ff} computes
$\sum_{\text{|n=A|}}^{\text{|n=B|}}$|\coeff{n}|. The initial and final indices
must obey the |\numexpr| constraint of expanding to numbers at most |2^31-1|.
The |\coeff| macro must be a one-parameter \fexpan dable macro, taking on
input an explicit number |n| and producing some number or fraction |\coeff{n}|;
it is expanded at the time it is
needed.%
%

\begin{everbatim*}
\def\coeff #1{\xintiiMON{#1}/#1.5} % (-1)^n/(n+1/2)
\fdef\w {\xintSeries {0}{50}{\coeff}} % we want to re-use it
\fdef\z {\xintJrr {\w}[0]} % the [0] for a microsecond gain.
% \xintJrr preferred to \xintIrr: a big common factor is suspected.
% But numbers much bigger would be needed to show the greater efficiency.
\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12} = \xintTeXFrac\z \]
\end{everbatim*}

The definition of |\coeff| as |\xintiiMON{#1}/#1.5| is quite suboptimal. It
allows |#1| to be a big integer, but anyhow only small integers are accepted
as initial and final indices (they are of the \numx{} type). Second, when the
\xintfracname parser sees the |#1.5| it will remove the dot hence create a
denominator with one digit more. For example |1/3.5| turns internally into
|10/35| whereas it would be more efficient to have |2/7|. For info here is the
non-reduced |\w|:
\[\xintTeXFrac\w\]
It would have been bigger still in releases earlier than |1.1|: now, the
\xintfracname \csbxint{Add} routine does not multiply blindly denominators
anymore, it checks if one is a multiple of the other. However it does not
practice systematic reduction to lowest terms.

A more efficient way to code |\coeff| is illustrated next.
\begin{everbatim*}
\def\coeff #1{\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}%
% The [0] in \coeff is a tiny optimization: in its presence the \xintfracname parser
% sees something which is already in internal format.
\fdef\w {\xintSeries {0}{50}{\coeff}}
\[\sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12}=\xintTeXFrac\w\]
\end{everbatim*}
The reduced form |\z| as displayed above only differs from this one by a
factor of \dtt{\xintNum {\xintDenominator\w/\xintDenominator\z}}.

\setlength{\columnsep}{0pt}
\everb|@
\def\coeffleibnitz #1{\the\numexpr\ifodd #1 1\else-1\fi\relax/#1[0]}
\cnta 1
\loop
% in this loop we recompute from scratch each partial sum!
% we can afford that, as \xintSeries is fast enough.
\noindent\hbox to 2em{\hfil\texttt{\the\cnta.} }%
         \xintTrunc {12}{\xintSeries {1}{\cnta}{\coeffleibnitz}}\dots
\endgraf
\ifnum\cnta < 30 \advance\cnta 1 \repeat
|

\begin{multicols}{3}
  \def\coeffleibnitz #1{\the\numexpr\ifodd #1 1\else-1\fi\relax/#1[0]} \cnta 1
  \loop
  \noindent\hbox to 2em{\hfil\dtt{\the\cnta.} }%
  \xintTrunc {12}{\xintSeries {1}{\cnta}{\coeffleibnitz}}\dots
    \endgraf
    \ifnum\cnta < 30 \advance\cnta 1 \repeat
\end{multicols}

\subsection{\csh{xintiSeries}}\label{xintiSeries}

\def\coeff #1{\xintiTrunc {40}
   {\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}}%

\csa{xintiSeries}|{A}{B}{\coeff}|\etype{\numx\numx f} computes
 $\sum_{\text{|n=A|}}^{\text{|n=B|}}$|\coeff{n}| where |\coeff{n}|
 must \fexpan d to a (possibly long) integer in the strict format.
\everb|@
\def\coeff #1{\xintiTrunc {40}{\xintiiMON{#1}/#1.5}}%
% better:
\def\coeff #1{\xintiTrunc {40}
   {\the\numexpr 2*\xintiiMON{#1}\relax/\the\numexpr 2*#1+1\relax [0]}}%
% better still:
\def\coeff #1{\xintiTrunc {40}
 {\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}}%
% (-1)^n/(n+1/2) times 10^40, truncated to an integer.
\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12} \approx
        \xintTrunc {40}{\xintiSeries {0}{50}{\coeff}[-40]}\dots\]
|

\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12} \approx \xintTrunc
{40}{\xintiSeries {0}{50}{\coeff}[-40]}\]

We should have cut out at
least the last two digits: truncating errors originating with the first
coefficients of the sum will never go away, and each truncation
introduces an uncertainty in the last digit, so as we have 40 terms, we
should trash the last two digits, or at least round at 38 digits. It is
interesting to compare with the computation where rounding rather than
truncation is used, and with the decimal
expansion of the exactly computed partial sum of the series:
\everb|@
\def\coeff #1{\xintiRound {40} % rounding at 40
  {\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}}%
% (-1)^n/(n+1/2) times 10^40, rounded to an integer.
\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12} \approx
        \xintTrunc {40}{\xintiSeries {0}{50}{\coeff}[-40]}\]
\def\exactcoeff #1%
  {\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}%
\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12}
   = \xintTrunc {50}{\xintSeries {0}{50}{\exactcoeff}}\dots\]
|

\def\coeff #1{\xintiRound {40}
   {\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}}%
% (-1)^n/(n+1/2) times 10^40, rounded to an integer.
\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12} \approx
        \xintTrunc {40}{\xintiSeries {0}{50}{\coeff}[-40]}\]
\def\exactcoeff #1%
  {\the\numexpr\ifodd #1 -2\else2\fi\relax/\the\numexpr 2*#1+1\relax [0]}%
\[ \sum_{n=0}^{n=50} \frac{(-1)^n}{n+\frac12}
   = \xintTrunc {50}{\xintSeries {0}{50}{\exactcoeff}}\dots\]
This shows indeed that our sum of truncated terms
estimated wrongly the 39th and 40th digits of the exact result%
%
\footnote{as the series is alternating, we can roughly expect an error
  of $\sqrt{40}$ and the last two digits are off by 4 units, which is
  not contradictory to our expectations.}
%
and that the sum of rounded terms fared a bit better.

\subsection{\csh{xintRationalSeries}}\label{xintRationalSeries}


\noindent \csa{xintRationalSeries}|{A}{B}{f}{\ratio}|\etype{\numx\numx\Ff\Ff}
evaluates $\sum_{\text{|n=A|}}^{\text{|n=B|}}$|F(n)|, where |F(n)| is specified
indirectly via the data of |f=F(A)| and the one-parameter macro |\ratio| which
must be such that |\macro{n}| expands to |F(n)/F(n-1)|. The name indicates that
\csa{xintRationalSeries} was designed to be useful in the cases where
|F(n)/F(n-1)| is a rational function of |n| but it may be anything expanding to
a fraction. The macro |\ratio| must be an expandable-only compatible macro and
expand to its value after iterated full expansion of its first item. |A| and
|B| are fed to a |\numexpr| hence may be count registers or arithmetic
expressions built with such; they must obey the \TeX{} bound. The initial term
|f| may be a macro |\f|, it will be expanded to its value representing |F(A)|.

\begin{everbatim*}
\def\ratio #1{2/#1[0]}% 2/n, to compute exp(2)
\cnta 0 % previously declared count
\begin{quote}
\loop \fdef\z {\xintRationalSeries {0}{\cnta}{1}{\ratio }}%
\noindent$\sum_{n=0}^{\the\cnta} \frac{2^n}{n!}=
           \xintTrunc{12}\z\dots=
           \xintTeXFrac\z=\xintTeXFrac{\xintIrr\z}$\vtop to 5pt{}\par
\ifnum\cnta<20 \advance\cnta 1 \repeat
\end{quote}
\end{everbatim*}

\begin{everbatim*}
\def\ratio #1{-1/#1[0]}% -1/n, comes from the series of exp(-1)
\cnta 0 % previously declared count
\begin{quote}
\loop
\fdef\z {\xintRationalSeries {0}{\cnta}{1}{\ratio }}%
\noindent$\sum_{n=0}^{\the\cnta} \frac{(-1)^n}{n!}=
  \xintTrunc{20}\z\dots=\xintTeXFrac{\z}=\xintTeXFrac{\xintIrr\z}$%
         \vtop to 5pt{}\par
\ifnum\cnta<20 \advance\cnta 1 \repeat
\end{quote}
\end{everbatim*}


 \def\ratioexp #1#2{\xintDiv{#1}{#2}}% #1/#2

\medskip We can incorporate an indeterminate if we define |\ratio| to be
a macro with two parameters: |\def\ratioexp
  #1#2{\xintDiv{#1}{#2}}|\texttt{\%}| x/n: x=#1, n=#2|.
Then, if |\x| expands to some fraction |x|, the
macro %
%
\leftedline{|\xintRationalSeries {0}{b}{1}{\ratioexp{\x}}|}
will compute $\sum_{n=0}^{n=b} x^n/n!$:\par
\begin{everbatim*}
\cnta 0
\def\ratioexp #1#2{\xintDiv{#1}{#2}}% #1/#2
\loop
\noindent
$\sum_{n=0}^{\the\cnta} (.57)^n/n! = \xintTrunc {50}
     {\xintRationalSeries {0}{\cnta}{1}{\ratioexp{.57}}}\dots$
     \vtop to 5pt {}\endgraf
\ifnum\cnta<50 \advance\cnta 10 \repeat
\end{everbatim*}

Observe that in this last example the |x| was directly inserted; if it
had been a more complicated explicit fraction it would have been
worthwile to use |\ratioexp\x| with |\x| defined to expand to its value.
In the further situation where this fraction |x| is not explicit but
itself defined via a complicated, and time-costly, formula, it should be
noted that \csa{xintRationalSeries} will do again the evaluation of |\x|
for each term of the partial sum. The easiest is thus when |x| can be
defined as an |\edef|. If however, you are in an expandable-only context
and cannot store in a macro like |\x| the value to be used, a variant of
\csa{xintRationalSeries} is needed which will first evaluate this |\x| and then
use this result without recomputing it. This is \csbxint{RationalSeriesX},
documented next.

Here is a slightly more complicated evaluation:
\begin{everbatim*}
\cnta 1
\begin{multicols}{2}
\loop \fdef\z {\xintRationalSeries
                   {\cnta}
                   {2*\cnta-1}
                   {\xintiiPow {\the\cnta}{\cnta}/\xintiiFac{\cnta}}
                   {\ratioexp{\the\cnta}}}%
\fdef\w {\xintRationalSeries {0}{2*\cnta-1}{1}{\ratioexp{\the\cnta}}}%
\noindent
$\sum_{n=\the\cnta}^{\the\numexpr 2*\cnta-1\relax} \frac{\the\cnta^n}{n!}/%
          \sum_{n=0}^{\the\numexpr 2*\cnta-1\relax} \frac{\the\cnta^n}{n!} =
          \xintTrunc{8}{\xintDiv\z\w}\dots$ \vtop to 5pt{}\endgraf
\ifnum\cnta<20 \advance\cnta 1 \repeat
\end{multicols}
\end{everbatim*}


\subsection{\csh{xintRationalSeriesX}}\label{xintRationalSeriesX}


\noindent\csa{xintRationalSeriesX}|{A}{B}{\first}{\ratio}{\g}|%
\etype{\numx\numx\Ff\Ff f} is a parametrized version of \csa{xintRationalSeries}
where |\first| is now a one-parameter macro such that |\first{\g}| gives the
initial term and |\ratio| is a two-parameter macro such that |\ratio{n}{\g}|
represents the ratio of one term to the previous one. The parameter |\g| is
evaluated only once at the beginning of the computation, and can thus itself be
the yet unevaluated result of a previous computation.

Let |\ratio| be such a two-parameter macro; note the subtle differences
between%
%
\leftedline{|\xintRationalSeries {A}{B}{\first}{\ratio{\g}}|}
%
\leftedline{and |\xintRationalSeriesX {A}{B}{\first}{\ratio}{\g}|.} First the
location of braces differ...\@ then, in the former case |\first| is a
\emph{no-parameter} macro expanding to a fractional number, and in the latter,
it is a
\emph{one-parameter} macro which will use |\g|. Furthermore the |X| variant
will expand |\g| at the very beginning whereas the former non-|X| former variant
will evaluate it each time it needs it (which is bad if this
evaluation is time-costly, but good if |\g| is a big explicit fraction
encapsulated in a macro).

The example will use the macro \csbxint{PowerSeries} which computes
efficiently exact partial sums of power series, and is discussed in the
next section.
\begin{everbatim*}
\def\firstterm #1{1[0]}% first term of the exponential series
% although it is the constant 1, here it must be defined as a
% one-parameter macro. Next comes the ratio function for exp:
\def\ratioexp  #1#2{\xintDiv {#1}{#2}}% x/n
% These are the (-1)^{n-1}/n of the log(1+h) series:
\def\coefflog #1{\the\numexpr\ifodd #1 1\else-1\fi\relax/#1[0]}%
% Let L(h) be the first 10 terms of the log(1+h) series and
% let E(t) be the first 10 terms of the exp(t) series.
% The following computes E(L(a/10)) for a=1,...,12.
\begin{multicols}{3}\raggedcolumns
\cnta 0
\loop
\noindent\xintTrunc {18}{%
     \xintRationalSeriesX {0}{9}{\firstterm}{\ratioexp}
         {\xintPowerSeries{1}{10}{\coefflog}{\the\cnta[-1]}}}\dots
\endgraf
\ifnum\cnta < 12 \advance \cnta 1 \repeat
\end{multicols}
\end{everbatim*}


These completely exact operations rapidly create numbers with many digits. Let
us print in full the raw fractions created by the operation illustrated above:

\fdef\z{\xintRationalSeriesX {0}{9}{\firstterm}
{\ratioexp}{\xintPowerSeries{1}{10}{\coefflog}{1[-1]}}}

|E(L(1[-1]))=|\dtt{\printnumber{\z}} (length of numerator:
\xintLen {\xintNumerator \z})

\fdef\z{\xintRationalSeriesX {0}{9}{\firstterm}
{\ratioexp}{\xintPowerSeries{1}{10}{\coefflog}{12[-2]}}}

|E(L(12[-2]))=|\dtt{\printnumber{\z}} (length of numerator:
\xintLen {\xintNumerator \z})

\fdef\z{\xintRationalSeriesX {0}{9}{\firstterm}
{\ratioexp}{\xintPowerSeries{1}{10}{\coefflog}{123[-3]}}}

|E(L(123[-3]))=|\dtt{\printnumber{\z}} (length of numerator:
\xintLen {\xintNumerator \z})

We see that the denominators here remain the same, as our input only had various
powers of ten as denominators, and \xintfracname efficiently assemble (some
only, as we can see) powers of ten. Notice that 1 more digit in an input
denominator seems to mean 90 more in the raw output. We can check that with some
other test cases:

\fdef\z{\xintRationalSeriesX {0}{9}{\firstterm}
{\ratioexp}{\xintPowerSeries{1}{10}{\coefflog}{1/7}}}

|E(L(1/7))=|\dtt{\printnumber{\z}} (length of numerator:
\xintLen {\xintNumerator \z}; length of denominator:
\xintLen {\xintDenominator \z})

\fdef\z{\xintRationalSeriesX {0}{9}{\firstterm}
{\ratioexp}{\xintPowerSeries{1}{10}{\coefflog}{1/71}}}

|E(L(1/71))=|\dtt{\printnumber{\z}} (length of numerator:
\xintLen {\xintNumerator \z}; length of denominator:
\xintLen {\xintDenominator \z})

\fdef\z{\xintRationalSeriesX {0}{9}{\firstterm}
{\ratioexp}{\xintPowerSeries{1}{10}{\coefflog}{1/712}}}

|E(L(1/712))=|\dtt{\printnumber{\z}} (length of numerator:
\xintLen {\xintNumerator \z}; length of denominator:
\xintLen {\xintDenominator \z})


Thus
decimal numbers such as |0.123| (equivalently
|123[-3]|) give less computing intensive tasks than fractions such as |1/712|:
in the case of decimal numbers the (raw) denominators originate in the
coefficients of the series themselves, powers of ten of the input within
brackets being treated separately. And even then the
numerators will grow with the size of the input in a sort of linear way, the
coefficient being given by the order of series: here 10 from the log and 9 from
the exp, so 90. One more digit in the input means 90 more digits in the
numerator of the output: obviously we can not go on composing such partial sums
of series and hope that \xintname will joyfully do all at the speed of light!

Hence, truncating the output (or better, rounding) is the only way to go if one
needs a general calculus of special functions. This is why the package
\xintseriesname provides, besides \csbxint{Series}, \csbxint{RationalSeries}, or
\csbxint{PowerSeries} which compute \emph{exact} sums,
\csbxint{FxPtPowerSeries} for fixed-point computations and a (tentative naive)
\csbxint{FloatPowerSeries}.

\subsection{\csh{xintPowerSeries}}\label{xintPowerSeries}

\csa{xintPowerSeries}|{A}{B}{\coeff}{f}|\etype{\numx\numx\Ff\Ff}
evaluates the sum
$\sum_{\text{|n=A|}}^{\text{|n=B|}}$|\coeff{n}|${}\cdot |f|^{\text{|n|}}$. The
initial and final indices are given to a |\numexpr| expression. The |\coeff|
macro (which, as argument to \csa{xintPowerSeries} is expanded only at the time
|\coeff{n}| is needed) should be defined as a one-parameter expandable macro,
its input will be an explicit number.

The |f| can be either a fraction directly input or a macro |\f| expanding to
such a fraction. It is actually more efficient to encapsulate an explicit
fraction |f| in such a macro, if it has big numerators and denominators (`big'
means hundreds of digits) as it will then take less space in the processing
until being (repeatedly) used.

This macro computes the \emph{exact} result (one can use it also for
polynomial evaluation), using a Horner scheme which helps avoiding a
denominator build-up (this problem however,  even if using a naive additive
approach, is much less acute since release |1.1| and its new policy regarding
\csbxint{Add}).

\begin{everbatim*}
\def\geom #1{1[0]} % the geometric series
\def\f {5/17[0]}
\[ \sum_{n=0}^{n=20} \Bigl(\frac 5{17}\Bigr)^n
 =\xintTeXFrac{\xintIrr{\xintPowerSeries {0}{20}{\geom}{\f}}}
 =\xintTeXFrac{\xinttheexpr (17^21-5^21)/12/17^20\relax}\]
\end{everbatim*}

\begin{everbatim*}
\def\coefflog #1{1/#1[0]}% 1/n
\def\f {1/2[0]}%
\[ \log 2 \approx \sum_{n=1}^{20} \frac1{n\cdot 2^n}
    = \xintTeXFrac {\xintIrr {\xintPowerSeries {1}{20}{\coefflog}{\f}}}\]
\[ \log 2 \approx \sum_{n=1}^{50} \frac1{n\cdot 2^n}
    = \xintTeXFrac {\xintIrr {\xintPowerSeries {1}{50}{\coefflog}{\f}}}\]
\end{everbatim*}


\begin{everbatim*}
\setlength{\columnsep}{0pt}
\begin{multicols}{3}
\cnta 1 % previously declared count
\loop   % in this loop we recompute from scratch each partial sum!
% we can afford that, as \xintPowerSeries is fast enough.
\noindent\hbox to 2em{\hfil\texttt{\the\cnta.} }%
         \xintTrunc {12}
             {\xintPowerSeries {1}{\cnta}{\coefflog}{\f}}\dots
\endgraf
\ifnum \cnta < 30 \advance\cnta 1 \repeat
\end{multicols}
\end{everbatim*}


\begin{everbatim*}
\def\coeffarctg  #1{1/\the\numexpr\ifodd #1 -2*#1-1\else2*#1+1\fi\relax }%
% the above gives (-1)^n/(2n+1). The sign being in the denominator,
%             **** no [0] should be added ****,
% else nothing is guaranteed to work (even if it could by sheer luck)
% Notice in passing this aspect of \numexpr:
%         ****  \numexpr -(1)\relax is ilegal !!! ****
\def\f {1/25[0]}% 1/5^2
\[\mathrm{Arctg}(\frac15)\approx \frac15\sum_{n=0}^{15} \frac{(-1)^n}{(2n+1)25^n}
= \xintTeXFrac{\xintIrr {\xintDiv {\xintPowerSeries {0}{15}{\coeffarctg}{\f}}{5}}}\]
\end{everbatim*}


\subsection{\csh{xintPowerSeriesX}}\label{xintPowerSeriesX}

%{\small\hspace*{\parindent}New with release |1.04|.\par}

\noindent This is the same as \csbxint{PowerSeries}\ntype{\numx\numx\Ff\Ff}
apart
from the fact that the last parameter |f| is expanded once and for all before
being then used repeatedly. If the |f| parameter is to be an explicit big
fraction with many (dozens) digits, rather than using it directly it is slightly
better to have some macro |\g| defined to expand to the explicit fraction and
then use \csbxint{PowerSeries} with |\g|; but if |f| has not yet been evaluated
and will be the output of a complicated expansion of some |\f|, and if, due to
an expanding only context, doing |\edef\g{\f}| is no option, then
\csa{xintPowerSeriesX} should be used with |\f| as last parameter.
%
\begin{everbatim*}
\def\ratioexp #1#2{\xintDiv {#1}{#2}}% x/n
% These are the (-1)^{n-1}/n of the log(1+h) series:
\def\coefflog #1{\the\numexpr\ifodd #1 1\else-1\fi\relax/#1[0]}%
% Let L(h) be the first 10 terms of the log(1+h) series and
% let E(t) be the first 10 terms of the exp(t) series.
% The following computes L(E(a/10)-1) for a=1,..., 12.
\begin{multicols}{3}\raggedcolumns
\cnta 1
\loop
\noindent\xintTrunc {18}{%
   \xintPowerSeriesX {1}{10}{\coefflog}
  {\xintSub
      {\xintRationalSeries {0}{9}{1[0]}{\ratioexp{\the\cnta[-1]}}}
      {1}}}\dots
\endgraf
\ifnum\cnta < 12 \advance \cnta 1 \repeat
\end{multicols}
\end{everbatim*}


\subsection{\csh{xintFxPtPowerSeries}}\label{xintFxPtPowerSeries}

\csa{xintFxPtPowerSeries}|{A}{B}{\coeff}{f}{D}|\etype{\numx\numx}
computes
$\sum_{\text{|n=A|}}^{\text{|n=B|}}$|\coeff{n}|${}\cdot |f|^{\,\text{|n|}}$ with each
    term of the series truncated to |D| digits\etype{\Ff\Ff\numx}
 after the decimal point. As
    usual, |A| and |B| are completely expanded through their inclusion in a
    |\numexpr| expression. Regarding |D| it will be similarly be expanded each
    time it is used inside an \csa{xintTrunc}. The one-parameter macro |\coeff|
    is similarly  expanded at the time it is used inside the
    computations. Idem for |f|. If |f| itself is some complicated macro it is
    thus better to use the variant \csbxint{FxPtPowerSeriesX} which expands it
    first and then uses the result of that expansion.

The current (|1.04|) implementation is: the first power |f^A| is
computed exactly, then \emph{truncated}. Then each successive power is
obtained from the previous one by multiplication by the exact value of
|f|, and truncated. And |\coeff{n}|\raisebox{.5ex}{|.|}|f^n| is obtained
from that by multiplying by |\coeff{n}| (untruncated) and then
truncating. Finally the sum is computed exactly. Apart from that
\csa{xintFxPtPowerSeries} (where |FxPt| means `fixed-point') is like
\csa{xintPowerSeries}.

There should be a variant for things of the type $\sum c_n \frac {f^n}{n!}$ to
avoid having to compute the factorial from scratch at each coefficient, the same
way \csa{xintFxPtPowerSeries} does not compute |f^n| from scratch at each |n|.
Perhaps in the next package release.

\def\coeffexp #1{1/\xintiiFac {#1}[0]}% [0] for faster parsing
\def\f {-1/2[0]}%
\newcount\cnta

\setlength{\multicolsep}{0pt}

\begin{multicols}{3}[%
\centeredline{$e^{-\frac12}\approx{}$}]%
\cnta 0
\noindent\loop
$\xintFxPtPowerSeries {0}{\cnta}{\coeffexp}{\f}{20}$\\
\ifnum\cnta<19
\advance\cnta 1
\repeat\par
\end{multicols}
\everb|@
\def\coeffexp #1{1/\xintiiFac {#1}[0]}% 1/n!
\def\f {-1/2[0]}% [0] for faster input parsing
\cnta 0 % previously declared \count register
\noindent\loop
$\xintFxPtPowerSeries {0}{\cnta}{\coeffexp}{\f}{20}$\\
\ifnum\cnta<19 \advance\cnta 1 \repeat\par
|


%
\leftedline{|\xintFxPtPowerSeries {0}{19}{\coeffexp}{\f}{25}=|
\dtt{\xintFxPtPowerSeries {0}{19}{\coeffexp}{\f}{25}}}
\fdef\z{\xintIrr {\xintPowerSeries {0}{19}{\coeffexp}{\f}}}
%

\texttt{\hyphenchar\font45 }%
It is no difficulty for \xintfracname to compute exactly, with the help
of \csa{xintPowerSeries}, the nineteenth partial sum, and to then give
(the start of) its exact decimal expansion:
%
\leftedline{|\xintPowerSeries {0}{19}{\coeffexp}{\f}| ${}=
  \displaystyle\xintTeXFrac{\z}$%
  \vphantom{\vrule height 20pt depth 12pt}}%
%
\leftedline{${}=\xintTrunc {30}{\z}\dots$} Thus, one should always
estimate a priori how many ending digits are not reliable: if there are
|N| terms and |N| has |k| digits, then digits up to but excluding the
last |k| may usually be trusted. If we are optimistic and the series is
alternating we may even replace |N| with $\sqrt{|N|}$ to get the number |k|
of digits possibly of dubious significance.

\subsection{\csh{xintFxPtPowerSeriesX}}\label{xintFxPtPowerSeriesX}


\noindent\csa{xintFxPtPowerSeriesX}|{A}{B}{\coeff}{\f}{D}|%
\ntype{\numx\numx}
computes, exactly as
\csa{xintFxPtPowerSeries}, the sum of
|\coeff{n}|\raisebox{.5ex}{|.|}|\f^n|\etype{\Ff\Ff\numx} from |n=A| to |n=B| with each term
of the series being \emph{truncated} to |D| digits after the decimal
point. The sole difference is that |\f| is first expanded and it
is the result of this which is used in the computations.


Let us illustrate this on the numerical exploration of the identity
%
\leftedline{|log(1+x) = -log(1/(1+x))|}
%
Let |L(h)=log(1+h)|, and |D(h)=L(h)+L(-h/(1+h))|. Theoretically thus,
|D(h)=0| but we shall evaluate |L(h)| and |-h/(1+h)| keeping only 10
terms of their respective series. We will assume $|h|<0.5$. With only
ten terms kept in the power series we do not have quite 3 digits
precision as $2^{10}=1024$. So it wouldn't make sense to evaluate things
more precisely than, say circa 5 digits after the decimal points.
\begin{everbatim*}
\cnta 0
\def\coefflog #1{\the\numexpr\ifodd#1 1\else-1\fi\relax/#1[0]}% (-1)^{n-1}/n
\def\coeffalt #1{\the\numexpr\ifodd#1 -1\else1\fi\relax [0]}%   (-1)^n
\begin{multicols}2
\loop
\noindent \hbox to 2.5cm {\hss\texttt{D(\the\cnta/100): }}%
\xintAdd {\xintFxPtPowerSeriesX {1}{10}{\coefflog}{\the\cnta [-2]}{5}}
         {\xintFxPtPowerSeriesX {1}{10}{\coefflog}
             {\xintFxPtPowerSeriesX {1}{10}{\coeffalt}{\the\cnta [-2]}{5}}
          {5}}\endgraf
\ifnum\cnta < 49 \advance\cnta 7 \repeat
\end{multicols}
\end{everbatim*}


Let's say we evaluate functions on |[-1/2,+1/2]| with values more or less also
in |[-1/2,+1/2]| and we want to keep 4 digits of precision. So, roughly we need
at least 14 terms in series like the geometric or log series. Let's make this
15. Then it doesn't make sense to compute intermediate summands with more than 6
digits precision. So we compute with 6 digits
precision but return only 4 digits (rounded) after the decimal point.
This result with 4 post-decimal points precision is then used as input
to the next evaluation.
\begin{everbatim*}
\begin{multicols}2
\loop
\noindent \hbox to 2.5cm {\hss\texttt{D(\the\cnta/100): }}%
\dtt{\xintRound{4}
 {\xintAdd {\xintFxPtPowerSeriesX {1}{15}{\coefflog}{\the\cnta [-2]}{6}}
           {\xintFxPtPowerSeriesX {1}{15}{\coefflog}
                  {\xintRound {4}{\xintFxPtPowerSeriesX {1}{15}{\coeffalt}
                                 {\the\cnta [-2]}{6}}}
            {6}}%
 }}\endgraf
\ifnum\cnta < 49 \advance\cnta 7 \repeat
\end{multicols}
\end{everbatim*}

Not bad... I have cheated a bit: the `four-digits precise' numeric
evaluations were left unrounded in the final addition. However the inner
rounding to four digits worked fine and made the next step faster than
it would have been with longer inputs. The morale is that one should not
use the raw results of \csa{xintFxPtPowerSeriesX} with the |D| digits
with which it was computed, as the last are to be considered garbage.
Rather, one should keep from the output only some smaller number of
digits. This will make further computations faster and not less precise.
I guess there should be some macro to do this final truncating, or
better, rounding, at a given number |D'<D| of digits. Maybe for the next
release.

\subsection{\csh{xintFloatPowerSeries}}\label{xintFloatPowerSeries}


\noindent\csa{xintFloatPowerSeries}|[P]{A}{B}{\coeff}{f}|%
\ntype{{\upshape[\numx]}\numx\numx}
 computes
$\sum_{\text{|n=A|}}^{\text{|n=B|}}$|\coeff{n}|${}\cdot |f|^{\,\text{|n|}}$
with a floating point
precision given by the optional parameter |P| or by the current setting of
|\xintDigits|.\etype{\Ff\Ff}

In the current, preliminary, version, no attempt has been made to try to
guarantee to the final result the precision |P|. Rather, |P| is used for all
intermediate floating point evaluations. So
rounding errors will make some of the last printed digits invalid. The
operations done are first the evaluation of |f^A| using \csa{xintFloatPow}, then
each successive power is obtained from this first one by multiplication by |f|
using \csa{xintFloatMul}, then again with \csa{xintFloatMul} this is multiplied
with |\coeff{n}|, and the sum is done adding one term at a time with
\csa{xintFloatAdd}. To sum up, this is just the naive transformation of
\csa{xintFxPtPowerSeries} from fixed point to floating point.

\def\coefflog #1{\the\numexpr\ifodd#1 1\else-1\fi\relax/#1[0]}%

\everb+@
\def\coefflog #1{\the\numexpr\ifodd#1 1\else-1\fi\relax/#1[0]}%
\xintFloatPowerSeries [8]{1}{30}{\coefflog}{-1/2[0]}
+

%
\leftedline{\dtt{\xintFloatPowerSeries [8]{1}{30}{\coefflog}{-1/2[0]}}}

\subsection{\csh{xintFloatPowerSeriesX}}\label{xintFloatPowerSeriesX}


\noindent\csa{xintFloatPowerSeriesX}|[P]{A}{B}{\coeff}{f}|%
\ntype{{\upshape[\numx]}\numx\numx}
is like
\csa{xintFloatPowerSeries} with the difference that |f| is
expanded once\etype{\Ff\Ff}
and for all at the start of the computation, thus allowing
efficient chaining of such series evaluations.
\def\coefflog #1{\the\numexpr\ifodd#1 1\else-1\fi\relax/#1[0]}%

\everb+@
\def\coeffexp #1{1/\xintiiFac {#1}[0]}% 1/n! (exact, not float)
\def\coefflog #1{\the\numexpr\ifodd#1 1\else-1\fi\relax/#1[0]}%
\xintFloatPowerSeriesX [8]{0}{30}{\coeffexp}
    {\xintFloatPowerSeries [8]{1}{30}{\coefflog}{-1/2[0]}}
+

%
\leftedline{\dtt{\xintFloatPowerSeriesX [8]{0}{30}{\coeffexp}
    {\xintFloatPowerSeries [8]{1}{30}{\coefflog}{-1/2[0]}}}}

\subsection{Computing \texorpdfstring{$\log 2$}{log(2)} and \texorpdfstring{$\pi$}{pi}}\label{ssec:Machin}

In this final section, the use of \csbxint{FxPtPowerSeries} (and
\csbxint{PowerSeries}) will be
illustrated on the (expandable...\@ why make things simple when it is so easy to
make them difficult!) computations of the first digits of the decimal expansion
of the familiar constants $\log 2$ and $\pi$.

Let us start with $\log 2$. We will get it from this formula (which is
left as an exercise): %
%
\leftedline{\dtt{log(2)=-2\,log(1-13/256)-%
 5\,log(1-1/9)}}
%
The number of terms to be kept in the log series, for a desired
precision of |10^{-D}| was roughly estimated without much theoretical
analysis. Computing exactly the partial sums with \csa{xintPowerSeries}
and then printing the truncated values, from |D=0| up to |D=100| showed
that it worked in terms of quality of the approximation. Because of
possible strings of zeroes or nines in the exact decimal expansion (in
the present case of $\log 2$, strings of zeroes around the fourtieth and
the sixtieth decimals), this
does not mean though that all digits printed were always exact. In
the end one always end up having to compute at some higher level of
desired precision to validate the earlier result.

Then we tried with \csa{xintFxPtPowerSeries}: this is worthwile only for
|D|'s at least 50, as the exact evaluations are faster (with these
short-length |f|'s) for a lower
number of digits. And as expected the degradation in the quality of
approximation was in this range of the order of two or three digits.
This meant roughly that the 3+1=4 ending digits were wrong. Again, we ended
up having to compute with five more digits and compare with the earlier
value to validate it. We use truncation rather than rounding because our
goal is not to obtain the correct rounded decimal expansion but the
correct exact truncated one.

% 693147180559945309417232121458176568075500134360255254120680009493

\begin{everbatim*}
\def\coefflog #1{1/#1[0]}% 1/n
\def\xa {13/256[0]}%  we will compute log(1-13/256)
\def\xb {1/9[0]}%     we will compute log(1-1/9)
\def\LogTwo #1%
%  get log(2)=-2log(1-13/256)- 5log(1-1/9)
{% we want to use \printnumber, hence need something expanding in two steps
 % only, so we use here the \romannumeral0 method
  \romannumeral0\expandafter\LogTwoDoIt \expandafter
    % Nb Terms for 1/9:
  {\the\numexpr #1*150/143\expandafter}\expandafter
    % Nb Terms for 13/256:
  {\the\numexpr #1*100/129\expandafter}\expandafter
    % We print #1 digits, but we know the ending ones are garbage
  {\the\numexpr #1\relax}% allows #1 to be a count register
}%
\def\LogTwoDoIt #1#2#3%
%  #1=nb of terms for 1/9, #2=nb of terms for 13/256,
{% #3=nb of digits for computations, also used for printing
 \xinttrunc {#3} % lowercase form to stop the \romannumeral0 expansion!
 {\xintAdd
  {\xintMul {2}{\xintFxPtPowerSeries {1}{#2}{\coefflog}{\xa}{#3}}}
  {\xintMul {5}{\xintFxPtPowerSeries {1}{#1}{\coefflog}{\xb}{#3}}}%
 }%
}%
\noindent $\log 2 \approx \LogTwo {60}\dots$\endgraf
\noindent\phantom{$\log 2$}${}\approx{}$\printnumber{\LogTwo {65}}\dots\endgraf
\noindent\phantom{$\log 2$}${}\approx{}$\printnumber{\LogTwo {70}}\dots\endgraf
\end{everbatim*}

Here is the code doing an exact evaluation of the partial sums. We have
added a |+1| to the number of digits for estimating the number of terms
to keep from the log series: we experimented that this gets exactly the
first |D| digits, for all values from |D=0| to |D=100|, except in one
case (|D=40|) where the last digit is wrong. For values of |D|
higher than |100| it is more efficient to use the code using
\csa{xintFxPtPowerSeries}.
\everb|@
\def\LogTwo #1% get log(2)=-2log(1-13/256)- 5log(1-1/9)
{%
    \romannumeral0\expandafter\LogTwoDoIt \expandafter
    {\the\numexpr (#1+1)*150/143\expandafter}\expandafter
    {\the\numexpr (#1+1)*100/129\expandafter}\expandafter
    {\the\numexpr #1\relax}%
}%
\def\LogTwoDoIt #1#2#3%
{%   #3=nb of digits for truncating an EXACT partial sum
  \xinttrunc {#3}
    {\xintAdd
      {\xintMul {2}{\xintPowerSeries {1}{#2}{\coefflog}{\xa}}}
      {\xintMul {5}{\xintPowerSeries {1}{#1}{\coefflog}{\xb}}}%
    }%
}%
|

Let us turn now to Pi, computed with the Machin formula (but see also the
approach via the \hyperlink{BrentSalamin}{Brent-Salamin algorithm} with
\csa{xintfloatexpr}) Again the numbers of terms to keep in the two |arctg|
series were roughly estimated, and some experimentations showed that removing
the last three digits was enough (at least for |D=0-100| range). And the
algorithm does print the correct digits when used with |D=1000| (to be
convinced of that one needs to run it for |D=1000| and again, say for
|D=1010|.) A theoretical analysis could help confirm that this algorithm
always gets better than |10^{-D}| precision, but again, strings of zeroes or
nines encountered in the decimal expansion may falsify the ending digits,
nines may be zeroes (and the last non-nine one should be increased) and zeroes
may be nine (and the last non-zero one should be decreased).

\hypertarget{MachinCode}{}
\begin{everbatim*}
\def\coeffarctg #1{\the\numexpr\ifodd#1 -1\else1\fi\relax/%
                                       \the\numexpr 2*#1+1\relax [0]}%
%\def\coeffarctg #1{\romannumeral0\xintmon{#1}/\the\numexpr 2*#1+1\relax }%
\def\xa {1/25[0]}%      1/5^2, the [0] for faster parsing
\def\xb {1/57121[0]}% 1/239^2, the [0] for faster parsing
\def\Machin #1{% #1 may be a count register, \Machin {\mycount} is allowed
    \romannumeral0\expandafter\MachinA \expandafter
     % number of terms for arctg(1/5):
    {\the\numexpr (#1+3)*5/7\expandafter}\expandafter
     % number of terms for arctg(1/239):
    {\the\numexpr (#1+3)*10/45\expandafter}\expandafter
     % do the computations with 3 additional digits:
    {\the\numexpr #1+3\expandafter}\expandafter
     % allow #1 to be a count register:
    {\the\numexpr #1\relax }}%
\def\MachinA #1#2#3#4%
{\xinttrunc {#4}
 {\xintSub
  {\xintMul {16/5}{\xintFxPtPowerSeries {0}{#1}{\coeffarctg}{\xa}{#3}}}
  {\xintMul{4/239}{\xintFxPtPowerSeries {0}{#2}{\coeffarctg}{\xb}{#3}}}%
 }}%
\begin{framed}
  \[ \pi = \Machin {60}\dots \]
\end{framed}
\end{everbatim*}

Here is a variant|\MachinBis|,
which evaluates the partial sums \emph{exactly} using
\csa{xintPowerSeries}, before their final truncation. No need for a
``|+3|'' then.
\begin{everbatim*}
\def\MachinBis #1{% #1 may be a count register,
% the final result will be truncated to #1 digits post decimal point
    \romannumeral0\expandafter\MachinBisA \expandafter
     % number of terms for arctg(1/5):
    {\the\numexpr #1*5/7\expandafter}\expandafter
     % number of terms for arctg(1/239):
    {\the\numexpr #1*10/45\expandafter}\expandafter
      % allow #1 to be a count register:
    {\the\numexpr #1\relax }}%
\def\MachinBisA #1#2#3%
{\xinttrunc {#3} %
 {\xintSub
   {\xintMul {16/5}{\xintPowerSeries {0}{#1}{\coeffarctg}{\xa}}}
   {\xintMul{4/239}{\xintPowerSeries {0}{#2}{\coeffarctg}{\xb}}}%
}}%
\end{everbatim*}

Let us use this variant for a loop showing the build-up of digits:
\begin{everbatim*}
\begin{multicols}{2}
  \cnta 0 % previously declared \count register
  \loop \noindent
        \centeredline{\dtt{\MachinBis{\cnta}}}%
  \ifnum\cnta < 30
  \advance\cnta 1 \repeat
\end{multicols}
\end{everbatim*}

\hypertarget{Machin1000}{}
%
You want more digits and have some time? compile this copy of the
\hyperlink{MachinCode}{|\Machin|} with |etex| (or |pdftex|):
%
\everb|@
% Compile with e-TeX extensions enabled (etex, pdftex, ...)
\input xintfrac.sty
\input xintseries.sty
% pi = 16 Arctg(1/5) - 4 Arctg(1/239) (John Machin's formula)
\def\coeffarctg #1{\the\numexpr\ifodd#1 -1\else1\fi\relax/%
                                       \the\numexpr 2*#1+1\relax [0]}%
\def\xa {1/25[0]}%
\def\xb {1/57121[0]}%
\def\Machin #1{%
    \romannumeral0\expandafter\MachinA \expandafter
    {\the\numexpr (#1+3)*5/7\expandafter}\expandafter
    {\the\numexpr (#1+3)*10/45\expandafter}\expandafter
    {\the\numexpr #1+3\expandafter}\expandafter
    {\the\numexpr #1\relax }}%
\def\MachinA #1#2#3#4%
{\xinttrunc {#4}
 {\xintSub
  {\xintMul {16/5}{\xintFxPtPowerSeries {0}{#1}{\coeffarctg}{\xa}{#3}}}
  {\xintMul {4/239}{\xintFxPtPowerSeries {0}{#2}{\coeffarctg}{\xb}{#3}}}%
}}%
\pdfresettimer
\fdef\Z {\Machin {1000}}
\odef\W {\the\pdfelapsedtime}
\message{\Z}
\message{computed in \xintRound {2}{\W/65536} seconds.}
\bye
|

This will log the first 1000 digits of $\pi$ after the decimal point. On my
laptop (a 2012 model) this took about $5.05$ seconds last time I tried.%
%
\footnote{With \texttt{1.09i} and earlier \xintname, this used to be \dtt{42}
  seconds; starting with \texttt{1.09j}, and prior to \texttt{1.2}, it was
  \dtt{16} seconds (this was probably due to a more efficient division with
  denominators at most $9999$). The |1.2| \xintcorename achieves a further
  gain at \dtt{5.6} seconds.}
%
\footnote{With |\xintDigits:=1001\relax|, the non-optimized implementation with the
  |iter| of \xintexprname fame using the
  \hyperlink{BrentSalamin}{Brent-Salamin algorithm}, took, last time I tried
  (1.2i), about \dtt{7} seconds on my laptop (the last two digits were wrong,
  which is ok as they serve as guard digits), and for obtaining about
  \dtt{500} digits, it was about \dtt{1.7}s. This is not bad, taking into
  account that the syntax is almost free rolling speech, contrarily to the
  code above for the Machin formula computation; we would like to use the
  quadratically convergent Brent-Salamin algorithm for more digits, but with
  such computations with numbers of one thousand digits we are beyond the
  border of the reasonable range for \xintname. Innocent people not knowing
  what it means to compute with \TeX, and with the extra constraint of
  expandability will wonder why this is at least thousands of times slower
  than with any other language (with a little Python program using the
  |Decimal| library, I timed the Brent-Salamin algorithm to \dtt{4.4ms} for
  about |1000| digits and \dtt{1.14ms} for |500| digits.) I will just say that
  for example digits are represented and manipulated via their ascii-code !
  all computations must convert from ascii-code to cpu words; furthermore
  nothing can be stored away. And there is no memory storage with |O(1)| time
  access...\@ if expandability is to be verified.}
%


As mentioned in the
introduction, the file \href{https://ctan.org/pkg/pi}{pi.tex} by \textsc{D.
  Roegel} shows that orders of magnitude faster computations are possible within
\TeX{}, but recall our constraints of complete expandability and be merciful,
please.

\textbf{Why truncating rather than rounding?} One of our main competitors
on the market of scientific computing, a canadian product (not
encumbered with expandability constraints, and having barely ever heard
of \TeX{} ;-), prints numbers rounded in the last digit. Why didn't we
follow suit in the macros \csa{xintFxPtPowerSeries} and
\csa{xintFxPtPowerSeriesX}? To round at |D| digits, and excluding a
rewrite or cloning of the division algorithm which anyhow would add to
it some overhead in its final steps, \xintfracname needs to truncate at
|D+1|, then round. And rounding loses information! So, with more time
spent, we obtain a worst result than the one truncated at |D+1| (one
could imagine that additions and so on, done with only |D| digits, cost
less; true, but this is a negligeable effect per summand compared to the
additional cost for this term of having been truncated at |D+1| then
rounded). Rounding is the way to go when setting up algorithms to
evaluate functions destined to be composed one after the other: exact
algebraic operations with many summands and an |f| variable which is a
fraction are costly and create an even bigger fraction; replacing |f|
with a reasonable rounding, and rounding the result, is necessary to
allow arbitrary chaining.

But, for the
computation of a single constant, we are really interested in the exact
decimal expansion, so we truncate and compute more terms until the
earlier result gets validated. Finally if we do want the rounding we can
always do it on a value computed with |D+1| truncation.

\clearpage
\let\xintseriesnameUp\undefined
\csname xintcfracnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xintcfracname package}
\RaisedLabel{sec:cfrac}

First version of this package was included in release |1.04| (|2013/04/25|) of the
\xintname bundle. It was kept almost unchanged until |1.09m| of |2014/02/26|
which brought some new macros: \csbxint{FtoC}, \csbxint{CtoF}, \csbxint{CtoCv},
dealing with sequences of braced partial quotients rather than comma separated
ones, \csbxint{FGtoC} which is to produce ``guaranteed'' coefficients of some
real number known approximately, and \csbxint{GGCFrac} for displaying arbitrary
material as a continued fraction; also, some changes to existing macros:
\csbxint{FtoCs} and \csbxint{CntoCs} insert spaces after the commas,
\csbxint{CstoF} and \csbxint{CstoCv} authorize spaces in the input also before
the commas.

Note: \csbxint{CstoF} and \csbxint{CstoCv} create a partial dependency on
\xinttoolsname (its \csbxint{CSVtoList}.)

\localtableofcontents

% This section contains:
% \begin{enumerate}
% \item an \hyperref[ssec:cfracoverview]{overview} of the package functionalities,
% \item a description of each one of the package macros,
% \item further illustration of their use via the study of the
%   \hyperref[ssec:e-convergents]{convergents of $e$}.
% \end{enumerate}

\subsection{Package overview}\label{ssec:cfracoverview}

The package computes partial quotients and convergents of a fraction, or
conversely start from coefficients and obtain the corresponding fraction; three
macros \csbxint {CFrac}, \csbxint {GCFrac} and \csbxint {GGCFrac} are
for typesetting, the others
can be nested (if applicable) or see their outputs further processed by other
macros from the \xintname bundle, particularly the macros of \xinttoolsname
dealing with sequences of braced items or comma separated lists.

A \emph{simple} continued fraction has coefficients
|[c0,c1,...,cN]| (usually called partial quotients, but I
dislike this entrenched terminology), where |c0| is a positive or
negative integer and the others are positive integers.

Typesetting is usually done via the |amsmath| macro |\cfrac|:
\begin{everbatim*}
\[ c_0 + \cfrac{1}{c_1+\cfrac1{c_2+\cfrac1{c_3+\cfrac1{\ddots}}}}\]
\end{everbatim*}

Here is a concrete example:
\begin{everbatim*}
\[ \xintTeXFrac {208341/66317}=\xintCFrac {208341/66317}\]%
\end{everbatim*}
But it is the macro \csbxint{CFrac} which did all the work of \emph{computing}
the continued fraction \emph{and} using |\cfrac| from |amsmath| to typeset
it.

A \emph{generalized} continued fraction has the same structure but the
numerators are not restricted to be $1$, and numbers used in the continued
fraction may be arbitrary, also fractions, irrationals, complex,
indeterminates.%
%
\footnote{\xintcfracname may be used with indeterminates,
  for basic conversions from one inline format to another, but not for
  actual computations. See \csbxint{GGCFrac}.}
%
The \emph{centered} continued fraction is an
example:
\begin{everbatim*}
\[ \xintTeXFrac {915286/188421}=\xintGCFrac {5+-1/7+1/39+-1/53+-1/13}
  =\xintCFrac {915286/188421}\]
\end{everbatim*}

The macro \csbxint{GCFrac}, contrarily to
\csbxint{CFrac}, does not compute anything, it just typesets starting from a
generalized continued fraction in inline format, which in this example
was input literally.  We also used \csa{xintCFrac}
for comparison of the two types of continued fractions.

To let \TeX{} compute the centered continued fraction of |f| there is
\csbxint{FtoCC}:
\begin{everbatim*}
\[\xintTeXFrac {915286/188421}\to\xintFtoCC {915286/188421}\]
\end{everbatim*}
The package macros are expandable and may be nested (naturally \csa{xintCFrac}
and \csa{xintGCFrac} must be at the top level, as they deal with typesetting).
\begin{everbatim*}
\[\xintGCFrac {\xintFtoCC{915286/188421}}\]
\end{everbatim*}

The `inline' format expected on input by \csbxint{GCFrac} is
%
\leftedline{$a_0+b_0/a_1+b_1/a_2+b_2/a_3+\cdots+b_{n-2}/a_{n-1}+b_{n-1}/a_n$}
%
Fractions among the coefficients are allowed but they must be enclosed
within braces. Signed integers may be left without braces (but the |+|
signs are mandatory). No spaces are allowed around the plus and fraction
symbols. The coefficients may themselves be macros, as long as these
macros are \fexpan dable.
\begin{everbatim*}
\[ \xintTeXFrac{\xintGCtoF {1+-1/57+\xintPow {-3}{7}/\xintiiQuo {132}{25}}}
    = \xintGCFrac {1+-1/57+\xintPow {-3}{7}/\xintiiQuo {132}{25}}\]
\end{everbatim*}
To compute the actual fraction one has \csbxint{GCtoF}:
\begin{everbatim*}
\[\xintTeXFrac{\xintGCtoF {1+-1/57+\xintPow {-3}{7}/\xintiiQuo {132}{25}}}\]
\end{everbatim*}
For non-numeric input  there is \csbxint{GGCFrac}.
\begin{everbatim*}
\[\xintGGCFrac {a_0+b_0/a_1+b_1/a_2+b_2/\ddots+\ddots/a_{n-1}+b_{n-1}/a_n}\]
\end{everbatim*}
For regular continued fractions, there is a simpler comma separated format:
\begin{everbatim*}
\[-7,6,19,1,33\to\xintTeXFrac{\xintCstoF{-7,6,19,1,33}}=\xintCFrac{\xintCstoF{-7,6,19,1,33}}\]
\end{everbatim*}
The macro \csbxint{FtoCs} produces from a fraction |f| the comma separated
list of its coefficients.
\begin{everbatim*}
\[\xintTeXFrac{1084483/398959}=[\xintFtoCs{1084483/398959}]\]
\end{everbatim*}
If one prefers other separators, one can use the two arguments macros
\csbxint{FtoCx} whose first argument is the separator (which may consist of more
than one token) which is to be used.
\begin{everbatim*}
\[\xintTeXFrac{2721/1001}=\xintFtoCx {+1/(}{2721/1001})\cdots)\]
\end{everbatim*}
This allows under Plain  \TeX{} with |amstex| to obtain the same effect
as with \LaTeX{}+|\amsmath|+\csbxint{CFrac}:
%
\leftedline{|$$\xintTeXOver{2721/1001}=\xintFtoCx {+\cfrac1\\ }{2721/1001}\endcfrac$$|}

As a shortcut to \csa{xintFtoCx} with separator |1+/|, there is
\csbxint{FtoGC}:
\begin{everbatim*}
2721/1001=\xintFtoGC {2721/1001}
\end{everbatim*}
Let us compare in that case with the output of \csbxint{FtoCC}:
\begin{everbatim*}
2721/1001=\xintFtoCC {2721/1001}
\end{everbatim*}
To obtain the coefficients as a sequence of braced numbers, there is
\csbxint{FtoC} (this is a shortcut for |\xintFtoCx {}|). This list
(sequence) may then be manipulated using the various macros of \xinttoolsname
such as the non-expandable macro \csbxint{AssignArray} or the expandable
\csbxint{Apply} and \csbxint{ListWithSep}.

Conversely to go from such a sequence of braced coefficients to the
corresponding fraction there is \csbxint{CtoF}.

The `|\printnumber|' (\autoref{ssec:printnumber}) macro which we use in this
document to print long numbers can also be useful on long continued fractions.
%
\begin{everbatim*}
\printnumber{\xintFtoCC {35037018906350720204351049/244241737886197404558180}}
\end{everbatim*}
%
If we apply \csbxint{GCtoF} to this generalized continued fraction, we
discover that the original fraction was reducible:
%
\leftedline{|\xintGCtoF
  {143+1/2+...+-1/9}|\dtt{=\xintGCtoF{143+1/2+1/5+-1/4+-1/4+-1/4+-1/3+1/2+1/2+1/6+-1/22+1/2+1/10+-1/5+-1/11+-1/3+1/4+-1/2+1/2+1/4+-1/2+1/23+1/3+1/8+-1/6+-1/9}}}

\def\mymacro #1{$\xintTeXFrac{#1}=[\xintFtoCs{#1}]$\vtop to 6pt{}}

\begingroup
\catcode`^\active
\def^#1^{\hbox{#1}}%

When a generalized continued fraction is built with integers, and
numerators are only |1|'s or |-1|'s, the produced fraction is
irreducible. And if we compute it again with the last sub-fraction
omitted we get another irreducible fraction related to the bigger one by
a Bézout identity. Doing this here we get:
%
\leftedline{|\xintGCtoF {143+1/2+...+-1/6}|\dtt{=\xintGCtoF{143+1/2+1/5+-1/4+-1/4+-1/4+-1/3+1/2+1/2+1/6+-1/22+1/2+1/10+-1/5+-1/11+-1/3+1/4+-1/2+1/2+1/4+-1/2+1/23+1/3+1/8+-1/6}}}
and indeed:
\[\begin{vmatrix}
    ^2897319801297630107^ & ^328124887710626729^\\
      ^20197107104701740^ & ^2287346221788023^
   \end{vmatrix} = \mbox{\dtt{\xintiiSub {\xintiiMul {2897319801297630107}{2287346221788023}}{\xintiiMul{20197107104701740}{328124887710626729}}}}\]

\endgroup

The various fractions obtained from the truncation of a continued fraction to
its initial terms are called the convergents. The macros of \xintcfracname
such as \csbxint{FtoCv}, \csbxint{FtoCCv}, and others which compute such
convergents, return them as a list of braced items, with no separator (as does
\csbxint {FtoC} for the partial quotients). Here is an example:

\begin{everbatim*}
\[\xintTeXFrac{915286/188421}\to
  \xintListWithSep{,}{\xintApply\xintTeXFrac{\xintFtoCv{915286/188421}}}\]
\end{everbatim*}
\begin{everbatim*}
\[\xintTeXFrac{915286/188421}\to
  \xintListWithSep{,}{\xintApply\xintTeXFrac{\xintFtoCCv{915286/188421}}}\]
\end{everbatim*}
%
We thus see that the `centered convergents' obtained with \csbxint{FtoCCv} are
among the fuller list of convergents as returned by \csbxint{FtoCv}.

Here is a more complicated use of \csa{xintApply}
and \csa{xintListWithSep}. We first define a macro which will be applied to each
convergent:%
%
\leftedline{|\newcommand{\mymacro}[1]{$\xintTeXFrac{#1}=[\xintFtoCs{#1}]$\vtop to 6pt{}}|}
%
Next, we use the following code:
%
\leftedline{|$\xintTeXFrac{49171/18089}\to{}$|}
%
\leftedline{|\xintListWithSep {,
  }{\xintApply{\mymacro}{\xintFtoCv{49171/18089}}}|}
It produces:\par
\noindent$ \xintTeXFrac{49171/18089}\to {}$\xintListWithSep {,
  }{\xintApply{\mymacro}{\xintFtoCv{49171/18089}}}.

The macro \csbxint{CntoF} allows to specify the coefficients as a function given
by a one-parameter macro. The produced values do not have to be integers.
\begin{everbatim*}
\def\cn #1{\xintiiPow {2}{#1}}% 2^n
  \[\xintTeXFrac{\xintCntoF {6}{\cn}}=\xintCFrac [l]{\xintCntoF {6}{\cn}}\]
\end{everbatim*}

Notice  the use of the optional argument |[l]| to \csa{xintCFrac}. Other
possibilities are |[r]| and (default) |[c]|.
\begin{everbatim*}
\def\cn #1{\xintPow {2}{-#1}}%
  \[\xintTeXFrac{\xintCntoF {6}{\cn}}=\xintGCFrac [r]{\xintCntoGC {6}{\cn}}=
      [\xintFtoCs {\xintCntoF {6}{\cn}}]\]
\end{everbatim*}
We used \csbxint{CntoGC} as we wanted to display also the continued fraction and
not only the fraction returned by \csa{xintCntoF}.

There are also \csbxint{GCntoF} and \csbxint{GCntoGC} which allow the same for
generalized fractions. An initial portion of a generalized continued
fraction for $\pi$ is obtained like this
\begin{everbatim*}
\def\an #1{\the\numexpr 2*#1+1\relax }%
\def\bn #1{\the\numexpr (#1+1)*(#1+1)\relax }%
\[\xintTeXFrac{\xintDiv {4}{\xintGCntoF {5}{\an}{\bn}}} =
        \cfrac{4}{\xintGCFrac{\xintGCntoGC {5}{\an}{\bn}}} =
                  \xintTrunc {10}{\xintDiv {4}{\xintGCntoF {5}{\an}{\bn}}}\dots\]
\end{everbatim*}

We see that the quality of approximation is not fantastic compared to the simple
continued fraction of $\pi$ with about as many terms:
\begin{everbatim*}
\[\xintTeXFrac{\xintCstoF{3,7,15,1,292,1,1}}=
             \xintGCFrac{3+1/7+1/15+1/1+1/292+1/1+1/1}=
                       \xintTrunc{10}{\xintCstoF{3,7,15,1,292,1,1}}\dots\]
\end{everbatim*}

When studying the continued fraction of some real number, there is always
some doubt about how many terms are valid, when computed starting from some
approximation. If $f\leqslant x\leqslant g$ and $f, g$ both have the
same first $K$ partial quotients, then $x$ also has the same first $K$ quotients
and convergents. The macro \csbxint{FGtoC} outputs as a sequence of braced items
the common partial quotients of its two arguments. We can thus use it to produce
a sure list of valid convergents of $\pi$ for example, starting from some proven
lower and upper bound:
\begin{everbatim*}
$$\pi\to [\xintListWithSep{,}
         {\xintFGtoC {3.14159265358979323}{3.14159265358979324}}, \dots]$$
\noindent$\pi\to\xintListWithSep{,\allowbreak\;}
   {\xintApply{\xintTeXFrac}
   {\xintCtoCv{\xintFGtoC {3.14159265358979323}{3.14159265358979324}}}}, \dots$
\end{everbatim*}


\subsection{\csh{xintCFrac}}\label{xintCFrac}

\csa{xintCFrac}|{f}|\ntype{\Ff} is a math-mode only, \LaTeX{} with |amsmath|
only, macro which first computes then displays with the help of |\cfrac| the
simple continued fraction corresponding to the given fraction. It admits an
optional argument which may be |[l]|, |[r]| or (the default) |[c]| to specify
the location of the one's in the numerators of the sub-fractions.
This macro is \fexpan dable in the sense that it prepares expandably
the whole expression with the multiple |\cfrac|'s, but it is not completely
expandable naturally as |\cfrac| isn't.

\subsection{\csh{xintGCFrac}}\label{xintGCFrac}

\csa{xintGCFrac}|{a+b/c+d/e+f/g+h/...+x/y}|\ntype{f} uses similarly |\cfrac|
to prepare the typesetting with the |amsmath| |\cfrac| (\LaTeX{}) of a
generalized continued fraction given in inline format (or as macro which
will \fexpan d to it). It admits the
same optional argument as \csa{xintCFrac}. Plain \TeX{} with |amstex|
users, see \csbxint{GCtoGCx}.
\begin{everbatim*}
\[\xintGCFrac {1+\xintPow{1.5}{3}/{1/7}+{-3/5}/\xintiiFac {6}}\]
\end{everbatim*}
This is mostly a typesetting macro, although it does trigger the
expansion of the coefficients. See \csbxint{GCtoF} if you are impatient
to see this specific fraction computed.

It admits an optional argument within square brackets which may be
either |[l]|, |[c]| or |[r]|. Default is |[c]| (numerators are centered).

Numerators and denominators are made arguments to the \csbxint{TeXFrac}
macro. This allows them to be themselves fractions or anything \fexpan
dable giving numbers or fractions, but also means however that they can
not be arbitrary material, they can not contain color changing macros
for example. One of the reasons is that \csa{xintGCFrac} tries to
determine the signs of the numerators and chooses accordingly to use
$+$ or $-$.

\subsection{\csh{xintGGCFrac}}\label{xintGGCFrac}

\csa{xintGGCFrac}|{a+b/c+d/e+f/g+h/...+x/y}|\ntype{f} is a clone of
\csbxint{GCFrac}, hence again \LaTeX{} specific with package
|amsmath|.
It does not assume the coefficients to be numbers as understood by
\xintfracname. The macro can be used for displaying arbitrary content as
a continued fraction with |\cfrac|, using only plus signs though. Note
though that it will first \fexpan d its argument, which may be thus be
one of the \xintcfracname macros producing a (general) continued
fraction in inline format, see \csbxint{FtoCx} for an example. If this
expansion is not wished, it is enough to start the argument with a
space.
\begin{everbatim*}
\[\xintGGCFrac {1+q/1+q^2/1+q^3/1+q^4/1+q^5/\ddots}\]
\end{everbatim*}

\subsection{\csh{xintGCtoGCx}}\label{xintGCtoGCx}
%{\small New with release |1.05|.\par}

\csa{xintGCtoGCx}|{sepa}{sepb}{a+b/c+d/e+f/...+x/y}|\etype{nnf} returns the list
of the coefficients of the generalized continued fraction of |f|, each one
within a pair of braces, and separated with the help of |sepa| and |sepb|. Thus
%
\leftedline{|\xintGCtoGCx :;{1+2/3+4/5+6/7}| gives \xintGCtoGCx
  :;{1+2/3+4/5+6/7}}
%
The following can be used byt Plain \TeX{}+|amstex| users to obtain an
output similar as the ones produced by \csbxint{GCFrac} and
\csbxint{GGCFrac}:\par
\everb|@
$$\xintGCtoGCx {+\cfrac}{\\}{a+b/...}\endcfrac$$
$$\xintGCtoGCx {+\cfrac\xintTeXOver}{\\\xintTeXOver}{a+b/...}\endcfrac$$
|

\subsection{\csh{xintFtoC}}\label{xintFtoC}

\csa{xintFtoC}|{f}|\etype{\Ff} computes the
coefficients of the simple continued fraction of |f| and returns them as a list
(sequence) of braced items.

\begin{everbatim*}
\fdef\test{\xintFtoC{-5262046/89233}}\texttt{\meaning\test}
\end{everbatim*}

\subsection{\csh{xintFtoCs}}\label{xintFtoCs}

\csa{xintFtoCs}|{f}|\etype{\Ff} returns the comma separated list of the
coefficients of the simple continued fraction of |f|. Notice that starting with
|1.09m| a space follows each comma (mainly for usage in text mode, as in math
mode spaces are produced in the typeset output by \TeX{} itself).
\begin{everbatim*}
\[ \xintTeXsignedFrac{-5262046/89233} \to [\xintFtoCs{-5262046/89233}]\]
\end{everbatim*}

\subsection{\csh{xintFtoCx}}\label{xintFtoCx}

\csa{xintFtoCx}|{sep}{f}|\etype{n\Ff} returns the list of the
coefficients of the simple continued fraction of |f| separated with the
help of |sep|, which may be anything (and is kept unexpanded). For
example, with Plain \TeX{} and |amstex|,
%
\leftedline{|$$\xintFtoCx {+\cfrac1\\ }{-5262046/89233}\endcfrac$$|}
%
will display the continued fraction using
|\cfrac|. Each coefficient is inside a brace pair \hbox{|{ }|}, allowing
a macro to end the separator and fetch it as argument,
for example, again with Plain \TeX{} and |amstex|:
\everb|@
  \def\highlight #1{\ifnum #1>200 \textcolor{red}{#1}\else #1\fi}
  $$\xintFtoCx {+\cfrac1\\ \highlight}{104348/33215}\endcfrac$$
|

Due to the different and extremely cumbersome syntax of |\cfrac| under
\LaTeX{} it proves a bit tortuous to obtain there the same effect.
Actually, it is partly for this purpose that |1.09m| added \csbxint
{GGCFrac}. We thus use \csa{xintFtoCx} with a suitable separator, and\;
then the whole thing as argument to \csbxint{GGCFrac}:
\begin{everbatim*}
\def\highlight #1{\ifnum #1>200 \fcolorbox{blue}{white}{\boldmath\color{red}$#1$}%
    \else #1\fi}
\[\xintGGCFrac {\xintFtoCx {+1/\highlight}{208341/66317}}\]
\end{everbatim*}

\subsection{\csh{xintFtoGC}}\label{xintFtoGC}

\csa{xintFtoGC}|{f}|\etype{\Ff} does the same as \csa{xintFtoCx}|{+1/}{f}|. Its
output may thus be used in the package macros expecting such an `inline
format'.
% This continued fraction is a \emph{simple} one, not a
% \emph{generalized} one, but as it is produced in the format used for
% user input of generalized continued fractions, the macro was called
% \csa{xintFtoGC} rather than \csa{xintFtoC} for example.
%
\begin{everbatim*}
566827/208524=\xintFtoGC {566827/208524}
\end{everbatim*}

\subsection{\csh{xintFGtoC}}\label{xintFGtoC}

\csa{xintFGtoC}|{f}{g}|\etype{\Ff\Ff} computes the common initial coefficients
to
two given fractions |f| and |g|. Notice that any real number |f<x<g| or |f>x>g|
will then necessarily share with |f| and |g| these common initial coefficients
for its regular continued fraction. The coefficients are output as a sequence of
braced numbers. This list can then be manipulated via macros from
\xinttoolsname, or other macros of \xintcfracname.

\begin{everbatim*}
\fdef\test{\xintFGtoC{-5262046/89233}{-5314647/90125}}\texttt{\meaning\test}
\end{everbatim*}
\begin{everbatim*}
\fdef\test{\xintFGtoC{3.141592653}{3.141592654}}\texttt{\meaning\test}
\end{everbatim*}
\begin{everbatim*}
\fdef\test{\xintFGtoC{3.1415926535897932384}{3.1415926535897932385}}\meaning\test
\end{everbatim*}
\begin{everbatim*}
\xintRound {30}{\xintCstoF{\xintListWithSep{,}{\test}}}
\end{everbatim*}
\begin{everbatim*}
\xintRound {30}{\xintCtoF{\test}}
\end{everbatim*}
\begin{everbatim*}
\fdef\test{\xintFGtoC{1.41421356237309}{1.4142135623731}}\meaning\test
\end{everbatim*}

\subsection{\csh{xintFtoCC}}\label{xintFtoCC}

\csa{xintFtoCC}|{f}|\etype{\Ff} returns the `centered' continued fraction of
|f|, in `inline format'. %
\begin{everbatim*}
566827/208524=\xintFtoCC {566827/208524}
\end{everbatim*}
\begin{everbatim*}
\[\xintTeXFrac{566827/208524} = \xintGCFrac{\xintFtoCC{566827/208524}}\]
\end{everbatim*}

\subsection{\csh{xintCstoF}}\label{xintCstoF}

\csa{xintCstoF}|{a,b,c,d,...,z}|\etype{f} computes the fraction corresponding to
the coefficients, which may be fractions or even macros expanding to such
fractions. The final fraction may then be highly reducible.

\emph{Usage of this macro requires the user to load} \xinttoolsname.\IMPORTANT

Starting with
release |1.09m| spaces before commas are allowed and trimmed automatically
(spaces after commas were already silently handled in earlier releases).
\begin{everbatim*}
\[\xintGCFrac {-1+1/3+1/-5+1/7+1/-9+1/11+1/-13}=
  \xintTeXsignedFrac{\xintCstoF {-1,3,-5,7,-9,11,-13}}=\xintTeXsignedFrac{\xintGCtoF
    {-1+1/3+1/-5+1/7+1/-9+1/11+1/-13}}\]
\end{everbatim*}
\begin{everbatim*}
\[\xintGCFrac{{1/2}+1/{1/3}+1/{1/4}+1/{1/5}}=\xintTeXFrac{\xintCstoF {1/2,1/3,1/4,1/5}}\]
\end{everbatim*}
%
A generalized continued fraction may produce a reducible fraction
(\csa{xintCstoF} tries its best not to accumulate in a silly way superfluous
factors but will not do simplifications which would be obvious to a human, like
simplification by 3 in the result above).

\subsection{\csh{xintCtoF}}\label{xintCtoF}

\csa{xintCtoF}|{{a}{b}{c}...{z}}|\etype{f} computes the fraction corresponding
to the coefficients, which may be fractions or even macros.
\begin{everbatim*}
\xintCtoF {\xintApply {\xintiiPow 3}{\xintSeq {1}{5}}}
\end{everbatim*}
\begin{everbatim*}
\[ \xintTeXFrac{14946960/4805083}=\xintCFrac {14946960/4805083}\]
\end{everbatim*}
In the example above the power of $3$ was already pre-computed via the expansion
done by |\xintApply|, but if we try with |\xintApply { \xintiiPow 3}| where the
space will stop this expansion, we can check that |\xintCtoF| will itself
provoke the needed coefficient expansion.% ok

\subsection{\csh{xintGCtoF}}\label{xintGCtoF}

\csa{xintGCtoF}|{a+b/c+d/e+f/g+......+v/w+x/y}|\etype{f} computes the fraction
defined by the inline generalized continued fraction. Coefficients may be
fractions but must then be put within braces. They can be macros. The plus signs
are mandatory.
\begin{everbatim*}
\[\xintGCFrac {1+\xintPow{1.5}{3}/{1/7}+{-3/5}/\xintiiFac {6}} =
\xintTeXFrac{\xintGCtoF {1+\xintPow{1.5}{3}/{1/7}+{-3/5}/\xintiiFac {6}}} =
\xintTeXFrac{\xintIrr{\xintGCtoF
                  {1+\xintPow{1.5}{3}/{1/7}+{-3/5}/\xintiiFac {6}}}}\]
\end{everbatim*}

\begin{everbatim*}
\[ \xintGCFrac{{1/2}+{2/3}/{4/5}+{1/2}/{1/5}+{3/2}/{5/3}} =
   \xintTeXFrac{\xintGCtoF {{1/2}+{2/3}/{4/5}+{1/2}/{1/5}+{3/2}/{5/3}}} \]
\end{everbatim*}

The macro tries its best not to accumulate superfluous factor in the
denominators, but doesn't reduce the fraction to irreducible form before
returning it and does not do simplifications which would be obvious to a human.

\subsection{\csh{xintCstoCv}}\label{xintCstoCv}

\csa{xintCstoCv}|{a,b,c,d,...,z}|\etype{f} returns the sequence of the
corresponding convergents, each one within braces.

\emph{Usage of this macro requires the user to load} \xinttoolsname.\IMPORTANT

It is allowed to use fractions as coefficients (the computed
convergents have then no reason to be the real convergents of the final
fraction). When the coefficients are integers, the convergents are irreducible
fractions, but otherwise it is not necessarily the case.
\begin{everbatim*}
\xintListWithSep:{\xintCstoCv{1,2,3,4,5,6}}
\end{everbatim*}
\begin{everbatim*}
\xintListWithSep:{\xintCstoCv{1,1/2,1/3,1/4,1/5,1/6}}
\end{everbatim*}
\begin{everbatim*}
\[\xintListWithSep{\to}{\xintApply\xintTeXFrac{\xintCstoCv {\xintPow
        {-.3}{-5},7.3/4.57,\xintCstoF{3/4,9,-1/3}}}}\]
\end{everbatim*}

\subsection{\csh{xintCtoCv}}\label{xintCtoCv}

\csa{xintCtoCv}|{{a}{b}{c}...{z}}|\etype{f} returns the sequence of the
corresponding convergents, each one within braces.
\begin{everbatim*}
\fdef\test{\xintCtoCv {11111111111}}\texttt{\meaning\test}
\end{everbatim*}

\subsection{\csh{xintGCtoCv}}\label{xintGCtoCv}

\csa{xintGCtoCv}|{a+b/c+d/e+f/g+......+v/w+x/y}|\etype{f} returns the list of
the corresponding convergents. The coefficients may be fractions, but must then
be inside braces. Or they may be macros, too.

The convergents will in the general case be reducible. To put them into
irreducible form, one needs one more step, for example it can be done
with |\xintApply\xintIrr|.
\begin{everbatim*}
\[\xintListWithSep{,}{\xintApply\xintTeXFrac
                {\xintGCtoCv{3+{-2}/{7/2}+{3/4}/12+{-56}/3}}}\]
\[\xintListWithSep{,}{\xintApply\xintTeXFrac{\xintApply\xintIrr
                {\xintGCtoCv{3+{-2}/{7/2}+{3/4}/12+{-56}/3}}}}\]
\end{everbatim*}


\subsection{\csh{xintFtoCv}}\label{xintFtoCv}

\csa{xintFtoCv}|{f}|\etype{\Ff} returns the list of the (braced) convergents of
|f|, with no separator. To be treated with \csbxint{AssignArray} or
\csbxint{ListWithSep}.
\begin{everbatim*}
\[\xintListWithSep{\to}{\xintApply\xintTeXFrac{\xintFtoCv{5211/3748}}}\]
\end{everbatim*}

\subsection{\csh{xintFtoCCv}}\label{xintFtoCCv}

\csa{xintFtoCCv}|{f}|\etype{\Ff} returns the list of the (braced) centered
convergents of |f|, with no separator. To be treated with \csbxint{AssignArray}
or \csbxint{ListWithSep}.
\begin{everbatim*}
\[\xintListWithSep{\to}{\xintApply\xintTeXFrac{\xintFtoCCv{5211/3748}}}\]
\end{everbatim*}

\subsection{\csh{xintCntoF}}\label{xintCntoF}


\csa{xintCntoF}|{N}{\macro}|\etype{\numx f} computes the fraction |f| having
coefficients |c(j)=\macro{j}| for |j=0,1,...,N|. The |N| parameter is given to a
|\numexpr|. The values of the coefficients, as returned by |\macro| do not have
to be positive, nor integers, and it is thus not necessarily the case that the
original |c(j)| are the true coefficients of the final |f|.
\begin{everbatim*}
\def\macro #1{\the\numexpr 1+#1*#1\relax} \xintCntoF {5}{\macro}
\end{everbatim*}

This example shows that the fraction is output with a trailing number in square
brackets (representing a power of ten), this is for consistency with what do
most macros of \xintfracname, and does not have to be always this annoying |[0]|
as the coefficients may for example be numbers in scientific notation. To avoid
these trailing square brackets, for example if the coefficients are known to be integers, there is always the possibility to filter the output via
\csbxint{PRaw}, or \csbxint{Irr} (the latter is overkill in the case of integer
coefficients, as the fraction is guaranteed to be irreducible then).

\subsection{\csh{xintGCntoF}}\label{xintGCntoF}

\csa{xintGCntoF}|{N}{\macroA}{\macroB}|\etype{\numx ff} returns the fraction |f|
corresponding to the inline generalized continued fraction
|a0+b0/a1+b1/a2+....+b(N-1)/aN|, with |a(j)=\macroA{j}| and |b(j)=\macroB{j}|.
The |N| parameter is given to a |\numexpr|.
\begin{everbatim*}
\def\coeffA #1{\the\numexpr #1+4-3*((#1+2)/3)\relax }%
\def\coeffB #1{\the\numexpr \ifodd #1 -\fi 1\relax }% (-1)^n
\[\xintGCFrac{\xintGCntoGC {6}{\coeffA}{\coeffB}} =
                                \xintTeXFrac{\xintGCntoF {6}{\coeffA}{\coeffB}}\]
\end{everbatim*}
There is also \csbxint{GCntoGC} to get the `inline format' continued
fraction.

\subsection{\csh{xintCntoCs}}\label{xintCntoCs}

\csa{xintCntoCs}|{N}{\macro}|\etype{\numx f} produces the comma separated list
of the corresponding coefficients, from |n=0| to |n=N|. The |N| is given to a
|\numexpr|. %
\begin{everbatim*}
\xintCntoCs {5}{\macro}
\end{everbatim*}
\begin{everbatim*}
\[ \xintTeXFrac{\xintCntoF{5}{\macro}}=\xintCFrac{\xintCntoF {5}{\macro}}\]
\end{everbatim*}

\subsection{\csh{xintCntoGC}}\label{xintCntoGC}

%
\csa{xintCntoGC}|{N}{\macro}|\etype{\numx f} evaluates the |c(j)=\macro{j}| from
|j=0| to |j=N| and returns a continued fraction written in inline format:
|{c(0)}+1/{c(1)}+1/...+1/{c(N)}|. The parameter |N| is given to a |\numexpr|.
The coefficients, after expansion, are, as shown, being enclosed in an added
pair of braces, they may thus be fractions.
\begin{everbatim*}
\def\macro #1{\the\numexpr\ifodd#1 -1-#1\else1+#1\fi\relax/\the\numexpr 1+#1*#1\relax}
\fdef\x{\xintCntoGC {5}{\macro}}\meaning\x
\[\xintGCFrac{\xintCntoGC {5}{\macro}}\]
\end{everbatim*}

\subsection{\csh{xintGCntoGC}}\label{xintGCntoGC}

\csa{xintGCntoGC}|{N}{\macroA}{\macroB}|\etype{\numx ff} evaluates the
coefficients and then returns the corresponding
|{a0}+{b0}/{a1}+{b1}/{a2}+...+{b(N-1)}/{aN}| inline generalized fraction. |N| is
givent to a |\numexpr|. The coefficients are enclosed into pairs
of braces, and may thus be fractions, the fraction slash will not be
confused in further processing by the continued fraction slashes.
%
\begin{everbatim*}
\def\an #1{\the\numexpr  #1*#1*#1+1\relax}%
\def\bn #1{\the\numexpr \ifodd#1 -\fi 1*(#1+1)\relax}%
$\xintGCntoGC {5}{\an}{\bn}=\xintGCFrac {\xintGCntoGC {5}{\an}{\bn}} =
\displaystyle\xintTeXFrac {\xintGCntoF {5}{\an}{\bn}}$\par
\end{everbatim*}

\subsection{\csh{xintCstoGC}}\label{xintCstoGC}

\csa{xintCstoGC}|{a,b,..,z}|\etype{f} transforms a comma separated list (or
something expanding to such a list) into an `inline format' continued fraction
|{a}+1/{b}+1/...+1/{z}|. The coefficients are just copied and put within braces,
without expansion. The output can then be used in \csbxint{GCFrac} for example.
\begin{everbatim*}
\[\xintGCFrac {\xintCstoGC {-1,1/2,-1/3,1/4,-1/5}}=\xintTeXsignedFrac{\xintCstoF {-1,1/2,-1/3,1/4,-1/5}}\]
\end{everbatim*}
\subsection{\csh{xintiCstoF}, \csh{xintiGCtoF}, \csh{xintiCstoCv}, \csh{xintiGCtoCv}}\label{xintiCstoF}
\label{xintiGCtoF}
\label{xintiCstoCv}
\label{xintiGCtoCv}

Essentially\etype{f} the same as the corresponding macros without the
`i', but for integer-only input. Infinitesimally faster, mainly for
internal use by the package.

\subsection{\csh{xintGCtoGC}}\label{xintGCtoGC}

\csa{xintGCtoGC}|{a+b/c+d/e+f/g+......+v/w+x/y}|\etype{f} expands (with the
usual meaning) each one of the coefficients and returns an inline continued
fraction of the same type, each expanded coefficient being enclosed within
braces.
%
\begin{everbatim*}
\fdef\x {\xintGCtoGC {1+\xintPow{1.5}{3}/{1/7}+{-3/5}/%
                        \xintiiFac {6}+\xintCstoF {2,-7,-5}/16}} \meaning\x
\end{everbatim*}

To be honest I have forgotten for which purpose I wrote this macro in the first
place.

\subsection{Euler's number \texorpdfstring{$e$}{e}}\label{ssec:e-convergents}

Let us explore
the convergents of Euler's number $e$.
\smallskip The volume of computation is kept minimal by the following steps:
\begin{itemize}
\item a comma separated list of the first 36 coefficients is produced by
  \csbxint{CntoCs},
\item this is then given to \csbxint{iCstoCv} which produces the list of the
  convergents (there is also \csbxint{CstoCv}, but our
  coefficients being integers we used the infinitesimally
  faster \csbxint{iCstoCv}),
\item then the whole list was converted into a sequence of one-line paragraphs,
  each convergent becomes the argument to a  macro printing it
  together with its decimal expansion with 30 digits after the decimal point.
\item A count register |\cnta| was used to give a line count serving as a visual
  aid: we could also have done that in an expandable way, but well, let's relax
  from time to time\dots
\end{itemize}

\begin{everbatim*}
\def\cn #1{\the\numexpr\ifcase \numexpr #1+3-3*((#1+2)/3)\relax
                           1\or1\or2*(#1/3)\fi\relax }
% produces the pattern 1,1,2,1,1,4,1,1,6,1,1,8,... which are the
% coefficients of the simple continued fraction of e-1.
\cnta 0
\def\mymacro #1{\advance\cnta by 1
                \noindent
                \hbox to 3em {\hfil\small\dtt{\the\cnta.} }%
                $\xintTrunc {30}{\xintAdd {1[0]}{#1}}\dots=
                 \xintTeXFrac{\xintAdd {1[0]}{#1}}$}%
\xintListWithSep{\vtop to 6pt{}\vbox to 12pt{}\par}
    {\xintApply\mymacro{\xintiCstoCv{\xintCntoCs {35}{\cn}}}}
\end{everbatim*}


\smallskip

% The actual computation of the list of all 36 convergents accounts for
% only 8\% of the total time (total time equal to about 5 hundredths of a second
% in my testing, on my laptop): another 80\% is occupied with the computation of
% the truncated decimal expansions (and the addition of 1 to everything as the
% formula gives the continued fraction of $e-1$).

One can with no problem compute
much bigger convergents. Let's get the 200th convergent. It turns out to
have the same first 268 digits after the decimal point as $e-1$. Higher
convergents get more and more digits in proportion to their index: the 500th
convergent already gets 799 digits correct! To allow speedy compilation of the
source of this document when the need arises, I limit here to the 200th
convergent.
%  (getting the 500th took about 1.2s on my laptop last time I tried,
% and the 200th convergent is obtained ten times faster).
\begin{everbatim*}
\fdef\z {\xintCntoF {199}{\cn}}%
\begingroup\parindent 0pt \leftskip 2.5cm
\indent\llap {Numerator = }\printnumber{\xintNumerator\z}\par
\indent\llap {Denominator = }\printnumber{\xintDenominator\z}\par
\indent\llap {Expansion = }\printnumber{\xintTrunc{268}\z}\dots\par\endgroup
\end{everbatim*}


One can also use a centered continued fraction: we get more digits but there are
also more computations as the numerators may be either
$1$ or $-1$.

\clearpage
\let\xintcfracnameUp\undefined
\csname xinttoolsnameUp\endcsname
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}
\section{Macros of the \xinttoolsname package}

\RaisedLabel{sec:tools}

These utilities used to be provided within the \xintname package; since |1.09g|
(|2013/11/22|) they have been moved to an independently usable package
\xinttoolsname, which has none of the \xintname facilities regarding big
numbers. Whenever relevant release |1.09h| has made the macros |\long| so they
accept |\par| tokens on input.

The completely expandable utilities (up to \csbxint{iloop}) are documented
first, then the non expandable utilities.

\autoref{sec:examples}
gives additional (also dated) examples of use of macros of this package.

\localtableofcontents

\subsection{\csh{xintRevWithBraces}}\label{xintRevWithBraces}

%{\small New in release |1.06|.\par}

\edef\X{\xintRevWithBraces{12345}}
\edef\y{\xintRevWithBraces\X}
\expandafter\def\expandafter\w\expandafter
     {\romannumeral0\xintrevwithbraces{{\A}{\B}{\C}{\D}{\E}}}

%
\csa{xintRevWithBraces}\marg{list}\etype{f} first does the \fexpan sion of its
argument then it reverses the order of the tokens, or braced material, it
encounters, maintaining existing braces and adding a brace pair around each
naked token encountered. Space tokens (in-between top level braces or naked
tokens) are gobbled. This macro is mainly thought out for use on a \meta{list}
of such braced material; with such a list as argument the \fexpan sion will only
hit against the first opening brace, hence do nothing, and the braced stuff may
thus be macros one does not want to expand.
%
\leftedline{|\edef\x{\xintRevWithBraces{12345}}|}
%
\leftedline{|\meaning\x:|\dtt{\meaning\X}}
%
\leftedline{|\edef\y{\xintRevWithBraces\x}|}
%
\leftedline{|\meaning\y:|\dtt{\meaning\y}}
%
The examples above could be defined with |\edef|'s because the braced material
did not contain macros. Alternatively:
%
\leftedline{|\expandafter\def\expandafter\w\expandafter|}
%
\leftedline{|{\romannumeral0\xintrevwithbraces{{\A}{\B}{\C}{\D}{\E}}}|}
%
\leftedline{|\meaning\w:|\dtt{\meaning\w}}
%
The macro \csa{xintReverseWithBracesNoExpand}\etype{n} does the same job
without the initial expansion of its argument.


\subsection{\csh{xintZapFirstSpaces}, \csh{xintZapLastSpaces}, \csh{xintZapSpaces}, \csh{xintZapSpacesB}}
\label{xintZapFirstSpaces}
\label{xintZapLastSpaces}
\label{xintZapSpaces}
\label{xintZapSpacesB}
%{\small New with release |1.09f|.\par}

\csa{xintZapFirstSpaces}\marg{stuff}\etype{n} does not do \emph{any} expansion
of its argument, nor brace removal of any sort, nor does it alter \meta{stuff}
in anyway apart from stripping away all \emph{leading} spaces.

This macro will be mostly of interest to programmers who will know what I will
now be talking about. \emph{The essential points, naturally, are the complete
  expandability and the fact that no brace removal nor any other alteration is
  done to the input.}

\TeX's input scanner already converts consecutive blanks into single space
tokens, but |\xintZapFirstSpaces| handles successfully also inputs with
consecutive multiple space tokens.
However, it is assumed that \meta{stuff} does not contain (except inside braced
sub-material) space tokens of character code distinct from $32$.

It expands in two steps, and if the goal is to apply it to the
expansion text of |\x| to define |\y|, then one can do:
|\odef\y{\romannumeral0\expandafter\xintzapfirstspaces\expandafter{\x}}|
(one can also define a wrapper macro to |\xintZapFirstSpaces| in order to
expand once the argument first, but \xinttoolsname not being a programming
layer, it provides no «Generate Variants» facilities). 

Other use case: inside a macro which received a parameter |#1|, one can do
|\oodef\x{\xintZapFirstSpaces {#1}}|, or, if |#1|, after leading spaces have
been stripped can accept |\edef| expansion, one can do
|\edef\x{\xintZapFirstSpaces{#1}}|.

\begingroup
\def\x {  \a {  \X } {  \b  \Y }  }
%
\leftedline{|\xintZapFirstSpaces {  \a {  \X } {  \b  \Y }  }->|%
\dtt{\color{magenta}{}\expandafter\detokenize\expandafter
{\romannumeral0\expandafter\xintzapfirstspaces\expandafter{\x}}}+++}
\endgroup

\medskip

\noindent\csbxint{ZapLastSpaces}\marg{stuff}\etype{n}  does not do \emph{any} expansion of
its argument, nor brace removal of any sort, nor does it alter \meta{stuff} in
anyway apart from stripping away all \emph{ending} spaces. The same remarks as
for \csbxint{ZapFirstSpaces} apply.

\begingroup
\def\x {  \a {  \X } {  \b  \Y }  }
%
\leftedline{|\xintZapLastSpaces {  \a {  \X } {  \b  \Y }  }->|%
\dtt{\color{magenta}{}\expandafter\detokenize\expandafter
{\romannumeral0\expandafter\xintzaplastspaces\expandafter{\x}}}+++}
\endgroup

\medskip

\noindent\csbxint{ZapSpaces}\marg{stuff}\etype{n}  does not do \emph{any}
expansion of its
argument, nor brace removal of any sort, nor does it alter \meta{stuff} in
anyway apart from stripping away all \emph{leading} and all \emph{ending}
spaces. The same remarks as for \csbxint{ZapFirstSpaces} apply.

\begingroup
\def\x {  \a {  \X } {  \b  \Y }  }
%
\leftedline{|\xintZapSpaces {  \a {  \X } {  \b  \Y }  }->|%
\dtt{\color{magenta}{}\expandafter\detokenize\expandafter
{\romannumeral0\expandafter\xintzapspaces\expandafter{\x}}}+++}
\endgroup

\medskip

\noindent\csbxint{ZapSpacesB}\marg{stuff}\etype{n}  does not do \emph{any}
expansion of
its argument, nor does it alter \meta{stuff} in anyway apart from stripping away
all leading and all ending spaces and possibly removing one level of braces if
\meta{stuff} had the shape |<spaces>{braced}<spaces>|. The same remarks as for
\csbxint{ZapFirstSpaces} apply.

\begingroup
\def\x {  \a {  \X } {  \b  \Y }  }
%
\leftedline{|\xintZapSpacesB {  \a {  \X } {  \b  \Y }  }->|%
\dtt{\color{magenta}{}\expandafter\detokenize\expandafter
{\romannumeral0\expandafter\xintzapspacesb\expandafter{\x}}}+++}
\def\x {  { \a {  \X } {  \b  \Y } }  }
%
\leftedline{|\xintZapSpacesB {  { \a {  \X } {  \b  \Y } }  }->|%
\dtt{\color{magenta}{}\expandafter\detokenize\expandafter
{\romannumeral0\expandafter\xintzapspacesb\expandafter{\x}}}+++}
\endgroup
 The spaces here at the start and end of the output come from the braced
 material, and are not removed (one would need a second application for that;
 recall though that the \xintname zapping macros do not expand their argument).

\subsection{\csh{xintCSVtoList}}
\label{xintCSVtoList}
\label{xintCSVtoListNoExpand}


\csa{xintCSVtoList}|{a,b,c...,z}|\etype{f}  returns |{a}{b}{c}...{z}|. A
\emph{list} is by
convention in this manual simply a succession of tokens, where each braced thing
will count as one item (``items'' are defined according to the rules of \TeX{}
for fetching undelimited parameters of a macro, which are exactly the same rules
as for \LaTeX{} and macro arguments [they are the same things]). The word
`list' in `comma separated list of items' has its usual linguistic meaning,
and then an ``item'' is what is delimited by commas.

So \csa{xintCSVtoList} takes on input a `comma separated list of items' and
converts it into a `\TeX{} list of braced items'. The argument to
|\xintCSVtoList| may be a macro: it will first be
\hyperref[ssec:expansions]{\fexpan ded}. Hence the item before the first comma,
if it is itself a macro, will be expanded which may or may not be a good thing.
A space inserted at the start of the first item serves to stop that expansion
(and disappears). The macro \csbxint{CSVtoListNoExpand}\etype{n} does the same
job without
the initial expansion of the list argument.

Apart from that no expansion of the items is done and the list items may thus be
completely arbitrary (and even contain perilous stuff such as unmatched |\if|
and |\fi| tokens).

Contiguous spaces and tab characters, are collapsed by \TeX{}
into single spaces. All such spaces around commas%
%
\footnote{and multiple space tokens are not a problem; but those at the
  top level (not hidden inside braces) \emph{must} be of character code
  |32|.}
%
\fbox{are removed}, as well as
the spaces at the start and the spaces at the end of the list.%
%
\footnote{let us recall that this is all done completely expandably...
  There is absolutely no alteration of any sort of the item apart from
  the stripping of initial and final space tokens (of character code
  |32|) and brace removal if and only if the item apart from intial and
  final spaces (or more generally multiple |char 32| space tokens) is
  braced.}
%
The items may contain explicit |\par|'s or
empty lines (converted by the \TeX{} input parsing into |\par| tokens).

\begingroup

\edef\X{\xintCSVtoList { 1 ,{ 2 , 3 , 4 , 5 }, a , {b,T} U , { c , d } , { {x ,
        y} } }}

%
\leftedline{|\xintCSVtoList { 1 ,{ 2 , 3 , 4 , 5 }, a , {b,T} U , { c , d } ,
    { {x , y} } }|}
%
\leftedline{|->|%
{\makeatletter\dtt{\expandafter\strip@prefix\meaning\X}}}

One sees on this example how braces protect commas from
sub-lists to be perceived as delimiters of the top list. Braces around an entire
item are removed, even when surrounded by spaces before and/or after. Braces for
sub-parts of an item are not removed.

We observe also that there is a slight difference regarding the brace stripping
of an item: if the braces were not surrounded by spaces, also the initial and
final (but no other) spaces of the \emph{enclosed} material are removed. This is
the only situation where spaces protected by braces are nevertheless removed.

From the rules above: for an empty argument (only spaces, no braces, no comma)
the output is
\dtt{\expandafter\detokenize\expandafter{\romannumeral0\xintcsvtolist { }}}
(a list with one empty item),
for ``|<opt. spaces>{}<opt.
spaces>|'' the output is
\dtt{\expandafter\detokenize\expandafter
   {\romannumeral0\xintcsvtolist { {} }}}
(again a list with one empty item, the braces were removed),
for ``|{ }|'' the output is
\dtt{\expandafter\detokenize\expandafter
 {\romannumeral0\xintcsvtolist {{ }}}}
(again a list with one empty item, the braces were removed and then
the inner space was removed),
for ``| { }|'' the output is
\dtt{\expandafter\detokenize\expandafter
{\romannumeral0\xintcsvtolist { { }}}} (again a list with one empty item, the initial space served only to stop the expansion, so this was like ``|{ }|'' as input, the braces were removed and the inner space was stripped),
for ``\texttt{\ \{\ \ \}\ }'' the output is
\dtt{\expandafter\detokenize\expandafter
{\romannumeral0\xintcsvtolist { {  } }}} (this time the ending space of the first
item meant that after brace removal the inner spaces were kept; recall though
that \TeX{} collapses on input consecutive blanks into one space token),
for ``|,|'' the output consists of two consecutive
empty items
\dtt{\expandafter\detokenize\expandafter{\romannumeral0\xintcsvtolist
    {,}}}. Recall that on output everything is braced, a |{}| is an ``empty''
item.
%
Most of the above is mainly irrelevant for every day use, apart perhaps from the
fact to be noted that an empty input does not give an empty output but a
one-empty-item list (it is as if an ending comma was always added at the end of
the input).

\def\y { \a,\b,\c,\d,\e}
\expandafter\def\expandafter\Y\expandafter{\romannumeral0\xintcsvtolist{\y}}
\def\t {{\if},\ifnum,\ifx,\ifdim,\ifcat,\ifmmode}
\expandafter\def\expandafter\T\expandafter{\romannumeral0\xintcsvtolist{\t}}

%
\leftedline{|\def\y{ \a,\b,\c,\d,\e} \xintCSVtoList\y->|%
  {\makeatletter\dtt{\expandafter\strip@prefix\meaning\Y}}}
%
\leftedline{|\def\t {{\if},\ifnum,\ifx,\ifdim,\ifcat,\ifmmode}|}
%
\leftedline
{|\xintCSVtoList\t->|\makeatletter\dtt{\expandafter\strip@prefix\meaning\T}}
%
The results above were automatically displayed using \TeX's primitive
\csa{meaning}, which adds a space after each control sequence name. These spaces
are not in the actual braced items of the produced lists. The first items |\a|
and |\if| were either preceded by a space or braced to prevent expansion. The
macro \csa{xintCSVtoListNoExpand} would have done the same job without the
initial expansion of the list argument, hence no need for such protection but if
|\y| is defined as |\def\y{\a,\b,\c,\d,\e}| we then must do:
%
\leftedline{|\expandafter\xintCSVtoListNoExpand\expandafter {\y}|} Else, we
may have direct use: %
%
\leftedline{|\xintCSVtoListNoExpand
 {\if,\ifnum,\ifx,\ifdim,\ifcat,\ifmmode}|}
%
\leftedline{|->|\dtt{\expandafter\detokenize\expandafter
    {\romannumeral0\xintcsvtolistnoexpand
      {\if,\ifnum,\ifx,\ifdim,\ifcat,\ifmmode}}}}
%
Again these spaces are an artefact from the use in the source of the document of
\csa{meaning} (or rather here, \csa{detokenize}) to display the result of using
\csa{xintCSVtoListNoExpand} (which is done for real in this document
source).

For the similar conversion from comma separated list to braced items list, but
without removal of spaces around the commas, there is
\csa{xintCSVtoListNonStripped}\etype{f} and
\csa{xintCSVtoListNonStrippedNoExpand}\etype{n}.

\endgroup

\subsection{\csh{xintNthElt}}\label{xintNthElt}


\def\macro #1{\the\numexpr 9-#1\relax}

\csa{xintNthElt\x}\marg{list}\etype{\numx f} gets (expandably) the |x|th
item of the \meta{list}. A braced item will lose one level of brace
pairs. The token list is first \fexpan ded.

Items are counted starting at one.

\leftedline{|\xintNthElt {3}{{agh}\u{zzz}\v{Z}}| is
    \texttt{\xintNthElt {3}{{agh}\u{zzz}\v{Z}}}}
%
\leftedline{|\xintNthElt {3}{{agh}\u{{zzz}}\v{Z}}| is
    \texttt{\expandafter\expandafter\expandafter
      \detokenize\expandafter\expandafter\expandafter {\xintNthElt
        {3}{{agh}\u{{zzz}}\v{Z}}}}}
%
\leftedline{|\xintNthElt {2}{{agh}\u{{zzz}}\v{Z}}| is
    \texttt{\expandafter\expandafter\expandafter
      \detokenize\expandafter\expandafter\expandafter {\xintNthElt
        {2}{{agh}\u{{zzz}}\v{Z}}}}}
%
\leftedline{|\xintNthElt {37}{\xintiiFac {100}}|\dtt{=\xintNthElt
      {37}{\xintiiFac {100}}} is the thirty-seventh digit of $100!$.}
%
\leftedline{|\xintNthElt {10}{\xintFtoCv
      {566827/208524}}|\dtt{=\xintNthElt {10}{\xintFtoCv
        {566827/208524}}}}
\leftedline{is the tenth convergent of $566827/208524$ (uses \xintcfracname
  package).}
%
\leftedline{|\xintNthElt {7}{\xintCSVtoList {1,2,3,4,5,6,7,8,9}}|%
    \dtt{=\xintNthElt {7}{\xintCSVtoList {1,2,3,4,5,6,7,8,9}}}}
%
\leftedline{|\xintNthElt {0}{\xintCSVtoList {1,2,3,4,5,6,7,8,9}}|%
    \dtt{=\xintNthElt {0}{\xintCSVtoList {1,2,3,4,5,6,7,8,9}}}}
%
\leftedline{|\xintNthElt {-3}{\xintCSVtoList {1,2,3,4,5,6,7,8,9}}|%
    \dtt{=\xintNthElt {-3}{\xintCSVtoList {1,2,3,4,5,6,7,8,9}}}}

If |x=0|,
the macro returns the \emph{length} of the expanded list: this is not equivalent
to \csbxint{Length} which does no pre-expansion. And it is different from
\csbxint{Len} which is to be used only on integers or fractions.

If |x<0|, the macro returns the \verb+|x|+th element from the end of the list.
Thus for example |x=-1| will fetch the last item of the list.
%
\leftedline {|\xintNthElt {-5}{{{agh}}\u{zzz}\v{Z}}| is
  \texttt{\expandafter\expandafter\expandafter \detokenize
  \expandafter\expandafter\expandafter{\xintNthElt {-5}{{{agh}}\u{zzz}\v{Z}}}}}

The macro \csa{xintNthEltNoExpand}\etype{\numx n} does the same job but without
first expanding the list argument: |\xintNthEltNoExpand {-4}{\u\v\w T\x\y\z}| is
\xintNthEltNoExpand {-4}{\a\b\c\u\v\w T\x\y\z}.

If |x| is strictly larger (in absolute value) than the length of the list
then |\xintNthElt| produces empty contents.

\subsection{\csh{xintNthOnePy}}
\label{xintNthOnePy}

\csa{xintNthOnePy\x}\marg{list}\etype{\numx f} gets (expandably) the |x|th
item of the \meta{list}, adding a brace pair if there wasn't one.\NewWith{1.4}

Attention, items are counted starting at zero. For negative index, behaves
as \csbxint{NthElt}.

If the index is out of range, the empty output is returned. If the input list
was empty (had no items) the empty output is returned.

\subsection{\csh{xintKeep}}\label{xintKeep}

\csa{xintKeep\x}\marg{list}\etype{\numx f} expands the token list argument |L|
and produces a new list, depending on the value of |x|:
\begin{itemize}[nosep]
\item if |x>0|, the new list contains the first |x| items from |L| (counting
  starts at one.) \emph{Each
    such item will be output within a brace pair.} Use \csbxint{KeepUnbraced} if
  this is not desired. This means that if the list item was braced to start
  with, there is no modification, but if it was a token without braces,
  then it acquires them.
\item if |x>=length(L)|, the new list is the old one with all its items now
  braced.
\item if |x=0| the empty list is returned.
\item if |x<0| the last \verb+|x|+ elements compose the output in the same
  order as in the initial list; as the macro proceeds by removing head items
  the kept items end up in output as they were in input: no added braces.
\item if |x<=-length(L)| the output is identical with the input.
\end{itemize}

\csa{xintKeepNoExpand} does the same without first \fexpan ding its list
argument.
%
\begin{everbatim*}
\fdef\test {\xintKeep {17}{\xintKeep {-69}{\xintSeq {1}{100}}}}\meaning\test\par
\noindent\fdef\test {\xintKeep {7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintKeep {-7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintKeep {7}{123456789}}\meaning\test\par
\noindent\fdef\test {\xintKeep {-7}{123456789}}\meaning\test\par
\end{everbatim*}


\subsection{\csh{xintKeepUnbraced}}\label{xintKeepUnbraced}

Same as \csbxint{Keep} but no brace pairs are added around the kept items from
the head of the list in the case |x>0|: each such item will lose one level of
braces. Thus, to remove braces from all items of the list, one can use
\csbxint{KeepUnbraced} with its first argument larger than the length of the
list; the same is obtained from \csbxint{ListWithSep}|{}|\marg{list}. But the
new list will then have generally many more items than the original ones,
corresponding to the unbraced original items.

For |x<0| the macro is no different from \csbxint{Keep}. Hence the name is a
bit misleading because brace removal will happen only if |x>0|.

\csa{xintKeepUnbracedNoExpand} does the same without first \fexpan ding
its list argument.
%
\begin{everbatim*}
\fdef\test {\xintKeepUnbraced {10}{\xintSeq {1}{100}}}\meaning\test\par
\noindent\fdef\test {\xintKeepUnbraced {7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintKeepUnbraced {-7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintKeepUnbraced {7}{123456789}}\meaning\test\par
\noindent\fdef\test {\xintKeepUnbraced {-7}{123456789}}\meaning\test\par
\end{everbatim*}

\subsection{\csh{xintTrim}}\label{xintTrim}

\csa{xintTrim\x}\marg{list}\etype{\numx f} expands the list argument and
gobbles its first |x| elements.
\begin{itemize}[nosep]
\item if |x>0|, the first |x| items from |L| are gobbled. The remaining items
  are not modified.
\item if |x>=length(L)|, the returned list is empty.
\item if |x=0| the original list is returned (with no added braces.)
\item if |x<0| the last \verb+|x|+ items of the list are removed. \emph{The
    head items end up braced in the output.} Use \csbxint{TrimUnbraced} if
  this is not desired.
\item if |x<=-length(L)| the output is empty.
\end{itemize}

\csa{xintTrimNoExpand} does the same without first \fexpan ding its list
argument.
\begin{everbatim*}
\fdef\test {\xintTrim {17}{\xintTrim {-69}{\xintSeq {1}{100}}}}\meaning\test\par
\noindent\fdef\test {\xintTrim {7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintTrim {-7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintTrim {7}{123456789}}\meaning\test\par
\noindent\fdef\test {\xintTrim {-7}{123456789}}\meaning\test\par
\end{everbatim*}

\subsection{\csh{xintTrimUnbraced}}\label{xintTrimUnbraced}

Same as \csbxint{Trim} but in case of a negative |x| (cutting items from
the tail), the kept items from the head are not enclosed in brace pairs. They
will lose one level of braces. The name is a bit misleading
because when |x>0| there is no brace-stripping done on the kept items, because
the macro works simply by gobbling the head ones.

\csa{xintTrimUnbracedNoExpand} does the same without first \fexpan ding its list
argument.

\begin{everbatim*}
\fdef\test {\xintTrimUnbraced {-90}{\xintSeq {1}{100}}}\meaning\test\par
\noindent\fdef\test {\xintTrimUnbraced {7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintTrimUnbraced {-7}{{1}{2}{3}{4}{5}{6}{7}{8}{9}}}\meaning\test\par
\noindent\fdef\test {\xintTrimUnbraced {7}{123456789}}\meaning\test\par
\noindent\fdef\test {\xintTrimUnbraced {-7}{123456789}}\meaning\test\par
\end{everbatim*}

\subsection{\csh{xintListWithSep}}\label{xintListWithSep}


\def\macro #1{\the\numexpr 9-#1\relax}

\csa{xintListWithSep}\marg{sep}\marg{list}\etype{nf} inserts the separator
\meta{sep} in-between all items of the given list of braced items (or
individual tokens). The items are fetched as does \TeX\ with undelimited macro
arguments, thus they end up unbraced in output. If the \meta{list} is only one
(or multiple) space tokens, the output is empty.

The list argument \meta{list} gets \fexpan ded first (thus if it is a macro
whose contents are braced items, the first opening brace stops the expansion,
and it is as if the macro had been expanded once.) The separator \meta{sep} is
not pre-expanded, it ends up as is in the output (if the \meta{list} contained
at least two items.)

The variant \csa{xintListWithSepNoExpand}\etype{nn} does the same
job without the initial expansion of the \meta{list} argument.
\begin{everbatim*}
\edef\foo{\xintListWithSep{, }{123456789{10}{11}{12}}}\meaning\foo\newline
\edef\foo{\xintListWithSep{:}{\xintiiFac{20}}}\meaning\foo\newline
\oodef\FOO{\xintListWithSepNoExpand{\FOO}{\bat\baz\biz\buz}}\meaning\FOO\newline
% a braced item or a space stops the f-expansion:
\oodef\foo{\xintListWithSep{\FOO}{{\bat}\baz\biz\buz}}\meaning\foo\newline
\oodef\foo{\xintListWithSep{\FOO}{ \bat\baz\biz\buz}}\meaning\foo\par
\end{everbatim*}

\subsection{\csh{xintApply}}\label{xintApply}


\def\macro #1{\the\numexpr 9-#1\relax}

\csa{xintApply}|{\macro}|\marg{list}\etype{ff} expandably applies the one
parameter macro |\macro| to each item in the \meta{list} given as second
argument and returns a new list with these outputs: each item is given one after
the other as parameter to |\macro| which is expanded at that time (as usual,
\emph{i.e.\@} fully for what comes first), the results are braced and output
together as a succession of braced items (if |\macro| is defined to start with a
space, the space will be gobbled and the |\macro| will not be expanded; it is
allowed to have its own arguments, the list items serve as last arguments to
|\macro|). Hence |\xintApply{\macro}{{1}{2}{3}}| returns
|{\macro{1}}{\macro{2}}{\macro{3}}| where all instances of |\macro| have been
already \fexpan ded.

Being expandable, |\xintApply| is useful for example inside alignments where
implicit groups make standard loops constructs usually fail. In such situation
it is often not wished that the new list elements be braced, see
\csbxint{ApplyUnbraced}. The |\macro| does not have to be expandable:
|\xintApply| will try to expand it, the expansion may remain partial.

The \meta{list} may
itself be some macro expanding (in the previously described way) to the list of
tokens to which the macro |\macro| will be applied. For example, if the
\meta{list} expands to some positive number, then each digit will be replaced by
the result of applying |\macro| on it. %
%
\leftedline{|\def\macro #1{\the\numexpr
    9-#1\relax}|} %
%
\leftedline{|\xintApply\macro{\xintiiFac
 {20}}|\dtt{=\xintApply\macro{\xintiiFac {20}}}}

The macro \csa{xintApplyNoExpand}\etype{fn} does the same job without the first
initial expansion which gave the \meta{list} of braced tokens to which |\macro|
is applied.

\subsection{\csh{xintApplyUnbraced}}\label{xintApplyUnbraced}


\csa{xintApplyUnbraced}|{\macro}|\marg{list}\etype{ff} is like \csbxint{Apply}.
The difference is that after having expanded its list argument, and applied
|\macro| in turn to each item from the list, it reassembles the outputs without
enclosing them in braces. The net effect is the same as doing
%
\leftedline{|\xintListWithSep {}{\xintApply {\macro}|\marg{list}|}|} This is
useful for preparing a macro which will itself define some other macros or make
assignments, as the scope will not be limited by brace pairs.
%
\begin{everbatim*}
\def\macro #1{\expandafter\def\csname myself#1\endcsname {#1}}
\xintApplyUnbraced\macro{{elta}{eltb}{eltc}}
\begin{enumerate}[nosep,label=(\arabic{*})]
\item \meaning\myselfelta
\item \meaning\myselfeltb
\item \meaning\myselfeltc
\end{enumerate}
\end{everbatim*}

%
The macro \csa{xintApplyUnbracedNoExpand}\etype{fn} does the same job without
the first initial expansion which gave the \meta{list} of braced tokens to which
|\macro| is applied.

\subsection{\csh{xintSeq}}\label{xintSeq}

\csa{xintSeq}|[d]{x}{y}|\etype{{{\upshape[\numx]}}\numx\numx} generates
expandably |{x}{x+d}...| up to and possibly including |{y}| if |d>0| or down
to and including |{y}| if |d<0|. Naturally |{y}| is omitted if |y-x| is not a
multiple of |d|. If |d=0| the macro returns |{x}|. If |y-x| and |d| have
opposite signs, the macro returns nothing. If the optional argument |d| is
omitted it is taken to be the sign of |y-x|. Hence |\xintSeq {1}{0}| is not
empty but |{1}{0}|. But |\xintSeq [1]{1}{0}| is empty.


The arguments |x| and |y| are expanded inside a |\numexpr| so they may be
count registers or a \LaTeX{} |\value{countername}|, or arithmetic with such
things.

%
\begin{everbatim*}
\xintListWithSep{,\hskip2pt plus 1pt minus 1pt }{\xintSeq {12}{-25}}
\end{everbatim*}
%
\begin{everbatim*}
\xintiiSum{\xintSeq [3]{1}{1000}}
\end{everbatim*}

When the macro is used without the optional argument |d|, it can only generate
up to about $5000$ numbers\IMPORTANT, the precise value depends upon some
\TeX{} memory parameter (input save stack).

With the optional argument |d| the macro proceeds differently (but less
efficiently) and does not stress the input save stack.



\subsection{\csh{xintloop}, \csh{xintbreakloop}, \csh{xintbreakloopanddo}, \csh{xintloopskiptonext}}
\label{xintloop}
\label{xintbreakloop}
\label{xintbreakloopanddo}
\label{xintloopskiptonext}

|\xintloop|\meta{stuff}|\if<test>...\repeat|\xtype{} is an expandable loop
compatible with nesting. However to break out of the loop one almost always need
some un-expandable step. The cousin \csbxint{iloop} is \csbxint{loop} with an
embedded expandable mechanism allowing to exit from the loop. The iterated
macros may contain |\par| tokens or empty lines.

If a sub-loop is to be used all the material from the start of the main loop and
up to the end of the entire subloop should be braced; these braces will be
removed and do not create a group. The simplest to allow the nesting of one or
more sub-loops is to brace everything between \csa{xintloop} and \csa{repeat},
being careful not to leave a space between the closing brace and |\repeat|.

As this loop and \csbxint{iloop} will primarily be of interest to experienced
\TeX{} macro programmers, my description will assume that the user is
knowledgeable enough. Some examples in this document will be perhaps more
illustrative than my attemps at explanation of use.

One can abort the loop with \csbxint{breakloop}; this should not be used inside
the final test, and one should expand the |\fi| from the corresponding test
before. One has also \csbxint{breakloopanddo} whose first argument will be
inserted in the token stream after the loop; one may need a macro such as
|\xint_afterfi| to move the whole thing after the |\fi|, as a simple
|\expandafter| will not be enough.

One will usually employ some count registers to manage the exit test from the
loop; this breaks expandability, see \csbxint{iloop} for an expandable integer
indexed loop. Use in alignments will be complicated by the fact that cells
create groups, and also from the fact that any encountered unexpandable material
will cause the \TeX{} input scanner to insert |\endtemplate| on each encountered
|&| or |\cr|; thus |\xintbreakloop| may not work as expected, but the situation
can be resolved via |\xint_firstofone{&}| or use of |\TAB| with |\def\TAB{&}|.
It is thus simpler for alignments to use rather than \csbxint{loop} either the
expandable \csbxint{ApplyUnbraced} or the non-expandable but alignment
compatible \csbxint{ApplyInline}, \csbxint{For} or \csbxint{For*}.

As an example, let us suppose we have two macros |\A|\marg{i}\marg{j} and
|\B|\marg{i}\marg{j} behaving like (small) integer valued matrix entries, and we
want to define a macro |\C|\marg{i}\marg{j} giving the matrix product (|i| and
|j| may be count registers). We will assume that |\A[I]| expands to the number
of rows, |\A[J]| to the number of columns and want the produced |\C| to act in
the same manner. The code is very dispendious in use of |\count| registers, not
optimized in any way, not made very robust (the defined macro can not have the
same name as the first two matrices for example), we just wanted to quickly
illustrate use of the nesting capabilities of |\xintloop|.%
%
\footnote{for a more sophisticated implementation of matrix
  multiplication, inclusive of determinants, inverses, and display
  utilities, with entries big integers or decimal numbers or even
  fractions see some code online posted
  from
  November 11, 2013.}
%


\begin{everbatim*}
\newcount\rowmax   \newcount\colmax   \newcount\summax
\newcount\rowindex \newcount\colindex \newcount\sumindex
\newcount\tmpcount
\makeatletter
\def\MatrixMultiplication #1#2#3{%
    \rowmax #1[I]\relax
    \colmax #2[J]\relax
    \summax #1[J]\relax
    \rowindex 1
    \xintloop % loop over row index i
    {\colindex 1
     \xintloop % loop over col index k
     {\tmpcount 0
      \sumindex 1
      \xintloop % loop over intermediate index j
      \advance\tmpcount \numexpr #1\rowindex\sumindex*#2\sumindex\colindex\relax
      \ifnum\sumindex<\summax
         \advance\sumindex 1
      \repeat }%
     \expandafter\edef\csname\string#3{\the\rowindex.\the\colindex}\endcsname
      {\the\tmpcount}%
     \ifnum\colindex<\colmax
         \advance\colindex 1
     \repeat }%
    \ifnum\rowindex<\rowmax
    \advance\rowindex 1
    \repeat
    \expandafter\edef\csname\string#3{I}\endcsname{\the\rowmax}%
    \expandafter\edef\csname\string#3{J}\endcsname{\the\colmax}%
    \def #3##1{\ifx[##1\expandafter\Matrix@helper@size
                    \else\expandafter\Matrix@helper@entry\fi #3{##1}}%
}%
\def\Matrix@helper@size #1#2#3]{\csname\string#1{#3}\endcsname }%
\def\Matrix@helper@entry #1#2#3%
   {\csname\string#1{\the\numexpr#2.\the\numexpr#3}\endcsname }%
\def\A #1{\ifx[#1\expandafter\A@size
            \else\expandafter\A@entry\fi {#1}}%
\def\A@size #1#2]{\ifx I#23\else4\fi}% 3rows, 4columns
\def\A@entry #1#2{\the\numexpr #1+#2-1\relax}% not pre-computed...
\def\B #1{\ifx[#1\expandafter\B@size
            \else\expandafter\B@entry\fi {#1}}%
\def\B@size #1#2]{\ifx I#24\else3\fi}% 4rows, 3columns
\def\B@entry #1#2{\the\numexpr #1-#2\relax}% not pre-computed...
\makeatother
\MatrixMultiplication\A\B\C \MatrixMultiplication\C\C\D
\MatrixMultiplication\C\D\E \MatrixMultiplication\C\E\F
\begin{multicols}2
  \[\begin{pmatrix}
    \A11&\A12&\A13&\A14\\
    \A21&\A22&\A23&\A24\\
    \A31&\A32&\A33&\A34
  \end{pmatrix}
  \times
  \begin{pmatrix}
    \B11&\B12&\B13\\
    \B21&\B22&\B23\\
    \B31&\B32&\B33\\
    \B41&\B42&\B43
  \end{pmatrix}
  =
  \begin{pmatrix}
    \C11&\C12&\C13\\
    \C21&\C22&\C23\\
    \C31&\C32&\C33
  \end{pmatrix}\]
  \[\begin{pmatrix}
    \C11&\C12&\C13\\
    \C21&\C22&\C23\\
    \C31&\C32&\C33
  \end{pmatrix}^2 = \begin{pmatrix}
    \D11&\D12&\D13\\
    \D21&\D22&\D23\\
    \D31&\D32&\D33
  \end{pmatrix}\]
  \[\begin{pmatrix}
    \C11&\C12&\C13\\
    \C21&\C22&\C23\\
    \C31&\C32&\C33
  \end{pmatrix}^3 = \begin{pmatrix}
    \E11&\E12&\E13\\
    \E21&\E22&\E23\\
    \E31&\E32&\E33
  \end{pmatrix}\]
  \[\begin{pmatrix}
    \C11&\C12&\C13\\
    \C21&\C22&\C23\\
    \C31&\C32&\C33
  \end{pmatrix}^4 = \begin{pmatrix}
    \F11&\F12&\F13\\
    \F21&\F22&\F23\\
    \F31&\F32&\F33
  \end{pmatrix}\]
\end{multicols}
\end{everbatim*}


\subsection{\csh{xintiloop}, \csh{xintiloopindex}, \csh{xintouteriloopindex},
  \csh{xintbreakiloop}, \csh{xintbreakiloopanddo}, \csh{xintiloopskiptonext},
\csh{xintiloopskipandredo}}
\label{xintiloop}
\label{xintbreakiloop}
\label{xintbreakiloopanddo}
\label{xintiloopskiptonext}
\label{xintiloopskipandredo}
\label{xintiloopindex}
\label{xintouteriloopindex}

\csa{xintiloop}|[start+delta]|\meta{stuff}|\if<test> ... \repeat|\xtype{} is a
completely expandable nestable loop. complete expandability depends naturally on
the actual iterated contents, and complete expansion will not be achievable
under a sole \fexpan sion, as is indicated by the hollow star in the margin;
thus the loop can be used inside an |\edef| but not inside arguments to the
package macros. It can be used inside an |\xintexpr..\relax|. The
|[start+delta]| is mandatory, not optional.

This loop benefits via \csbxint{iloopindex} to (a limited access to) the integer
index of the iteration. The starting value |start| (which may be a |\count|) and
increment |delta| (\emph{id.}) are mandatory arguments. A space after the
closing square bracket is not significant, it will be ignored. Spaces inside the
square brackets will also be ignored as the two arguments are first given to a
|\numexpr...\relax|. Empty lines and explicit |\par| tokens are accepted.

As with \csbxint{loop}, this tool will mostly be of interest to advanced users.
For nesting, one puts inside braces all the
material from the start (immediately after |[start+delta]|) and up to and
inclusive of the inner loop, these braces will be removed and do not create a
loop. In case of nesting, \csbxint{outeriloopindex} gives access to the index of
the outer loop. If needed one could write on its model a macro giving access to
the index of the outer outer loop (or even to the |nth| outer loop).

The \csa{xintiloopindex} and \csa{xintouteriloopindex} can not be used inside
braces, and generally speaking this means they should be expanded first when
given as argument to a macro, and that this macro receives them as delimited
arguments, not braced ones. Or, but naturally this will break expandability, one
can assign the value of \csa{xintiloopindex} to some |\count|. Both
\csa{xintiloopindex} and \csa{xintouteriloopindex} extend to the litteral
representation of the index, thus in |\ifnum| tests, if it comes last one has to
correctly end the macro with a |\space|, or encapsulate it in a
|\numexpr..\relax|.

When the repeat-test of the loop is, for example, |\ifnum\xintiloopindex<10
\repeat|, this means that the last iteration will be with |\xintiloopindex=10|
(assuming |delta=1|). There is also |\ifnum\xintiloopindex=10 \else\repeat| to
get the last iteration to be the one with |\xintiloopindex=10|.

One has \csbxint{breakiloop} and \csbxint{breakiloopanddo} to abort the loop.
The syntax of |\xintbreakiloopanddo| is a bit surprising, the sequence of tokens
to be executed after breaking the loop is not within braces but is delimited by
a dot as in:
%
\leftedline{|\xintbreakiloopanddo <afterloop>.etc.. etc... \repeat|}
%
The reason is that one may wish to use the then current value of
|\xintiloopindex| in |<afterloop>| but it can't be within braces at the time it
is evaluated. However, it is not that easy as |\xintiloopindex| must be expanded
before, so one ends up with code like this:
%
\leftedline
{|\expandafter\xintbreakiloopanddo\expandafter\macro\xintiloopindex.%|}
%
\leftedline{|etc.. etc.. \repeat|}
%
As moreover the |\fi| from the test leading to the decision of breaking out of
the loop must be cleared out of the way, the above should be
a branch of an expandable conditional test, else one needs something such
as:
%
\leftedline
{|\xint_afterfi{\expandafter\xintbreakiloopanddo\expandafter\macro\xintiloopindex.}%|}
%
\leftedline{|\fi etc..etc.. \repeat|}

There is \csbxint{iloopskiptonext} to abort the current iteration and skip to
the next, \hyperref[xintiloopskipandredo]{\ttfamily\hyphenchar\font45 \char92
  xintiloopskip\-and\-redo} to skip to the end of the current iteration and redo
it with the same value of the index (something else will have to change for this
not to become an eternal loop\dots ).

Inside alignments, if the looped-over text contains a |&| or a |\cr|, any
un-expandable material before a \csbxint{iloopindex} will make it fail because
of |\endtemplate|; in such cases one can always either replace |&| by a macro
expanding to it or replace it by a suitable |\firstofone{&}|, and similarly for
|\cr|.

\phantomsection\label{edefprimes}
As an example, let us construct an |\edef\z{...}| which will define |\z| to be a
list of prime numbers:
\begin{everbatim*}
\begingroup
\edef\z
{\xintiloop [10001+2]
  {\xintiloop [3+2]
   \ifnum\xintouteriloopindex<\numexpr\xintiloopindex*\xintiloopindex\relax
          \xintouteriloopindex,
          \expandafter\xintbreakiloop
   \fi
   \ifnum\xintouteriloopindex=\numexpr
        (\xintouteriloopindex/\xintiloopindex)*\xintiloopindex\relax
   \else
   \repeat
  }% no space here
 \ifnum \xintiloopindex < 10999 \repeat }%
\meaning\z\endgroup
\end{everbatim*}and we should have taken
some steps to not have a trailing comma, but
the point was to show that one can do that in an |\edef|\,! See also
\autoref{ssec:primesII} which extracts from this code its way of testing
primality.

Let us create an alignment where each row will contain all divisors of its
first entry.
Here is the output, thus obtained without any count register:
\begin{everbatim*}
\begin{multicols}2
\tabskip1ex \normalcolor
\halign{&\hfil#\hfil\cr
    \xintiloop [1+1]
    {\expandafter\bfseries\xintiloopindex &
     \xintiloop [1+1]
     \ifnum\xintouteriloopindex=\numexpr
           (\xintouteriloopindex/\xintiloopindex)*\xintiloopindex\relax
     \xintiloopindex&\fi
     \ifnum\xintiloopindex<\xintouteriloopindex\space % CRUCIAL \space HERE
     \repeat \cr }%
    \ifnum\xintiloopindex<30
    \repeat
}
\end{multicols}
\end{everbatim*}
We wanted this first entry in bold face, but |\bfseries| leads to
unexpandable tokens, so the |\expandafter| was necessary for |\xintiloopindex|
and |\xintouteriloopindex| not to be confronted with a hard to digest
|\endtemplate|. An alternative way of coding:
%
\begin{everbatim}
\tabskip1ex
\def\firstofone #1{#1}%
\halign{&\hfil#\hfil\cr
  \xintiloop [1+1]
    {\bfseries\xintiloopindex\firstofone{&}%
    \xintiloop [1+1] \ifnum\xintouteriloopindex=\numexpr
    (\xintouteriloopindex/\xintiloopindex)*\xintiloopindex\relax
    \xintiloopindex\firstofone{&}\fi
    \ifnum\xintiloopindex<\xintouteriloopindex\space % \space is CRUCIAL
    \repeat \firstofone{\cr}}%
  \ifnum\xintiloopindex<30 \repeat }
\end{everbatim}

\begin{framed}
  The next utilities are not compatible with expansion-only context.
\end{framed}

\subsection{\csh{xintApplyInline}}\label{xintApplyInline}


\csa{xintApplyInline}|{\macro}|\marg{list}\ntype{o{\lowast f}} works non
expandably. It applies the one-parameter |\macro| to the first element of the
expanded list (|\macro| may have itself some arguments, the list item will be
appended as last argument), and is then re-inserted in the input stream after
the tokens resulting from this first expansion of |\macro|. The next item is
then handled.

This is to be used in situations where one needs to do some repetitive
things. It is not expandable and can not be completely expanded inside a
macro definition, to prepare material for later execution, contrarily to what
\csbxint{Apply} or \csbxint{ApplyUnbraced} achieve.

\begin{everbatim*}
\def\Macro #1{\advance\cnta #1 , \the\cnta}
\cnta 0
0\xintApplyInline\Macro {3141592653}.
\end{everbatim*}
The first argument |\macro| does not have to be an expandable macro.

\csa{xintApplyInline} submits its second, token list parameter to an
\hyperref[ssec:expansions]{\fexpan
sion}. Then, each \emph{unbraced} item will also be \fexpan ded. This provides
an easy way to insert one list inside another. \emph{Braced} items are not
expanded. Spaces in-between items are gobbled (as well as those at the start
or the end of the list), but not the spaces \emph{inside} the braced items.

\csa{xintApplyInline}, despite being non-expandable, does survive to
contexts where the executed |\macro| closes groups, as happens inside
alignments with the tabulation character |&|.
This tabular provides an example:\par
\begin{everbatim*}
\centerline{\normalcolor\begin{tabular}{ccc}
     $N$ & $N^2$ & $N^3$ \\ \hline
     \def\Row #1{ #1 & \xintiiSqr {#1} & \xintiiPow {#1}{3} \\ \hline }%
     \xintApplyInline \Row {\xintCSVtoList{17,28,39,50,61}}
\end{tabular}}\medskip
\end{everbatim*}

We see that despite the fact that the first encountered tabulation character in
the first row close a group and thus erases |\Row| from \TeX's memory,
|\xintApplyInline| knows how to deal with this.

Using \csbxint{ApplyUnbraced} is an alternative: the difference is that
this would have prepared all rows first and only put them back into the
token stream once they are all assembled, whereas with |\xintApplyInline|
each row is constructed and immediately fed back into the token stream: when
one does things with numbers having hundreds of digits, one learns that
keeping on hold and shuffling around hundreds of tokens has an impact on
\TeX{}'s speed (make this ``thousands of tokens'' for the impact to be
noticeable).

One may nest various |\xintApplyInline|'s. For example (see the
\hyperref[float]{table} \vpageref{float}):\par
\begin{everbatim*}
\begin{figure*}[ht!]
  \centering\phantomsection\label{float}
  \def\Row #1{#1:\xintApplyInline {\Item {#1}}{0123456789}\\ }%
  \def\Item #1#2{&\xintiiPow {#1}{#2}}%
  \centeredline {\begin{tabular}{ccccccccccc} &0&1&2&3&4&5&6&7&8&9\\ \hline
      \xintApplyInline \Row {0123456789}
    \end{tabular}}
\end{figure*}
\end{everbatim*}

One could not move the definition of |\Item| inside the tabular,
as it would get lost after the first |&|. But this
works:
\everb|@
\begin{tabular}{ccccccccccc}
    &0&1&2&3&4&5&6&7&8&9\\ \hline
    \def\Row #1{#1:\xintApplyInline {&\xintiiPow {#1}}{0123456789}\\ }%
    \xintApplyInline \Row {0123456789}
\end{tabular}
|

A limitation is that, contrarily to what one may have expected, the
|\macro| for an |\xintApplyInline| can not be used to define
the |\macro| for a nested sub-|\xintApplyInline|. For example,
this does not work:\par
\everb|@
  \def\Row #1{#1:\def\Item ##1{&\xintiiPow {#1}{##1}}%
                 \xintApplyInline \Item {0123456789}\\ }%
  \xintApplyInline \Row {0123456789} % does not work
|
\noindent But see \csbxint{For}.

\subsection{\csh{xintFor}, \csh{xintFor*}}\label{xintFor}\label{xintFor*}

\csbxint{For}\ntype{on} is a new kind of for loop.\footnote{first introduced
  with \xintname |1.09c| of |2013/10/09|.} Rather than using macros
for encapsulating list items, its behaviour is like a macro with parameters:
|#1|, |#2|, \dots, |#9| are used to represent the items for up to nine levels of
nested loops. Here is an example:
%
\everb|@
\xintFor #9 in {1,2,3} \do {%
  \xintFor #1 in {4,5,6} \do {%
    \xintFor #3 in {7,8,9} \do {%
      \xintFor #2 in {10,11,12} \do {%
      $$#9\times#1\times#3\times#2=\xintiiPrd{{#1}{#2}{#3}{#9}}$$}}}}
|
\noindent This example illustrates that one does not have to use |#1| as the
first one:
the order is arbitrary. But each level of nesting should have its specific macro
parameter. Nine levels of nesting is presumably overkill, but I did not know
where it was reasonable to stop. |\par| tokens are accepted in both the comma
separated list and the replacement text.

\begin{framed}
  \TeX nical notes:

\begin{itemize}
  \item The |#1| is replaced in the iterated-over text exactly as in general
    \TeX\ macros or \LaTeX\ commands. This spares the user quite a few
    |\expandafter|'s or other tricks needed with loops which have the
    values encapsulated in macros, like \LaTeX's |\@for| and |\@tfor|.

  \item \csa{xintFor} (and \csa{xintFor*}) isn't purely expandable: one can
    not use it inside an |\edef|. But it may be used, as will be shown in
    examples, in some contexts such as \LaTeX's |tabular| which are usually
    hostile to non-expandable loops.
  
  \item \csa{xintFor} (and \csa{xintFor*}) does some assignments prior to
    executing each iteration of the replacement text, but it acts purely
    expandably after the last iteration, hence if for example the replacement
    text ends with a |\\|, the loop can be used insided a tabular and be
    followed by a |\hline| without creating the dreaded ``|Misplaced
    \noalign|'' error.

  \item As stated in previous item the first iteration follows some
    non-expandable internal dealings. This means for example that in \LaTeX{},
    one can not inject a |\multicolumn| in the first iteration. Sometimes one
    way work around this by injecting father |&\multicolumn| or |\\
    \multicolumn|.

  \item It does not create groups.

  \item It makes no global assignments.

  \item The iterated replacement text may close a group which was opened even
    before the start of the loop (typical example being with |&| in
    alignments).
\begin{everbatim*}
\begin{tabular}{rccccc}
    \hline
    \xintFor #1 in {A, B, C} \do {%
      #1:\xintFor #2 in {a, b, c, d, e} \do {&($ #2 \to #1 $)}\\ }%
    \hline
\end{tabular}
\end{everbatim*}
  
  \item There is no facility provided which would give access to a count of
    the number of iterations as it is technically not easy to do so it in a
    way working with nested loops while maintaining the ``expandable after
    done'' property; something in the spirit of \csbxint{iloopindex} is
    possible but this approach would bring its own limitations and
    complications. Hence the user is invited to update her own count or
    \LaTeX{} counter or macro at each iteration, if needed.

  \item A |\macro| whose definition uses internally an \csbxint{For} loop
    may be used inside another \csbxint{For} loop even if the two loops both
    use the same macro parameter. The loop definition inside |\macro|
    must use |##| as is the general rule for definitions done inside macros.

  \item \csbxint{For} is for comma separated values and \csbxint{For*} for
    lists of braced items; their respective expansion policies differ. They
    are described later.
\end{itemize}
\unskip
\end{framed}

\noindent Regarding \csbxint{For}:
\begin{itemize}[nosep, listparindent=\leftmarginiii]
\item the spaces between the various declarative elements are all optional,
\item in the list of comma separated values,  spaces around the commas or at
  the start and end are ignored,
\item if an item must contain itself its own commas, then it should
  be braced, and the braces will be removed before feeding the iterated-over
  text,
\item the list may be a macro, it is expanded only once,
\item items are not pre-expanded. The first item should be braced or start
  with a space if the list is explicit and the item should not be
  pre-expanded,
\item empty items give empty |#1|'s in the replacement text, they are not
  skipped,
\item an empty list executes once the replacement text with an empty parameter
  value,
\item the list, if not a macro, \fbox{must be braced.}
\end{itemize}

\noindent Regarding \csbxint{For*}:\ntype{{\lowast f}n}
\begin{itemize}[nosep, listparindent=\leftmarginiii]
\item it handles lists of braced items (or naked tokens),
\item it \hyperref[ssec:expansions]{\fexpan ds} the list,
\item and more generally it \hyperref[ssec:expansions]{\fexpan ds} each naked
  token encountered 
  before assigning the |#1| values (gobbling spaces in the process);
  this
  makes it easy to simulate concatenation of multiple lists|\x|, |\y|:
  if |\x| expands to |{1}{2}{3}| and |\y| expands to |{4}{5}{6}| then |{\x\y}|
  as argument to |\xintFor*| has the same effect as |{{1}{2}{3}{4}{5}{6}}|.

  For a further illustration see the use of |\xintFor*| at the end of
  \autoref{ssec:fibonacci}.
\item spaces at the start, end, or in-between items are gobbled (but naturally
  not the spaces inside \emph{braced} items),
\item except if the list argument is a macro (with no parameters), \fbox{it
    must be braced.},
\item an empty list leads to an empty result.
\end{itemize}

The macro \csbxint{Seq} which generates arithmetic sequences is to be used
with \csbxint{For*} as its output consists of successive braced numbers (given
as digit tokens).
\begin{everbatim*}
\xintFor* #1 in {\xintSeq [+2]{-7}{+2}}\do {stuff
    with #1\xintifForLast{\par}{\newline}}
\end{everbatim*}


When nesting \csa{xintFor*} loops, using \csa{xintSeq} in the inner loops is
inefficient, as the arithmetic sequence will be re-created each time. A more
efficient style is:
%
\begin{everbatim}
    \edef\innersequence {\xintSeq[+2]{-50}{50}}%
    \xintFor* #1 in {\xintSeq {13}{27}} \do
        {\xintFor* #2 in \innersequence \do {stuff with #1 and #2}%
         .. some other macros .. }
\end{everbatim}

This is a general remark applying for any nesting of loops, one should avoid
recreating the inner lists of arguments at each iteration of the outer loop.


When the loop is defined inside a macro for later execution the |#| characters
must be doubled.%
%
\footnote{sometimes what seems to be a macro argument isn't really; in
  \csa{raisebox\{1cm\}\{}\csa{xintFor \#1 in \{a,b,c\} }\csa{do
    \{\#1\}\}} no doubling should be done.}
%
For example:
%
\begin{everbatim*}
\def\T{\def\z {}%
  \xintFor* ##1 in {{u}{v}{w}} \do {%
    \xintFor ##2 in {x,y,z} \do {%
      \expandafter\def\expandafter\z\expandafter {\z\sep (##1,##2)} }%
  }%
}%
\T\def\sep {\def\sep{, }}\z
\end{everbatim*}

Similarly when the replacement text
of |\xintFor| defines a macro with parameters, the macro character |#| must be
doubled.


The iterated macros as well as the list items are allowed to contain explicit
|\par| tokens.


\subsection{\csh{xintifForFirst}, \csh{xintifForLast}}
\label{xintifForFirst}\label{xintifForLast}

\csbxint{ifForFirst}\,\texttt{\{YES branch\}\{NO branch\}}\etype{nn}
 and \csbxint{ifForLast}\,\texttt{\{YES
  branch\}\hskip 0pt plus 0.2em \{NO branch\}}\etype{nn} execute the |YES| or
|NO| branch
if the
\csbxint{For}
or \csbxint{For*} loop is currently in its first, respectively last, iteration.

Designed to work as expected under nesting (but see frame next.) Don't forget
an empty brace pair |{}| if a branch is to do nothing. May be used multiple
times in the replacement text of the loop.

\begin{framed}
  \noindent Pay attention to these implementation features:
  \begin{itemize}[nosep, listparindent=\leftmarginiii]
  \item \emph{if an inner \csbxint{For} loop is positioned before the
    \csb{xintifForFirst} or \csb{xintifForLast} of the outer loop it will
    contaminate their settings. This applies also naturally if the inner loop
    arises from the expansion of some macro located before the outer
    conditionals.}

    One fix is to make sure that the outer conditionals are expanded before the
    inner loop is executed, e.g. this will be the case if the inner loop is
    located inside one of the branches of the conditional.

    Another approach is to enclose, if feasible, the inner loop in a group of
    its own.
  \item \emph{if the replacement text closes a group (e.g. from a |&| inside an
    alignment), the conditionals will lose their ascribed meanings and end up
    possibly undefined, depending whether there is some outer loop whose
    execution started before the opening of the group.}

    The fix is to arrange things so that the conditionals are expanded
    before \TeX\ encounters the closing-group token.
  \end{itemize}
\end{framed}

\subsection{ \csh{xintBreakFor}, \csh{xintBreakForAndDo}}
\label{xintBreakFor}\label{xintBreakForAndDo}

One may immediately terminate an \csbxint{For} or \csbxint{For*} loop with
\csbxint{BreakFor}.

\begin{framed}
  As it acts by clearing up all the rest of the replacement text when
  encountered, it will not work from inside some |\if...\fi| without
  suitable |\expandafter| or swapping technique.

  Also it can't be used from inside braces as from there it can't see the end
  of the replacement text.
\end{framed}

There is also \csbxint{BreakForAndDo}. Both are illustrated by various examples
in the next section which is devoted to ``forever'' loops.

\subsection{\csh{xintintegers}, \csh{xintdimensions}, \csh{xintrationals}}
\label{xintegers}\label{xintintegers}
\label{xintdimensions}\label{xintrationals}

If the list argument to \csbxint{For} (or \csbxint{For*}, both are equivalent in
this context) is \csbxint{integers} (equivalently \csbxint{egers}) or more
generally \csbxint{integers}|[||start|\allowbreak|+|\allowbreak|delta||]|
(\emph{the whole within braces}!)%
%
\footnote{the |start+delta| optional specification may have extra spaces
  around the plus sign of near the square brackets, such spaces are
  removed. The same applies with \csa{xintdimensions} and
  \csa{xintrationals}.},
%
then \csbxint{For} does an infinite iteration where
|#1| (or |#2|, \dots, |#9|) will run through the arithmetic sequence of (short)
integers with initial value |start| and increment |delta| (default values:
|start=1|, |delta=1|; if the optional argument is present it must contains both
of them, and they may be explicit integers, or macros or count registers). The
|#1| (or |#2|, \dots, |#9|) will stand for |\numexpr <opt sign><digits>\relax|,
and the litteral representation as a string of digits can thus be obtained as
\fbox{\csa{the\#1}} or |\number#1|. Such a |#1| can be used in an |\ifnum| test
with no need to be postfixed with a space or a |\relax| and one should
\emph{not} add them.

If the list argument is \csbxint{dimensions} or more generally
\csbxint{dimensions}|[||start|\allowbreak|+|\allowbreak|delta||]|  (\emph{within
  braces}!), then
\csbxint{For} does an infinite iteration where |#1| (or |#2|, \dots, |#9|) will
run through the arithmetic sequence of dimensions with initial value
|start| and increment |delta|. Default values: |start=0pt|, |delta=1pt|; if
the optional argument is present it must contain both of them, and they may
be explicit specifications, or macros, or dimen registers, or length macros
in \LaTeX{} (the stretch and shrink components will be discarded). The |#1|
will be |\dimexpr <opt sign><digits>sp\relax|, from which one can get the
litteral (approximate) representation in points via |\the#1|. So |#1| can be
used anywhere \TeX{} expects a dimension (and there is no need in conditionals
to insert a |\relax|, and one should \emph{not} do it), and to print its value
one uses \fbox{\csa{the\#1}}. The chosen representation guarantees exact
incrementation with no rounding errors accumulating from converting into
points at each step.






If the list argument to \csbxint{For} (or \csbxint{For*}) is \csbxint{rationals}
or more generally
\csbxint{rationals}|[||start|\allowbreak|+|\allowbreak|delta||]| (\emph{within
  braces}!), then \csbxint{For} does an infinite iteration where |#1| (or |#2|,
\dots, |#9|) will run through the arithmetic sequence of \xintfracname fractions
with initial value |start| and increment |delta| (default values: |start=1/1|,
|delta=1/1|). This loop works \emph{only with \xintfracname loaded}. if the
optional argument is present it must contain both of them, and they may be given
in any of the formats recognized by \xintfracname (fractions, decimal
numbers, numbers in scientific notations, numerators and denominators in
scientific notation, etc...) , or as macros or count registers (if they are
short integers). The |#1| (or |#2|, \dots, |#9|) will be an |a/b| fraction
(without a |[n]| part), where
the denominator |b| is the product of the denominators of
|start| and |delta| (for reasons of speed |#1| is not reduced to irreducible
form, and for another reason explained later  |start| and |delta| are not put
either into irreducible form; the input may use explicitely \csa{xintIrr} to
achieve that).
\begin{everbatim*}
\begingroup\small
\noindent\parbox{\dimexpr\linewidth-3em}{\color[named]{OrangeRed}%
\xintFor #1 in {\xintrationals [10/21+1/21]} \do
{#1=\xintifInt {#1}
    {\textcolor{blue}{\xintTrunc{10}{#1}}}
    {\xintTrunc{10}{#1}}% display in blue if an integer
    \xintifGt {#1}{1.123}{\xintBreakFor}{, }%
  }}
\endgroup\smallskip
\end{everbatim*}

\smallskip The example above confirms that computations are done exactly, and
illustrates that the two initial (reduced) denominators are not multiplied when
they are found to be equal.  It is thus recommended to input |start| and |delta|
with a common smallest possible denominator, or as fixed point numbers with the
same numbers of digits after the decimal mark;  and this is also the reason why
|start| and |delta| are not by default made irreducible. As internally the
computations are done with numerators and denominators completely expanded, one
should be careful not to input numbers in scientific notation with exponents in
the hundreds, as they will get converted into as many zeroes.

\begin{everbatim*}
\noindent\parbox{\dimexpr.7\linewidth}{\raggedright
\xintFor #1 in {\xintrationals [0.000+0.125]} \do
{\edef\tmp{\xintTrunc{3}{#1}}%
 \xintifInt {#1}
    {\textcolor{blue}{\tmp}}
    {\tmp}%
    \xintifGt {#1}{2}{\xintBreakFor}{, }%
  }}\smallskip
\end{everbatim*}

We see here that \csbxint{Trunc} outputs (deliberately) zero as $0$, not (here)
$0.000$, the idea being not to lose the information that the truncated thing was
truly zero. Perhaps this behaviour should be changed? or made optional? Anyhow
printing of fixed points numbers should be dealt with via dedicated packages
such as |numprint| or |siunitx|.\par


\subsection{\csh{xintForpair}, \csh{xintForthree}, \csh{xintForfour}}\label{xintForpair}\label{xintForthree}\label{xintForfour}

The syntax\ntype{on} is illustrated in this
example. The notation is the usual one for |n|-uples, with parentheses and
commas. Spaces around commas and parentheses are ignored.
%
\begin{everbatim*}
{\centering\begin{tabular}{cccc}
    \xintForpair #1#2 in { ( A , a ) , ( B , b ) , ( C , c ) } \do {%
      \xintForpair #3#4 in { ( X , x ) , ( Y , y ) , ( Z , z ) } \do {%
        $\Biggl($\begin{tabular}{cc}
          -#1- & -#3-\\
          -#4- & -#2-\\
        \end{tabular}$\Biggr)$&}\\\noalign{\vskip1\jot}}%
\end{tabular}\\}
\end{everbatim*}

\csbxint{Forpair} must be followed by either |#1#2|, |#2#3|, |#3#4|, \dots, or
|#8#9| with |#1| usable as an alias for |#1#2|, |#2| as alias for |#2#3|,
etc \dots\ and similarly for \csbxint{Forthree} (using |#1#2#3| or simply
|#1|, |#2#3#4| or simply |#2|, \dots) and \csbxint{Forfour} (with |#1#2#3#4|
etc\dots).

Nesting works as long as the macro parameters are distinct among |#1|, |#2|,
..., |#9|. A macro which expands to an \csa{xintFor} or a
\csa{xintFor(pair,three,four)} can be used in another one with no constraint
about using distinct macro parameters.

|\par| tokens are accepted in both the comma separated list and the
replacement text.


\subsection{\csh{xintAssign}}\label{xintAssign}

\csa{xintAssign}\meta{braced things}\csa{to}%
\meta{as many cs as they are things} %\ntype{{(f$\to$\lowast [x)}{\lowast N}}
%
defines (without checking if something gets overwritten) the control sequences
on the right of \csa{to} to expand to the successive tokens or braced items
located to the left of \csa{to}. \csa{xintAssign} is not an expandable macro.

\fexpan sion is first applied to the material in front of \csa{xintAssign}
which is fetched as one argument if it is braced. Then the expansion of this
argument is examined and successive items are assigned to the macros following
|\to|. There must be exactly as many macros as items. No check is done. The
macro assignments are done with removal of one level of brace pairs from each
item.

After the initial \fexpan sion, each assigned (brace-stripped) item will be
expanded according to the setting of the optional parameter.

For example |\xintAssign [e]...| means that all assignments are done using
|\edef|. With |[f]| the assignments will be made using
\hyperref[fdef]{\ttfamily\char92fdef}. The default is simply to make the
definitions with |\def|, corresponding to an empty optional paramter |[]|.
Possibilities for the optional parameter are: |[], [g], [e], [x], [o], [go],
[oo], [goo], [f], [gf]|. For example |[oo]| means a double expansion.
\begin{everbatim*}
\xintAssign \xintiiDivision{1000000000000}{133333333}\to\Q\R
\meaning\Q\newline
\meaning\R\newline
\xintAssign {{\xintiiDivision{1000000000000}{133333333}}}\to\X
\meaning\X\newline
\xintAssign [oo]{{\xintiiDivision{1000000000000}{133333333}}}\to\X
\meaning\X\newline
\xintAssign \xintiiPow{7}{13}\to\SevenToThePowerThirteen
\meaning\SevenToThePowerThirteen\par
\end{everbatim*}

Two special cases:
\begin{itemize}[nosep]
\item if after this initial expansion no brace is found immediately after
  \csa{xintAssign}, it is assumed that there is only one control sequence
  following |\to|, and this control sequence is then defined via |\def| (or
  what is set-up by the optional parameter) to expand to the material between
  \csa{xintAssign} and \csa{to}.
\item if the material between \csa{xintAssign} and |\to| is enclosed in two
  brace pairs, the first brace pair is removed, then the \fexpan sion is
  immediately stopped by the inner brace pair, hence \csa{xintAssign} now
  finds a unique item and thus defines only a single macro to be this item,
  which is now stripped of the second pair of braces.
\end{itemize}


\emph{Note:} prior to release |1.09j|, |\xintAssign| did an |\edef| by default
for each item assignment but it now does |\def| corresponding to no or empty
optional parameter.

It is allowed for the successive braced items to be separated by spaces. They
are removed during the assignments. But if a single macro is defined (which
happens if the argument after \fexpan sion does not start with a brace),
naturally the scooped up material has all intervening spaces, as it is
considered a
single item. But an upfront initial space will have been absorbed by \fexpan
sion.
\begin{everbatim*}
\def\X{ {a}  {b} {c}   {d} }\def\Y { u {a}  {b} {c}   {d} }
\xintAssign\X\to\A\B\C\D
\xintAssign\Y\to\Z
\meaning\A, \meaning\B, \meaning\C, \meaning\D+++\newline
\meaning\Z+++\par
\end{everbatim*}
As usual successive space characters in input make for a single \TeX\ space token.


\subsection{\csh{xintAssignArray}}\label{xintAssignArray}

\xintAssignArray \xintBezout {1000}{113}\to\Bez

\csa{xintAssignArray}\meta{braced things}\csa{to}\csa{myArray}
%
%\ntype{{(f$\to$\lowast x)}N}
%
first expands fully what comes immediately after |\xintAssignArray| and
expects to find a list of braced things |{A}{B}...| (or tokens). It then
defines \csa{myArray} as a macro with one parameter, such that \csa{myArray\x}
expands to give the |x|th braced thing of this original
list (the argument \texttt{\x} itself is fed to a |\numexpr| by |\myArray|,
and |\myArray| expands in two steps to its output). With |0| as parameter,
\csa{myArray}|{0}| returns the number |M| of elements of the array so that the
successive elements are \csa{myArray}|{1}|, \dots, \csa{myArray}|{M}|.
%
\leftedline{|\xintAssignArray \xintBezout {1000}{113}\to\Bez|} will set
|\Bez{0}| to \dtt{\Bez0}, |\Bez{1}| to \dtt{\Bez1}, |\Bez{2}| to
\dtt{\Bez2}, and |\Bez{3}| to \dtt{\Bez3}:
\dtt{$\Bez1\times1000+\Bez2\times113=\Bez3$.}
This macro is incompatible with expansion-only contexts.

\csa{xintAssignArray} admits an optional parameter, for example
|\xintAssignArray [e]| means that the definitions of the macros will be made
with |\edef|. The empty optional parameter (default) means that definitions
are done with |\def|. Other possibilities: |[], [o], [oo], [f]|. Contrarily to
\csbxint{Assign} one can not use the |g| here to make the definitions global.
For this, one should rather do |\xintAssignArray| within a group starting with
|\globaldefs 1|.


\subsection{\csh{xintDigitsOf}}\label{xintDigitsOf}

This is a synonym for \csbxint{AssignArray},\ntype{fN} to be used to define
an array giving all the digits of a given (positive, else the minus sign will
be treated as first item) number.
\begingroup\xintDigitsOf\xintiiPow {7}{500}\to\digits
%
\leftedline{|\xintDigitsOf\xintiiPow {7}{500}\to\digits|}
\noindent $7^{500}$ has |\digits{0}=|\digits{0} digits, and the 123rd among them
(starting from the most significant) is
|\digits{123}=|\digits{123}.
\endgroup

\subsection{\csh{xintRelaxArray}}\label{xintRelaxArray}

\csa{xintRelaxArray}\csa{myArray} %\ntype{N}
%
(globally) sets to \csa{relax} all macros which were defined by the previous
\csa{xintAssignArray} with \csa{myArray} as array macro.

\clearpage
\let\xinttoolsnameUp\undefined
\def\n{|{N}|}
\def\m{|{M}|}
\def\x{|{x}|}

\csname xintexprnameUp\endcsname
\section{Macros of the \xintexprname package}%
\RaisedLabel{sec:oldxintexpr}

\localtableofcontents

The \xintexprname package was first released with version |1.07|
(|2013/05/25|) of the \xintname bundle. It was substantially enhanced with
release |1.1| from |2014/10/28|.

The |1.4| release from |2020/01/31| maintains the same general architecture
but needed adapting all the code base for the switch from |\csname| to
|\expanded| techniques. On this occasion the mechanism for defining functions
was substantially strengthened.  The parser core mechanisms were improved too.

The package loads automatically \xintfracname and \xinttoolsname.

This section should be trimmed to contain only information not
already covered in \autoref{sec:expr}.

\subsection{The \csh{xintexpr} expressions}
\label{xintexpr}
\label{xinttheexpr}
\label{thexintexpr}
\label{xintthe}

An \xintexprname{}ession is a construct
\csbxint{expr}\meta{expandable\_expression}|\relax|\etype{x} where the
expandable expression is read and completely expanded from left to right.

An |\xintexpr...\relax| \emph{must} end in a |\relax| (which will be absorbed).
Contrarily to a |\numexpr| expression, it is printable\CHANGED{1.4} as is without a prefix
|\the| or |\number| (don't use them with |\xintexpr| this will raise an
error).

But one can use |\xintthe| prefix if one does need the explicit digits and
other characters as in the final typesetted result.

As an alternative and equivalent syntax to
\begin{everbatim}
\xintexpr round(<expression>, D)\relax
\end{everbatim}
there is
\begin{everbatim}
\xintiexpr [D] <expression> \relax
\end{everbatim}
For |D>0| this produces a decimal number with |D| figures after the decimal
mark, which is the rounding of the expression. For |D=0| the rounding to an
integer is produced. For |D<0| (and this was changed at |1.4f|), the rounded
quotient of the expression by \verb=1e|D|= is produced.
\begin{itemize}
\item the expression may contain arbitrarily many levels of nested parenthesized
  sub-expressions,
\item the expression may contain explicitely or from a macro expansion a
  sub-expression |\xintexpr...\relax|, which itself may contain a
  sub-expressions etc\dots
\item to let sub-contents evaluate as a sub-unit it should thus be either
   \begin{enumerate}
   \item parenthesized,
   \item or a sub-expression |\xintexpr...\relax|.
   \end{enumerate}
 \item to use an expression as argument to macros from \xintfracname,
   or more generally to macros which expand their arguments, one must use the
   |\xinttheexpr...\relax| or |\xintthe\xintexpr...\relax| forms.
 \item one should not use |\xintthe\xintexpr...\relax| as a sub-constituent of
   another expression but only the
   |\xintexpr...\relax| form which is more efficient in this context.
 \item each \xintexprname{}ession, whether prefixed or not with |\xintthe|, is
   completely expandable and obtains its result in two expansion steps.
\end{itemize}

The information now following is possibly in need of updates.

\begin{itemize}[parsep=0pt, labelwidth=\leftmarginii,
  itemindent=0pt, listparindent=\leftmarginiii, leftmargin=\leftmarginii]
\item An expression is built the standard way with opening and closing
  parentheses, infix operators, and (big) numbers, with possibly a fractional
  part, and/or scientific notation (except for \csbxint{iiexpr} which only
  admits big integers). All variants work with comma separated expressions. On
  output each comma will be followed by a space. A decimal number must have
  digits either before or after the decimal mark.

\item As everything gets expanded, the characters |.|, |+|, |-|, |*|, |/|, |^|,
  |!|, |&|, \verb+|+, |?|, |:|, |<|, |>|, |=|, |(|, |)|, |"|, |]|, |[|, |@|
  and the comma |,| should not (if used in the expression) be active. For
  example, the French language in |Babel| system, for pdf\LaTeX, activates |!|,
  |?|, |;| and |:|. Turn off the activity before expressions using such characters.

  Alternatively the macro \csbxint{exprSafeCatcodes} resets all
  characters potentially needed by \csbxint{expr} to their standard catcodes
  and \csbxint{exprRestoreCatcodes} restores the former status.

\item Count registers and |\numexpr|-essions are accepted (LaTeX{}'s counters
  can be inserted using |\value|) natively without |\the| or |\number| as
  prefix. Also dimen registers and control sequences, skip registers and
  control sequences (\LaTeX{}'s lengths), |\dimexpr|-essions,
  |\glueexpr|-essions are automatically unpacked using |\number|, discarding
  the stretch and shrink components and giving the dimension value in |sp|
  units ($1/65536$th of a \TeX{} point). Furthermore, tacit multiplication is
  implied, when the (count or dimen or glue) register or variable, or the
  (|\numexpr| or |\dimexpr| or |\glueexpr|) expression is immediately prefixed
  by a (decimal) number. See \autoref{ssec:tacit multiplication} for the complete rules
  of tacit multiplication.\IMPORTANT

\item With a macro |\x| defined like this:
  %
  \leftedline{|\def\x {\xintexpr \a + \b \relax}| or |\edef\x {\xintexpr
      \a+\b\relax}|}
  %
  one may then do |\xintthe\x|, either for printing the result on the page or
  to use it in some other macros expanding their arguments. The |\edef| does
  the computation immediately but keeps it in a protected form.
  Naturally, the |\edef| is only possible if |\a| and |\b| are already
  defined. With both approaches the |\x| can be inserted in other expressions,
  as for example (assuming naturally as we use an |\edef| that in the
  `yet-to-be computed' case the |\a| and |\b| now have some suitable meaning):
  %
  \leftedline {|\edef\y {\xintexpr \x^3\relax}|}

\item There is also \csbxint{boolexpr}| ... \relax| and
  \csbxint{theboolexpr}| ... \relax|.

\item  See also
  \csbxint{ifboolexpr} (\autoref{xintifboolexpr}) and the
  \func{bool} and \func{togl} functions
  in \autoref{sec:expr}. Here is an example. Well in fact the
  example ended up using only \csbxint{boolexpr} so it 
  was modified to use \csbxint{ifboolexpr}.
\catcode`| 12 %
\begin{everbatim*}
\xintdeffunc A(p,q,r) =  p && (q || r) ;
\xintdeffunc B(p,q,r) =  p || (q && r) ;
\xintdeffunc C(p,q,r) =  xor(p, q, r)  ;

\centeredline{\normalcolor
\begin{tabular}{ccrclcl}
        \xintFor* #1 in  {{False}{True}} \do {%
        \xintFor* #2 in  {{False}{True}} \do {%
        \xintFor* #3 in  {{False}{True}} \do {%
    #1 &AND &(#2 &OR  &#3)&is&\textcolor[named]{OrangeRed}
                              {\xintifboolexpr{A(#1,#2,#3)}{true}{false}}\\
    #1 &OR  &(#2 &AND &#3)&is&\textcolor[named]{OrangeRed}
                              {\xintifboolexpr{B(#1,#2,#3)}{yes}{no}}\\
    #1 &XOR & #2 &XOR &#3 &is&\textcolor[named]{OrangeRed}
                              {\xintifboolexpr{C(#1,#2,#3)}{oui}{non}}\\
}}}
\end{tabular}%
}
\end{everbatim*}\catcode`| 13

\item See also \csbxint{ifsgnexpr}.

\item There is  \csbxint{floatexpr}| ... \relax| where the algebra is done
  in floating point approximation (also for each intermediate result). Use the
  syntax |\xintDigits:=N\relax| to set the precision. Default: $16$ digits.
  %
  \leftedline{|\xintthefloatexpr 2^100000\relax:| \dtt{\xintthefloatexpr
      2^100000\relax }}
  %
  The square-root operation can be used in |\xintexpr|, it is computed
  as a float with the precision set by |\xintDigits| or by the optional
  second argument:
  %
\begin{everbatim*}
\xinttheexpr sqrt(2,60)\relax\newline
Here the [60] is to avoid truncation to |\xinttheDigits| of precision on output.\newline
\printnumber{\xintthefloatexpr [60] sqrt(2,60)\relax}
\end{everbatim*}

  Floats are quickly indispensable when using the power function, as exact
  results will easily have hundreds, even thousands of digits.
  %
\begin{everbatim*}
\xintDigits:=48\relax \xintthefloatexpr 2^100000\relax
\end{everbatim*}

  Only integer and (in |\xintfloatexpr...\relax|) half-integer exponents are
  allowed.

\item if one uses \emph{macros} within |\xintexpr..\relax| one should
  obviously take into account that the parser will \emph{not} see the macro
  arguments, hence one cannot use the syntax there, except if the arguments
  are themselves wrapped as |\xinttheexpr...\relax| and assuming the macro
  \fexpan ds these arguments.
\end{itemize}
\subsection{\texorpdfstring{\texttt{\protect\string\numexpr}}{\textbackslash
    numexpr} or \texorpdfstring{\texttt{\protect\string\dimexpr}}{\textbackslash
    dimexpr} expressions, count and dimension registers and variables}
\label{ssec:countinexpr}

Count registers, count control sequences, dimen registers, dimen control
sequences (like |\parindent|), skips and skip control sequences, |\numexpr|,
|\dimexpr|, |\glueexpr|, |\fontdimen| can be inserted directly, they will be
unpacked using |\number| which gives the internal value in terms of scaled
points for the dimensional variables: $1$\,|pt|${}=65536$\,|sp| (stretch and
shrink components are thus discarded).

Tacit multiplication (see \autoref{ssec:tacit multiplication}) is implied,
when a number or decimal number prefixes such a register or control sequence.
\LaTeX{} lengths are skip control sequences and \LaTeX{} counters should be
inserted using |\value|.

Release |1.2| of the |\xintexpr| parser also recognizes and prefixes with
|\number| the |\ht|, |\dp|, and |\wd| \TeX{} primitives as well as the
|\fontcharht|, |\fontcharwd|, |\fontchardp| and |\fontcharic| \eTeX{}
primitives.

In the case of numbered registers like |\count255| or |\dimen0| (or |\ht0|),
the resulting digits will be re-parsed, so for example |\count255 0| is like
|100| if |\the\count255| would give |10|. The same happens with inputs such
as |\fontdimen6\font|. And |\numexpr 35+52\relax| will be exactly as if |87|
as been encountered by the parser, thus more digits may follow: |\numexpr
35+52\relax 000| is like |87000|. If a new |\numexpr| follows, it is treated
as what would happen when |\xintexpr| scans a number and finds a non-digit: it
does a tacit multiplication.
\begin{everbatim*}
\xinttheexpr \numexpr 351+877\relax\numexpr 1000-125\relax\relax{} is the same
as \xinttheexpr 1228*875\relax.
\end{everbatim*}

Control sequences however (such as |\parindent|) are picked up as a whole by
|\xintexpr|, and the numbers they define cannot be extended extra digits, a
syntax error is raised if the parser finds digits rather than a legal
operation after such a control sequence.

A token list variable must be prefixed by |\the|, it will not be unpacked
automatically (the parser will actually try |\number|, and thus fail). Do not
use |\the| but only |\number| with a dimen or skip, as the |\xintexpr| parser
doesn't understand |pt| and its presence is a syntax error. To use a dimension
expressed in terms of points or other \TeX{} recognized units, incorporate it in
|\dimexpr...\relax|.

Regarding how dimensional expressions are converted by \TeX{} into scaled points
see also \autoref{sec:Dimensions}.

\subsection{Catcodes and spaces}

The main problems are caused by active characters, because \csbxint{expr} et
al.\@ expand forward whatever comes from token stream; they apply |\string|
only in a second step.  For example the catcode of |&| from |&&| Boolean
disjunction is not really important as long as it is not active, or comment,
or escape...\@ or brace...\@ or ignored...\@ in brief, as long as it is reasonable,
and in particular whether |@| is of catcode letter or other does not matter.

It is always possible to insert manually the |\string| in the expression
before a problematic (but reasonable) character catcode, or even to use
|\detokenize| for a big chunk.  

\subsubsection{\csh{xintexprSafeCatcodes}}
\label{xintexprSafeCatcodes}

For an even more radical way, there is \csbxint{exprSafeCatcodes} which sets
the catcodes of many characters to safe values. This is a non-expandable step
as it changes catcodes.
%  This is used
% internally by \csbxint{NewExpr} (restoring the catcodes on exit), hence it
% does not have to be protected against active characters when used at
% top-level.

\csbxint{defvar}, \csbxint{deffunc}, et al., execute it before fetching their
semi-colon delimited arguments, so they can be used (also in the document
body) for example with Babel+French (which makes the semi-colon active in the
(\LaTeX) document body).  This applies also to \csbxint{NewExpr}.

% As \csbxint{NewExpr} and \csbxint{deffunc} and variants use internally some
% |\scantokens|, they will (reasonably) succeed in sanitizing catcodes in the
% expressions, even if all is from the replacement text of some macro whose
% definition was done under some special catcode regime.

But, if used in the body of macro definitions problems may arise from the
catcode regime at that location. This applies in particular to the
semi-colon as used by \csbxint{deffunc}, \csbxint{defvar} and variants as
delimiter. Thus make sure the semi-colon has its normal catcode when issueing
\csbxint{deffunc} inside some macro definition.

\csbxint{deffunc} is more lenient than \csbxint{defvar} regarding catcodes of
characters in expression bodies as it does some |\scantokens| which will reset
compatible catcodes. And also, characters inside the
expression may usually be prefixed with |\string|; but some aspects of the parsing
use delimited macros which need the comma, equality sign and closing
parenthesis to have standard catcodes.

Even if used in a context where catcodes are already set, \csbxint{deffunc},
\csbxint{defvar} and variants ignore completely the colon in |:=| so it can
have any (reasonable) catcode. Moreover it is optional.

The semi-colon in the syntax of \csbxint{Digits} is no real problem either
(cf. \csbxint{Digits} documentation).

\begin{framed}
  It is important to ALWAYS shortly let \csbxint{exprSafeCatcodes} be followed
  by \csbxint{exprRestoreCatcodes}.\IMPORTANTf{} If one uses twice
  \csbxint{exprSafeCatcodes} then the next \csbxint{exprRestoreCatcodes} will
  restore the ancient catcode regime at time of the first one.
\end{framed}

\subsubsection{\csh{xintexprRestoreCatcodes}}
\label{xintexprRestoreCatcodes}

Restores the catcodes to the earlier state. More precisely,
\csbxint{exprSafeCatcodes} sets a toggle (with local scope). If the toggle is
set already it does not restore the current catcodes. The next
\csa{xintexprRestoreCatcodes} unsets the toggle.
So, in case of nesting, the
catcodes are restored to what they were when the \emph{first} un-paired
\csbxint{exprSafeCatcodes} got executed.

\bigskip

Spaces inside an |\xinttheexpr...\relax| should mostly be
innocuous (except inside macro arguments).

|\xintexpr| and |\xinttheexpr| are for the most part agnostic regarding
catcodes: (unbraced) digits, binary operators, minus and plus signs as
prefixes, dot as decimal mark, parentheses, may be indifferently of catcode
letter or other or subscript or superscript, ..., it doesn't matter.%
%
\footnote{Furthermore, although \csbxint{expr} uses \csa{string}, it is
  escape-char agnostic. It should work with any \csa{escapechar} setting
  including -1.}

The characters |+|, |-|, |*|, |/|, |^|, |!|, |&|, \verb+|+, |?|, |:|, |<|, |>|,
|=|, |(|, |)|, |"|, |[|, |]|, |;|, the dot and the comma should not be active if
in the expression, as everything is expanded along the way. If one of them is
active, it should be prefixed with |\string|.

The exclamation mark |!| should have its standard catcode: with catcode letter
it is used internally and hence will confuse the parsers if it comes from the
expression.

Digits, slash, square brackets, minus sign, in the output from an
|\xinttheexpr| are all of catcode 12. For |\xintthefloatexpr| the `e' in the
output has its standard catcode ``letter''.

A macro with arguments will expand and grab its arguments before the
parser may get a chance to see them, so the situation with catcodes and spaces
is not the same within such macro arguments.



\subsection{Expandability, \csh{xintexpro}}

As is the case with all other package macros |\xintexpr| \fexpan ds (in two
steps) to its final (somewhat protected) result; and |\xinttheexpr| \fexpan ds
(in two steps) to the chain of digits (and possibly minus sign |-|, decimal
mark |.|, fraction slash |/|, scientific |e|, square brackets |[|, |]|)
representing the result.

The once expanded |\xintexpr| is |\romannumeral0\xintexpro|.  is similarly
|\xintiexpro| |\xintiiexpro| and |\xintfloatexpro|. For an example see
\autoref{ssec:fibonacci}.

An expression can only be legally finished by a |\relax| token, which
will be absorbed.

It is quite possible to nest expressions among themselves; for example, if one
needs inside an |\xintiiexpr...\relax| to do some computations with fractions,
rounding the final result to an integer, one just has to insert
|\xintiexpr...\relax|. The functioning of the infix operators will not be in
the least affected from the fact that the outer ``environment'' is the
|\xintiiexpr| one.


\subsection{\csh{xintDigits*}, \csh{xintSetDigits*}}
\label{xintDigits*}
\label{xintSetDigits*}

These starred variants of \csbxint{Digits} and \csbxint{SetDigits} execute
\csbxint{reloadxinttrig} and \csbxint{reloadxintlog}.

\subsection{\csh{xintiexpr}, \csh{xinttheiexpr}}
\label{xintiexpr}\label{xinttheiexpr}\label{thexintiexpr}

Equivalent\etype{x} to doing |\xintexpr round(...)\relax| (more precisely,
|round| is applied to each leaf item of the |ople| independently of its
depth).

Intermediate calculations are exact, only the final output gets
rounded. Half integers are rounded towards $+\infty$ for positive
numbers and towards $-\infty$ for negative ones.

An optional parameter |D| within brackets, immediately after |\xintiexpr| is
allowed: it instructs (for |D>0|) the expression to do its final rounding to
the nearest value with that many digits after the decimal mark, i.e.\@
|\xintiexpr [D] <expression>\relax| is equivalent (in case of a single
expression) to |\xintexpr round(<expression>, D)\relax|.

|\xintiexpr [0] ...| is the same as |\xintiexpr ...| and rounds to an integer.

The case of negative |D| gives quantization to an integer multiple of
\dtt{1e-D}.\NewWith{1.4a} This was modified at |1.4f| and the produced value
is now the rounded quotient by \dtt{1e-D} (i.e. no trailing zeros nor
scientific exponent in the output).

If truncation rather than rounding is needed on can use |\xintexpr
trunc(...)\relax| for truncation to an integer or |\xintexpr
trunc(...,D)\relax| for quantization to an integer multiple or \dtt{1eD} (if
|D>0|, for |D<0| the analog would be |trunc((...)/1e-D)|).  But
this works only for a single scalar value.


\subsection{\csh{xintiiexpr}, \csh{xinttheiiexpr}}
\label{xintiiexpr}\label{xinttheiiexpr}\label{thexintiiexpr}

This variant\etype{x} does not know fractions. It deals almost only with long
integers. Comma separated lists of expressions are allowed.

\begin{framed}
  It maps |/| to the \emph{rounded} quotient. The operator
  |//| is, like in |\xintexpr...\relax|, mapped to \emph{truncated} division.
  The Euclidean quotient (which for positive operands is like the truncated
  quotient) was, prior to release |1.1|, associated to |/|. The function
  |quo(a,b)| can still be employed.
\end{framed}

The \csbxint{iiexpr}-essions use the `ii' macros for addition, subtraction,
multiplication, power, square, sums, products, Euclidean quotient and
remainder.

The |round|, |trunc|, |floor|, |ceil| functions are still available, and are
about the only places where fractions can be used, but |/| within, if not
somehow hidden will be executed as integer rounded division. To avoid this one
can wrap the input in \dtt{qfrac}: this means however that none of the normal
expression parsing will be executed on the argument.

To understand the illustrative examples, recall that |round| and |trunc| have
a second (non negative) optional argument. In a normal \csbxint{expr}-essions,
|round| and |trunc| are mapped to \csbxint{Round} and \csbxint{Trunc}, in
\csbxint{iiexpr}-essions, they are mapped to \csbxint{iRound} and
\csbxint{iTrunc}.


\begin{everbatim*}
\xinttheiiexpr 5/3, round(5/3,3), trunc(5/3,3), trunc(\xintDiv {5}{3},3),
trunc(\xintRaw {5/3},3)\relax{} are problematic, but
%
\xinttheiiexpr 5/3,  round(qfrac(5/3),3), trunc(qfrac(5/3),3), floor(qfrac(5/3)),
ceil(qfrac(5/3))\relax{} work!
\end{everbatim*}

On the other hand decimal numbers and scientific numbers can be used directly
as arguments to the |num|, |round|, or any function producing an integer.

\begin{framed}
  Scientific numbers will be
  represented with as many zeroes as necessary, thus one does not want to
  insert \dtt{num(1e100000)} for example in an \csa{xintiiexpr}ession!
\end{framed}

%
\begin{everbatim*}
\xinttheiiexpr num(13.4567e3)+num(10000123e-3)\relax % should (num truncates) compute 13456+10000
\end{everbatim*}
%

The |reduce| function is not available and will raise an error. The |frac|
function also. The |sqrt| function is mapped to \csbxint{iiSqrt} which gives
a truncated square root. The |sqrtr| function is mapped to \csbxint{iiSqrtR}
which gives a rounded square root.

One can use the Float macros if one is careful to use |num|, or |round|
etc\dots on their output.

\begin{everbatim*}
\xinttheiiexpr \xintFloatSqrt [20]{2}, \xintFloatSqrt [20]{3}\relax % no operations

\noindent The next example requires the |round|, and one could not put the |+| inside it:

\xinttheiiexpr round(\xintFloatSqrt [20]{2},19)+round(\xintFloatSqrt [20]{3},19)\relax

(the second argument of |round| and |trunc| tells how many digits from after the
decimal mark one should keep.)
\end{everbatim*}

The whole point of \csbxint{iiexpr} is to gain some speed in
\emph{integer-only} algorithms, and the above explanations related to how to
nevertheless use fractions therein are a bit peripheral. We observed
(2013/12/18) of the order of $30$\% speed gain when dealing with numbers with
circa one hundred digits (1.2: this info may be obsolete).


\subsection{\csh{xintboolexpr}, \csh{xinttheboolexpr}}
\label{xintboolexpr}\label{xinttheboolexpr}\label{thexintboolexpr}

Equivalent\etype{x} to doing |\xintexpr ...\relax| and returning $True$ if the
result does not vanish, and $False$ if the result is zero. As |\xintexpr|, this
can be used on comma separated lists of expressions, and even bracketed lists.
\CHANGED{1.4}

It can be customized, one only needs to modify the following:
\begin{everbatim}
\def\xintboolexprPrintOne#1{\xintiiifNotZero{#1}{True}{False}}%
\end{everbatim}
Not only are |True| and |False| usable in input, also |true| and |false| are
pre-declared variables.

Maybe obsolete:

There is slight quirk in case it is used as a sub-expression: the boolean
expression needs at least one logic operation else the value is not
standardized to |1| or |0|, for example we get from
\begin{everbatim*}
\xinttheexpr \xintboolexpr 1.23\relax\relax\newline
\end{everbatim*}which is to be compared with
\begin{everbatim*}
\xinttheboolexpr 1.23\relax
\end{everbatim*}

% A related issue existed with
% |\xinttheexpr \xintiexpr 1.23\relax\relax|, which was fixed with |1.1|
% release, and I decided back then not to add the needed overhead also to the
% |\xintboolexpr| context, as one only needs to use |?(1.23)| for example or
% involve the |1.23| in any logic operation like |1.23 'and' 3.45|, or involve
% the |\xintboolexpr ..\relax | itself with any logical operation, contrarily to
% the sub-|\xintiexpr| case where |\xinttheexpr 1+\xintiexpr 1.23\relax\relax|
% did behave contrarily to expectations until |1.1|.


\subsection{\csh{xintfloatexpr},
  \csh{xintthefloatexpr}}
\label{xintfloatexpr}\label{xintthefloatexpr}\label{thexintfloatexpr}

\csbxint{floatexpr}|...\relax|\etype{x} is exactly like |\xintexpr...\relax|
but with the four binary operations and the power function are mapped to
\csa{xintFloatAdd}, \csa{xintFloatSub}, \csa{xintFloatMul}, \csa{xintFloatDiv}
and \csa{xintFloatPower}, respectively.\footnote{Since |1.2f| the \string^
  handles half-integer exponents, contrarily to \csa{xintFloatPower}.}

The target precision for the computation is from the
current setting of |\xintDigits|. Comma separated lists of expressions are
allowed.

An optional parameter within brackets is allowed:
\begin{itemize}
\item if positive it instructs the macro to round the result to that many
  digits of precision. It thus makes sense to employ it only if this parameter is
  less than the \csbxint{theDigits} precision.
\item if negative it means to trim off that many digits (of course, in the
  sense of rounding
  the values to shorter mantissas). Don't use it to trim all digits (or more than all)!
\end{itemize}

Since |1.2f| all float operations first round their arguments; a parsed number
is not rounded prior to its use as operand to such a float operation.

|\thexintfloatexpr| is synonym to |\xintthefloatexpr|.

|\xintDigits:=36\relax|\xintDigits:=36\relax 
%
\leftedline{|\xintthefloatexpr
  (1/13+1/121)*(1/179-1/173)/(1/19-1/18)\relax|}
%
\leftedline{\dtt{\xintthefloatexpr
  (1/13+1/121)*(1/179-1/173)/(1/19-1/18)\relax}}
% 0.00564487459334466559166166079096852897
%
\leftedline{|\xintthefloatexpr\xintexpr
  (1/13+1/121)*(1/179-1/173)/(1/19-1/18)\relax\relax|}
%
\leftedline{\dtt{\xintthefloatexpr\xintexpr
  (1/13+1/121)*(1/179-1/173)/(1/19-1/18)\relax\relax}}

\xintDigits := 16;

The latter is the rounding of the exact result. The former one has
its last three digits wrong due to the cumulative effect of rounding errors
in the intermediate computations, as compared to exact evaluations.




I recall here from \autoref{ssec:floatingpoint} that with release |1.2f| the
float macros for addition, subtraction, multiplication and division round
their arguments first to |P| significant places with |P| the asked-for
precision of the output; and similarly the power macros and the
square root macro. This does not modify anything for computations with
arguments having at most |P| significant places already.

\subsection{\csh{xinteval}, \csh{xintieval}, \csh{xintiieval},
  \csh{xintfloateval}}
\label{xinteval}\label{xintieval}\label{xintiieval}\label{xintfloateval}

\csbxint{eval}\etype{x} is an \fexpan dable macro which is basically defined
like this (DON'T BELIEVE THIS; it has been entirely revamped at |1.4|):
\begin{everbatim}
\def\xinteval#1{\romannumeral-`0\xinttheexpr#1\relax}% OLD DEFINITION < 1.4
\end{everbatim}
thus expands in two steps (its exact definition differs from the one given
above in order to achieve a slight optimization).
\begin{everbatim*}
\xinteval{add(x^2, x = 100..110), add(x^3, x = 100..110)}
\end{everbatim*}

\csbxint{ieval}\etype{x} is similarly related to \csbxint{theiexpr}. Its optional
argument must be located inside the braces:
\begin{everbatim*}
\xintieval{[7] 355/113}
\end{everbatim*}

\csbxint{iieval}\etype{x} is similarly related to \csbxint{theiiexpr}.
\begin{everbatim*}
\xintiieval{add(x^2, x = 100..110), add(x^3, x = 100..110)}
\end{everbatim*}

\csbxint{floateval}\etype{x} is similarly related to \csbxint{thefloatexpr}. Its optional
argument must be located inside the braces:
\begin{everbatim*}
\xintfloateval{[7] 355/113}
\end{everbatim*}

When negative it tells how many digits to remove from the prevailing precision
(\csbxint{theDigits}):
\begin{everbatim*}
\xintfloateval{[-2] 355/113} has \xinttheDigits\ minus 2 digits.
\end{everbatim*}

These macros are useful when one uses some extra wrapper doing some parsing of
its input, like the |\num| macro of
\href{http://ctan.org/pkg/siunitx}{siunitx}, which would choke on some of the
syntax elements allowed inside \csb{xintexpr}|...\relax| (for example
brackets).
As shown in the above examples, these macros, like the underlying parsers
accept arbitrarily many comma separated expressions.


\subsection{Using an expression parser within another one}

This was already illustrated before. In the following:
\begin{everbatim*}
\xintfloatexpr \xintexpr add(1/i, i=1234..1243)\relax ^100\relax
\end{everbatim*},
the inner sum is computed exactly. Then it will be rounded to |\xinttheDigits|
significant digits, and then its power will be evaluated as a float operation.
One should avoid the "|\xintthe|" parsers in inner positions as this induces
digit by digit parsing of the inner computation result by the outer parser.
Here is the same computation done with floats all the way:
\begin{everbatim*}
\xintfloatexpr add(1/i, i=1234..1243)^100\relax
\end{everbatim*}

Not surprisingly this differs from the previous one which was exact until
raising to the |100|th power.

The fact that the inner expression occurs inside a bigger one has nil
influence on its behaviour. There is the limitation though that the outputs
from \csbxint{expr} and \csbxint{floatexpr} can not be used directly in
\csbxint{theiiexpr} integer-only parser. But one can do:
\begin{everbatim*}
\xintiiexpr round(\xintfloatexpr 3.14^10\relax)\relax % or trunc
\end{everbatim*}


\subsection{The \csh{xintthecoords} macro}
\label{xintthecoords}

It converts (in two expansion steps) the expansion result of
\csbxint{floatexpr} (or \csbxint{expr} or \csbxint{iiexpr}) into the |(a, b)
(c, d) ...| format for list of coordinates as expected by the |TikZ|
|coordinates| syntax.%
\begin{everbatim*}
\begin{figure}[htbp]
\centering\begin{tikzpicture}[scale=10]\xintDigits:=8\relax 
  \clip (-1.1,-.25) rectangle (.3,.25);
  \draw [blue] (-1.1,0)--(1,0);
  \draw [blue] (0,-1)--(0,+1);
  \draw [red] plot[smooth] coordinates {%
%%% (\xintthecoords converts output of next expression into (x1, y1) (x2, y2)... format)
    \xintthecoords\xintfloatexpr
%%%    This syntax -1+[0..4]/2 is currenty dropped at xint 1.4
%%%    seq((x^2-1,mul(x-t,t=-1+[0..4]/2)),x=-1.2..[0.1]..+1.2)\relax
%%%    Use this:
    seq((x^2-1,mul(x-t,t=seq(-1+u/2, u=0..4))),x=-1.2..[0.1]..+1.2)
    \relax
     };
\end{tikzpicture}
\caption{Coordinates with \cs{xintthecoords}.}
\end{figure}
\end{everbatim*}

% Notice: if x goes not take exactly value 1 or -1, the origin appears slightly
% off the curve, not MY fault!!!
It is currently undecided how \csa{xintthecoords} should handle bracketed
data.\UNSTABLE{} Currently, it (or |TikZ|) will break it the input contains
nested structures. One can use it with \func{flat} which removes all nesting.
And in combination with \func{zip} it is easy to plot data given by some
mechanism in separate
lists of x- and y-coordinates (see an example in next section)

\subsection{The \csh{xintthespaceseparated} macro}
\label{xintthespaceseparated}

It converts (in two expansion steps)\NewWith{1.4a} the expansion result of
\csbxint{floatexpr} (or \csbxint{expr} or \csbxint{iiexpr}) into the space
separated format suitable for usage with |PS-Tricks| |\listplot| macro.

Here is for example some syntax (the replacement text of |\foo|, which is used
here only to show that indeed complete expansion is attained in two steps)
which can be used as argument to |\listplot|. Using 4 fractional decimal
digits is sufficient when unit is the centimeter (it gives a fixed point
precision of one micron, amply enough for plots...).
\begin{everbatim*}
\oodef\foo{%
\xintthespaceseparated\xintiexpr[4]\xintfloatexpr seq((i, log10(i)), i=1..[0.5]..10)\relax\relax
}\meaning\foo
\end{everbatim*}

Here we don't really need the inner |\xintfloatexpr...\relax| because the
\func{log10} function works the same in the exact parser |\xintexpr| but in
general this is recommended.

It is currently undecided how \csa{xintthespaceseparated} should handle
bracketed data.\UNSTABLE{} Currently, it (or |\listplot|) will break if the
input contains nested structures. One can use it with \func{flat} which
removes all nesting. And in combination with \func{zip} it is easy to plot
data given by some mechanism in separate lists of x- and y-coordinates.

\begin{everbatim*}
% let's imagine we have something like this
\def\Xcoordinates{1, 3, 5, 7, 9}
\def\Ycoordinates{1, 9, 25, 49, 81}
% then:
|\xintthespaceseparated\xintexpr flat(zip([\Xcoordinates], [\Ycoordinates]))\relax|
is suitable to use as argument to |\listplot|, as it expands to
\xintthespaceseparated\xintexpr flat(zip([\Xcoordinates], [\Ycoordinates]))\relax
\end{everbatim*}

\subsection{\csh{xintifboolexpr}, \csh{xintifboolfloatexpr}, \csh{xintifbooliiexpr}}
\label{xintifboolexpr}
\label{xintifboolfloatexpr}
\label{xintifbooliiexpr}

\csh{xintifboolexpr}\marg{expr}\marg{YES}\marg{NO}\etype{xnn} does
\csbxint{theexpr}<expr>|\relax| and then executes the \meta{YES} or the
\meta{NO} branch depending on whether the outcome was non-zero or zero. Thus
one can read \emph{if bool expr} as meaning \emph{if not zero}:
\centeredline{if \meta{expr}-ession does not vanish do \meta{YES} else do
  \meta{NO}}

The expression is not limited to using only comparison operators and Boolean
logic (|<|, |>|, |==|, |!=|, |&&|, \verb+||+, \func{all}, \func{any},
\func{xor}, \func{bool}, \func{togl}, ...), it can be the most general
computation.

\csh{xintifboolfloatexpr}\marg{expr}\marg{YES}\marg{NO}\etype{xnn} does
\csbxint{thefloatexpr}\meta{expr}|\relax| and then executes the \meta{YES} or the
\meta{NO} branch depending on whether the outcome was non zero or zero.

\csh{xintifbooliiexpr}\marg{expr}\marg{YES}\marg{NO}\etype{xnn} does
\csbxint{theiiexpr}\meta{expr}|\relax| and then executes the \meta{YES} or the
\meta{NO} branch depending on whether the outcome was non zero or zero.

The expression argument must be a single one, comma separated sub-expressions
will cause low-level errors.

\subsection{\csh{xintifsgnexpr}, \csh{xintifsgnfloatexpr}, \csh{xintifsgniiexpr}}
\label{xintifsgnexpr}
\label{xintifsgnfloatexpr}
\label{xintifsgniiexpr}

\csh{xintifsgnexpr}\marg{expr}\marg{<0}\marg{=0}\marg{>0}\etype{xnnn} evaluates
the \csbxint{expr}ession and chooses the branch corresponding to its sign.

\csh{xintifsgnfloatexpr}\marg{expr}\marg{<0}\marg{=0}\marg{>0}\etype{xnnn} evaluates
the \csbxint{floatexpr}ession and chooses the branch corresponding to its sign.

\csh{xintifsgniiexpr}\marg{expr}\marg{<0}\marg{=0}\marg{>0}\etype{xnnn} evaluates
the \csbxint{iiexpr}ession and chooses the branch corresponding to its sign.

The expression argument must be a single one, comma separated sub-expressions
will cause low-level errors.

\subsection{The \csh{xintNewExpr}, \csh{xintNewIIExpr},
  \csh{xintNewFloatExpr}, \csh{xintNewIExpr},  and \csh{xintNewBoolExpr}
  macros}
\label{xintNewExpr}
\label{xintNewIIExpr}
\label{xintNewFloatExpr}
\label{xintNewIExpr}
\label{xintNewBoolExpr}

\csbxint{NewExpr} macro is used as:
%
\leftedline{|\xintNewExpr{\myformula}[n]|\marg{stuff}, where}
\begin{itemize}
\item \meta{stuff} will be inserted inside |\xinttheexpr . . . \relax|,
\item |n| is an integer between zero and nine, inclusive, which is the number
  of parameters of |\myformula|,
\item the placeholders |#1|, |#2|, ..., |#n| are used inside \meta{stuff} in
  their usual r\^ole,%
%
\catcode`# 12
\footnote{if \csa{xintNewExpr} is used inside a macro,
    the |#|'s must be doubled as usual.}
  \footnote{the |#|'s will in pratice have their usual
    catcode, but  category code other |#|'s are accepted too.}
\catcode`# 6
%
\item the |[n]| is \emph{mandatory}, even for |n=0|.%
\footnote{there is some use for \csa{xintNewExpr}|[0]| compared to an
    \csa{edef} as \csa{xintNewExpr} has some built-in catcode protection.}
\item the macro |\myformula| is defined without checking if it already exists,
  \LaTeX{} users might prefer to do first |\newcommand*\myformula {}| to get a
  reasonable error message in case |\myformula| already exists,
\item the protection against active characters is done automatically (as long
  as the whole thing has not already been fetched as a macro argument and
  the catcodes correspondingly already frozen).
\end{itemize}

It (if it succeeds) will be a completely expandable macro entirely built-up using |\xintAdd|,
|\xintSub|, |\xintMul|, |\xintDiv|, |\xintPow|, etc\dots as corresponds to the
expression written with the infix operators.
Macros created by |\xintNewExpr| can thus be nested.

\begin{everbatim*}
    \xintNewFloatExpr \FA [2]{(#1+#2)^10}
    \xintNewFloatExpr \FB [2]{sqrt(#1*#2)}
\begin{enumerate}[nosep]
    \item \FA {5}{5}
    \item \FB {30}{10}
    \item \FA {\FB {30}{10}}{\FB {40}{20}}
\end{enumerate}
\end{everbatim*}

The documentation is much shortened here because \csbxint{NewExpr} and \csbxint{deffunc}
are very much related one with the other.

\begin{framed}
  ATTENTION!

  The original spirit of \csbxint{NewExpr} was to define a (possibly very big)
  macro using only \xintfracname, and this means in particular that it must be
  used only with arguments compatible with the \xintfracname input
  format.\IMPORTANTf

  Thus an |\xintexpr| declared variable has no chance to work, it must be
  wrapped explicitly in |\xinteval{...}| to be fetched as argument to a macro
  constructed by \csbxint{NewExpr}.
\end{framed}

They share essentially the same limitations.

Notice though that \csbxint{NewFloatExpr} accepts and recognizes the optional
argument |[Q]| of \csbxint{floatexpr}, contrarily to \csbxint{deffloatfunc}.
Use an |\empty| in case the contents are not known in advance.


Historical note: prior to |1.4|, \xintexprname used a |\csname..\endcsname|
encapsulation technique which impacted the string pool memory. The
\csbxint{NewExpr} was designed as a method to pre-parse the expression and
produce one single, gigantic, nested usage of the relevant \xintfracname
macros. This way, only those macros were expanded which had nil impact on the
\TeX{} string pool.

Later on it was found that this mechanism could be employed to define
functions. Basically underneath |98%| of \csbxint{NewExpr} and
\csbxint{deffunc} are using the same shared code.

\xintDigits:= 16\relax 

\subsection{Analogies and differences of \csh{xintiiexpr} with \csh{numexpr}}

\csbxint{iiexpr}|..\relax| is a parser of expressions knowing only (big)
integers. There are, besides the enlarged range of allowable inputs, some
important differences of syntax between |\numexpr| and |\xintiiexpr| and
variants:
\begin{itemize}
\item Contrarily to |\numexpr|, the |\xintiiexpr| parser will stop expanding
  only after having encountered (and swallowed) a \emph{mandatory} |\relax|
  token.
\item In particular, spaces between digits (and not only around infix
  operators or parentheses) do not stop |\xintiiexpr|, contrarily to the
  situation with |numexpr|: |\the\numexpr 7 + 3 5\relax| expands (in one
  step)%
%
\footnote {The |\numexpr| triggers continued expansion after the space
    following the |3| to check if some operator like |+| is upstream. But
    after having found the |5| it treats it as and end-marker.}
%
  to \dtt{\detokenize\expandafter{\the\numexpr 7 + 3 5\relax}\unskip}, whereas
  |\xintthe\xintiiexpr 7 + 3 5\relax| expands (in two steps) to
  \dtt{\detokenize\expandafter\expandafter\expandafter {\xintthe\xintiiexpr 7
      + 3 5\relax}}.%
%
\footnote {Since |1.2l| one can also use the underscore |_| to separate digits
for readability of long numbers.}

\item Inside an |\edef|, an expression |\xintiiexpr...\relax| get fully
  evaluated, whereas |\numexpr| without |\the| or |\number| prefix would not,
  if not itself embedded in another |\the\numexpr| or similar context.
\item (ctd.) The private format to which |\xintiiexpr...\relax| (et al.)
  evaluates may use |\xintthe| prefix to turn into explicit digits,
  (for example in arguments to some macros which expand their arguments). The |\the| \TeX\ primitive prefix would
  not work here.
\item (ctd.) One can embed a |\numexpr...\relax| (with its |\relax|!) inside an
  |\xintiiexpr...\relax| without |\the| or |\number|, but the reverse situation
  requires usage of |\xintthe| or \csbxint{eval} user interface,
\item |\numexpr -(1)\relax| is illegal. In contrast |\xintiiexpr -(1)\relax| is
  perfectly legal and gives the expected result (what else ?).
\item |\numexpr 2\cnta\relax| is illegal (with |\cnta| a |\count| register.)
  In contrast
  |\xintiiexpr 2\cnta\relax| is perfectly legal and will do the tacit
  multiplication.
\item |\the\numexpr| or |\number\numexpr| expands in one step, but
  |\xintthe\xintiiexpr| or |\xinttheiiexpr| needs two steps.
\end{itemize}

\subsection{Chaining expressions for expandable algorithmics}
\label{ssec:fibonacci}

We will see in this section how to chain |\xintexpr|-essions with
|\expandafter|'s, like it is possible with |\numexpr|. For this it is
convenient to use |\romannumeral0\xintexpro| which is the once-expanded form of
|\xintexpr|, as we can then chain using only one |\expandafter| each time.

For example, here is the code employed
on the title page to compute (expandably, of course!) the 1250th Fibonacci
number:

\begin{everbatim*}
\catcode`_ 11
\def\Fibonacci #1{%  \Fibonacci{N} computes F(N) with F(0)=0, F(1)=1.
    \expandafter\Fibonacci_a\expandafter
        {\the\numexpr #1\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 1\expandafter\relax\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 1\expandafter\relax\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 1\expandafter\relax\expandafter}\expandafter
        {\romannumeral0\xintiiexpro 0\relax}}
%
\def\Fibonacci_a #1{%
    \ifcase #1
          \expandafter\Fibonacci_end_i
    \or
          \expandafter\Fibonacci_end_ii
    \else
          \ifodd #1
              \expandafter\expandafter\expandafter\Fibonacci_b_ii
          \else
              \expandafter\expandafter\expandafter\Fibonacci_b_i
          \fi
    \fi {#1}%
}% * signs are omitted from the next macros, tacit multiplications
\def\Fibonacci_b_i #1#2#3{\expandafter\Fibonacci_a\expandafter
  {\the\numexpr #1/2\expandafter}\expandafter
  {\romannumeral0\xintiiexpro sqr(#2)+sqr(#3)\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro (2#2-#3)#3\relax}%
}% end of Fibonacci_b_i
\def\Fibonacci_b_ii #1#2#3#4#5{\expandafter\Fibonacci_a\expandafter
  {\the\numexpr (#1-1)/2\expandafter}\expandafter
  {\romannumeral0\xintiiexpro sqr(#2)+sqr(#3)\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro (2#2-#3)#3\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro #2#4+#3#5\expandafter\relax\expandafter}\expandafter
  {\romannumeral0\xintiiexpro #2#5+#3(#4-#5)\relax}%
}% end of Fibonacci_b_ii
%         code as used on title page:
%\def\Fibonacci_end_i  #1#2#3#4#5{\xintthe#5}
%\def\Fibonacci_end_ii #1#2#3#4#5{\xinttheiiexpr #2#5+#3(#4-#5)\relax}
%         new definitions:
\def\Fibonacci_end_i  #1#2#3#4#5{{#4}{#5}}% {F(N+1)}{F(N)} in \xintexpr format
\def\Fibonacci_end_ii #1#2#3#4#5%
    {\expandafter
     {\romannumeral0\xintiiexpro #2#4+#3#5\expandafter\relax
      \expandafter}\expandafter
     {\romannumeral0\xintiiexpro #2#5+#3(#4-#5)\relax}}% idem.
% \FibonacciN returns F(N) (in encapsulated format: needs \xintthe for printing)
\def\FibonacciN {\expandafter\xint_secondoftwo\romannumeral-`0\Fibonacci }%
\catcode`_ 8
\end{everbatim*}


The macro |\Fibonacci| produces not one specific value |F(N)| but a pair of
successive values |{F(N)}{F(N+1)}| which can then serve as starting point of
another routine devoted to compute a whole sequence |F(N), F(N+1),
F(N+2),....|. Each of |F(N)| and |F(N+1)| is kept in the encapsulated internal
\xintexprname format.

|\FibonacciN| produces the single |F(N)|. It also keeps it in the private
format; thus printing it will need the |\xintthe| prefix.

\begingroup\footnotesize\sffamily\baselineskip 10pt
Here a code snippet which
checks the routine via a \string\message\ of the first $51$ Fibonacci
numbers (this is not an efficient way to generate a sequence of such
numbers, it is only for validating \csa{FibonacciN}).
%
\begin{everbatim}
\def\Fibo #1.{\xintthe\FibonacciN {#1}}%
\message{\xintiloop [0+1] \expandafter\Fibo\xintiloopindex.,
                          \ifnum\xintiloopindex<49 \repeat \xintthe\FibonacciN{50}.}
\end{everbatim}
\endgroup

The way we use |\expandafter|'s to chain successive |\xintiiexpro| evaluations
is exactly analogous to what is possible with |\numexpr|. The various
|\romannumeral0\xintiiexpro| could very well all have been |\xintiiexpr|'s but
then we would have needed |\expandafter\expandafter\expandafter| each
time.

\begin{framed}
  There is a difference though: |\numexpr| does \emph{NOT} expand inside an
  |\edef|, and to force its expansion we must prefix it with |\the| or
  |\number| or |\romannumeral| or another |\numexpr| which is itself prefixed,
  etc\dots.

  But |\xintexpr|, |\xintiexpr|, ..., expand fully in an |\edef|, with the
  completely expanded
  result encapsulated in a private format.

  Using |\xintthe| as prefix is necessary to print the result (like |\the| or
  |\number| in the case of |\numexpr|), but it is not necessary to get the
  computation done (contrarily to the situation with |\numexpr|).
\end{framed}


Our |\Fibonacci| expands completely under \fexpan sion, so we can use
\hyperref[fdef]{\ttfamily\char92fdef} rather than |\edef| in a situation such
as
%
\leftedline {|\fdef \X {\FibonacciN {100}}|}
%
but it is usually about as efficient to employ |\edef|. And if we want
%
\leftedline{|\edef \Y {(\FibonacciN{100},\FibonacciN{200})}|,}
%
then |\edef| is necessary.

Allright, so let's now give the code to generate |{F(N)}{F(N+1)}{F(N+2)}...|,
using |\Fibonacci| for the first two and then using the standard recursion
|F(N+2)=F(N+1)+F(N)|:

\catcode`_ 11
\def\FibonacciSeq #1#2{%#1=starting index, #2>#1=ending index
    \expandafter\Fibonacci_Seq\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2-1}%
}%
\def\Fibonacci_Seq #1#2{%
     \expandafter\Fibonacci_Seq_loop\expandafter
                {\the\numexpr #1\expandafter}\romannumeral0\Fibonacci {#1}{#2}%
}%
\def\Fibonacci_Seq_loop #1#2#3#4{% standard Fibonacci recursion
    {#3}\unless\ifnum #1<#4  \Fibonacci_Seq_end\fi
        \expandafter\Fibonacci_Seq_loop\expandafter
        {\the\numexpr #1+1\expandafter}\expandafter
        {\romannumeral0\xintiiexpro #2+#3\relax}{#2}{#4}%
}%
\def\Fibonacci_Seq_end\fi\expandafter\Fibonacci_Seq_loop\expandafter
    #1\expandafter #2#3#4{\fi {#3}}%
\catcode`_ 8

\begingroup\footnotesize\baselineskip10pt
\everb|@
\catcode`_ 11
\def\FibonacciSeq #1#2{%#1=starting index, #2>#1=ending index
    \expandafter\Fibonacci_Seq\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2-1}%
}%
\def\Fibonacci_Seq #1#2{%
     \expandafter\Fibonacci_Seq_loop\expandafter
                {\the\numexpr #1\expandafter}\romannumeral0\Fibonacci {#1}{#2}%
}%
\def\Fibonacci_Seq_loop #1#2#3#4{% standard Fibonacci recursion
    {#3}\unless\ifnum #1<#4  \Fibonacci_Seq_end\fi
        \expandafter\Fibonacci_Seq_loop\expandafter
        {\the\numexpr #1+1\expandafter}\expandafter
        {\romannumeral0\xintiiexpro #2+#3\relax}{#2}{#4}%
}%
\def\Fibonacci_Seq_end\fi\expandafter\Fibonacci_Seq_loop\expandafter
    #1\expandafter #2#3#4{\fi {#3}}%
\catcode`_ 8
|
\endgroup

This |\FibonacciSeq| macro is
completely expandable but it is not \fexpan dable.

This is not a problem in the next example which uses \csbxint{For*} as the
latter applies repeatedly full expansion to what comes next each time it
fetches an item from its list argument. Thus \csbxint{For*} still manages to
generate the list via iterated full expansion.


\begin{figure*}[ht!]
  \phantomsection\label{fibonacci}
  \newcounter{myindex}
  \fdef\Fibxxx{\FibonacciN {30}}%
  \setcounter{myindex}{30}%
\centeredline{\tabskip 1ex
\vbox{\halign{\bfseries#.\hfil&#\hfil &\hfil #\cr
  \xintFor* #1 in {\FibonacciSeq {30}{59}}\do
  {\themyindex &\xintthe#1 &
    \xintiiRem{\xintthe#1}{\xintthe\Fibxxx}\stepcounter{myindex}\cr }}%
}\vrule
\vbox{\halign{\bfseries#.\hfil&#\hfil &\hfil #\cr
  \xintFor* #1 in {\FibonacciSeq {60}{89}}\do
  {\themyindex &\xintthe#1 &
    \xintiiRem{\xintthe#1}{\xintthe\Fibxxx}\stepcounter{myindex}\cr }}%
}\vrule
\vbox{\halign{\bfseries#.\hfil&#\hfil &\hfil #\cr
  \xintFor* #1 in {\FibonacciSeq {90}{119}}\do
  {\themyindex &\xintthe#1 &
   \xintiiRem{\xintthe#1}{\xintthe\Fibxxx}\stepcounter{myindex}\cr }}%
}}%
%
\centeredline{Some Fibonacci numbers together with their residues modulo
  |F(30)|\dtt{=\xintthe\Fibxxx}}
\end{figure*}

\begingroup\footnotesize\baselineskip10pt
\everb|@
\newcounter{myindex}% not "index", which would overwrite theindex environment!
% (many have probably been bitten by this trap)
\tabskip 1ex
  \fdef\Fibxxx{\FibonacciN {30}}%
  \setcounter{myindex}{30}%
\vbox{\halign{\bfseries#.\hfil&#\hfil &\hfil #\cr
  \xintFor* #1 in {\FibonacciSeq {30}{59}}\do
  {\themyindex &\xintthe#1 &
    \xintiiRem{\xintthe#1}{\xintthe\Fibxxx}\stepcounter{myindex}\cr }}%
}\vrule
\vbox{\halign{\bfseries#.\hfil&#\hfil &\hfil #\cr
  \xintFor* #1 in {\FibonacciSeq {60}{89}}\do
  {\themyindex &\xintthe#1 &
    \xintiiRem{\xintthe#1}{\xintthe\Fibxxx}\stepcounter{myindex}\cr }}%
}\vrule
\vbox{\halign{\bfseries#.\hfil&#\hfil &\hfil #\cr
  \xintFor* #1 in {\FibonacciSeq {90}{119}}\do
  {\themyindex &\xintthe#1 &
   \xintiiRem{\xintthe#1}{\xintthe\Fibxxx}\stepcounter{myindex}\cr }}%
}%
|
\endgroup

This produces the Fibonacci numbers from |F(30)| to |F(119)|, and
computes also  all the
congruence classes modulo |F(30)|.  The output has
been put in a \hyperref[fibonacci]{float}, which appears
\vpageref[above]{fibonacci}. I leave to the mathematically inclined
readers the task to explain the visible patterns\dots |;-)|.

\subsection{When expandability is too much}

Let's use the macros of \autoref{ssec:fibonacci} related to Fibonacci numbers.
Notice that the $47$th Fibonacci number is \dtt{\xintthe\FibonacciN {47}} thus
already too big for \TeX{} and \eTeX{}.


The |\FibonacciN| macro found in \autoref{ssec:fibonacci} is completely
expandable, it is even \fexpan dable. We need a wrapper with |\xintthe|
prefix
\begin{everbatim*}
\def\theFibonacciN{\xintthe\FibonacciN}
\end{everbatim*}
to print in the document or to use within |\message| (or \LaTeX\ |typeout|) to
write to the log and terminal.

\begingroup
  \def\A {1859}  \def\B {1573}
  \edef\X {\theFibonacciN\A}  \edef\Y {\theFibonacciN\B}
  \edef\GCDAB {\xintiiGCD\A\B}\edef\Z {\theFibonacciN\GCDAB}
  \edef\GCDXY{\xintiiGCD\X\Y}

  The |\xintthe| prefix also allows its use it as argument to the \xintname
  macros: for example if we are interested in knowing how many digits
  $F(1250)$ has, it suffices to issue |\xintLen {\theFibonacciN {1250}}|
  (which expands to \dtt{\xintLen {\theFibonacciN {1250}}}). Or if we want to
  check the formula $gcd(F(1859),F(1573))=F(gcd(1859,1573))=F(143)$, we only
  need%
%
\footnote{The
  \csa{xintiiGCD} macro is provided by both the \xintgcdname package (since
  |1.0|) and by the \xintname package (since |1.3d|).}
%
\begin{everbatim}
$\xintiiGCD{\theFibonacciN{1859}}{\theFibonacciN{1573}}=%
 \theFibonacciN{\xintiiGCD{1859}{1573}}$
\end{everbatim}
%
which produces:
%
\leftedline{$\dtt{\xintiiGCD{\X}{\Y}}=\dtt{\theFibonacciN{\GCDAB}}$}

The |\theFibonacciN| macro expanded its |\xintiiGCD{1859}{1573}| argument via the
services of |\numexpr|: this step allows only things obeying the \TeX{} bound,
naturally! (but \dtt{F(\xintiiPow2{31}}) would be rather big anyhow...).

This is very convenient but of course it repeats the complete evaluation each
time it is done. In practice, it is often useful to store the result of such
evaluations in macros. Any |\edef| will break expandability, but if the goal
is at some point to print something to the |dvi| or |pdf| output, and not only
to the |log| file, then expandability has to be broken one day or another!

Hence, in practice, if we want to print in the document some computation
results, we can proceed like this and avoid having to repeat identical
evaluations:
\begin{everbatim}
\begingroup
  \def\A {1859}  \def\B {1573}
  \edef\X {\theFibonacciN\A}  \edef\Y {\theFibonacciN\B}
  \edef\GCDAB {\xintiiGCD\A\B}\edef\Z {\theFibonacciN\GCDAB}
  \edef\GCDXY{\xintiiGCD\X\Y}
The identity $\gcd(F(\A),F(\B))=F(\gcd(\A,\B))$ can be checked via evaluation
of both sides: $\gcd(F(\A),F(\B))=\gcd(\printnumber\X,\printnumber\Y)=
\printnumber{\GCDXY} = F(\gcd(\A,\B)) = F(\GCDAB) =\printnumber\Z$.\par
          % some further computations involving \A, \B, \X, \Y
\endgroup % closing the group removes assignments to \A, \B, ...
% or choose longer names less susceptible to overwrite something.
% Note: there is no LaTeX \newecommand which would be to \edef like \newcommand is to \def
\end{everbatim}
The identity $\gcd(F(\A),F(\B))=F(\gcd(\A,\B))$ can be checked via evaluation
of both sides: $\gcd(F(\A),F(\B))=\gcd(\printnumber\X,\printnumber\Y)=
\printnumber{\GCDXY} = F(\gcd(\A,\B)) = F(\GCDAB) =\printnumber\Z$.\par
\endgroup

One may legitimately ask the author: why expandability
to such extremes, for things such as big fractions or floating point numbers
(even continued fractions...) which anyhow can not be used directly within
\TeX's primitives such as |\ifnum|? Why insist on a concept
which is foreign to the vast majority of \TeX\ users and even programmers?

I have no answer: it made definitely sense at the start of \xintname (see
\autoref{ssec:origins}) and once started I could not stop.


\subsection{Acknowledgements (2013/05/25)}

I was greatly helped in my preparatory thinking, prior to producing such an
expandable parser, by the commented source of the
\href{https://ctan.org/pkg/l3kernel}{l3fp} package, specifically the
|l3fp-parse.dtx| file (in the version of April-May 2013; I think there was in
particular a text called ``roadmap'' which was helpful). Also the source of the
|calc| package was instructive, despite the fact that here for |\xintexpr| the
principles are necessarily different due to the aim of achieving expandability.

\clearpage
\expandafter\let\csname xintexprnameUp\endcsname\undefined
\csname ExamplesnameUp\endcsname
\section {More examples with \xinttoolsname or \xintexprname or both}
\RaisedLabel{sec:examples}

Note: \xintexprname.sty automatically loads \xinttoolsname.sty.

The examples given here start to feel dated and are currently in need of some
rewrite to better illustrate newer features of the package.

 
\localtableofcontents

\subsection{More examples with dummy variables}
\label{ssec:moredummies}

These examples were first added to this manual at the time of the |1.1|
release (|2014/10/29|).

\begin{everbatim*}
Prime numbers are always cool
\xinttheiiexpr seq((seq((subs((x/:m)?{(m*m>x)?{1}{0}}{-1},m=2n+1))
                        ??{break(0)}{omit}{break(1)},n=1++))?{x}{omit},
               x=10001..[2]..10200)\relax
\end{everbatim*}

The syntax in this last example may look a bit involved (...\@ and it is so I
admit). First |x/:m| computes |x modulo m| (this is the modulo with respect to
floored division). The |(x)?{yes}{no}| construct checks if |x| (which
\emph{must} be within parentheses) is true or false, i.e.\@ non zero or zero. It
then executes either the |yes| or the |no| branch, the non chosen branch is
\emph{not} evaluated. Thus if |m| divides |x| we are in the second (``false'')
branch. This gives a |-1|. This |-1| is the argument to a |??| branch which is
of the type |(y)??{y<0}{y=0}{y>0}|, thus here the |y<0|, i.e.\@, |break(0)| is
chosen. This |0| is thus given to another |?| which consequently chooses
|omit|, hence the number is not kept in the list. The numbers which survive
are the prime numbers.

\begin{everbatim*}
The first Fibonacci number beyond |2^64|, |2^64|, and the index are respectively
\xinttheiiexpr subs(iterr(0,1;(@1>N)?{break(@1,N,i)}{@1+@2},i=1++),N=2^64)\relax.
\end{everbatim*}

% A006877 In the `3x+1' problem, these values for the starting value set new
% records for number of steps to reach 1. (Formerly M0748) 14 1, 2, 3, 6, 7,
% 9, 18, 25, 27, 54, 73, 97, 129, 171, 231, 313, 327, 649, 703, 871, 1161,
% 2223, 2463, 2919, 3711, 6171, 10971, 13255, 17647, 23529, 26623, 34239,
% 35655, 52527, 77031, 106239, 142587, 156159, 216367, 230631, 410011, 511935,
% 626331, 837799

One more recursion:
\begin{everbatim*}
\def\syr #1{\xinttheiiexpr rseq(#1; (@<=1)?{break(i)}{odd(@)?{3@+1}{@//2}},i=0++)\relax}
The 3x+1 problem: \syr{231}\par
\end{everbatim*}

OK, a final one:%
\begin{everbatim*}
\def\syrMax #1{\xintiiexpr iterr(#1,#1;even(i)?
                                       {(@2<=1)?{break(@1,i//2)}{odd(@2)?{3@2+1}{@2//2}}}
                                       {(@1>@2)?{@1}{@2}},i=0++)\relax }
With initial value 1161, the maximal intermediate value and the number of steps
needed to reach 1 are respectively \syrMax{1161}.\par
\end{everbatim*}

Look at the
  \hyperlink{BrentSalamin}{Brent-Salamin algorithm implementation} for a more
  interesting recursion.

% \begin{everbatim*}
% \newcommand\Factors [1]{\xinttheiiexpr
%     subs(seq((i/:3=1)?{omit}{[L][i]},i=0..len(L)-1),
%     L=rseq(#1;(p^2>[@][0])?{([@][0]>1)?{break(1,[@][0],1)}{abort}}
%                           {(([@][0])/:p)?{omit}
%     {iter(([@][0])//p; (@/:p)?{break(@,p,e)}{@//p},e=1++)}},p=2++))\relax }
% \Factors {41^4*59^2*29^3*13^5*17^8*29^2*59^4*37^6}
% \end{everbatim*}

% This might look a bit scary, I admit.%
% %
% \footnote{Look at the
%   \hyperlink{BrentSalamin}{Brent-Salamin algorithm implementation} for a much
%   saner example.} 
% %

% \xintexprname has minimal tools and
% is obstinate about doing everything expandably! We are hampered by absence of a
% notion of ``nuple''. The algorithm divides |N| by |2| until no more possible,
% then by |3|, then by |4| (which is silly), then by |5|, then by |6| (silly
% again), \dots.

% The variable |L=rseq(#1;...)| expands, if one follows the steps, to a comma
% separated list starting with the initial (evaluated) |N=#1| and then
% pseudo-triplets where the first item is |N| trimmed of small primes, the
% second item is the last prime divisor found, and the third item is its
% exponent in original |N|.

% The algorithm needs to keep handy the last computed quotient by prime powers,
% hence all of them, but at the very end it will be cleaner to get rid of them
% (this corresponds to the first line in the code above). This is achieved in a
% cumbersome inefficient way; indeed each item extraction |[L][i]| is costly: it
% is not like accessing an array stored in memory, due to expandability, nothing
% can be stored in memory! Nevertheless, this step could be done here in a far
% less inefficient manner if there was a variant of |seq| which, in the spirit
% of \csbxint{iloopindex}, would know how many steps it had been through so far.
% This is a feature to be added to |\xintexpr|! (as well as a |++| construct
% allowing a non unit step).

% Notice that in |iter(([@][0])//p;| the |@| refers to the previous triplet (or
% in the first step to |N|), but the latter |@| showing up in |(@/:p)?| refers
% to the previous value computed by |iter|.

% \begin{snugframed}
%   Parentheses are essential in |..([y][0])| else the parser will see |..[| and
%   end up in ultimate confusion, and also in |([@][0])/:p| else the parser will
%   see the itemwise operator |]/| on lists and again be very confused (I could
%   implement a |]/:| on lists, but in this situation this would also be very
%   confusing to the parser.)
% \end{snugframed}

% See \autoref{ssec:factorize} for a routine |\Factorize| written directly with
% \xintname macros. Last time I checked |\Factors| was about seven times slower
% than |\Factorize| in test cases such as
% |16246355912554185673266068721806243461403654781833| and others. Among the
% various things explaining the speed difference, there is fact that the
% |\Factorize| algorithm step by increments of two, not one, and also it divides
% only once, obtaining quotient and remainder in one go. These two things
% already make for a speed-up factor of about four. Thus, |\Factors| is not
% completely inefficient in comparison, and was quite easier to come up with
% than |\Factorize|.

\subsection{Completely expandable prime test}
\label{ssec:primesI}

Let us now construct a completely expandable macro which returns $1$ if its
given input is prime and $0$ if not:
\everb|@
\def\remainder #1#2{\the\numexpr #1-(#1/#2)*#2\relax }
\def\IsPrime #1%
 {\xintANDof {\xintApply {\remainder {#1}}{\xintSeq {2}{\xintiiSqrt{#1}}}}}
|

This uses \csbxint{iiSqrt} and assumes its input is at least $5$. Rather than
\xintname's own \csbxint{iiRem} we used a quicker |\numexpr| expression as we
are dealing with short integers. Also we used \csbxint{ANDof} which will
return $1$ only if all the items are non-zero. The macro is a bit
silly with an even input, ok, let's enhance it to detect an even input:
\everb|@
\def\IsPrime #1%
   {\xintiiifOdd {#1}
        {\xintANDof % odd case
            {\xintApply {\remainder {#1}}
                        {\xintSeq [2]{3}{\xintiiSqrt{#1}}}%
            }%
        }
        {\xintifEq {#1}{2}{1}{0}}%
   }
|

We used the \xintname expandable tests (on big integers or fractions)
in order for |\IsPrime| to be \fexpan dable.

Our integers are short, but without |\expandafter|'s with
|\@firstoftwo|, % @ n'est plus actif dans le dtx 1.1 !
or some other related techniques,
direct use of |\ifnum..\fi| tests is dangerous. So to make the macro more
efficient we are going to use the expandable tests provided by the package
\href{http://ctan.org/pkg/etoolbox}{etoolbox}%
%
\footnote{\url{http://ctan.org/pkg/etoolbox}}.
%
The macro becomes:
%
\everb|@
\def\IsPrime #1%
   {\ifnumodd {#1}
    {\xintANDof % odd case
     {\xintApply {\remainder {#1}}{\xintSeq [2]{3}{\xintiiSqrt{#1}}}}}
    {\ifnumequal {#1}{2}{1}{0}}}
|

In the odd case however we have to assume the integer is at least $7$, as
|\xintSeq| generates an empty list if |#1=3| or |5|, and |\xintANDof| returns
$1$ when supplied an empty list. Let us ease up a bit |\xintANDof|'s work by
letting it work on only $0$'s and $1$'s. We could use:
%
\everb|@
\def\IsNotDivisibleBy #1#2%
  {\ifnum\numexpr #1-(#1/#2)*#2=0 \expandafter 0\else \expandafter1\fi}
|
\noindent
where the |\expandafter|'s are crucial for this macro to be \fexpan dable and
hence work within the applied \csbxint{ANDof}. Anyhow, now that we have loaded
\href{http://ctan.org/pkg/etoolbox}{etoolbox}, we might as well use:
%
\everb|@
\newcommand{\IsNotDivisibleBy}[2]{\ifnumequal{#1-(#1/#2)*#2}{0}{0}{1}}
|
\noindent
Let us enhance our prime macro to work also on the small primes:
\everb|@
\newcommand{\IsPrime}[1] % returns 1 if #1 is prime, and 0 if not
  {\ifnumodd {#1}
    {\ifnumless {#1}{8}
      {\ifnumequal{#1}{1}{0}{1}}% 3,5,7 are primes
      {\xintANDof
         {\xintApply
        { \IsNotDivisibleBy {#1}}{\xintSeq [2]{3}{\xintiiSqrt{#1}}}}%
        }}% END OF THE ODD BRANCH
    {\ifnumequal {#1}{2}{1}{0}}% EVEN BRANCH
}
|

The input is still assumed positive. There is a deliberate blank before
\csa{IsNotDivisibleBy} to use this feature of \csbxint{Apply}: a space stops the
expansion of the applied macro (and disappears). This expansion will be done by
\csbxint{ANDof}, which has been designed to skip everything as soon as it finds
a false (i.e.\@ zero) input. This way, the efficiency is considerably improved.

We did generate via the \csbxint{Seq} too many potential divisors though. Later
sections give two variants: one with \csbxint{iloop} (\autoref{ssec:primesII})
which is still expandable and another one (\autoref{ssec:primesIII}) which is a
close variant of the |\IsPrime| code above but with the \csbxint{For} loop, thus
breaking expandability. The \hyperref[ssec:primesII]{xintiloop variant} does not
first evaluate the integer square root, the \hyperref[ssec:primesIII]{xintFor
  variant} still does. I did not compare their efficiencies.


Let us construct with this expandable primality test a table of the prime
numbers up to $1000$. We need to count how many we have in order to know how
many tab stops one shoud add in the last row.%
%
\footnote{although a tabular row may have less tabs than in the
  preamble, there is a problem with the \char`\|\space\space vertical
  rule, if one does that.}
%
There is some subtlety for this
last row. Turns out to be better to insert a |\\| only when we know for sure we
are starting a new row; this is how we have designed the |\OneCell| macro. And
for the last row, there are many ways, we use again |\xintApplyUnbraced| but
with a macro which gobbles its argument and replaces it with a tabulation
character. The \csbxint{For*} macro would be more elegant here.
%
\everb?@
\newcounter{primecount}
\newcounter{cellcount}
\newcommand{\NbOfColumns}{13}
\newcommand{\OneCell}[1]{%
    \ifnumequal{\IsPrime{#1}}{1}
     {\stepcounter{primecount}
      \ifnumequal{\value{cellcount}}{\NbOfColumns}
       {\\\setcounter{cellcount}{1}#1}
       {&\stepcounter{cellcount}#1}%
     } % was prime
  {}% not a prime, nothing to do
}
\newcommand{\OneTab}[1]{&}
\begin{tabular}{|*{\NbOfColumns}{r}|}
\hline
2  \setcounter{cellcount}{1}\setcounter{primecount}{1}%
   \xintApplyUnbraced \OneCell {\xintSeq [2]{3}{999}}%
   \xintApplyUnbraced \OneTab
      {\xintSeq [1]{1}{\the\numexpr\NbOfColumns-\value{cellcount}\relax}}%
    \\
\hline
\end{tabular}
There are \arabic{primecount} prime numbers up to 1000.
?

The table has been put in \hyperref[primesupto1000]{float} which appears
\vpageref{primesupto1000}.
We had to be careful to use in the last row \csbxint{Seq} with its optional
argument |[1]| so as to not generate a decreasing sequence from |1| to |0|, but
really an empty sequence in case the row turns out to already have all its
cells (which doesn't happen here but would with a number of columns dividing
$168$).
%
\newcommand{\IsNotDivisibleBy}[2]{\ifnumequal{#1-(#1/#2)*#2}{0}{0}{1}}

\newcommand{\IsPrime}[1]
   {\ifnumodd {#1}
        {\ifnumless {#1}{8}
          {\ifnumequal{#1}{1}{0}{1}}% 3,5,7 are primes
          {\xintANDof
             {\xintApply
                { \IsNotDivisibleBy {#1}}{\xintSeq [2]{3}{\xintiiSqrt{#1}}}}%
            }}% END OF THE ODD BRANCH
        {\ifnumequal {#1}{2}{1}{0}}% EVEN BRANCH
}

\newcounter{primecount}
\newcounter{cellcount}
\newcommand{\NbOfColumns}{13}
\newcommand{\OneCell}[1]
     {\ifnumequal{\IsPrime{#1}}{1}
        {\stepcounter{primecount}
         \ifnumequal{\value{cellcount}}{\NbOfColumns}
            {\\\setcounter{cellcount}{1}#1}
            {&\stepcounter{cellcount}#1}%
        } % was prime
        {}% not a prime nothing to do
}
\newcommand{\OneTab}[1]{&}
\begin{figure*}[ht!]
  \centering
  \phantomsection\label{primesupto1000}
  \begin{tabular}{|*{\NbOfColumns}{r}|}
    \hline
    2\setcounter{cellcount}{1}\setcounter{primecount}{1}%
    \xintApplyUnbraced \OneCell {\xintSeq [2]{3}{999}}%
    \xintApplyUnbraced \OneTab
    {\xintSeq [1]{1}{\the\numexpr\NbOfColumns-\value{cellcount}\relax}}%
    \\
    \hline
  \end{tabular}
\smallskip
\centeredline{There are \arabic{primecount} prime numbers up to 1000.}
\end{figure*}

\subsection{Another completely expandable prime test}
\label{ssec:primesII}

The |\IsPrime| macro from \autoref{ssec:primesI} checked expandably if a (short)
integer was prime, here is a partial rewrite using \csbxint{iloop}. We use the
|etoolbox| expandable conditionals for convenience, but not everywhere as
|\xintiloopindex| can not be evaluated while being braced. This is also the
reason why |\xintbreakiloopanddo| is delimited, and the next macro
|\SmallestFactor| which returns the smallest prime factor examplifies that. One
could write more efficient completely expandable routines, the aim here was only
to illustrate use of the general purpose \csbxint{iloop}. A little table giving
the first values of |\SmallestFactor| follows, its coding uses \csbxint{For},
which is described later; none of this uses count registers.
%


\begin{everbatim*}
\let\IsPrime\undefined \let\SmallestFactor\undefined % clean up possible previous mess
\newcommand{\IsPrime}[1] % returns 1 if #1 is prime, and 0 if not
  {\ifnumodd {#1}
    {\ifnumless {#1}{8}
      {\ifnumequal{#1}{1}{0}{1}}% 3,5,7 are primes
      {\if
       \xintiloop [3+2]
       \ifnum#1<\numexpr\xintiloopindex*\xintiloopindex\relax
           \expandafter\xintbreakiloopanddo\expandafter1\expandafter.%
       \fi
       \ifnum#1=\numexpr (#1/\xintiloopindex)*\xintiloopindex\relax
       \else
       \repeat 00\expandafter0\else\expandafter1\fi
      }%
    }% END OF THE ODD BRANCH
    {\ifnumequal {#1}{2}{1}{0}}% EVEN BRANCH
}%
\catcode`_ 11
\newcommand{\SmallestFactor}[1] % returns the smallest prime factor of #1>1
  {\ifnumodd {#1}
    {\ifnumless {#1}{8}
      {#1}% 3,5,7 are primes
      {\xintiloop [3+2]
       \ifnum#1<\numexpr\xintiloopindex*\xintiloopindex\relax
           \xint_afterfi{\xintbreakiloopanddo#1.}%
       \fi
       \ifnum#1=\numexpr (#1/\xintiloopindex)*\xintiloopindex\relax
           \xint_afterfi{\expandafter\xintbreakiloopanddo\xintiloopindex.}%
       \fi
       \iftrue\repeat
      }%
     }% END OF THE ODD BRANCH
   {2}% EVEN BRANCH
}%
\catcode`_ 8
{\centering
  \begin{tabular}{|c|*{10}c|}
    \hline
    \xintFor #1 in {0,1,2,3,4,5,6,7,8,9}\do {&\bfseries #1}\\
    \hline
    \bfseries 0&--&--&2&3&2&5&2&7&2&3\\
    \xintFor #1 in {1,2,3,4,5,6,7,8,9}\do
    {\bfseries #1%
      \xintFor #2 in {0,1,2,3,4,5,6,7,8,9}\do
      {&\SmallestFactor{#1#2}}\\}%
    \hline
  \end{tabular}\par
}
\end{everbatim*}

\subsection{Miller-Rabin Pseudo-Primality expandably}
\label{ssec:PrimesIV}


% At the time of writing, the code at the link above is still the version from
% April 2016 and it needed some hacks to get recursive (pseudo)-functions
% defined. Since |1.2h| of |2016/11/20| there is \csbxint{NewFunction} which
% allows us here to avoid such internal hacking.

% And since |1.3| of |2018/03/01|, it is possible to use \csbxint{defiifunc}
% also for recursive definitions, so we use it here, but we can benefit from it
% only for modular exponentiation as the rest of the code uses |iter| or |break|
% statements which are not yet compatible with \csbxint{defiifunc}.

The |isPseudoPrime(n)| is usable in \csbxint{iiexpr}-essions and establishes
if its (positive) argument is a Miller-Rabin PseudoPrime to the bases $2, 3,
5, 7, 11, 13, 17$. If this is true and $n<341550071728321$ (which has 15
digits) then $n$ really is a prime number.

Similarly $n=3825123056546413051$ (19 digits) is the smallest composite number
which is a strong pseudo prime for bases $2, 3, 5, 7, 11, 13, 17, 19$ and
$23$. It is easy to extend the code below to include these additional tests
(we could make the list of tested bases an argument too, now that I think
about it.)

For more information see
  \centeredline{\url{https://en.wikipedia.org/wiki/Miller%E2%80%93Rabin_primality_test#Deterministic_variants_of_the_test}}
  and
\centeredline{\url{http://primes.utm.edu/prove/prove2_3.html}}

In particular, according to \textsc{Jaeschke} \emph{On strong pseudoprimes to
  several bases,} Math. Comp., 61 (1993) 915-926, if $n < 4,759,123,141$ it is
enough to establish Rabin-Miller pseudo-primality to bases $a = 2, 7, 61$ to
prove that $n$ is prime. This range is enough for \TeX\ numbers and we could
then write a very fast expandable primality test for such numbers using only
|\numexpr|. Left as an exercise\dots

\begin{everbatim*}
% I -------------------------------- Modular Exponentiation
% Computes x^m modulo n (with m non negative).
% We will always use it with 1 < x < n
%
% With xint 1.4 we should use ? and ?? (although in the case at hand ifsgn()
% and if() would be ok but I should not say that).
%
\xintdefiifunc powmod_a(x, m, n) :=
       isone(m)?
           % m=1, return x modulo n
           {   x /: n  }
           % m > 1 test if odd or even and do recursive call
           {   odd(m)? {  x*sqr(powmod_a(x, m//2, n)) /: n }
                       {    sqr(powmod_a(x, m//2, n)) /: n }
            }
         ;
\xintdefiifunc powmod(x, m, n) := (m)?{powmod_a(x, m, n)}{1};

%% Syntax used before xint 1.4:
% \xintdefiifunc powmod_a(x, m, n) :=
%        ifone(m,
%            % m=1, return x modulo n
%               x /: n,
%            % m > 1 test if odd or even and do recursive call
%               if(odd(m), (x*sqr(powmod_a(x, m//2, n))) /: n,
%                             sqr(powmod_a(x, m//2, n))  /: n
%                 )
%          );
% \xintdefiifunc powmod(x, m, n) := if(m, powmod_a(x, m, n), 1);

% II ------------------------------ Miller-Rabin compositeness witness

% n=2^k m + 1 with m odd and k at least 1

% Choose 1<x<n.
% compute y=x^m modulo n
% if equals 1 we can't say anything
% if equals n-1 we can't say anything
% else put j=1, and
% compute repeatedly the square, incrementing j by 1 each time,
% thus always we have y^{2^{j-1}}
%   -> if at some point n-1 mod n found, we can't say anything and break out
%   -> if however we never find n-1 mod n before reaching
%        z=y^{2^{k-1}} with j=k
%        we then have z^2=x^{n-1}.
    % Suppose z is not -1 mod n. If z^2 is 1 mod n, then n can be prime only if
    % z is 1 mod n, and we can go back up, until initial y, and we have already
    % excluded y=1. Thus if z is not -1 mod n and z^2 is 1 then n is not prime.
    % But if z^2 is not 1, then n is not prime by Fermat. Hence (z not -1 mod n)
    % implies (n is composite). (Miller test)

% let's use again xintexpr indecipherable (except to author) syntax. Of course
% doing it with macros only would be faster.

% Here \xintdefiifunc is not usable because not compatible with iter, break, ... 
% but \xintNewFunction comes to the rescue.

\xintNewFunction{isCompositeWitness}[4]{% x=#1, n=#2, m=#3, k=#4
   subs((y==1)?{0}
         {iter(y;(j==#4)?{break(!(@==#2-1))}
                        {(@==#2-1)?{break(0)}{sqr(@)/:#2}},j=1++)}
         ,y=powmod(#1,#3,#2))}

% added note (2018/03/07) it is possible in the above that m=#3 is never
% zero, so we should rather call powmod_a for a small gain, but I don't
% have time to re-read the code comments and settle this.
 
% III ------------------------------------- Strong Pseudo Primes

% cf
%  http://oeis.org/A014233
%     <http://mathworld.wolfram.com/Rabin-MillerStrongPseudoprimeTest.html>
%     <http://mathworld.wolfram.com/StrongPseudoprime.html>

% check if positive integer <49 si a prime.
% 2,3,5,7,11,13,17,19,23,29,31,37,41,43,47
\def\IsVerySmallPrime #1%
    {\ifnum#1=1 \xintdothis0\fi
     \ifnum#1=2 \xintdothis1\fi
     \ifnum#1=3 \xintdothis1\fi
     \ifnum#1=5 \xintdothis1\fi
     \ifnum#1=\numexpr (#1/2)*2\relax\xintdothis0\fi
     \ifnum#1=\numexpr (#1/3)*3\relax\xintdothis0\fi
     \ifnum#1=\numexpr (#1/5)*5\relax\xintdothis0\fi
     \xintorthat 1}

\xintNewFunction{isPseudoPrime}[1]{% n = #1
     (#1<49)?% use ? syntax to evaluate only what is needed
% prior to 1.4 we had \xintthe#1 here but the actual tokens represented
% by this #1 when isPseudoPrime() function expands have changed and 
% the correct way is now \xintiieval{#1} to hand over explicit digits to
% the \IsVerySmallPrime macro.
       {\IsVerySmallPrime{\xintiieval{#1}}}
       {(even(#1))?
        {0}
        {subs(%
         % L expands to two values m, k hence isCompositeWitness does get
         % its four variables x, n, m, k
         isCompositeWitness(2, #1, L)?
          {0}%
          {isCompositeWitness(3, #1, L)?
           {0}%
           {isCompositeWitness(5, #1, L)?
            {0}%
            {isCompositeWitness(7, #1, L)?
             {0}%
% above enough for N<3215031751 hence all TeX numbers
             {isCompositeWitness(11, #1, L)?
              {0}%
% above enough for N<2152302898747, hence all 12-digits numbers
              {isCompositeWitness(13, #1, L)?
               {0}%
% above enough for N<3474749660383
               {isCompositeWitness(17, #1, L)?
                {0}%
% above enough for N<341550071728321
                {1}%
               }% not needed to comment-out end of lines spaces inside
              }%  \xintexpr but this is too much of a habit for me with TeX!
             }%   I left some after the ? characters.
            }%
           }%
          }% this computes (m, k) such that n = 2^k m + 1, m odd, k>=1
          , L=iter(#1//2;(even(@))?{@//2}{break(@,k)},k=1++))%
         }%
        }%
}

% if needed:
%\def\IsPseudoPrime #1{\xinttheiiexpr isPseudoPrime(#1)\relax}

\noindent The smallest prime number at least equal to 3141592653589 is
\xintiiexpr 
   seq(isPseudoPrime(3141592653589+n)?
                    {break(3141592653589+n)}{omit}, n=0++)\relax.
% we could not use 3141592653589++ syntax because it works only with TeX numbers
\par
\end{everbatim*}





\subsection{A table of factorizations}
\label{ssec:factorizationtable}

As one more example with \csbxint{iloop} let us use an alignment to display the
factorization of some numbers. The loop will actually only play a minor r\^ole
here, just handling the row index, the row contents being almost entirely
produced via a macro |\factorize|. The factorizing macro does not use
|\xintiloop| as it didn't appear to be the convenient tool. As |\factorize| will
have to be used on |\xintiloopindex|, it has been defined as a delimited macro.

To spare some fractions of a second in the compilation time of this document
(which has many many other things to do), \number"7FFFFFED{} and
\number"7FFFFFFF, which turn out to be prime numbers, are not given to
|factorize| but just typeset directly; this illustrates use of
\csbxint{iloopskiptonext}.

The code next generates a \hyperref[floatfactorize]{table} which has
been made into a float appearing \vpageref{floatfactorize}. Here is now
the code for factorization; the conditionals use the package provided
|\xint_firstoftwo| and |\xint_secondoftwo|, one could have employed
rather \LaTeX{}'s own |\@firstoftwo| and |\@secondoftwo|, or, simpler
still in \LaTeX{} context, the |\ifnumequal|, |\ifnumless| \dots,
utilities from the package |etoolbox| which do exactly that under the
hood. Only \TeX{} acceptable numbers are treated here, but it would be
easy to make a translation and use the \xintname macros, thus extending
the scope to big numbers; naturally up to a cost in speed.

The reason for some strange looking expressions is to avoid arithmetic overflow.

\begin{everbatim*}
\catcode`_ 11
\def\abortfactorize #1\xint_secondoftwo\fi #2#3{\fi}

\def\factorize #1.{\ifnum#1=1 \abortfactorize\fi
          \ifnum\numexpr #1-2=\numexpr ((#1/2)-1)*2\relax
               \expandafter\xint_firstoftwo
          \else\expandafter\xint_secondoftwo
          \fi
         {2&\expandafter\factorize\the\numexpr#1/2.}%
         {\factorize_b #1.3.}}%

\def\factorize_b #1.#2.{\ifnum#1=1 \abortfactorize\fi
         \ifnum\numexpr #1-(#2-1)*#2<#2
                 #1\abortfactorize
         \fi
         \ifnum \numexpr #1-#2=\numexpr ((#1/#2)-1)*#2\relax
              \expandafter\xint_firstoftwo
         \else\expandafter\xint_secondoftwo
         \fi
         {#2&\expandafter\factorize_b\the\numexpr#1/#2.#2.}%
         {\expandafter\factorize_b\the\numexpr #1\expandafter.%
                                  \the\numexpr #2+2.}}%
\catcode`_ 8
\begin{figure*}[ht!]
\centering\phantomsection\label{floatfactorize}\normalcolor
\tabskip1ex
\centeredline{\vbox{\halign {\hfil\strut#\hfil&&\hfil#\hfil\cr\noalign{\hrule}
         \xintiloop ["7FFFFFE0+1]
         \expandafter\bfseries\xintiloopindex &
         \ifnum\xintiloopindex="7FFFFFED
              \number"7FFFFFED\cr\noalign{\hrule}
         \expandafter\xintiloopskiptonext
         \fi
         \expandafter\factorize\xintiloopindex.\cr\noalign{\hrule}
         \ifnum\xintiloopindex<"7FFFFFFE
         \repeat
         \bfseries \number"7FFFFFFF&\number "7FFFFFFF\cr\noalign{\hrule}
}}}
\centeredline{A table of factorizations}
\end{figure*}
\end{everbatim*}

\subsection{Another table of primes}
\label{ssec:primesIII}

As a further example, let us dynamically generate a tabular with the first $50$
prime numbers after $12345$. First we need a macro to test if a (short) number
is prime. Such a completely expandable macro was given in \autoref{ssec:primesI},
here we consider a variant which will be slightly more efficient. This new
|\IsPrime| has two parameters. The first one is a macro which it redefines to
expand to the result of the primality test applied to the second argument. For
convenience we use the \href{http://ctan.org/pkg/etoolbox}{etoolbox} wrappers to
various |\ifnum| tests, although here there isn't anymore the constraint of
complete expandability (but using explicit |\if..\fi| in tabulars has its
quirks); equivalent tests are provided by \xintname, but they have some overhead
as they are able to deal with arbitrarily big integers.

\def\IsPrime #1#2%
{\edef\TheNumber {\the\numexpr #2}% positive integer
 \ifnumodd {\TheNumber}
 {\ifnumgreater {\TheNumber}{1}
  {\edef\ItsSquareRoot{\xintiiSqrt \TheNumber}%
    \xintFor ##1 in {\xintintegers [3+2]}\do
    {\ifnumgreater {##1}{\ItsSquareRoot}
               {\def#1{1}\xintBreakFor}
               {}%
     \ifnumequal {\TheNumber}{(\TheNumber/##1)*##1}
                 {\def#1{0}\xintBreakFor }
                 {}%
    }}
  {\def#1{0}}}% 1 is not prime
 {\ifnumequal {\TheNumber}{2}{\def#1{1}}{\def#1{0}}}%
}%

\everb|@
\def\IsPrime #1#2% """color[named]{PineGreen}#1=\Result, #2=tested number (assumed >0).;!
{\edef\TheNumber {\the\numexpr #2}%"""color[named]{PineGreen} hence #2 may be a count or \numexpr.;!
 \ifnumodd {\TheNumber}
 {\ifnumgreater {\TheNumber}{1}
  {\edef\ItsSquareRoot{\xintiiSqrt \TheNumber}%
    \xintFor """color{red}##1;! in {"""color{red}\xintintegers;! [3+2]}\do
    {\ifnumgreater {"""color{red}##1;!}{\ItsSquareRoot} """color[named]{PineGreen}% "textcolor{red}{##1} is a \numexpr.;!
               {\def#1{1}\xintBreakFor}
               {}%
     \ifnumequal {\TheNumber}{(\TheNumber/##1)*##1}
                 {\def#1{0}\xintBreakFor }
                 {}%
    }}
  {\def#1{0}}}% 1 is not prime
 {\ifnumequal {\TheNumber}{2}{\def#1{1}}{\def#1{0}}}%
}
|

As we used \csbxint{For} inside a macro we had to double the |#| in its |#1|
parameter. Here is now the code which creates the prime table (the table has
been put in a \hyperref[primes]{float}, which should be found on page
\pageref{primes}):

\everb?@
\newcounter{primecount}
\newcounter{cellcount}
\begin{figure*}[ht!]
  \centering
  \begin{tabular}{|*{7}c|}
  \hline
  \setcounter{primecount}{0}\setcounter{cellcount}{0}%
  \xintFor """color{red}#1;! in {"""color{red}\xintintegers;! [12345+2]} \do
"""color[named]{PineGreen}% "textcolor{red}{#1} is a \numexpr.;!
  {\IsPrime\Result{#1}%
   \ifnumgreater{\Result}{0}
   {\stepcounter{primecount}%
    \stepcounter{cellcount}%
    \ifnumequal {\value{cellcount}}{7}
       {"""color{red}\the#1;! \\\setcounter{cellcount}{0}}
       {"""color{red}\the#1;! &}}
   {}%
    \ifnumequal {\value{primecount}}{50}
     {\xintBreakForAndDo
      {\multicolumn {6}{l|}{These are the first 50 primes after 12345.}\\}}
     {}%
  }\hline
\end{tabular}
\end{figure*}
?

\begin{figure*}[ht!]
  \centering\phantomsection\label{primes}
  \begin{tabular}{|*{7}c|}
  \hline
  \setcounter{primecount}{0}\setcounter{cellcount}{0}%
  \xintFor #1 in {\xintintegers [12345+2]} \do
  {\IsPrime\Result{#1}%
   \ifnumgreater{\Result}{0}
   {\stepcounter{primecount}%
    \stepcounter{cellcount}%
    \ifnumequal {\value{cellcount}}{7}
       {\the#1 \\\setcounter{cellcount}{0}}
       {\the#1 &}}
   {}%
    \ifnumequal {\value{primecount}}{50}
     {\xintBreakForAndDo
      {\multicolumn {6}{l|}{These are the first 50 primes after 12345.}\\}}
     {}%
  }\hline
\end{tabular}
\end{figure*}

\subsection{Factorizing again}
\label{ssec:factorize}

Here is an \fexpan dable macro which computes the factors of an integer. It
uses the \xintname macros only.
\begin{everbatim*}
\catcode`\@ 11
\let\factorize\relax
\newcommand\Factorize [1]
      {\romannumeral0\expandafter\factorize\expandafter{\romannumeral-`0#1}}%
\newcommand\factorize [1]{\xintiiifOne{#1}{ 1}{\factors@a #1.{#1};}}%
\def\factors@a #1.{\xintiiifOdd{#1}
   {\factors@c 3.#1.}%
   {\expandafter\factors@b \expandafter1\expandafter.\romannumeral0\xinthalf{#1}.}}%
\def\factors@b #1.#2.{\xintiiifOne{#2}
   {\factors@end {2, #1}}%
   {\xintiiifOdd{#2}{\factors@c 3.#2.{2, #1}}%
                     {\expandafter\factors@b \the\numexpr #1+\@ne\expandafter.%
                         \romannumeral0\xinthalf{#2}.}}%
}%
\def\factors@c #1.#2.{%
    \expandafter\factors@d\romannumeral0\xintiidivision {#2}{#1}{#1}{#2}%
}%
\def\factors@d #1#2#3#4{\xintiiifNotZero{#2}
   {\xintiiifGt{#3}{#1}
        {\factors@end {#4, 1}}% ultimate quotient is a prime with power 1
        {\expandafter\factors@c\the\numexpr #3+\tw@.#4.}}%
   {\factors@e 1.#3.#1.}%
}%
\def\factors@e #1.#2.#3.{\xintiiifOne{#3}
   {\factors@end {#2, #1}}%
   {\expandafter\factors@f\romannumeral0\xintiidivision {#3}{#2}{#1}{#2}{#3}}%
}%
\def\factors@f #1#2#3#4#5{\xintiiifNotZero{#2}
   {\expandafter\factors@c\the\numexpr #4+\tw@.#5.{#4, #3}}%
   {\expandafter\factors@e\the\numexpr #3+\@ne.#4.#1.}%
}%
\def\factors@end #1;{\xintlistwithsep{, }{\xintRevWithBraces {#1}}}%
\catcode`@ 12
\end{everbatim*}
The macro will be acceptably efficient only with numbers having somewhat small
prime factors.
\begin{everbatim}
\Factorize{16246355912554185673266068721806243461403654781833}
\end{everbatim}
\begingroup\fdef\Z
{\Factorize{16246355912554185673266068721806243461403654781833}}
\noindent{\small\dtt{\Z}}


It puts a little stress on the input save stack in order
not be bothered with previously gathered things.\footnote{2015/11/18 I have
  not revisited this code for a long time, and perhaps I could improve it now
  with some new techniques.}

Its output is a comma separated list with the number first, then its prime
factors with multiplicity. Let's produce something prettier:
\begin{everbatim*}
\catcode`_ 11
\def\ShowFactors #1{\expandafter\ShowFactors_a\romannumeral-`0\Factorize{#1},\relax,\relax,}
\def\ShowFactors_a #1,{#1=\ShowFactors_b}
\def\ShowFactors_b #1,#2,{\if\relax#1\else#1^{#2}\expandafter\ShowFactors_b\fi}
\catcode`_ 8
\end{everbatim*}
\begin{everbatim}
$$\ShowFactors{16246355912554185673266068721806243461403654781833}$$
\end{everbatim}
$$\csname ShowFactors_a\expandafter\endcsname\Z,\relax,\relax,$$
\endgroup

If we only considered small integers, we could write pure |\numexpr| methods
which would be very much faster (especially if we had a table of small primes
prepared first) but still ridiculously slow compared to any non expandable
implementation, not to mention use of programming languages directly accessing
the CPU registers\dots

\subsection{The Quick Sort algorithm illustrated}\label{ssec:quicksort}

First a completely expandable macro which sorts a comma separated list of
numbers.%
%
\footnote{The code in earlier versions of this manual handled inputs composed
  of braced items. I have switched to comma separated inputs on the occasion
  of (link removed) 
  The version here is like
  |code 3| on 
  (link removed) (which is about |3x| faster
  than the earlier code it replaced in this manual) with a modification to
  make it more efficient if the data has many repeated values.

  A faster routine (for sorting hundreds of values) is provided as |code 6| at
  the link mentioned in the footnote, it is based on Merge Sort, but limited
  to inputs which one can handle as \TeX{} dimensions.%

  This |code 6| could be extended to handle more general numbers, as
  acceptable by \xintfracname. I have also written a non expandable version,
  which is even faster, but this matters really only when handling hundreds or
  rather thousands of values.}
%

The |\QSx| macro expands its list argument, which may thus be a macro; its
comma separated items must expand to integers or decimal numbers or fractions
or scientific notation as acceptable to \xintfracname, but if an item is
itself some (expandable) macro, this macro will be expanded each time the item
is considered in a comparison test! This is actually good if the macro expands
in one step to the digits, and there are many many digits, but bad if the macro
needs to do many computations. Thus |\QSx| should be used with either explicit
numbers or with items being macros expanding in one step to the numbers
(particularly if these numbers are very big).

If the interest is only in \TeX{} integers, then one should replace the
|\xintifCmp| macro with a suitable conditional, possibly helped by tools such as
|\ifnumgreater|, |\ifnumequal| and |\ifnumless| from
\href{http://ctan.org/pkg/etoolbox}{etoolbox} (\LaTeX{} only; I didn't see a
direct equivalent to |\xintifCmp|.) Or, if we are dealing with decimal numbers
with at most four+four digits, then one should use suitable |\ifdim| tests.
Naturally this will boost consequently the speed, from having skipped all the
overhead in parsing fractions and scientific numbers as are acceptable by
\xintfracname macros, and subsequent treatment.

\begin{everbatim*}
% THE QUICK SORT ALGORITHM EXPANDABLY
% \usepackage{xintfrac} in the preamble (latex)
\makeatletter
% use extra safe delimiters
\catcode`! 3 \catcode`? 3
\def\QSx {\romannumeral0\qsx }%
% first we check if empty list (else \qsx@finish will not find a comma)
\def\qsx   #1{\expandafter\qsx@a\romannumeral-`0#1,!,?}%
\def\qsx@a #1{\ifx,#1\expandafter\qsx@abort\else
                     \expandafter\qsx@start\fi #1}%
\def\qsx@abort #1?{ }%
\def\qsx@start {\expandafter\qsx@finish\romannumeral0\qsx@b,}%
\def\qsx@finish ,#1{ #1}%
%
% we check if empty of single and if not pick up the first as Pivot:
\def\qsx@b ,#1#2,#3{\ifx?#3\xintdothis\qsx@empty\fi
                    \ifx!#3\xintdothis\qsx@single\fi
                    \xintorthat\qsx@separate {#1#2}{}{}{#1#2}#3}%
\def\qsx@empty  #1#2#3#4#5{ }%
\def\qsx@single #1#2#3#4#5?{, #4}%
\def\qsx@separate #1#2#3#4#5#6,%
{%
    \ifx!#5\expandafter\qsx@separate@done\fi
    \xintifCmp {#5#6}{#4}%
          \qsx@separate@appendtosmaller
          \qsx@separate@appendtoequal
          \qsx@separate@appendtogreater {#5#6}{#1}{#2}{#3}{#4}%
}%
%
\def\qsx@separate@appendtoequal   #1#2{\qsx@separate {#2,#1}}%
\def\qsx@separate@appendtogreater #1#2#3{\qsx@separate {#2}{#3,#1}}%
\def\qsx@separate@appendtosmaller #1#2#3#4{\qsx@separate {#2}{#3}{#4,#1}}%
%
\def\qsx@separate@done\xintifCmp #1%
          \qsx@separate@appendtosmaller
          \qsx@separate@appendtoequal
          \qsx@separate@appendtogreater #2#3#4#5#6#7?%
{%
    \expandafter\qsx@f\expandafter {\romannumeral0\qsx@b #4,!,?}{\qsx@b #5,!,?}{#3}%
}%
%
\def\qsx@f #1#2#3{#2, #3#1}%
%
\catcode`! 12 \catcode`? 12
\makeatother

% EXAMPLE
\begingroup
\edef\z {\QSx {1.0, 0.5, 0.3, 1.5, 1.8, 2.0, 1.7, 0.4, 1.2, 1.4,
               1.3, 1.1, 0.7, 1.6, 0.6, 0.9, 0.8, 0.2, 0.1, 1.9}}
\meaning\z

\def\a {3.123456789123456789}\def\b {3.123456789123456788}
\def\c {3.123456789123456790}\def\d {3.123456789123456787}
\oodef\z {\QSx { \a, \b, \c, \d}}%
% The space before \a to let it not be expanded during the conversion from CSV
% values to List. The \oodef expands exactly twice (via a bunch of \expandafter's)
\meaning\z
\endgroup
\end{everbatim*} (the spaces after \string\d, etc...\@ come from the use of the
|\meaning| primitive.)

The choice of pivot as first element is bad if the list is already almost
sorted. Let's add a variant which will pick up the pivot index randomly. The
previous routine worked also internally with comma separated lists, but for a
change this one will use internally lists of braced items (the initial
conversion via \csbxint{CSVtoList} handles all potential spurious space
problems).

\unless\ifxetex % pour tester compilation de xint.dtx avec xetex qui n'a pas
                % \pdfuniformdeviate
\begin{everbatim*}
% QuickSort expandably on comma separated values with random choice of pivots
% ====> Requires availability of \pdfuniformdeviate <====
% \usepackage{xintfrac, xinttools} in preamble
\makeatletter
\def\QSx {\romannumeral0\qsx }% This is a f-expandable macro.
% This converts from comma separated values on input and back on output.
% **** NOTE: these steps (and the other ones too, actually) are costly if input
%            has thousands of items.
\def\qsx #1{\xintlistwithsep{, }%
            {\expandafter\qsx@sort@a\expandafter{\romannumeral0\xintcsvtolist{#1}}}}%
%
% we check if empty or single or double and if not pick up the first as Pivot:
\def\qsx@sort@a #1%
    {\expandafter\qsx@sort@b\expandafter{\romannumeral0\xintlength{#1}}{#1}}%
\def\qsx@sort@b #1{\ifcase #1
                      \expandafter\qsx@sort@empty
                      \or\expandafter\qsx@sort@single
                      \or\expandafter\qsx@sort@double
                      \else\expandafter\qsx@sort@c\fi {#1}}%
\def\qsx@sort@empty  #1#2{ }%
\def\qsx@sort@single #1#2{#2}%
\catcode`_ 11
\def\qsx@sort@double #1#2{\xintifGt #2{\xint_exchangetwo_keepbraces}{}#2}%
\catcode`_ 8
\def\qsx@sort@c      #1#2{%
    \expandafter\qsx@sort@sep@a\expandafter
                {\romannumeral0\xintnthelt{\pdfuniformdeviate #1+\@ne}{#2}}#2?}%
\def\qsx@sort@sep@a #1{\qsx@sort@sep@loop {}{}{}{#1}}%
\def\qsx@sort@sep@loop #1#2#3#4#5%
{%
    \ifx?#5\expandafter\qsx@sort@sep@done\fi
    \xintifCmp {#5}{#4}%
          \qsx@sort@sep@appendtosmaller
          \qsx@sort@sep@appendtoequal
          \qsx@sort@sep@appendtogreater {#5}{#1}{#2}{#3}{#4}%
}%
%
\def\qsx@sort@sep@appendtoequal   #1#2{\qsx@sort@sep@loop {#2{#1}}}%
\def\qsx@sort@sep@appendtogreater #1#2#3{\qsx@sort@sep@loop {#2}{#3{#1}}}%
\def\qsx@sort@sep@appendtosmaller #1#2#3#4{\qsx@sort@sep@loop {#2}{#3}{#4{#1}}}%
%
\def\qsx@sort@sep@done\xintifCmp #1%
          \qsx@sort@sep@appendtosmaller
          \qsx@sort@sep@appendtoequal
          \qsx@sort@sep@appendtogreater #2#3#4#5#6%
{%
    \expandafter\qsx@sort@recurse\expandafter
               {\romannumeral0\qsx@sort@a {#4}}{\qsx@sort@a {#5}}{#3}%
}%
%
\def\qsx@sort@recurse #1#2#3{#2#3#1}%
%
\makeatother

% EXAMPLES
\begingroup
\edef\z {\QSx {1.0, 0.5, 0.3, 1.5, 1.8, 2.0, 1.7, 0.4, 1.2, 1.4,
               1.3, 1.1, 0.7, 1.6, 0.6, 0.9, 0.8, 0.2, 0.1, 1.9}}
\meaning\z

\def\a {3.123456789123456789}\def\b {3.123456789123456788}
\def\c {3.123456789123456790}\def\d {3.123456789123456787}
\oodef\z {\QSx { \a, \b, \c, \d}}%
% The space before \a to let it not be expanded during the conversion from CSV
% values to List. The \oodef expands exactly twice (via a bunch of \expandafter's)
\meaning\z

\def\somenumbers{%
3997.6421, 8809.9358, 1805.4976, 5673.6478, 3179.1328, 1425.4503, 4417.7691,
2166.9040, 9279.7159, 3797.6992, 8057.1926, 2971.9166, 9372.2699, 9128.4052,
1228.0931, 3859.5459, 8561.7670, 2949.6929, 3512.1873, 1698.3952, 5282.9359,
1055.2154, 8760.8428, 7543.6015, 4934.4302, 7526.2729, 6246.0052, 9512.4667,
7423.1124, 5601.8436, 4433.5361, 9970.4849, 1519.3302, 7944.4953, 4910.7662,
3679.1515, 8167.6824, 2644.4325, 8239.4799, 4595.1908, 1560.2458, 6098.9677,
3116.3850, 9130.5298, 3236.2895, 3177.6830, 5373.1193, 5118.4922, 2743.8513,
8008.5975, 4189.2614, 1883.2764, 9090.9641, 2625.5400, 2899.3257, 9157.1094,
8048.4216, 3875.6233, 5684.3375, 8399.4277, 4528.5308, 6926.7729, 6941.6278,
9745.4137, 1875.1205, 2755.0443, 9161.1524, 9491.1593, 8857.3519, 4290.0451,
2382.4218, 3678.2963, 5647.0379, 1528.7301, 2627.8957, 9007.9860, 1988.5417,
2405.1911, 5065.8063, 5856.2141, 8989.8105, 9349.7840, 9970.3013, 8105.4062,
3041.7779, 5058.0480, 8165.0721, 9637.7196, 1795.0894, 7275.3838, 5997.0429,
7562.6481, 8084.0163, 3481.6319, 8078.8512, 2983.7624, 3925.4026, 4931.5812,
1323.1517, 6253.0945}%

\oodef\z {\QSx \somenumbers}% produced as a comma+space separated list
% black magic as workaround to the shrinkability of spaces in last line...
\hsize 87\fontcharwd\font`0
\lccode`~=32
\lowercase{\def~}{\discretionary{}{}{\kern\fontcharwd\font`0}}\catcode32 13
\noindent\phantom{00}\scantokens\expandafter{\meaning\z}\par
\endgroup
\end{everbatim*}
\fi % fin de si pas xetex


All the previous examples were with numbers which could have been handled via
|\ifdim| tests rather than the \csbxint{ifCmp} macro from \xintfracname; using
|\ifdim| tests would naturally be faster. Even faster routine is |code 6| at
(link removed) which uses |\pdfescapestring| and a
Merge Sort algorithm.

We then turn to a graphical illustration of the algorithm.%
%
\footnote{I have rewritten (2015/11/21) the routine to do only once (and not thrice) the
  needed calls to \csa{xintifCmp}, up to the price of one additional |\edef|,
  although due to the context execution time on our side is not an issue and
  moreover is anyhow overwhelmed by the TikZ's activities. Simultaneously I
  have updated the code.
  The
  variant with the choice of pivot on the right has more overhead: the reason
  is simply that we do not convert the data into an array, but maintain a list
  of tokens with self-reorganizing delimiters.}
%
For simplicity the pivot is always chosen as the first list item. Then we also
give a variant which picks up the last item as pivot.
\begin{everbatim*}
% in LaTeX preamble:
% \usepackage{xintfrac, xinttools}
% \usepackage{color}
% or, when using Plain TeX:
% \input xintfrac.sty \input xinttools.sty
% \input color.tex
%
% Color definitions
\definecolor{LEFT}{RGB}{216,195,88}
\definecolor{RIGHT}{RGB}{208,231,153}
\definecolor{INERT}{RGB}{199,200,194}
\definecolor{INERTpiv}{RGB}{237,237,237}
\definecolor{PIVOT}{RGB}{109,8,57}
% Start of macro defintions
\makeatletter
% \catcode`? 3 % a bit too paranoid. Normal ? will do.
%
% argument will never be empty
\def\QS@cmp@a    #1{\QS@cmp@b  #1??}%
\def\QS@cmp@b    #1{\noexpand\QS@sep@A\@ne{#1}\QS@cmp@d {#1}}%
\def\QS@cmp@d    #1#2{\ifx ?#2\expandafter\QS@cmp@done\fi
                      \xintifCmp {#1}{#2}\tw@\@ne\z@{#2}\QS@cmp@d {#1}}%
\def\QS@cmp@done #1?{?}%
%
\def\QS@sep@A #1?{\QSLr\QS@sep@L #1\thr@@?#1\thr@@?#1\thr@@?}%
\def\QS@sep@L #1#2{\ifcase #1{#2}\or\or\else\expandafter\QS@sep@I@start\fi \QS@sep@L}%
\def\QS@sep@I@start\QS@sep@L {\noexpand\empty?\QSIr\QS@sep@I}%
\def\QS@sep@I #1#2{\ifcase#1\or{#2}\or\else\expandafter\QS@sep@R@start\fi\QS@sep@I}%
\def\QS@sep@R@start\QS@sep@I {\noexpand\empty?\QSRr\QS@sep@R}%
\def\QS@sep@R #1#2{\ifcase#1\or\or{#2}\else\expandafter\QS@sep@done\fi\QS@sep@R}%
\def\QS@sep@done\QS@sep@R {\noexpand\empty?}%
%
\def\QS@loop {%
    \xintloop
    % pivot phase
    \def\QS@pivotcount{0}%
    \let\QSLr\DecoLEFTwithPivot  \let\QSIr \DecoINERT
    \let\QSRr\DecoRIGHTwithPivot \let\QSIrr\DecoINERT
    \centerline{\QS@list}%
    % sorting phase
    \ifnum\QS@pivotcount>\z@
            \def\QSLr {\QS@cmp@a}\def\QSRr {\QS@cmp@a}%
            \def\QSIr {\QSIrr}\let\QSIrr\relax
                \edef\QS@list{\QS@list}% compare
            \let\QSLr\relax\let\QSRr\relax\let\QSIr\relax
                \edef\QS@list{\QS@list}% separate
            \def\QSLr ##1##2?{\ifx\empty##1\else\noexpand \QSLr {{##1}##2}\fi}%
            \def\QSIr ##1##2?{\ifx\empty##1\else\noexpand \QSIr {{##1}##2}\fi}%
            \def\QSRr ##1##2?{\ifx\empty##1\else\noexpand \QSRr {{##1}##2}\fi}%
                \edef\QS@list{\QS@list}% gather
            \let\QSLr\DecoLEFT \let\QSRr\DecoRIGHT
            \let\QSIr\DecoINERTwithPivot \let\QSIrr\DecoINERT
            \centerline{\QS@list}%
    \repeat }%
%
% \xintFor* loops handle gracefully empty lists.
\def\DecoLEFT  #1{\xintFor* ##1 in {#1} \do {\colorbox{LEFT}{##1}}}%
\def\DecoINERT #1{\xintFor* ##1 in {#1} \do {\colorbox{INERT}{##1}}}%
\def\DecoRIGHT #1{\xintFor* ##1 in {#1} \do {\colorbox{RIGHT}{##1}}}%
\def\DecoPivot #1{\begingroup\color{PIVOT}\advance\fboxsep-\fboxrule\fbox{#1}\endgroup}%
%
\def\DecoLEFTwithPivot #1{\xdef\QS@pivotcount{\the\numexpr\QS@pivotcount+\@ne}%
    \xintFor* ##1 in {#1} \do
        {\xintifForFirst {\DecoPivot {##1}}{\colorbox{LEFT}{##1}}}}%
\def\DecoINERTwithPivot #1{\xdef\QS@pivotcount{\the\numexpr\QS@pivotcount+\@ne}%
    \xintFor* ##1 in {#1} \do
        {\xintifForFirst {\colorbox{INERTpiv}{##1}}{\colorbox{INERT}{##1}}}}%
\def\DecoRIGHTwithPivot #1{\xdef\QS@pivotcount{\the\numexpr\QS@pivotcount+\@ne}%
    \xintFor* ##1 in {#1}  \do
        {\xintifForFirst {\DecoPivot {##1}}{\colorbox{RIGHT}{##1}}}}%
%
\def\QuickSort #1{% warning: not compatible with empty #1.
    % initialize, doing conversion from comma separated values to a list of braced items
    \edef\QS@list{\noexpand\QSRr{\xintCSVtoList{#1}}}% many \edef's are to follow anyhow
% earlier I did a first drawing of the list, here with the color of RIGHT elements,
% but the color should have been for example white, anyway I drop this first line
    %\let\QSRr\DecoRIGHT
    %\par\centerline{\QS@list}%
%
    % loop as many times as needed
    \QS@loop }%
%
% \catcode`? 12 % in case we had used a funny ? as delimiter.
\makeatother
%% End of macro definitions.
%% Start of Example
\begingroup\offinterlineskip
\small
% \QuickSort {1.0, 0.5, 0.3, 1.5, 1.8, 2.0, 1.7, 0.4, 1.2, 1.4,
%                1.3, 1.1, 0.7, 1.6, 0.6, 0.9, 0.8, 0.2, 0.1, 1.9}
% \medskip
% with repeated values
\QuickSort {1.0, 0.5, 0.3, 0.8, 1.5, 1.8, 2.0, 1.7, 0.4, 1.2, 1.4,
               1.3, 1.1, 0.7, 0.3, 1.6, 0.6, 0.3, 0.8, 0.2, 0.8, 0.7, 1.2}
\endgroup
\end{everbatim*}

Here is the variant which always picks the pivot as the rightmost element.

\begin{everbatim*}
\makeatletter
%
\def\QS@cmp@a #1{\noexpand\QS@sep@A\expandafter\QS@cmp@d\expandafter
                 {\romannumeral0\xintnthelt{-1}{#1}}#1??}%
%
\def\DecoLEFTwithPivot #1{\xdef\QS@pivotcount{\the\numexpr\QS@pivotcount+\@ne}%
    \xintFor* ##1 in {#1} \do
        {\xintifForLast {\DecoPivot {##1}}{\colorbox{LEFT}{##1}}}}
\def\DecoINERTwithPivot #1{\xdef\QS@pivotcount{\the\numexpr\QS@pivotcount+\@ne}%
    \xintFor* ##1 in {#1} \do
        {\xintifForLast {\colorbox{INERTpiv}{##1}}{\colorbox{INERT}{##1}}}}
\def\DecoRIGHTwithPivot #1{\xdef\QS@pivotcount{\the\numexpr\QS@pivotcount+\@ne}%
    \xintFor* ##1 in {#1}  \do
        {\xintifForLast {\DecoPivot {##1}}{\colorbox{RIGHT}{##1}}}}
\def\QuickSort #1{%
    % initialize, doing conversion from comma separated values to a list of braced items
    \edef\QS@list{\noexpand\QSLr {\xintCSVtoList{#1}}}% many \edef's are to follow anyhow
    %
    % loop as many times as needed
    \QS@loop }%
\makeatother
\begingroup\offinterlineskip
\small
% \QuickSort {1.0, 0.5, 0.3, 1.5, 1.8, 2.0, 1.7, 0.4, 1.2, 1.4,
%                1.3, 1.1, 0.7, 1.6, 0.6, 0.9, 0.8, 0.2, 0.1, 1.9}
% \medskip
% with repeated values
\QuickSort {1.0, 0.5, 0.3, 0.8, 1.5, 1.8, 2.0, 1.7, 0.4, 1.2, 1.4,
               1.3, 1.1, 0.7, 0.3, 1.6, 0.6, 0.3, 0.8, 0.2, 0.8, 0.7, 1.2}
\endgroup
\end{everbatim*}

The choice of the first or last item as pivot is not a good one as nearly
ordered lists will take quadratic time. But for explaining the algorithm via a
graphical interpretation, it is not that bad. If one wanted to pick up the
pivot randomly, the routine would have to be substantially rewritten: in
particular the |\Deco..withPivot| macros need to know where the pivot is, and
currently this is implemented by using either |\xintifForFirst| or
|\xintifForLast|.

\clearpage

\expandafter\let\csname ExamplesnameUp\endcsname\undefined

\ifnum\NoSourceCode=1
\bigskip
\begin{framed}
  \small This documentation has been compiled without the source code,
  which is available in the separate file:
  %
  \centeredline{|sourcexint.pdf|,}
  %
  which will open in a PDF viewer via |texdoc sourcexint.pdf|.

  To produce a single file including both the user documentation and the
  source code:
  \begin{itemize}
  \item run |etex| on |xint.dtx| to generate |xint.tex| among other files,
  \item edit |xint.tex| to set the |\NoSourceCode| toggle within it to |0|,
  \item run |make clean| and then |make xint.pdf|.
  \end{itemize}
  This will need |latexmk|; if not available you will need in replacement of
  the last step to execute manually |latex| on |xint.tex| (thrice)
  then |dvipdfmx|.
\end{framed}
\fi

\ifnum\dosourcexint=1
+fi
+catcode`\ 0
\catcode0 15 % retour à la normale, peu importe
\catcode`\+ 12
\etocignoredepthtags
\etocsetnexttocdepth{section}
\tableofcontents
\makeatletter
\@gobble\fi
\StopEventually{\end{document}\endinput}
\ifnum\dosourcexint=1
\renewcommand{\etocaftertochook}{\addvspace{\bigskipamount}}
\etocsettocstyle {}{}
\clearpage
% \newgeometry{%hmarginratio=4:3,
%              hscale=0.7,vscale=0.75}% ATTENTION \newgeometry fait
%                                 % un reset de vscale si on ne le
%                                 % précise pas ici !!!
\else
\clearpage
\fi

\makeatletter

\def\MARGEPAGENO{1.25em}
\etocsettocdepth{subsubsection}% 2015/09/15

\etocdepthtag.toc {implementation}
\addtocontents{toc}{\gdef\string\sectioncouleur{[named]{RoyalPurple}}}

\def\storedlinecounts {}
\def\StoreCodelineNo #1{\edef\storedlinecounts{%
                        \unexpanded\expandafter{\storedlinecounts}%
                        {{#1}{\the\c@CodelineNo}}}\c@CodelineNo\z@ }

% \macrocode
% ==========
% 2014/11/04 did some hack with active characters à la upquote for
% straight quotes, but this is now irrelevant as we use suitable font
% from newtxtt with straight quotes.

%
% Actually, I should not at all rely on the doc class, I should do it all by
% myself. As I don't use at all \DocInput (which caused me loads of problems
% back then when I was trying to get a workflow satisfying my views on how
% .dtx files should be structured), there is not much rationale for using the
% doc class.

\def\macrocode{\macro@code
               \frenchspacing \@vobeyspaces
               \makestarlowast
               \xmacro@code }

\def\macro@font {\ttbfamily }% slashed 0

% \lverb
% ======

% Updated 2021/05/24 to work-around macrocode bug
% https://github.com/latex3/latex2e/issues/563
% which shows when I have some \subsection directly
% followed by macrocode, then next time there is some actual
% text the spacing to next macrocode block will be wrong.
\AddToHook{env/macrocode/after}{\@nobreakfalse}
% I suspect the \par\medskip which was in \lverb code
% below was a poorman workaround, dating back to 2014
% or earlier.
%
% macrocode has another strange problem which is
% that it reacts differently if one leaves a "blank line"
% above it in the comments, or use there \par. I consider
% this to be another bug of macrocode (which is inherited
% from the usage of the \trivlist mechanism).
% The presence of \par will add \partopsep space.
% \setlength{\partopsep}{0pt}
\AddToHook{env/macrocode/begin}{\partopsep0pt\relax}

% Définition de \lverb
% Has become more complicated for 1.2l
\catcode`_ 11
{\catcode32\active%
\gdef\myobeyspaces{\catcode32\active\def {\leavevmode\kern\fontcharwd\font`X}}}
\def\lverbpercent {\catcode32\active\lverbpercent_a}%
\def\lverbpercent_a #1{%
  \if\XINT_sptoken\detokenize{#1}\xint_dothis{\catcode32 10 }\fi
  \if-\detokenize{#1}\xint_dothis{\par #1}\fi
  \if(\detokenize{#1}\xint_dothis{\par\bgroup\myobeyspaces\obeylines}\fi
  \if:\detokenize{#1}\xint_dothis{}\fi
  \if)\detokenize{#1}\xint_dothis{\egroup\everypar{\hskip-\parindent\everypar{}}}\fi
  \if!\detokenize{#1}\xint_dothis{\lverbpercent}\fi
  \ifx#1\lverbpercent\xint_dothis{\catcode32 10 \par #1}\fi
  \xint_orthat{\catcode32 10 #1}%
}
\catcode`_ 8
\long\def\lverb {%
  \relax\par\smallskip%\noindent\null
%  \begingroup
  \bgroup
% this mechanism (whose motivation I do not really recall, in particular
% probably the \medskip was related to counteract the macrocode bug
% latex2e#563 whose effects I must have seen in 2013/2014) now removed
% at 1.4g.  I vaguely remember this code was originally done
% in relation to coloured short verbatims?
%
% Removing the \medskip has reduced when it was done xint-source.pdf to 545
% from 550 pages. Update: and removing the bad extra spacing when a macrocode
% block follows immediately a \section or \subsection had the result of
% reducing further to 542 pages...
%    \aftergroup\@@par \aftergroup\endgroup \aftergroup\medskip
    \let\do\do@noligs  \verbatim@nolig@list
    \let\do\@makeother \dospecials
    \catcode32 10 \catcode`\& 14 \catcode`\$ 0
    \catcode`\% \active
    \begingroup\lccode`\~`\%\lowercase{\endgroup\let~\lverbpercent}%
  \MicroFont % sera donc en couleur.
    \@lverb
}

% let's use (7 years later) simply a \def and not \let~\egroup
\def\@lverb #1{\catcode`#1\active
               \lccode`\~`#1\lowercase{\def~}{\egroup\par}}%

% there is a group to handle restore
\def\MicroFont {%\ttzfamily
                \color[named]{Purple}\makestarlowast }

% privatecodecomments
% ===================
\newenvironment{privatecodecomments}
   {\par \textbf{\textcolor{red}{COMMENTAIRES PRIVÉS.}}\par
    \def\MicroFont{\color[named]{BlueGreen}\makestarlowast}% (<-- 2021/05/24...)
    \begingroup\lccode`\~`\%\lowercase{\endgroup\let~\lverbpercent}%
    \catcode`\%\active}
   {\par \textbf{\textcolor{red}{FIN DES COMMENTAIRES PRIVÉS.}}\par}

% \changed
% ========

% updated 2021/05/29 to use \xintreleasedate (always) and put
% second argument if non empty within square brackets.
% updated again same day to make \changed a LaTeX macro with optional
% parameter, so order is permuted
\newcommand\changed[2][]{%
    \par\smallskip\noindent
    \textbf{#2\space(\xintreleasedate{#2})\if\relax\detokenize{#1}\relax
                     \else\space[commented #1]\fi.}%
    \hangindent\parindent
}

% \xintreleasedate
% ================

% added 2021/05/29
% xint-dates.txt is produced from CHANGES.md by grep via Makefile rule
% 1.09 (2013/09/23) aajouté manuellement car pas dans CHANGES.md
% (il y avait un bug "a stray underscore in a code line of xintexpr.sty must
%  be removed" et CTAN a annoncé seulement 1.09a)
%
\begingroup\endlinechar-1
\catcode`\`=\active
\def`#1 (#2)`{\expandafter\gdef\csname xintreleasedate_#1\endcsname{#2}}
`1.09 (2013/09/23)`% pas dans CHANGES.md car remplacé immédiatement par 1.09a
\@@input xint-dates.txt
\endgroup

\def\xintreleasedate#1{%
    \ifcsname xintreleasedate_#1\endcsname
       \csname xintreleasedate_#1\endcsname
    \else
       \immediate\write128{Pas de date pour #1 en ligne \the\inputlineno}%
       \error
    \fi
}

\expandafter\edef\csname xintreleasedate_\xintbndlversion\endcsname{\xintbndldate}

% Hyperlinks
% ==========

% renew some definitions (new labels, prefixed with src-)

% hyperlink and slash
\DeclareRobustCommand\csbxint[1]
        {\hyperref[\detokenize{src-xint#1}]%
          {{\char92\mbox{xint}\-\endlinechar-1
              \makestarlowast \catcode`_ 12 \catcode`^ 12
                 \scantokens\expandafter{\detokenize{#1}}}}}

\DeclareRobustCommand\csbXINT[1]
        {\hyperref[\detokenize{src-XINT#1}]%
          {{\char92\mbox{XINT}\-\endlinechar-1
              \makestarlowast \catcode`_ 12 \catcode`^ 12
                 \scantokens\expandafter{\detokenize{#1}}}}}

\DeclareRobustCommand\csb [1]
        {\hyperref[\detokenize{src-#1}]%
          {{\char92 \endlinechar-1 
              \makestarlowast \catcode`_ 12 \catcode`^ 12
                 \scantokens\expandafter{\detokenize{#1}}}}}

% hyperlink and no slash
\DeclareRobustCommand\csbn[1]
        {\hyperref[\detokenize{src-#1}]%
          {{\endlinechar-1
              \makestarlowast \catcode`_ 12 \catcode`^ 12
                 \scantokens\expandafter{\detokenize{#1}}}}}

% HACK OF \@sect
% ==============
% goal is to add labels but without having to modify currently
% existing mark-up in sources. But KOMA annoyingly makes an extra
% step needed. 2018/06/11
\let\original@sect\@sect
\def\@sect#1#2#3#4#5#6[#7]#8{\original@sect{#1}{#2}{#3}{#4}{#5}{#6}[{#7}]%
    {\begingroup
       %not possible because of KOMA wrappers
       %\def\csh##1{\csa{##1}\label{\detokenize{src-##1}}}%
       \let\csh\cshintitle
       \let\cshn\cshnintitle
       #8%
     \endgroup}%
}%
\def\cshintitle#1{\csa{#1}%
                  \label{\detokenize{src-#1}}%
                  %\expandafter\DescribeMacro\csname#1\endcsname
                  }
% \csan: no backslash
\def\cshnintitle#1{\csan{#1}\label{\detokenize{src-#1}}}

%% END OF MACRO DEFINITIONS FOR SOURCEXINT

\def\xintImpRunningHeader{{\inheadertrue\catcode`,12\relax
          \DOCxintfrontpage,
          \xintkernelnameimp,
          \xinttoolsnameimp,
          \xintcorenameimp,
          \xintnameimp,
          \xintbinhexnameimp,
          \xintgcdnameimp,
          \xintfracnameimp,
          \xintseriesnameimp,
          \xintcfracnameimp,
          \xintexprnameimp,
          \xinttrignameimp, \xintlognameimp}}
\markboth{\makebox[0pt]{\xintImpRunningHeader}}{\makebox[0pt]{\xintImpRunningHeader}}

\makeatother

\section{Timeline (in brief)}

This is \expandafter|\xintbndlversion| of \expandafter|\xintbndldate|.

Please refer to |CHANGES.html| for a (very) detailed history.
\centeredline{Internet:
  \url{http://mirrors.ctan.org/macros/generic/xint/CHANGES.html}}

\begin{itemize}
\item Release |1.4i| of |2021/06/11|: extension of the
  «simultaneous assignments» concept (backwards compatible).
\item Release |1.4g| of |2021/05/25|:
  powers are now parsed in a right associative way. Removal of the
  single-character operators |&|, \verb=|=, and |=| (deprecated at |1.1|).
  Reformatted expandable error messages.
\item Release |1.4e| of |2021/05/05|: logarithms and exponentials up to 62
  digits, trigonometry still mainly done at high level but with guard digits
  so all digits up to the last one included can be trusted for faithful
  rounding and high probability of correct rounding.
\item Release |1.4| of |2020/01/31|: \xintexprnameimp overhaul to use
  |\expanded| based expansion control.  Many new features, in particular
  support for input and output of nested structures.  Breaking changes, main
  ones being the (provisory) drop of |x*[a, b,...]|, |x+[a, b,...]| et al.\@
  syntax and the requirement of |\expanded| primitive (currently required only
  by \xintexprnameimp).
\item Release |1.3e| of |2019/04/05|: packages \xinttrignameimp, \xintlognameimp;
  \csa{xintdefefunc} ``non-protected'' variant of \csbxint{deffunc} (at |1.4|
  the two got merged and \csa{xintdefefunc} became a deprecated alias for
  \csbxint{deffunc}).
  Indices removed from |sourcexint.pdf|.
%  Their functionality is advantageously
  % made available via the search function in PDF viewers. Already the local
  % tables of contents are useful enough most of the time.
\item Release |1.3d| of |2019/01/06|: fix of |1.2p| bug for division with a
  zero dividend and a one-digit divisor, \csbxint{eval} et al. wrappers,
  |gcd()| and |lcm()| work with fractions.
\item Release |1.3c| of |2018/06/17|: documentation better hyperlinked,
  indices added to |sourcexint.pdf|. Colon in |:=| now optional for
  \csbxint{defvar} and \csbxint{deffunc}.
\item Release |1.3b| of |2018/05/18|: randomness related additions (still WIP).
\item Release |1.3a| of |2018/03/07|: efficiency fix of the mechanism for
  recursive functions.
\item Release |1.3| of |2018/03/01|: addition and subtraction use
  systematically least common multiple of denominators. Extensive
  under-the-hood refactoring of \csbxint{NewExpr} and \csbxint{deffunc} which
  now allow recursive definitions. Removal of |1.2o| deprecated macros.
\item Release |1.2q| of |2018/02/06|: fix of |1.2l| subtraction bug
  in special situation; tacit multiplication extended to cases such as
  |10!20!30!|.
\item Release |1.2p| of |2017/12/05|: maps |//| and |/:| to the floored, not
  truncated, division. Simultaneous assignments possible with \csbxint{defvar}.
  Efficiency improvements in \xinttoolsnameimp.
\item Release |1.2o| of |2017/08/29|: massive deprecations of those macros
  from \xintcorenameimp and \xintnameimp which filtered their arguments via
  \csbxint{Num}.
\item Release |1.2n| of |2017/08/06|: improvements of \xintbinhexnameimp.
\item Release |1.2m| of |2017/07/31|: rewrite of \xintbinhexnameimp in the
  style of the |1.2| techniques.
\item Release |1.2l| of |2017/07/26|: under the hood efficiency improvements
  in the style of the |1.2| techniques; subtraction refactored. Compatibility
  of most \xintfracnameimp macros with arguments using non-delimited
  |\the\numexpr| or |\the\mathcode| etc...
\item Release |1.2i| of |2016/12/13|: under the hood efficiency improvements
  in the style of the |1.2| techniques.
\item Release |1.2| of |2015/10/10|: complete refactoring of the core
  arithmetic macros and faster \csbxint{expr} parser.
\item Release |1.1| of |2014/10/28|: extensive changes in \xintexprnameimp.
  Addition and subtraction do not multiply denominators blindly but sometimes
  produce smaller ones. Also with that release, packages \xintkernelnameimp
  and \xintcorenameimp got extracted from \xinttoolsnameimp and \xintnameimp.
\item Release |1.09g| of |2013/11/22|: the \xinttoolsnameimp package is
  extracted from \xintnameimp; addition of \csbxint{loop} and \csbxint{iloop}.
\item Release |1.09c| of |2013/10/09|: \csbxint{For}, \csa{xintNewNumExpr}
  (ancestor of \csbxint{NewExpr}/\csbxint{deffunc} mechanism).
\item Release |1.09a| of |2013/09/24|: support for functions by \xintexprnameimp.
\item Release |1.08| of |2013/06/07|: the \xintbinhexnameimp package.
\item Release |1.07| of |2013/05/25|: support for floating point numbers
  added to \xintfracnameimp and first release of the \xintexprnameimp package
  (provided
  \csbxint{expr} and \csbxint{floatexpr}).
\item Release |1.04| of |2013/04/25|: the \xintcfracnameimp package.
\item Release |1.03| of |2013/04/14|: the \xintfracnameimp and
  \xintseriesnameimp packages.
\item Release |1.0| of |2013/03/28|: initial release of the \xintnameimp and
  \xintgcdnameimp packages.
\end{itemize}

Some parts of the code still date back to the initial release, and
  at that time I was learning my trade in expandable TeX macro programming.
  At some point in the future, I will have to re-examine the older parts of
  the code.

Warning: pay attention when looking at the code to the catcode configuration
as found in \csbXINT{_setcatcodes}. Additional temporary configuration is used
at some locations. For example |!| is of catcode letter in \xintexprnameimp
and there are locations with funny catcodes e.g. using some letters with the
math shift catcode.

\MakePercentIgnore
%\def\gardesactifs {^^A
%\catcode`\<=0 \catcode`\>=11 \catcode`\*=11 \catcode`\/=11 }
%\def\gardesinactifs {^^A
%\catcode`\<=12 \catcode`\>=12 \catcode`\*=12 \catcode`\/=12 }
%\gardesactifs
%\let</dtx>\relax
%\let<*xintkernel>\gardesinactifs
%</dtx>^^A--------------------------------------------------------
%<*xintkernel>^^A-------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintkernelnameUp\endcsname
% \section {Package \xintkernelnameimp implementation}
% \RaisedLabel{sec:kernelimp}
%
% \localtableofcontents
%
% This package provides the common minimal code base for loading management
% and catcode control and also a few programming utilities. With |1.2| a few
% more helper macros and all |\chardef|'s have been moved here. The package is
% loaded by both |xintcore.sty| and |xinttools.sty| hence by all other
% packages.
%
% \changed{1.1}
% Separated package.
%
% \changed{1.2i}
% \csbxint{replicate}, \csbxint{gobble}, \csbxint{LengthUpTo}
% and \csbxint{LastItem}, and faster \csbxint{Length}.
%
% \changed{1.3b}
% \csbxint{UniformDeviate}.
%
% \changed[2020/01/11]{1.4}
% \csbxint{Replicate}, \csbxint{Gobble}, \csbxint{LastOne}, \csbxint{FirstOne}.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \catcode95=11   % _
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \let\z\relax
  \expandafter
    \ifx\csname numexpr\endcsname\relax
     \y{xintkernel}{\numexpr not available, aborting input}%
     \def\z{\endgroup\endinput}%
    \else
    \expandafter
     \ifx\csname XINTsetupcatcodes\endcsname\relax
      \else
       \y{xintkernel}{I was already loaded, aborting input}%
       \def\z{\endgroup\endinput}%
      \fi
    \fi
  \ifx\z\relax\else\expandafter\z\fi%
%    \end{macrocode}
% \subsubsection{\csh{XINTrestorecatcodes}, \csh{XINTsetcatcodes},
%                \csh{XINTrestorecatcodesendinput}}
% \lverb|Renamed at 1.4e without underscores, in connexion with
% easying up reloading process for xintlog.sty and xinttrig.sty.|
%    \begin{macrocode}
  \def\PrepareCatcodes
  {%
      \endgroup
      \def\XINTrestorecatcodes
      {% takes care of all, to allow more economical code in modules
           \catcode0=\the\catcode0 %
           \catcode59=\the\catcode59   % ; xintexpr
           \catcode126=\the\catcode126 % ~ xintexpr
           \catcode39=\the\catcode39   % ' xintexpr
           \catcode34=\the\catcode34   % " xintbinhex, and xintexpr
           \catcode63=\the\catcode63   % ? xintexpr
           \catcode124=\the\catcode124 % | xintexpr
           \catcode38=\the\catcode38   % & xintexpr
           \catcode64=\the\catcode64   % @ xintexpr
           \catcode33=\the\catcode33   % ! xintexpr
           \catcode93=\the\catcode93   % ] -, xintfrac, xintseries, xintcfrac
           \catcode91=\the\catcode91   % [ -, xintfrac, xintseries, xintcfrac
           \catcode36=\the\catcode36   % $ xintgcd only $
           \catcode94=\the\catcode94   % ^
           \catcode96=\the\catcode96   % `
           \catcode47=\the\catcode47   % /
           \catcode41=\the\catcode41   % )
           \catcode40=\the\catcode40   % (
           \catcode42=\the\catcode42   % *
           \catcode43=\the\catcode43   % +
           \catcode62=\the\catcode62   % >
           \catcode60=\the\catcode60   % <
           \catcode58=\the\catcode58   % :
           \catcode46=\the\catcode46   % .
           \catcode45=\the\catcode45   % -
           \catcode44=\the\catcode44   % ,
           \catcode35=\the\catcode35   % #
           \catcode95=\the\catcode95   % _
           \catcode125=\the\catcode125 % }
           \catcode123=\the\catcode123 % {
           \endlinechar=\the\endlinechar
           \catcode13=\the\catcode13   % ^^M
           \catcode32=\the\catcode32   %
           \catcode61=\the\catcode61\relax   % =
      }%
      \edef\XINTrestorecatcodesendinput
      {%
           \XINTrestorecatcodes\noexpand\endinput %
      }%
      \def\XINTsetcatcodes
      {%
        \catcode61=12   % =
        \catcode32=10   % space
        \catcode13=5    % ^^M
        \endlinechar=13 %
        \catcode123=1   % {
        \catcode125=2   % }
        \catcode95=11   % _ LETTER
        \catcode35=6    % #
        \catcode44=12   % ,
        \catcode45=12   % -
        \catcode46=12   % .
        \catcode58=11   % : LETTER
        \catcode60=12   % <
        \catcode62=12   % >
        \catcode43=12   % +
        \catcode42=12   % *
        \catcode40=12   % (
        \catcode41=12   % )
        \catcode47=12   % /
        \catcode96=12   % `
        \catcode94=11   % ^ LETTER
        \catcode36=3    % $
        \catcode91=12   % [
        \catcode93=12   % ]
        \catcode33=12   % ! (xintexpr.sty will use catcode 11)
        \catcode64=11   % @ LETTER
        \catcode38=7    % & for \romannumeral`&&@ trick.
        \catcode124=12  % |
        \catcode63=11   % ? LETTER
        \catcode34=12   % "
        \catcode39=12   % '
        \catcode126=3   % ~ MATH
        \catcode59=12   % ;
        \catcode0=12    % for \romannumeral`&&@ trick
        \catcode1=3     % for ultra-safe séparateur &&A
      }%
      \let\XINT_setcatcodes\XINTsetcatcodes
      \let\XINT_restorecatcodes\XINTrestorecatcodes
      \XINTsetcatcodes
  }%
\PrepareCatcodes
%    \end{macrocode}
% Other modules could possibly be loaded under a different catcode regime.
%    \begin{macrocode}
\def\XINTsetupcatcodes {% for use by other modules
      \edef\XINTrestorecatcodesendinput
      {%
           \XINTrestorecatcodes\noexpand\endinput %
      }%
      \XINTsetcatcodes
}%
%    \end{macrocode}
% \subsection{Package identification}
%
% Inspired from \textsc{Heiko Oberdiek}'s packages. Modified in |1.09b| to allow
% re-use in the other modules. Also I assume now that if |\ProvidesPackage|
% exists it then does define |\ver@<pkgname>.sty|, code of |HO| for some reason
% escaping me  (compatibility with LaTeX 2.09 or other things ??) seems to set
% extra precautions.
%
% |1.09c| uses e-\TeX{} |\ifdefined|.
%    \begin{macrocode}
\ifdefined\ProvidesPackage
  \let\XINT_providespackage\relax
\else
  \def\XINT_providespackage #1#2[#3]%
              {\immediate\write-1{Package: #2 #3}%
               \expandafter\xdef\csname ver@#2.sty\endcsname{#3}}%
\fi
\XINT_providespackage
\ProvidesPackage {xintkernel}%
  [2021/07/13 v1.4j Paraphernalia for the xint packages (JFB)]%
%    \end{macrocode}
% \subsection{Constants}
%    \begin{macrocode}
\chardef\xint_c_     0
\chardef\xint_c_i    1
\chardef\xint_c_ii   2
\chardef\xint_c_iii  3
\chardef\xint_c_iv   4
\chardef\xint_c_v    5
\chardef\xint_c_vi   6
\chardef\xint_c_vii  7
\chardef\xint_c_viii 8
\chardef\xint_c_ix     9
\chardef\xint_c_x     10
\chardef\xint_c_xii   12
\chardef\xint_c_xiv   14
\chardef\xint_c_xvi   16
\chardef\xint_c_xvii  17
\chardef\xint_c_xviii 18
\chardef\xint_c_xx    20
\chardef\xint_c_xxii  22
\chardef\xint_c_ii^v  32
\chardef\xint_c_ii^vi 64
\chardef\xint_c_ii^vii       128
\mathchardef\xint_c_ii^viii  256
\mathchardef\xint_c_ii^xii  4096
\mathchardef\xint_c_x^iv   10000
%    \end{macrocode}
% \subsection{(WIP) \csh{xint_texuniformdeviate} and needed counts}
%    \begin{macrocode}
\ifdefined\pdfuniformdeviate  \let\xint_texuniformdeviate\pdfuniformdeviate\fi
\ifdefined\uniformdeviate     \let\xint_texuniformdeviate\uniformdeviate   \fi
\ifx\xint_texuniformdeviate\relax\let\xint_texuniformdeviate\xint_undefined\fi
\ifdefined\xint_texuniformdeviate
  \csname newcount\endcsname\xint_c_ii^xiv
  \xint_c_ii^xiv   16384 % "4000, 2**14
  \csname newcount\endcsname\xint_c_ii^xxi
  \xint_c_ii^xxi 2097152 % "200000, 2**21
\fi
%    \end{macrocode}
% \subsection{Token management utilities}
% \changed{1.3b}
% |\xint_gobandstop_...| macros because this is handy for
% \csbxint{RandomDigits}.
% |1.3g| forces \cs{empty} and \cs{space} to have their standard meanings,
% rather than simply alerting user in the (theoretical) case they don't that
% nothing will work.
% If some \LaTeX{} user has \cs{renewcommand}ed them they will be long and
% this will trigger xint redefinitions and warnings. 
%    \begin{macrocode}
\def\XINT_tmpa { }%
\ifx\XINT_tmpa\space\else
   \immediate\write-1{Package xintkernel Warning:}%
   \immediate\write-1{\string\space\XINT_tmpa macro does not have its normal
     meaning from Plain or LaTeX, but:}%
   \immediate\write-1{\meaning\space}%
   \let\space\XINT_tmpa
   \immediate\write-1{\space\space\space\space
   % an exclam might let Emacs/AUCTeX think it is an error message, afair
                      Forcing \string\space\space to be the usual one.}%
\fi
\def\XINT_tmpa {}%
\ifx\XINT_tmpa\empty\else
   \immediate\write-1{Package xintkernel Warning:}%
   \immediate\write-1{\string\empty\space macro does not have its normal
     meaning from Plain or LaTeX, but:}%
   \immediate\write-1{\meaning\empty}%
   \let\empty\XINT_tmpa
   \immediate\write-1{\space\space\space\space
                      Forcing \string\empty\space to be the usual one.}%
\fi
\let\XINT_tmpa\relax
\let\xint_gobble_\empty
\long\def\xint_gobble_i    #1{}%
\long\def\xint_gobble_ii   #1#2{}%
\long\def\xint_gobble_iii  #1#2#3{}%
\long\def\xint_gobble_iv   #1#2#3#4{}%
\long\def\xint_gobble_v    #1#2#3#4#5{}%
\long\def\xint_gobble_vi   #1#2#3#4#5#6{}%
\long\def\xint_gobble_vii  #1#2#3#4#5#6#7{}%
\long\def\xint_gobble_viii #1#2#3#4#5#6#7#8{}%
\let\xint_gob_andstop_\space
\long\def\xint_gob_andstop_i    #1{ }%
\long\def\xint_gob_andstop_ii   #1#2{ }%
\long\def\xint_gob_andstop_iii  #1#2#3{ }%
\long\def\xint_gob_andstop_iv   #1#2#3#4{ }%
\long\def\xint_gob_andstop_v    #1#2#3#4#5{ }%
\long\def\xint_gob_andstop_vi   #1#2#3#4#5#6{ }%
\long\def\xint_gob_andstop_vii  #1#2#3#4#5#6#7{ }%
\long\def\xint_gob_andstop_viii #1#2#3#4#5#6#7#8{ }%
\long\def\xint_firstofone  #1{#1}%
\long\def\xint_firstoftwo  #1#2{#1}%
\long\def\xint_secondoftwo #1#2{#2}%
\long\def\xint_thirdofthree#1#2#3{#3}% 1.4d
\let\xint_stop_aftergobble\xint_gob_andstop_i
\long\def\xint_stop_atfirstofone  #1{ #1}%
\long\def\xint_stop_atfirstoftwo  #1#2{ #1}%
\long\def\xint_stop_atsecondoftwo #1#2{ #2}%
\long\def\xint_exchangetwo_keepbraces    #1#2{{#2}{#1}}%
%    \end{macrocode}
% \subsection{``gob til'' macros and UD style fork}
%    \begin{macrocode}
\long\def\xint_gob_til_R #1\R {}%
\long\def\xint_gob_til_W #1\W {}%
\long\def\xint_gob_til_Z #1\Z {}%
\long\def\xint_gob_til_zero #10{}%
\long\def\xint_gob_til_one  #11{}%
\long\def\xint_gob_til_zeros_iii   #1000{}%
\long\def\xint_gob_til_zeros_iv    #10000{}%
\long\def\xint_gob_til_eightzeroes #100000000{}%
\long\def\xint_gob_til_dot    #1.{}%
\long\def\xint_gob_til_G     #1G{}%
\long\def\xint_gob_til_minus #1-{}%
\long\def\xint_UDzerominusfork #10-#2#3\krof {#2}%
\long\def\xint_UDzerofork       #10#2#3\krof {#2}%
\long\def\xint_UDsignfork       #1-#2#3\krof {#2}%
\long\def\xint_UDwfork         #1\W#2#3\krof {#2}%
\long\def\xint_UDXINTWfork     #1\XINT_W#2#3\krof {#2}%
\long\def\xint_UDzerosfork     #100#2#3\krof {#2}%
\long\def\xint_UDonezerofork   #110#2#3\krof {#2}%
\long\def\xint_UDsignsfork     #1--#2#3\krof {#2}%
\let\xint:\char
\long\def\xint_gob_til_xint:#1\xint:{}%
\long\def\xint_gob_til_^#1^{}%
\def\xint_bracedstopper{\xint:}%
\long\def\xint_gob_til_exclam #1!{}% This ! has catcode 12
\long\def\xint_gob_til_sc #1;{}%
%    \end{macrocode}
% \subsection{\csh{xint_afterfi}}
%    \begin{macrocode}
\long\def\xint_afterfi #1#2\fi {\fi #1}%
%    \end{macrocode}
% \subsection{\csh{xint_bye}, \csh{xint_Bye}}
% \changed{1.09}
% |\xint_bye|
% \changed{1.2i}
% |\xint_Bye| for \csbxint{DSRr} and \csbxint{Round}. Also |\xint_stop_afterbye|.
%    \begin{macrocode}
\long\def\xint_bye #1\xint_bye {}%
\long\def\xint_Bye #1\xint_bye {}%
\long\def\xint_stop_afterbye #1\xint_bye { }%
%    \end{macrocode}
% \subsection{\csh{xintdothis}, \csh{xintorthat}}
% \changed{1.1}
% \changed{1.2} Names without underscores.
%
% To be used this way: 
% \lverb|
%( \if..\xint_dothis{..}\fi
%: \if..\xint_dothis{..}\fi
%: \if..\xint_dothis{..}\fi
%: ...more such...
%: \xint_orthat{...}
%) |
% Ancient testing indicated it is more efficient to list first the more
% improbable clauses.
%    \begin{macrocode}
\long\def\xint_dothis #1#2\xint_orthat #3{\fi #1}% 1.1
\let\xint_orthat \xint_firstofone
\long\def\xintdothis #1#2\xintorthat #3{\fi #1}%
\let\xintorthat \xint_firstofone
%    \end{macrocode}
% \subsection{\csh{xint_zapspaces}}
% \changed{1.1}
%
% This little (quite fragile in the normal sense i.e. non robust in the normal
% sense of programming lingua) utility zaps leading, intermediate, trailing,
% spaces in completely expanding context (|\edef|, |\csname...\endcsname|).
% \centeredline{Usage: |\xint_zapspaces foo<space>\xint_gobble_i|}
%
% Explanation: if there are leading spaces, then the first |#1| will be empty,
% and the first |#2| being undelimited will be stripped from all the remaining
% leading spaces, if there was more than one to start with. Of course
% brace-stripping may occur. And this iterates: each time a |#2| is removed,
% either we then have spaces and next |#1| will be empty, or we have no spaces
% and |#1| will end at the first space. Ultimately |#2| will be
% |\xint_gobble_i|.
%
% The
% |\zap@spaces| of LaTeX2e handles unexpectedly things such as
% \centeredline{|\zap@spaces 1 {22} 3 4 \@empty|} (spaces are not all
% removed). This does not happen with |\xint_zapspaces|.
%
% But for example |\foo{aa} {bb} {cc}| where |\foo| is a macro with three
% non-delimited arguments breaks expansion, as expansion of |\foo| will happen
% with |\xint_zapspaces| still around, and even if it wasn't it would have
% stripped the braces around |{bb}|, certainly breaking other things.
%
% Despite such obvious shortcomings it is enough for our purposes. It is
% currently used by \xintexprnameimp at various locations e.g. cleaning up
% optional argument of |\xintiexpr| and |\xintfloatexpr|; maybe in future
% internal usage will drop this in favour of a more robust utility.
%
% \changed{1.2e} |\xint_zapspaces_o|.
%
% \changed{1.2i} Made |\long|.
%
% ATTENTION THAT \xinttoolsnameimp HAS AN |\xintzapspaces| WHICH SHOULD NOT
% GET CONFUSED WITH THIS ONE.
%    \begin{macrocode}
\long\def\xint_zapspaces #1 #2{#1#2\xint_zapspaces }% 1.1
\long\def\xint_zapspaces_o #1{\expandafter\xint_zapspaces#1 \xint_gobble_i}%
%    \end{macrocode}
% \subsection{\csh{odef}, \csh{oodef}, \csh{fdef}}
% May be prefixed with |\global|. No parameter text.
%    \begin{macrocode}
\def\xintodef #1{\expandafter\def\expandafter#1\expandafter }%
\def\xintoodef #1{\expandafter\expandafter\expandafter\def
                  \expandafter\expandafter\expandafter#1%
                  \expandafter\expandafter\expandafter }%
\def\xintfdef #1#2%
    {\expandafter\def\expandafter#1\expandafter{\romannumeral`&&@#2}}%
\ifdefined\odef\else\let\odef\xintodef\fi
\ifdefined\oodef\else\let\oodef\xintoodef\fi
\ifdefined\fdef\else\let\fdef\xintfdef\fi
%    \end{macrocode}
% \subsection{\csh{xintReverseOrder}}
% \changed{1.0} Does not expand its argument. The whole of xint codebase now
% contains only two calls to |\XINT_rord_main| (in \xintgcdnameimp).
%
% Attention: removes brace pairs (and swallows spaces).
%
% For digit tokens a faster reverse macro is provided by (|1.2|)
% \csbxint{ReverseDigits} in \xintnameimp.
%
% For comma separated items, |1.2g| has \csbxint{CSVReverse} in
% \xinttoolsnameimp.
%    \begin{macrocode}
\def\xintReverseOrder {\romannumeral0\xintreverseorder }%
\long\def\xintreverseorder #1%
{%
    \XINT_rord_main {}#1%
      \xint:
        \xint_bye\xint_bye\xint_bye\xint_bye
        \xint_bye\xint_bye\xint_bye\xint_bye
      \xint:
}%
\long\def\XINT_rord_main #1#2#3#4#5#6#7#8#9%
{%
    \xint_bye #9\XINT_rord_cleanup\xint_bye
    \XINT_rord_main {#9#8#7#6#5#4#3#2#1}%
}%
\def\XINT_rord_cleanup #1{%
\long\def\XINT_rord_cleanup\xint_bye\XINT_rord_main ##1##2\xint:
{%
    \expandafter#1\xint_gob_til_xint: ##1%
}}\XINT_rord_cleanup { }%
%    \end{macrocode}
% \subsection{\csh{xintLength}}
% \changed{1.0} Does not expand its argument. See \csbxint{NthElt}|{0}| from
% \xinttoolsnameimp which f-expands its argument.
%
% \changed{1.2g} Added \csbxint{CSVLength} to \xinttoolsnameimp.
%
% \changed{1.2i} Rewrote this venerable macro. New code about 40\%
% faster across all lengths. Syntax with |\romannumeral0| adds some
% slight (negligible) overhead; it is done to fit some general
% principles of structure of the xint package macros but maybe
% at some point I should drop it. And in fact it is often called
% directly via the |\numexpr| access point. (bad coding...)
%    \begin{macrocode}
\def\xintLength {\romannumeral0\xintlength }%
\def\xintlength #1{%
\long\def\xintlength ##1%
{%
    \expandafter#1\the\numexpr\XINT_length_loop
    ##1\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
       \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
       \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    \relax
}}\xintlength{ }%
\long\def\XINT_length_loop #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9\XINT_length_finish_a\xint:
    \xint_c_ix+\XINT_length_loop
}%
\def\XINT_length_finish_a\xint:\xint_c_ix+\XINT_length_loop
    #1#2#3#4#5#6#7#8#9%
{%
    #9\xint_bye
}%
%    \end{macrocode}
% \subsection{\csh{xintLastItem}}
% \changed[2016/12/10]{1.2i}
% One level
% of braces removed in output. Output empty if input empty. Attention! 
% This means
% that an empty input or an input ending with a {} empty brace pair
% both give same output.
%
% The |\xint:| token must not be among items. |\xintFirstItem| added
% at 1.4 for usage in xintexpr. It must contain neither |\xint:|
% nor |\xint_bye| in its first item.
%
%    \begin{macrocode}
\def\xintLastItem {\romannumeral0\xintlastitem }%
\long\def\xintlastitem #1%
{%
    \XINT_last_loop {}.#1%
    {\xint:\XINT_last_loop_enda}{\xint:\XINT_last_loop_endb}%
    {\xint:\XINT_last_loop_endc}{\xint:\XINT_last_loop_endd}%
    {\xint:\XINT_last_loop_ende}{\xint:\XINT_last_loop_endf}%
    {\xint:\XINT_last_loop_endg}{\xint:\XINT_last_loop_endh}\xint_bye
}%
\long\def\XINT_last_loop #1.#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9%
        {#8}{#7}{#6}{#5}{#4}{#3}{#2}{#1}\xint:
    \XINT_last_loop {#9}.%
}%
\long\def\XINT_last_loop_enda #1#2\xint_bye{ #1}%
\long\def\XINT_last_loop_endb #1#2#3\xint_bye{ #2}%
\long\def\XINT_last_loop_endc #1#2#3#4\xint_bye{ #3}%
\long\def\XINT_last_loop_endd #1#2#3#4#5\xint_bye{ #4}%
\long\def\XINT_last_loop_ende #1#2#3#4#5#6\xint_bye{ #5}%
\long\def\XINT_last_loop_endf #1#2#3#4#5#6#7\xint_bye{ #6}%
\long\def\XINT_last_loop_endg #1#2#3#4#5#6#7#8\xint_bye{ #7}%
\long\def\XINT_last_loop_endh #1#2#3#4#5#6#7#8#9\xint_bye{ #8}%
%    \end{macrocode}
% \subsection{\csh{xintFirstItem}}
% \lverb|1.4. There must be neither \xint:
% nor \xint_bye in its first item.|
%    \begin{macrocode}
\def\xintFirstItem         {\romannumeral0\xintfirstitem }%
\long\def\xintfirstitem  #1{\XINT_firstitem #1{\xint:\XINT_firstitem_end}\xint_bye}%
\long\def\XINT_firstitem #1#2\xint_bye{\xint_gob_til_xint: #1\xint:\space #1}%
\def\XINT_firstitem_end\xint:{ }%
%    \end{macrocode}
% \subsection{\csh{xintLastOne}}
% \lverb|&
%
% As xintexpr 1.4 uses {c1}{c2}....{cN} storage when gathering comma separated
% values we need to not handle identically an empty list and a list with an
% empty item (as the above allows hierarchical structures). But \xintLastItem
% removed one level of brace pair so it is anadequate for the last() function.
%
% By the way it is logical to interpret «item» as meaning {cj} inclusive of
% the braces; but legacy xint user manual was not written in this spirit. And thus
% \xintLastItem did brace stripping, thus we need another name for maintaining
% backwards compatibility (although the cardinality of users is small).
%
% The \xint: token must not be found (visible) among the item contents.
%
% |
%    \begin{macrocode}
\def\xintLastOne {\romannumeral0\xintlastone }%
\long\def\xintlastone #1%
{%
    \XINT_lastone_loop {}.#1%
    {\xint:\XINT_lastone_loop_enda}{\xint:\XINT_lastone_loop_endb}%
    {\xint:\XINT_lastone_loop_endc}{\xint:\XINT_lastone_loop_endd}%
    {\xint:\XINT_lastone_loop_ende}{\xint:\XINT_lastone_loop_endf}%
    {\xint:\XINT_lastone_loop_endg}{\xint:\XINT_lastone_loop_endh}\xint_bye
}%
\long\def\XINT_lastone_loop #1.#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9%
        {#8}{#7}{#6}{#5}{#4}{#3}{#2}{#1}\xint:
    \XINT_lastone_loop {{#9}}.%
}%
\long\def\XINT_lastone_loop_enda #1#2\xint_bye{{#1}}%
\long\def\XINT_lastone_loop_endb #1#2#3\xint_bye{{#2}}%
\long\def\XINT_lastone_loop_endc #1#2#3#4\xint_bye{{#3}}%
\long\def\XINT_lastone_loop_endd #1#2#3#4#5\xint_bye{{#4}}%
\long\def\XINT_lastone_loop_ende #1#2#3#4#5#6\xint_bye{{#5}}%
\long\def\XINT_lastone_loop_endf #1#2#3#4#5#6#7\xint_bye{{#6}}%
\long\def\XINT_lastone_loop_endg #1#2#3#4#5#6#7#8\xint_bye{{#7}}%
\long\def\XINT_lastone_loop_endh #1#2#3#4#5#6#7#8#9\xint_bye{ #8}%
%    \end{macrocode}
% \subsection{\csh{xintFirstOne}}
% \lverb|&
% For xintexpr 1.4 too. Jan 3, 2020.
%
% This is an experimental macro, don't use it. If input is nil (empty set) it
% expands to nil, if not it fetches first item and braces it. Fetching will
% have stripped one brace pair if item was braced to start with, which is
% the case in non-symbolic xintexpr data objects.
%
% I have not given much thought to this (make it shorter, allow all tokens,
% (we could first test if empty via combination with \detokenize), etc...)
% as I need to get xint 1.4 out soon. So in particular attention that
% the macro assumes the \xint: token is absent from first item of input.
% |
%    \begin{macrocode}
\def\xintFirstOne  {\romannumeral0\xintfirstone }%
\long\def\xintfirstone  #1{\XINT_firstone #1{\xint:\XINT_firstone_empty}\xint:}%
\long\def\XINT_firstone #1#2\xint:{\xint_gob_til_xint: #1\xint:{#1}}%
\def\XINT_firstone_empty\xint:#1{ }%
%    \end{macrocode}
% \subsection{\csh{xintLengthUpTo}}
% \changed{1.2i} For use by \csbxint{Keep} and \csbxint{Trim}
% (\xinttoolsnameimp). The argument N **must be non-negative**.
%
% |\xintLengthUpTo{N}{List}| produces |-0| if length(List)>N, else it returns
% N-length(List). Hence subtracting it from N always computes min(N,length(List)).
% \changed{1.2j} Changed ending and interface to core loop.
%    \begin{macrocode}
\def\xintLengthUpTo {\romannumeral0\xintlengthupto}%
\long\def\xintlengthupto #1#2%
{%
    \expandafter\XINT_lengthupto_loop
    \the\numexpr#1.#2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
         \xint_c_vii\xint_c_vi\xint_c_v\xint_c_iv
         \xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye.%
}%
\def\XINT_lengthupto_loop_a #1%
{%
    \xint_UDsignfork
      #1\XINT_lengthupto_gt
      -\XINT_lengthupto_loop
    \krof #1%
}%
\long\def\XINT_lengthupto_gt #1\xint_bye.{-0}%
\long\def\XINT_lengthupto_loop #1.#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9\XINT_lengthupto_finish_a\xint:%
    \expandafter\XINT_lengthupto_loop_a\the\numexpr #1-\xint_c_viii.%
}%
\def\XINT_lengthupto_finish_a\xint:\expandafter\XINT_lengthupto_loop_a
    \the\numexpr #1-\xint_c_viii.#2#3#4#5#6#7#8#9%
{%
    \expandafter\XINT_lengthupto_finish_b\the\numexpr #1-#9\xint_bye
}%
\def\XINT_lengthupto_finish_b #1#2.%
{%
    \xint_UDsignfork
       #1{-0}%
        -{ #1#2}%
    \krof
}%
%    \end{macrocode}
% \subsection{\csh{xintreplicate}, \csh{xintReplicate}}
% \changed{1.2i}
%
% This is cloned from LaTeX3's |\prg_replicate:nn|, see Joseph's post
% at
% \centeredline{http://tex.stackexchange.com/questions/16189/repeat-command-n-times}
% I
% posted there an alternative not using the chained |\csname|'s but it is a bit
% less efficient (except perhaps for thousands of repetitions).
% The code in Joseph's post does |abs(#1)| replications when input |#1| is negative
% and then activates an error triggering macro; here we simply do nothing when
% |#1| is negative.
% \centeredline{Usage: |\romannumeral\xintreplicate{N}{stuff}|}
%
% When |N| is already explicit digits (even |N=0|, but non-negative) one can
% call the macro as 
% \centeredline{|\romannumeral\XINT_rep N\endcsname {foo}|}
% to skip the |\numexpr|.
%
% \changed[2020/01/11]{1.4}
% Added |\xintReplicate| ! The reason I did not before is that the prevailing
% habits in xint source code was to trigger with |\romannumeral0| not
% |\romannumeral| which is the lowercased named macros. Thus adding the
% camelcase one creates a couple |\xintReplicate/\xintreplicate| not obeying
% the general mold.
%    \begin{macrocode}
\def\xintReplicate{\romannumeral\xintreplicate}%
\def\xintreplicate#1%
   {\expandafter\XINT_replicate\the\numexpr#1\endcsname}%
\def\XINT_replicate #1{\xint_UDsignfork
                         #1\XINT_rep_neg
                          -\XINT_rep
                       \krof #1}%
\long\def\XINT_rep_neg #1\endcsname #2{\xint_c_}%
\def\XINT_rep   #1{\csname XINT_rep_f#1\XINT_rep_a}%
\def\XINT_rep_a #1{\csname XINT_rep_#1\XINT_rep_a}%
\def\XINT_rep_\XINT_rep_a{\endcsname}%
\long\expandafter\def\csname XINT_rep_0\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}}%
\long\expandafter\def\csname XINT_rep_1\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1}%
\long\expandafter\def\csname XINT_rep_2\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1}%
\long\expandafter\def\csname XINT_rep_3\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1}%
\long\expandafter\def\csname XINT_rep_4\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_5\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_6\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_7\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_8\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_9\endcsname #1%
    {\endcsname{#1#1#1#1#1#1#1#1#1#1}#1#1#1#1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_f0\endcsname #1%
    {\xint_c_}%
\long\expandafter\def\csname XINT_rep_f1\endcsname #1%
    {\xint_c_ #1}%
\long\expandafter\def\csname XINT_rep_f2\endcsname #1%
    {\xint_c_ #1#1}%
\long\expandafter\def\csname XINT_rep_f3\endcsname #1%
    {\xint_c_ #1#1#1}%
\long\expandafter\def\csname XINT_rep_f4\endcsname #1%
    {\xint_c_ #1#1#1#1}%
\long\expandafter\def\csname XINT_rep_f5\endcsname #1%
    {\xint_c_ #1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_f6\endcsname #1%
    {\xint_c_ #1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_f7\endcsname #1%
    {\xint_c_ #1#1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_f8\endcsname #1%
    {\xint_c_ #1#1#1#1#1#1#1#1}%
\long\expandafter\def\csname XINT_rep_f9\endcsname #1%
    {\xint_c_ #1#1#1#1#1#1#1#1#1}%
%    \end{macrocode}
% \subsection{\csh{xintgobble}, \csh{xintGobble}}
% \changed{1.2i}
%
% I hesitated about allowing as many as |9^6-1=531440| tokens to gobble, but
% |9^5-1=59058| is too low for playing with long decimal expansions.
% \centeredline{Usage: |\romannumeral\xintgobble{N}...|}
%
% \changed[2020/01/11]{1.4}
% Added |\xintGobble|.
%    \begin{macrocode}
\def\xintGobble{\romannumeral\xintgobble}%
\def\xintgobble #1%
   {\csname xint_c_\expandafter\XINT_gobble_a\the\numexpr#1.0}%
\def\XINT_gobble #1.{\csname xint_c_\XINT_gobble_a #1.0}%
\def\XINT_gobble_a #1{\xint_gob_til_zero#1\XINT_gobble_d0\XINT_gobble_b#1}%
\def\XINT_gobble_b #1.#2%
  {\expandafter\XINT_gobble_c
      \the\numexpr (#1+\xint_c_v)/\xint_c_ix-\xint_c_i\expandafter.%
      \the\numexpr #2+\xint_c_i.#1.}%
\def\XINT_gobble_c #1.#2.#3.%
  {\csname XINT_g#2\the\numexpr#3-\xint_c_ix*#1\relax\XINT_gobble_a #1.#2}%
\def\XINT_gobble_d0\XINT_gobble_b0.#1{\endcsname}%
\expandafter\let\csname XINT_g10\endcsname\endcsname
\long\expandafter\def\csname XINT_g11\endcsname#1{\endcsname}%
\long\expandafter\def\csname XINT_g12\endcsname#1#2{\endcsname}%
\long\expandafter\def\csname XINT_g13\endcsname#1#2#3{\endcsname}%
\long\expandafter\def\csname XINT_g14\endcsname#1#2#3#4{\endcsname}%
\long\expandafter\def\csname XINT_g15\endcsname#1#2#3#4#5{\endcsname}%
\long\expandafter\def\csname XINT_g16\endcsname#1#2#3#4#5#6{\endcsname}%
\long\expandafter\def\csname XINT_g17\endcsname#1#2#3#4#5#6#7{\endcsname}%
\long\expandafter\def\csname XINT_g18\endcsname#1#2#3#4#5#6#7#8{\endcsname}%
\expandafter\let\csname XINT_g20\endcsname\endcsname
\long\expandafter\def\csname XINT_g21\endcsname #1#2#3#4#5#6#7#8#9%
 {\endcsname}%
\long\expandafter\edef\csname XINT_g22\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g21\endcsname}%
\long\expandafter\edef\csname XINT_g23\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g22\endcsname}%
\long\expandafter\edef\csname XINT_g24\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g23\endcsname}%
\long\expandafter\edef\csname XINT_g25\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g24\endcsname}%
\long\expandafter\edef\csname XINT_g26\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g25\endcsname}%
\long\expandafter\edef\csname XINT_g27\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g26\endcsname}%
\long\expandafter\edef\csname XINT_g28\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g27\endcsname}%
\expandafter\let\csname XINT_g30\endcsname\endcsname
\long\expandafter\edef\csname XINT_g31\endcsname #1#2#3#4#5#6#7#8#9%
 {\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g32\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g31\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g33\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g32\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g34\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g33\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g35\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g34\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g36\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g35\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g37\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g36\expandafter\noexpand\csname XINT_g28\endcsname}%
\long\expandafter\edef\csname XINT_g38\endcsname #1#2#3#4#5#6#7#8#9%
 {\noexpand\csname XINT_g37\expandafter\noexpand\csname XINT_g28\endcsname}%
\expandafter\let\csname XINT_g40\endcsname\endcsname
\expandafter\edef\csname XINT_g41\endcsname
 {\noexpand\csname XINT_g38\expandafter\noexpand\csname XINT_g31\endcsname}%
\expandafter\edef\csname XINT_g42\endcsname
 {\noexpand\csname XINT_g41\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g43\endcsname
 {\noexpand\csname XINT_g42\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g44\endcsname
 {\noexpand\csname XINT_g43\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g45\endcsname
 {\noexpand\csname XINT_g44\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g46\endcsname
 {\noexpand\csname XINT_g45\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g47\endcsname
 {\noexpand\csname XINT_g46\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g48\endcsname
 {\noexpand\csname XINT_g47\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\let\csname XINT_g50\endcsname\endcsname
\expandafter\edef\csname XINT_g51\endcsname
 {\noexpand\csname XINT_g48\expandafter\noexpand\csname XINT_g41\endcsname}%
\expandafter\edef\csname XINT_g52\endcsname
 {\noexpand\csname XINT_g51\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g53\endcsname
 {\noexpand\csname XINT_g52\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g54\endcsname
 {\noexpand\csname XINT_g53\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g55\endcsname
 {\noexpand\csname XINT_g54\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g56\endcsname
 {\noexpand\csname XINT_g55\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g57\endcsname
 {\noexpand\csname XINT_g56\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g58\endcsname
 {\noexpand\csname XINT_g57\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\let\csname XINT_g60\endcsname\endcsname
\expandafter\edef\csname XINT_g61\endcsname
 {\noexpand\csname XINT_g58\expandafter\noexpand\csname XINT_g51\endcsname}%
\expandafter\edef\csname XINT_g62\endcsname
 {\noexpand\csname XINT_g61\expandafter\noexpand\csname XINT_g61\endcsname}%
\expandafter\edef\csname XINT_g63\endcsname
 {\noexpand\csname XINT_g62\expandafter\noexpand\csname XINT_g61\endcsname}%
\expandafter\edef\csname XINT_g64\endcsname
 {\noexpand\csname XINT_g63\expandafter\noexpand\csname XINT_g61\endcsname}%
\expandafter\edef\csname XINT_g65\endcsname
 {\noexpand\csname XINT_g64\expandafter\noexpand\csname XINT_g61\endcsname}%
\expandafter\edef\csname XINT_g66\endcsname
 {\noexpand\csname XINT_g65\expandafter\noexpand\csname XINT_g61\endcsname}%
\expandafter\edef\csname XINT_g67\endcsname
 {\noexpand\csname XINT_g66\expandafter\noexpand\csname XINT_g61\endcsname}%
\expandafter\edef\csname XINT_g68\endcsname
 {\noexpand\csname XINT_g67\expandafter\noexpand\csname XINT_g61\endcsname}%
%    \end{macrocode}
% \subsection{(WIP) \csh{xintUniformDeviate}}
% \changed{1.3b} See user manual for related information.
%    \begin{macrocode}
\ifdefined\xint_texuniformdeviate
     \expandafter\xint_firstoftwo
\else\expandafter\xint_secondoftwo
\fi
{% 
  \def\xintUniformDeviate#1%
     {\the\numexpr\expandafter\XINT_uniformdeviate_sgnfork\the\numexpr#1\xint:}%
  \def\XINT_uniformdeviate_sgnfork#1%
  {%
     \if-#1\XINT_uniformdeviate_neg\fi \XINT_uniformdeviate{}#1%
  }%
  \def\XINT_uniformdeviate_neg\fi\XINT_uniformdeviate#1-%
  {%
      \fi-\numexpr\XINT_uniformdeviate\relax
  }%
  \def\XINT_uniformdeviate#1#2\xint:
  {%(
      \expandafter\XINT_uniformdeviate_a\the\numexpr%
                   -\xint_texuniformdeviate\xint_c_ii^vii%
                   -\xint_c_ii^vii*\xint_texuniformdeviate\xint_c_ii^vii%
                   -\xint_c_ii^xiv*\xint_texuniformdeviate\xint_c_ii^vii%
                   -\xint_c_ii^xxi*\xint_texuniformdeviate\xint_c_ii^vii%
                   +\xint_texuniformdeviate#2\xint:/#2)*#2\xint:+#2\fi\relax#1%
  }%
  \def\XINT_uniformdeviate_a #1\xint:
  {%
      \expandafter\XINT_uniformdeviate_b\the\numexpr#1-(#1%
  }%
  \def\XINT_uniformdeviate_b#1#2\xint:{#1#2\if-#1}%
}%
{%
  \def\xintUniformDeviate#1%
  {%
      \the\numexpr
      \XINT_expandableerror{(xintkernel) No uniformdeviate primitive!}%
      0\relax
  }%
}%
%    \end{macrocode}
% \subsection{\csh{xintMessage}, \csh{ifxintverbose}}
% \changed{1.2c} For use by \csbxint{defvar} and \csbxint{deffunc} of
% \xintexprnameimp.
%
% \changed{1.2e} Uses |\write128| rather than |\write16| for compatibility
% with future extended range of output streams, in LuaTeX in particular.
%
% \changed{1.3e} Set the |\newlinechar|.
%    \begin{macrocode}
\def\xintMessage #1#2#3{%
    \edef\XINT_newlinechar{\the\newlinechar}%
    \newlinechar10
    \immediate\write128{Package #1 #2: (on line \the\inputlineno)}%
    \immediate\write128{\space\space\space\space#3}%
    \newlinechar\XINT_newlinechar\space
}%
\newif\ifxintverbose
%    \end{macrocode}
% \subsection{\csh{ifxintglobaldefs}, \csh{XINT_global}}\label{src-xintglobaldefstrue}
% \changed{1.3c}
%    \begin{macrocode}
\newif\ifxintglobaldefs
\def\XINT_global{\ifxintglobaldefs\global\fi}%
%    \end{macrocode}
% \subsection{(WIP) Expandable error message}
% \changed[2017/07/26]{1.2l} But really belongs to next major release beyond |1.3|.
% Basically copied over from l3kernel code. Using |\ ! /| control sequence,
% which must be left undefined. |\xintError:| would be 6 letters more.
%
% \changed[2020/01/25]{1.4} Finally rather than |\ ! /| I use |\xint/|.
%
% \changed[2021/05/19]{1.4g} Rewrote to use not an undefined control
% sequence but trigger "Use of |\xint/| doesn't match its definition." message.
%
% \changed[2021/05/20]{1.4g} Things evolve fast and I switch to a third
% method which will exploit "Paragraph ended before |\foo| was complete" style
% error. See
% \centeredline{\url{https://github.com/latex3/latex3/issues/931#issuecomment-845367201}}
% However I can not fully exploit this because xint may be used with
% Plain etex which does not set |\newlinechar|. 
% I can only use a poorman version with no usage of |^^J|. Also
% \ctanpackage{xintsession} could use the |^^J|, maybe I will integrate it
% there.
%
% \def\xintsession{\ctanpackage{xintsession}}
%
% \lverb`&
% I. Explanations on 2021/05/19 and 2021/05/20 before final change
%
% First I tried out things with undefined control sequence such as
%(\       an error was reported by xint ... %
%)
% whose output produces a nice symmetrical display with no \, and with ... both
% on left and right but this reduces drastically the available
% space for the actual error context. No go. But see 2021/05/20 update below!
%
% Having replaced \xint/ by "\xint ", I next opted provisorily for "\Hit
% RET at ?" control sequence, despite it being quite longer. And then I
% thought about using "\ xint error", possibly with an included ^^J in the name,
% or in the context.
%
% I experimented with ^^J in the context.  But the context size is much
% constrained, and when \errorcontextlines is at its default value of 5 for
% etex, not -1 as done by LaTeX, having the info shifted to the right makes it
% actually more visible. (however I have now updated $xintsession$ to 0.2b which
% sets \errorcontextlines to 0)
%
% So I was finally back here to square one, apart from having replaced "\xint/"
% by the more longish "\ xint error", hesitating with "\xinterrupt"...
%
% Then I had the idea to replace the undefined control sequence method by a
% method with a macro \foo defined as \def\foo.{} but used as \foo<space> for
% example.  This gives something like this (the first line will be otherwise
% if engine is run with -file-line-error):
%( ! Use of \xint/ doesn't match its definition.
%: <argument> \xint/  
%:                     Ooops, looks like we are missing a ] (hit RET) 
%)
% \xint/<space> (where the space is the unexpected token, the definition
% expecting rather a full stop) makes for 7 characters
% to compare to \ xint error which had 12, so I gained back 5.
%
% Back to ^^J: I had overlooked that TeX in the first part of the error
% message will display \macro fully, so inserting ^^J in its name allows
% arbitrarily long expandable error messages... as pointed out by BLF in
% latex3/issues#931 as I read on the morning of 2021/05/20. This is very nice
% but requires to predefine control sequences for each message, and also the
% actual arguments #1, #2, ... values can appear only in the context.
%
% And the situation with ^^J is somewhat complicated:
% 
% $xintsession$ sets the \newlinechar to 10, but this is not the case with bare
% usage of xintexpr with etex.  And this matters.  To discuss ^^J we have to
% separate two locations:
%(- it appears in the control sequence name,
%:- or in the context (which itself has two parts)
%)
%
% 1) When in the context, what happens with ^^J is independent of the setting of
% \newlinechar: and with TeXLive pdflatex the ^^J will induce a linebreak, but
% with xelatex it must be used with option -8bit.
%
% $noindent2) When in the control sequence name the behaviour in log/terminal of ^^J is
% influenced by the setting of \newlinechar.  Although with pdflatex it will
% always induce a linebreak, the actual count of characters where TeX will
% forcefully break is influenced by whether ^^J is or not \newlinechar. And
% with xelatex if it is \newlinechar, it does not depend then if -8bit or not,
% but if not \newlinechar then it does and TeX forceful breaks also change
% as for pdflatex.
%
% So, the control sequence name trick can be used to obtain arbitrarily
% long messages, but the \newlinechar must be set.
%
% And in the context, we can try to insert some ^^J but this would need
% with xetex the -8bit option, and anyhow the context size is limited,
% and there is apparently no trick to get it larger.
%
% So, in view of all the above I decided not to use ^^J (rather $&$&J
% here) at all, whether here in the control sequence or the context or
% inserted in \XINT_signalcondition in the context!
%
% I also have a problem with usage from bnumexpr or polexpr for example,
% they would need their own to avoid perhaps displaying \xint/ or analogous.
%
% II. Finally I modified again the method (completely, and no more need for
% funny catcode 7 space as delimiter) as this allows a longer context message,
% starting at start of line, and which obeys ^^J if \newlinechar is set to
% it. It also allows to incorporate non-limited generic explanations as a
% postfix, with linebreaks if \newlinechar is known.
%
% But as $xintexprnameimp can be used with Plain+etex which does not set the
% \newlinechar, I can't use ^^J out of thee box.  I can in $xintsession.
% What I decided finally is to make a conditional definition here.
%
% In both cases I include the "hit RET" (how rather "hit <return>") in the
% control sequence name serving to both provide extra information and trigger
% the error from being defined short and finding a \par.
%
% The maximal size was increased from 48 characters (method with \xint/ being
% badly delimited), to now 55 characters (using "! xint error:<^^J or space>"
% as prefix to the message). Longer messages are truncated at 56 characters
% with an appended "\ETC.".
%
% As it is late on this 2021/05/20, and in order to not have to change all
% usages, I keep \XINT_signalcondition (in $xintcorenameimp) as a one argument
% macro for time being, so will not include a more specific module name.
%
% The \par token has a special role here, and can't be (I)nserted without
% damage, but who would want to insert it in an expandable computation
% anyhow... and I don't need it in my custom error messages for sure.
%
% On 2021/05/21 I add a test about \newlinechar at time of package loading,
% and make two distinct definitions: one using ^^J in the control sequence,
% the other not using it.
%
% The -file-line-error toggle makes it impossible to control if the
% line-break on first line will match next lines.  In the ^^J branch I insert
% "| " (no, finally "$ $ " with two spaces) at start of continuation lines.
% Also I preferred to ensure a good-looking first line break for the case it
% starts with a "!  Paragraph ended ..."  because a priori error messages will
% be read if -file-line-error was emitted only a fortiori (this toggle
% suggests some IDE launched TeX and probably -interaction=nonstopmode).
%
% I will perhaps make another definition in $xintsession$ (it currently loads
% xintexpr prior to having set the \newlinechar, so the no ^^J definition will
% be used, if nothing else is modified there).
%
% With some hesitation I do not insert a ^^J after "! xint error:", as
% Emacs/AucTeX will display only the first line prominently and then the rest
% (which is in file:line:error mode) in one block under "--- TeX said ---".  I
% use the ^^J only in the generic helper message embedded in the control
% sequence. The cases with or without \newlinechar being 10 diverge a bit, as
% in the latter case I had to ensure acceptable linebreaks at 79 chars, and I
% did that first and then had spent enough time on the matter not to add more
% to backport the latest ^^J style message.`
%    \begin{macrocode}
\ifnum\newlinechar=10
\expandafter\def\csname
romannumeral (or \string\expanded,&&J \space
\string\numexpr, ...) expansion could produce its final output.&&J \space
See above the exception specifics.&&J \space
xint will try to recover (if in interactive mode, hit <return>&&J \space
at the ? prompt) and will go ahead hoping repair\endcsname
#1\xint:{}%
\def\XINT_expandableerror#1{%
\def\XINT_expandableerror##1{%
   \expandafter
   \XINT_expandableerrorcontinue
   #1! xint error: ##1\par
}}\expandafter\XINT_expandableerror\csname
romannumeral (or \string\expanded,&&J \space
\string\numexpr, ...) expansion could produce its final output.&&J \space
See above the exception specifics.&&J \space
xint will try to recover (if in interactive mode, hit <return>&&J \space
at the ? prompt) and will go ahead hoping repair\endcsname
\else
\expandafter\def\csname
numexpr or \string\expanded\space or \string\romannumeral\space expansion
could terminate because an exception was raised (see the above short explanation for
specifics). xint will now try to recover (hit <return> if in interactive
mode) and will go ahead hoping repair\endcsname
#1\xint:{}%
\def\XINT_expandableerror#1{%
\def\XINT_expandableerror##1{%
   \expandafter
   \XINT_expandableerrorcontinue
   #1! xint error: ##1\par
}}\expandafter\XINT_expandableerror\csname
numexpr or \string\expanded\space or \string\romannumeral\space expansion
could terminate because an exception was raised (see the above short explanation for
specifics). xint will now try to recover (hit <return> if in interactive
mode) and will go ahead hoping repair\endcsname
\fi
\def\XINT_expandableerrorcontinue#1\par{#1}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintkernel}
% \cleardoublepage\let\xintkernelnameUp\undefined
%\gardesactifs
%\let</xintkernel>\relax
%\let<*xinttools>\gardesinactifs
%</xintkernel>^^A-------------------------------------------------
%<*xinttools>^^A--------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xinttoolsnameUp\endcsname
% \section{Package \xinttoolsnameimp implementation}
% \RaisedLabel{sec:toolsimp}
%
% \localtableofcontents
%
% Release |1.09g| of |2013/11/22| splits off |xinttools.sty| from |xint.sty|.
% Starting with |1.1|, \xinttoolsnameimp ceases being loaded automatically by
% \xintnameimp.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xinttools.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintkernel.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xinttools}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xinttools.sty
      \ifx\w\relax % but xintkernel.sty not yet loaded.
         \def\z{\endgroup\input xintkernel.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintkernel.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xintkernel}}%
          \fi
      \else
        \aftergroup\endinput % xinttools already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xinttools}%
  [2021/07/13 v1.4j Expandable and non-expandable utilities (JFB)]%
%    \end{macrocode}
% \lverb|\XINT_toks is used in macros such as \xintFor. It is not used
% elsewhere in the xint bundle.|
%    \begin{macrocode}
\newtoks\XINT_toks
\xint_firstofone{\let\XINT_sptoken= } %<- space here!
%    \end{macrocode}
% \subsection{\csh{xintgodef}, \csh{xintgoodef}, \csh{xintgfdef}}
% \lverb|1.09i. For use in \xintAssign.|
%    \begin{macrocode}
\def\xintgodef  {\global\xintodef }%
\def\xintgoodef {\global\xintoodef }%
\def\xintgfdef  {\global\xintfdef }%
%    \end{macrocode}
% \subsection{\csh{xintRevWithBraces}}
% \lverb|New with 1.06. Makes the expansion of its argument and then reverses
% the resulting tokens or braced tokens, adding a pair of braces to each (thus,
% maintaining it when it was already there.) The reason for
% \xint:, here and in other locations, is in case #1 expands to nothing,
% the \romannumeral-`0 must be stopped|
%    \begin{macrocode}
\def\xintRevWithBraces         {\romannumeral0\xintrevwithbraces }%
\def\xintRevWithBracesNoExpand {\romannumeral0\xintrevwithbracesnoexpand }%
\long\def\xintrevwithbraces #1%
{%
    \expandafter\XINT_revwbr_loop\expandafter{\expandafter}%
    \romannumeral`&&@#1\xint:\xint:\xint:\xint:%
                      \xint:\xint:\xint:\xint:\xint_bye
}%
\long\def\xintrevwithbracesnoexpand #1%
{%
    \XINT_revwbr_loop {}%
    #1\xint:\xint:\xint:\xint:%
      \xint:\xint:\xint:\xint:\xint_bye
}%
\long\def\XINT_revwbr_loop #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9\XINT_revwbr_finish_a\xint:%
    \XINT_revwbr_loop {{#9}{#8}{#7}{#6}{#5}{#4}{#3}{#2}#1}%
}%
\long\def\XINT_revwbr_finish_a\xint:\XINT_revwbr_loop #1#2\xint_bye
{%
    \XINT_revwbr_finish_b #2\R\R\R\R\R\R\R\Z #1%
}%
\def\XINT_revwbr_finish_b #1#2#3#4#5#6#7#8\Z
{%
    \xint_gob_til_R
            #1\XINT_revwbr_finish_c \xint_gobble_viii
            #2\XINT_revwbr_finish_c \xint_gobble_vii
            #3\XINT_revwbr_finish_c \xint_gobble_vi
            #4\XINT_revwbr_finish_c \xint_gobble_v
            #5\XINT_revwbr_finish_c \xint_gobble_iv
            #6\XINT_revwbr_finish_c \xint_gobble_iii
            #7\XINT_revwbr_finish_c \xint_gobble_ii
            \R\XINT_revwbr_finish_c \xint_gobble_i\Z
}%
%    \end{macrocode}
% \lverb|1.1c revisited this old code and improved upon the earlier endings.|
%    \begin{macrocode}
\def\XINT_revwbr_finish_c#1{%
\def\XINT_revwbr_finish_c##1##2\Z{\expandafter#1##1}%
}\XINT_revwbr_finish_c{ }%
%    \end{macrocode}
% \subsection{\csh{xintZapFirstSpaces}}
% \lverb|1.09f, written [2013/11/01]. Modified (2014/10/21) for release 1.1 to
% correct the bug in case of an empty argument, or argument containing only
% spaces, which had been forgotten in first version. New version is simpler than
% the initial one. This macro does NOT expand its argument.|
%    \begin{macrocode}
\def\xintZapFirstSpaces {\romannumeral0\xintzapfirstspaces }%
\def\xintzapfirstspaces#1{\long
\def\xintzapfirstspaces ##1{\XINT_zapbsp_a #1##1\xint:#1#1\xint:}%
}\xintzapfirstspaces{ }%
%    \end{macrocode}
% \lverb|If the original #1 started with a space, the grabbed #1 is empty. Thus
% _again? will see #1=\xint_bye, and hand over control to _again which will loop
% back into \XINT_zapbsp_a, with one initial space less. If the original #1 did
% not start with a space, or was empty, then the #1 below will be a <sptoken>,
% then an extract of the original #1, not empty and not starting with a space,
% which contains what was up to the first <sp><sp> present in original #1, or,
% if none preexisted, <sptoken> and all of #1 (possibly empty) plus an ending
% \xint:. The added initial space will stop later the \romannumeral0. No
% brace stripping is possible. Control is handed over to \XINT_zapbsp_b which
% strips out the ending \xint:<sp><sp>\xint:|
%    \begin{macrocode}
\def\XINT_zapbsp_a#1{\long\def\XINT_zapbsp_a ##1#1#1{%
  \XINT_zapbsp_again?##1\xint_bye\XINT_zapbsp_b ##1#1#1}%
}\XINT_zapbsp_a{ }%
\long\def\XINT_zapbsp_again? #1{\xint_bye #1\XINT_zapbsp_again }%
\xint_firstofone{\def\XINT_zapbsp_again\XINT_zapbsp_b} {\XINT_zapbsp_a }%
\long\def\XINT_zapbsp_b #1\xint:#2\xint:{#1}%
%    \end{macrocode}
% \subsection{\csh{xintZapLastSpaces}}
% \lverb+1.09f, written [2013/11/01]. +
%    \begin{macrocode}
\def\xintZapLastSpaces {\romannumeral0\xintzaplastspaces }%
\def\xintzaplastspaces#1{\long
\def\xintzaplastspaces ##1{\XINT_zapesp_a {}\empty##1#1#1\xint_bye\xint:}%
}\xintzaplastspaces{ }%
%    \end{macrocode}
% \lverb|The \empty from \xintzaplastspaces is to prevent brace removal in the
% #2 below. The \expandafter chain removes it.|
%    \begin{macrocode}
\xint_firstofone {\long\def\XINT_zapesp_a #1#2 } %<- second space here
    {\expandafter\XINT_zapesp_b\expandafter{#2}{#1}}%
%    \end{macrocode}
% \lverb|Notice again an \empty added here. This is in preparation for possibly looping
% back to \XINT_zapesp_a. If the initial #1 had no <sp><sp>, the stuff however
% will not loop, because #3 will already be <some spaces>\xint_bye. Notice
% that this macro fetches all way to the ending \xint:. This looks not
% very efficient, but how often do we have to strip ending spaces from
% something which also has inner stretches of _multiple_ space tokens ?;-). |
%    \begin{macrocode}
\long\def\XINT_zapesp_b #1#2#3\xint:%
    {\XINT_zapesp_end? #3\XINT_zapesp_e {#2#1}\empty #3\xint:}%
%    \end{macrocode}
% \lverb|When we have been over all possible <sp><sp> things, we reach the
% ending space tokens, and #3 will be a bunch of spaces (possibly none)
% followed by \xint_bye. So the #1 in _end? will be \xint_bye. In all other cases
% #1 can not be \xint_bye (assuming naturally this token does nor arise in
% original input), hence control falls back to \XINT_zapesp_e which will loop back
% to \XINT_zapesp_a.|
%    \begin{macrocode}
\long\def\XINT_zapesp_end? #1{\xint_bye #1\XINT_zapesp_end }%
%    \end{macrocode}
% \lverb|We are done. The #1 here has accumulated all the previous material,
% and is stripped of its ending spaces, if any.|
%    \begin{macrocode}
\long\def\XINT_zapesp_end\XINT_zapesp_e #1#2\xint:{ #1}%
%    \end{macrocode}
% \lverb|We haven't yet reached the end, so we need to re-inject two space
% tokens after what we have gotten so far. Then we loop.|
%    \begin{macrocode}
\def\XINT_zapesp_e#1{%
\long\def\XINT_zapesp_e ##1{\XINT_zapesp_a {##1#1#1}}%
}\XINT_zapesp_e{ }%
%    \end{macrocode}
% \subsection{\csh{xintZapSpaces}}
% \lverb+1.09f, written [2013/11/01]. Modified for 1.1, 2014/10/21 as it has the
% same bug as \xintZapFirstSpaces. We in effect do first \xintZapFirstSpaces,
% then \xintZapLastSpaces.+
%    \begin{macrocode}
\def\xintZapSpaces {\romannumeral0\xintzapspaces }%
\def\xintzapspaces#1{%
\long\def\xintzapspaces ##1% like \xintZapFirstSpaces.
        {\XINT_zapsp_a #1##1\xint:#1#1\xint:}%
}\xintzapspaces{ }%
\def\XINT_zapsp_a#1{%
\long\def\XINT_zapsp_a ##1#1#1%
        {\XINT_zapsp_again?##1\xint_bye\XINT_zapsp_b##1#1#1}%
}\XINT_zapsp_a{ }%
\long\def\XINT_zapsp_again? #1{\xint_bye #1\XINT_zapsp_again }%
\xint_firstofone{\def\XINT_zapsp_again\XINT_zapsp_b} {\XINT_zapsp_a }%
\xint_firstofone{\def\XINT_zapsp_b} {\XINT_zapsp_c }%
\def\XINT_zapsp_c#1{%
\long\def\XINT_zapsp_c ##1\xint:##2\xint:%
        {\XINT_zapesp_a{}\empty ##1#1#1\xint_bye\xint:}%
}\XINT_zapsp_c{ }%
%    \end{macrocode}
% \subsection{\csh{xintZapSpacesB}}
% \lverb+1.09f, written [2013/11/01]. Strips up to one pair of braces (but then
% does not strip spaces inside).+
%    \begin{macrocode}
\def\xintZapSpacesB {\romannumeral0\xintzapspacesb }%
\long\def\xintzapspacesb #1{\XINT_zapspb_one? #1\xint:\xint:%
                         \xint_bye\xintzapspaces {#1}}%
\long\def\XINT_zapspb_one? #1#2%
   {\xint_gob_til_xint: #1\XINT_zapspb_onlyspaces\xint:%
    \xint_gob_til_xint: #2\XINT_zapspb_bracedorone\xint:%
    \xint_bye {#1}}%
\def\XINT_zapspb_onlyspaces\xint:%
    \xint_gob_til_xint:\xint:\XINT_zapspb_bracedorone\xint:%
    \xint_bye #1\xint_bye\xintzapspaces #2{ }%
\long\def\XINT_zapspb_bracedorone\xint:%
    \xint_bye #1\xint:\xint_bye\xintzapspaces #2{ #1}%
%    \end{macrocode}
% \subsection{\csh{xintCSVtoList}, \csh{xintCSVtoListNonStripped}}
% \lverb|\xintCSVtoList transforms a,b,..,z into {a}{b}...{z}. The comma
% separated list may be a macro which is first f-expanded. First included in
% release 1.06. Here, use of \Z (and \R) perfectly safe.
%
% [2013/11/02]: Starting with 1.09f, automatically filters items with
% \xintZapSpacesB to strip away all spaces around commas, and spaces at the start
% and end of the list. The original is kept as \xintCSVtoListNonStripped, and is
% faster. But ... it doesn't strip spaces.
%
% ATTENTION: if the input is empty the output contains one item (empty, of
% course). This means an \xintFor loop always executes at least once the
% iteration, contrarily to \xintFor*.|
%    \begin{macrocode}
\def\xintCSVtoList {\romannumeral0\xintcsvtolist }%
\long\def\xintcsvtolist #1{\expandafter\xintApply
           \expandafter\xintzapspacesb
           \expandafter{\romannumeral0\xintcsvtolistnonstripped{#1}}}%
\def\xintCSVtoListNoExpand {\romannumeral0\xintcsvtolistnoexpand }%
\long\def\xintcsvtolistnoexpand #1{\expandafter\xintApply
           \expandafter\xintzapspacesb
           \expandafter{\romannumeral0\xintcsvtolistnonstrippednoexpand{#1}}}%
\def\xintCSVtoListNonStripped {\romannumeral0\xintcsvtolistnonstripped }%
\def\xintCSVtoListNonStrippedNoExpand
         {\romannumeral0\xintcsvtolistnonstrippednoexpand }%
\long\def\xintcsvtolistnonstripped #1%
{%
    \expandafter\XINT_csvtol_loop_a\expandafter
    {\expandafter}\romannumeral`&&@#1%
        ,\xint_bye,\xint_bye,\xint_bye,\xint_bye
        ,\xint_bye,\xint_bye,\xint_bye,\xint_bye,\Z
}%
\long\def\xintcsvtolistnonstrippednoexpand #1%
{%
    \XINT_csvtol_loop_a
    {}#1,\xint_bye,\xint_bye,\xint_bye,\xint_bye
        ,\xint_bye,\xint_bye,\xint_bye,\xint_bye,\Z
}%
\long\def\XINT_csvtol_loop_a #1#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \xint_bye #9\XINT_csvtol_finish_a\xint_bye
    \XINT_csvtol_loop_b {#1}{{#2}{#3}{#4}{#5}{#6}{#7}{#8}{#9}}%
}%
\long\def\XINT_csvtol_loop_b #1#2{\XINT_csvtol_loop_a {#1#2}}%
\long\def\XINT_csvtol_finish_a\xint_bye\XINT_csvtol_loop_b #1#2#3\Z
{%
    \XINT_csvtol_finish_b #3\R,\R,\R,\R,\R,\R,\R,\Z #2{#1}%
}%
%    \end{macrocode}
% \lverb|1.1c revisits this old code and improves upon the earlier endings.
% But as the _d.. macros have already nine parameters, I needed the
% \expandafter and \xint_gob_til_Z in finish_b (compare \XINT_keep_endb, or
% also \XINT_RQ_end_b).|
%    \begin{macrocode}
\def\XINT_csvtol_finish_b #1,#2,#3,#4,#5,#6,#7,#8\Z
{%
    \xint_gob_til_R
            #1\expandafter\XINT_csvtol_finish_dviii\xint_gob_til_Z
            #2\expandafter\XINT_csvtol_finish_dvii \xint_gob_til_Z
            #3\expandafter\XINT_csvtol_finish_dvi  \xint_gob_til_Z
            #4\expandafter\XINT_csvtol_finish_dv   \xint_gob_til_Z
            #5\expandafter\XINT_csvtol_finish_div  \xint_gob_til_Z
            #6\expandafter\XINT_csvtol_finish_diii \xint_gob_til_Z
            #7\expandafter\XINT_csvtol_finish_dii  \xint_gob_til_Z
            \R\XINT_csvtol_finish_di \Z
}%
\long\def\XINT_csvtol_finish_dviii #1#2#3#4#5#6#7#8#9{ #9}%
\long\def\XINT_csvtol_finish_dvii  #1#2#3#4#5#6#7#8#9{ #9{#1}}%
\long\def\XINT_csvtol_finish_dvi   #1#2#3#4#5#6#7#8#9{ #9{#1}{#2}}%
\long\def\XINT_csvtol_finish_dv    #1#2#3#4#5#6#7#8#9{ #9{#1}{#2}{#3}}%
\long\def\XINT_csvtol_finish_div   #1#2#3#4#5#6#7#8#9{ #9{#1}{#2}{#3}{#4}}%
\long\def\XINT_csvtol_finish_diii  #1#2#3#4#5#6#7#8#9{ #9{#1}{#2}{#3}{#4}{#5}}%
\long\def\XINT_csvtol_finish_dii   #1#2#3#4#5#6#7#8#9%
                                            { #9{#1}{#2}{#3}{#4}{#5}{#6}}%
\long\def\XINT_csvtol_finish_di\Z  #1#2#3#4#5#6#7#8#9%
                                            { #9{#1}{#2}{#3}{#4}{#5}{#6}{#7}}%
%    \end{macrocode}
% \subsection{\csh{xintListWithSep}}
% \lverb|1.04.
% \xintListWithSep {\sep}{{a}{b}...{z}} returns a \sep b \sep ....\sep z. It
% f-expands its second argument. The 'sep' may be \par's: the macro
% \xintlistwithsep etc... are all declared long. 'sep' does not have to be a
% single token. It is not expanded. The "list" argument may be empty.
%
% \xintListWithSepNoExpand does not f-expand its second argument.
%
% This venerable macro from 1.04 remained unchanged for a long time and was
% finally refactored at 1.2p for increased speed. Tests done with a list of
% identical {\x} items and a sep of \z demonstrated a speed increase of about:
%( - 3x for 30 items,
%: - 4.5x for 100 items,
%: - 7.5x--8x for 1000 items.
%) |
%    \begin{macrocode}
\def\xintListWithSep         {\romannumeral0\xintlistwithsep }%
\def\xintListWithSepNoExpand {\romannumeral0\xintlistwithsepnoexpand }%
\long\def\xintlistwithsep #1#2%
    {\expandafter\XINT_lws\expandafter {\romannumeral`&&@#2}{#1}}%
\long\def\xintlistwithsepnoexpand #1#2%
{%
    \XINT_lws_loop_a {#1}#2{\xint_bye\XINT_lws_e_vi}%
       {\xint_bye\XINT_lws_e_v}{\xint_bye\XINT_lws_e_iv}%
       {\xint_bye\XINT_lws_e_iii}{\xint_bye\XINT_lws_e_ii}%
       {\xint_bye\XINT_lws_e_i}{\xint_bye\XINT_lws_e}%
       {\xint_bye\expandafter\space}\xint_bye
}%
\long\def\XINT_lws #1#2%
{%
    \XINT_lws_loop_a {#2}#1{\xint_bye\XINT_lws_e_vi}%
       {\xint_bye\XINT_lws_e_v}{\xint_bye\XINT_lws_e_iv}%
       {\xint_bye\XINT_lws_e_iii}{\xint_bye\XINT_lws_e_ii}%
       {\xint_bye\XINT_lws_e_i}{\xint_bye\XINT_lws_e}%
       {\xint_bye\expandafter\space}\xint_bye
}%
\long\def\XINT_lws_loop_a #1#2#3#4#5#6#7#8#9%
{%
    \xint_bye #9\xint_bye
    \XINT_lws_loop_b {#1}{#2}{#3}{#4}{#5}{#6}{#7}{#8}{#9}%
}%
\long\def\XINT_lws_loop_b #1#2#3#4#5#6#7#8#9%
{%
    \XINT_lws_loop_a {#1}{#2#1#3#1#4#1#5#1#6#1#7#1#8#1#9}%
}%
\long\def\XINT_lws_e_vi\xint_bye\XINT_lws_loop_b #1#2#3#4#5#6#7#8#9\xint_bye
    { #2#1#3#1#4#1#5#1#6#1#7#1#8}%
\long\def\XINT_lws_e_v\xint_bye\XINT_lws_loop_b #1#2#3#4#5#6#7#8\xint_bye
    { #2#1#3#1#4#1#5#1#6#1#7}%
\long\def\XINT_lws_e_iv\xint_bye\XINT_lws_loop_b #1#2#3#4#5#6#7\xint_bye
    { #2#1#3#1#4#1#5#1#6}%
\long\def\XINT_lws_e_iii\xint_bye\XINT_lws_loop_b #1#2#3#4#5#6\xint_bye
    { #2#1#3#1#4#1#5}%
\long\def\XINT_lws_e_ii\xint_bye\XINT_lws_loop_b #1#2#3#4#5\xint_bye
    { #2#1#3#1#4}%
\long\def\XINT_lws_e_i\xint_bye\XINT_lws_loop_b #1#2#3#4\xint_bye
    { #2#1#3}%
\long\def\XINT_lws_e\xint_bye\XINT_lws_loop_b #1#2#3\xint_bye
    { #2}%
%    \end{macrocode}
% \subsection{\csh{xintNthElt}}
% \lverb?First included in release 1.06. Last refactored in 1.2j.
%
% \xintNthElt {i}{List} returns the i th item from List (one pair of braces
% removed). The list is first f-expanded. The \xintNthEltNoExpand does no
% expansion of its second argument. Both variants expand i inside \numexpr.
%
% With i = 0, the number of items is returned using \xintLength but with the
% List argument f-expanded first.
%
% Negative values return the |i|th element from the end. 
%
% When i is out of range, an empty value is returned.
% ?
%    \begin{macrocode}
\def\xintNthElt         {\romannumeral0\xintnthelt }%
\def\xintNthEltNoExpand {\romannumeral0\xintntheltnoexpand }%
\long\def\xintnthelt #1#2{\expandafter\XINT_nthelt_a\the\numexpr #1\expandafter.%
                        \expandafter{\romannumeral`&&@#2}}%
\def\xintntheltnoexpand #1{\expandafter\XINT_nthelt_a\the\numexpr #1.}%
\def\XINT_nthelt_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_nthelt_zero
        0#1\XINT_nthelt_neg
         0-{\XINT_nthelt_pos #1}%
    \krof
}%
\def\XINT_nthelt_zero #1.{\xintlength }%
\long\def\XINT_nthelt_neg #1.#2%
{%
    \expandafter\XINT_nthelt_neg_a\the\numexpr\xint_c_i+\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    -#1.#2\xint_bye
}%
\def\XINT_nthelt_neg_a #1%
{%
    \xint_UDzerominusfork
        #1-\xint_stop_afterbye
        0#1\xint_stop_afterbye
         0-{}%
    \krof
    \expandafter\XINT_nthelt_neg_b
    \romannumeral\expandafter\XINT_gobble\the\numexpr-\xint_c_i+#1%
}%
\long\def\XINT_nthelt_neg_b #1#2\xint_bye{ #1}%
\long\def\XINT_nthelt_pos #1.#2%
{%
    \expandafter\XINT_nthelt_pos_done
    \romannumeral0\expandafter\XINT_trim_loop\the\numexpr#1-\xint_c_x.%
     #2\xint:\xint:\xint:\xint:\xint:%
       \xint:\xint:\xint:\xint:\xint:%
    \xint_bye
}%
\def\XINT_nthelt_pos_done #1{%
\long\def\XINT_nthelt_pos_done ##1##2\xint_bye{%
  \xint_gob_til_xint:##1\expandafter#1\xint_gobble_ii\xint:#1##1}%
}\XINT_nthelt_pos_done{ }%
%    \end{macrocode}
% \subsection{\csh{xintNthOnePy}}
% \lverb|
% First included in release 1.4. See relevant code comments in xintexpr.
% |
%    \begin{macrocode}
\def\xintNthOnePy         {\romannumeral0\xintnthonepy }%
\def\xintNthOnePyNoExpand {\romannumeral0\xintnthonepynoexpand }%
\long\def\xintnthonepy #1#2{\expandafter\XINT_nthonepy_a\the\numexpr #1\expandafter.%
                        \expandafter{\romannumeral`&&@#2}}%
\def\xintnthonepynoexpand #1{\expandafter\XINT_nthonepy_a\the\numexpr #1.}%
\def\XINT_nthonepy_a #1%
{%
    \xint_UDsignfork
        #1\XINT_nthonepy_neg
         -{\XINT_nthonepy_nonneg #1}%
    \krof
}%
\long\def\XINT_nthonepy_neg #1.#2%
{%
    \expandafter\XINT_nthonepy_neg_a\the\numexpr\xint_c_i+\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    -#1.#2\xint_bye
}%
\def\XINT_nthonepy_neg_a #1%
{%
    \xint_UDzerominusfork
        #1-\xint_stop_afterbye
        0#1\xint_stop_afterbye
         0-{}%
    \krof
    \expandafter\XINT_nthonepy_neg_b
    \romannumeral\expandafter\XINT_gobble\the\numexpr-\xint_c_i+#1%
}%
\long\def\XINT_nthonepy_neg_b #1#2\xint_bye{{#1}}%
\long\def\XINT_nthonepy_nonneg #1.#2%
{%
    \expandafter\XINT_nthonepy_nonneg_done
    \romannumeral0\expandafter\XINT_trim_loop\the\numexpr#1-\xint_c_ix.%
     #2\xint:\xint:\xint:\xint:\xint:%
       \xint:\xint:\xint:\xint:\xint:%
    \xint_bye
}%
\def\XINT_nthonepy_nonneg_done #1{%
\long\def\XINT_nthonepy_nonneg_done ##1##2\xint_bye{%
  \xint_gob_til_xint:##1\expandafter#1\xint_gobble_ii\xint:{##1}}%
}\XINT_nthonepy_nonneg_done{ }%
%    \end{macrocode}
% \subsection{\csh{xintKeep}}
% \lverb@&
%
% First included in release 1.09m.
%
% \xintKeep{i}{L} f-expands its second argument L. It then grabs the first i
% items from L and discards the rest. 
%
% ATTENTION: **each such kept item is returned inside a brace pair**
% Use \xintKeepUnbraced to avoid that.
%
% For i equal or larger to the number N of items in (expanded) L, the full L
% is returned (with braced items). For i=0, the macro returns an empty output.
% For i<0, the macro discards the first N-|i| items. No brace pairs added to
% the remaining items. For i is less or equal to -N, the full L is returned
% (with no braces added.)
%
% \xintKeepNoExpand does not expand the L argument.
%
%
%
% Prior to 1.2i the code proceeded along a loop with no pre-computation of
% the length of L, for the i>0 case. The faster 1.2i version takes advantage
% of novel \xintLengthUpTo from xintkernel.sty.
% @
%    \begin{macrocode}
\def\xintKeep         {\romannumeral0\xintkeep }%
\def\xintKeepNoExpand {\romannumeral0\xintkeepnoexpand }%
\long\def\xintkeep #1#2{\expandafter\XINT_keep_a\the\numexpr #1\expandafter.%
                        \expandafter{\romannumeral`&&@#2}}%
\def\xintkeepnoexpand #1{\expandafter\XINT_keep_a\the\numexpr #1.}%
\def\XINT_keep_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_keep_keepnone
        0#1\XINT_keep_neg
         0-{\XINT_keep_pos #1}%
    \krof
}%
\long\def\XINT_keep_keepnone .#1{ }%
\long\def\XINT_keep_neg #1.#2%
{%
    \expandafter\XINT_keep_neg_a\the\numexpr
    #1-\numexpr\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye.#2%
}%
\def\XINT_keep_neg_a #1%
{%
    \xint_UDsignfork
        #1{\expandafter\space\romannumeral\XINT_gobble}%
         -\XINT_keep_keepall
    \krof
}%
\def\XINT_keep_keepall #1.{ }%
\long\def\XINT_keep_pos #1.#2%
{%
    \expandafter\XINT_keep_loop
    \the\numexpr#1-\XINT_lengthupto_loop
    #1.#2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
         \xint_c_vii\xint_c_vi\xint_c_v\xint_c_iv
         \xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye.%
    -\xint_c_viii.{}#2\xint_bye%
}%
\def\XINT_keep_loop #1#2.%
{%
    \xint_gob_til_minus#1\XINT_keep_loop_end-%
    \expandafter\XINT_keep_loop
    \the\numexpr#1#2-\xint_c_viii\expandafter.\XINT_keep_loop_pickeight
}%
\long\def\XINT_keep_loop_pickeight 
     #1#2#3#4#5#6#7#8#9{{#1{#2}{#3}{#4}{#5}{#6}{#7}{#8}{#9}}}%
\def\XINT_keep_loop_end-\expandafter\XINT_keep_loop
    \the\numexpr-#1-\xint_c_viii\expandafter.\XINT_keep_loop_pickeight
    {\csname XINT_keep_end#1\endcsname}%
\long\expandafter\def\csname XINT_keep_end1\endcsname
   #1#2#3#4#5#6#7#8#9\xint_bye { #1{#2}{#3}{#4}{#5}{#6}{#7}{#8}}%
\long\expandafter\def\csname XINT_keep_end2\endcsname
   #1#2#3#4#5#6#7#8\xint_bye { #1{#2}{#3}{#4}{#5}{#6}{#7}}%
\long\expandafter\def\csname XINT_keep_end3\endcsname
   #1#2#3#4#5#6#7\xint_bye { #1{#2}{#3}{#4}{#5}{#6}}%
\long\expandafter\def\csname XINT_keep_end4\endcsname
   #1#2#3#4#5#6\xint_bye { #1{#2}{#3}{#4}{#5}}%
\long\expandafter\def\csname XINT_keep_end5\endcsname
   #1#2#3#4#5\xint_bye { #1{#2}{#3}{#4}}%
\long\expandafter\def\csname XINT_keep_end6\endcsname
   #1#2#3#4\xint_bye { #1{#2}{#3}}%
\long\expandafter\def\csname XINT_keep_end7\endcsname
   #1#2#3\xint_bye { #1{#2}}%
\long\expandafter\def\csname XINT_keep_end8\endcsname
   #1#2\xint_bye { #1}%
%    \end{macrocode}
% \subsection{\csh{xintKeepUnbraced}}
% \lverb?1.2a. Same as \xintKeep but will *not* add (or maintain) brace pairs
% around the kept items when length(L)>i>0.
%
% The name may cause a mis-understanding: for i<0, (i.e. keeping only
% trailing items), there is no brace removal at all happening.
%
% Modified for 1.2i like \xintKeep.
% ?
%    \begin{macrocode}
\def\xintKeepUnbraced         {\romannumeral0\xintkeepunbraced }%
\def\xintKeepUnbracedNoExpand {\romannumeral0\xintkeepunbracednoexpand }%
\long\def\xintkeepunbraced #1#2%
    {\expandafter\XINT_keepunbr_a\the\numexpr #1\expandafter.%
                        \expandafter{\romannumeral`&&@#2}}%
\def\xintkeepunbracednoexpand #1%
   {\expandafter\XINT_keepunbr_a\the\numexpr #1.}%
\def\XINT_keepunbr_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_keep_keepnone
        0#1\XINT_keep_neg
         0-{\XINT_keepunbr_pos #1}%
    \krof
}%
\long\def\XINT_keepunbr_pos #1.#2%
{%
    \expandafter\XINT_keepunbr_loop
    \the\numexpr#1-\XINT_lengthupto_loop
    #1.#2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
         \xint_c_vii\xint_c_vi\xint_c_v\xint_c_iv
         \xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye.%
    -\xint_c_viii.{}#2\xint_bye%
}%
\def\XINT_keepunbr_loop #1#2.%
{%
    \xint_gob_til_minus#1\XINT_keepunbr_loop_end-%
    \expandafter\XINT_keepunbr_loop
    \the\numexpr#1#2-\xint_c_viii\expandafter.\XINT_keepunbr_loop_pickeight
}%
\long\def\XINT_keepunbr_loop_pickeight 
     #1#2#3#4#5#6#7#8#9{{#1#2#3#4#5#6#7#8#9}}%
\def\XINT_keepunbr_loop_end-\expandafter\XINT_keepunbr_loop
    \the\numexpr-#1-\xint_c_viii\expandafter.\XINT_keepunbr_loop_pickeight
    {\csname XINT_keepunbr_end#1\endcsname}%
\long\expandafter\def\csname XINT_keepunbr_end1\endcsname
   #1#2#3#4#5#6#7#8#9\xint_bye { #1#2#3#4#5#6#7#8}%
\long\expandafter\def\csname XINT_keepunbr_end2\endcsname
   #1#2#3#4#5#6#7#8\xint_bye { #1#2#3#4#5#6#7}%
\long\expandafter\def\csname XINT_keepunbr_end3\endcsname
   #1#2#3#4#5#6#7\xint_bye { #1#2#3#4#5#6}%
\long\expandafter\def\csname XINT_keepunbr_end4\endcsname
   #1#2#3#4#5#6\xint_bye { #1#2#3#4#5}%
\long\expandafter\def\csname XINT_keepunbr_end5\endcsname
   #1#2#3#4#5\xint_bye { #1#2#3#4}%
\long\expandafter\def\csname XINT_keepunbr_end6\endcsname
   #1#2#3#4\xint_bye { #1#2#3}%
\long\expandafter\def\csname XINT_keepunbr_end7\endcsname
   #1#2#3\xint_bye { #1#2}%
\long\expandafter\def\csname XINT_keepunbr_end8\endcsname
   #1#2\xint_bye { #1}%
%    \end{macrocode}
% \subsection{\csh{xintTrim}}
% \lverb?&
%
% First included in release 1.09m.
%
% \xintTrim{i}{L} f-expands its second argument L. It then removes the first i
% items from L and keeps the rest. For i equal or larger to the number N of
% items in (expanded) L, the macro returns an empty output. For i=0, the
% original (expanded) L is returned. For i<0, the macro proceeds from the
% tail. It thus removes the last |i| items, i.e. it keeps the first N-|i|
% items. For |i|>= N, the empty list is returned.
%
% \xintTrimNoExpand does not expand the L argument.
%
% Speed improvements with 1.2i for i<0 branch (which hands over to
% \xintKeep). Speed improvements with 1.2j for i>0 branch which gobbles items
% nine by nine despite not knowing in advance if it will go too far.
% ?
%    \begin{macrocode}
\def\xintTrim         {\romannumeral0\xinttrim }%
\def\xintTrimNoExpand {\romannumeral0\xinttrimnoexpand }%
\long\def\xinttrim #1#2{\expandafter\XINT_trim_a\the\numexpr #1\expandafter.%
                        \expandafter{\romannumeral`&&@#2}}%
\def\xinttrimnoexpand #1{\expandafter\XINT_trim_a\the\numexpr #1.}%
\def\XINT_trim_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_trim_trimnone
        0#1\XINT_trim_neg
         0-{\XINT_trim_pos #1}%
    \krof
}%
\long\def\XINT_trim_trimnone .#1{ #1}%
\long\def\XINT_trim_neg #1.#2%
{%
    \expandafter\XINT_trim_neg_a\the\numexpr
    #1-\numexpr\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    .{}#2\xint_bye
}%
\def\XINT_trim_neg_a #1%
{%
    \xint_UDsignfork
        #1{\expandafter\XINT_keep_loop\the\numexpr-\xint_c_viii+}%
         -\XINT_trim_trimall
    \krof
}%
\def\XINT_trim_trimall#1{%
\def\XINT_trim_trimall {\expandafter#1\xint_bye}%
}\XINT_trim_trimall{ }%
%    \end{macrocode}
% \lverb|This branch doesn't pre-evaluate the length of the list argument.
% Redone again for 1.2j, manages to trim nine by nine. Some non optimal
% looking aspect of the code is for allowing sharing with \xintNthElt.|
%    \begin{macrocode}
\long\def\XINT_trim_pos #1.#2%
{%
    \expandafter\XINT_trim_pos_done\expandafter\space
    \romannumeral0\expandafter\XINT_trim_loop\the\numexpr#1-\xint_c_ix.%
     #2\xint:\xint:\xint:\xint:\xint:%
       \xint:\xint:\xint:\xint:\xint:%
    \xint_bye
}%
\def\XINT_trim_loop #1#2.%
{%
    \xint_gob_til_minus#1\XINT_trim_finish-%
    \expandafter\XINT_trim_loop\the\numexpr#1#2\XINT_trim_loop_trimnine
}%
\long\def\XINT_trim_loop_trimnine #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9\XINT_trim_toofew\xint:-\xint_c_ix.%
}%
\def\XINT_trim_toofew\xint:{*\xint_c_}%
\def\XINT_trim_finish#1{%
\def\XINT_trim_finish-%
    \expandafter\XINT_trim_loop\the\numexpr-##1\XINT_trim_loop_trimnine
{%
    \expandafter\expandafter\expandafter#1%
    \csname xint_gobble_\romannumeral\numexpr\xint_c_ix-##1\endcsname
}}\XINT_trim_finish{ }%
\long\def\XINT_trim_pos_done #1\xint:#2\xint_bye {#1}%
%    \end{macrocode}
% \subsection{\csh{xintTrimUnbraced}}
% \lverb?1.2a. Modified in 1.2i like \xintTrim?
%    \begin{macrocode}
\def\xintTrimUnbraced         {\romannumeral0\xinttrimunbraced }%
\def\xintTrimUnbracedNoExpand {\romannumeral0\xinttrimunbracednoexpand }%
\long\def\xinttrimunbraced #1#2%
    {\expandafter\XINT_trimunbr_a\the\numexpr #1\expandafter.%
                        \expandafter{\romannumeral`&&@#2}}%
\def\xinttrimunbracednoexpand #1%
    {\expandafter\XINT_trimunbr_a\the\numexpr #1.}%
\def\XINT_trimunbr_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_trim_trimnone
        0#1\XINT_trimunbr_neg
         0-{\XINT_trim_pos #1}%
    \krof
}%
\long\def\XINT_trimunbr_neg #1.#2%
{%
    \expandafter\XINT_trimunbr_neg_a\the\numexpr
    #1-\numexpr\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    .{}#2\xint_bye
}%
\def\XINT_trimunbr_neg_a #1%
{%
    \xint_UDsignfork
        #1{\expandafter\XINT_keepunbr_loop\the\numexpr-\xint_c_viii+}%
         -\XINT_trim_trimall
    \krof
}%
%    \end{macrocode}
% \subsection{\csh{xintApply}}
% \lverb|\xintApply {\macro}{{a}{b}...{z}} returns {\macro{a}}...{\macro{b}}
% where each instance of \macro is f-expanded. The list itself is first
% f-expanded and may thus be a macro. Introduced with release 1.04.|
%    \begin{macrocode}
\def\xintApply         {\romannumeral0\xintapply }%
\def\xintApplyNoExpand {\romannumeral0\xintapplynoexpand }%
\long\def\xintapply #1#2%
{%
    \expandafter\XINT_apply\expandafter {\romannumeral`&&@#2}%
    {#1}%
}%
\long\def\XINT_apply #1#2{\XINT_apply_loop_a {}{#2}#1\xint_bye }%
\long\def\xintapplynoexpand #1#2{\XINT_apply_loop_a {}{#1}#2\xint_bye }%
\long\def\XINT_apply_loop_a #1#2#3%
{%
    \xint_bye #3\XINT_apply_end\xint_bye
    \expandafter
    \XINT_apply_loop_b
    \expandafter {\romannumeral`&&@#2{#3}}{#1}{#2}%
}%
\long\def\XINT_apply_loop_b #1#2{\XINT_apply_loop_a {#2{#1}}}%
\long\def\XINT_apply_end\xint_bye\expandafter\XINT_apply_loop_b
    \expandafter #1#2#3{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintApply:x} (WIP, commented-out)}
% \lverb|Done for 1.4 (2020/01/27). For usage in the NumPy-like slicing
% routines. Well, actually, in the end I sticked with old-fashioned (quadratic
% cost) \xintApply for 1.4 2020/01/31 release. See comments there.
%
% (Comments mainly from 2020/01/27, but on 2020/02/24 I comment out
%  the code and add an alternative)
%
% To expand in \expanded context, and does not need to
% do any expansion of its second argument.
%
% This uses techniques I had developed for 1.2i/1.2j Keep, Trim, Length,
% LastItem like macros, and I should revamp venerable \xintApply probably too.
% But the latter f-expandability (if it does not have \expanded at disposal)
% complicates significantly matters as it has to store material and release at
% very end.
%
% Here it is simpler and I am doing it quickly as I really want to release
% 1.4. The \xint: token should not be located in looped over items. I could
% use something more exotic like the null char with catcode 3...
%
%( \long\def\xintApply:x #1#2$%
%: {$%
%:     \XINT_apply:x_loop {#1}#2$%
%:     {\xint:\XINT_apply:x_loop_enda}{\xint:\XINT_apply:x_loop_endb}$%
%:     {\xint:\XINT_apply:x_loop_endc}{\xint:\XINT_apply:x_loop_endd}$%
%:     {\xint:\XINT_apply:x_loop_ende}{\xint:\XINT_apply:x_loop_endf}$%
%:     {\xint:\XINT_apply:x_loop_endg}{\xint:\XINT_apply:x_loop_endh}\xint_bye
%: }$%
%: \long\def\XINT_apply:x_loop #1#2#3#4#5#6#7#8#9$%
%: {$%
%:     \xint_gob_til_xint: #9\xint:
%:     {#1{#2}}{#1{#3}}{#1{#4}}{#1{#5}}{#1{#6}}{#1{#7}}{#1{#8}}{#1{#9}}$%
%:     \XINT_apply:x_loop {#1}$%
%: }$%
%: \long\def\XINT_apply:x_loop_endh\xint: #1\xint_bye{}$%
%: \long\def\XINT_apply:x_loop_endg\xint: #1#2\xint_bye{{#1}}$%
%: \long\def\XINT_apply:x_loop_endf\xint: #1#2#3\xint_bye{{#1}{#2}}$%
%: \long\def\XINT_apply:x_loop_ende\xint: #1#2#3#4\xint_bye{{#1}{#2}{#3}}$%
%: \long\def\XINT_apply:x_loop_endd\xint: #1#2#3#4#5\xint_bye{{#1}{#2}{#3}{#4}}$%
%: \long\def\XINT_apply:x_loop_endc\xint: #1#2#3#4#5#6\xint_bye{{#1}{#2}{#3}{#4}{#5}}$%
%: \long\def\XINT_apply:x_loop_endb\xint: #1#2#3#4#5#6#7\xint_bye{{#1}{#2}{#3}{#4}{#5}{#6}}$%
%: \long\def\XINT_apply:x_loop_enda\xint: #1#2#3#4#5#6#7#8\xint_bye{{#1}{#2}{#3}{#4}{#5}{#6}{#7}}$%
%)
%
% For small number of items gain with respect to \xintApply is little if any
% (might even be a loss).
%
% Picking one by one is possibly better for small number of items. Like
% this for example, the natural simple minded thing:
%
%(\long\def\xintApply:x #1#2$%
%: {$%
%:     \XINT_apply:x_loop {#1}#2\xint_bye\xint_bye
%: }$%
%: \long\def\XINT_apply:x_loop #1#2$%
%: {$%
%:     \xint_bye #2\xint_bye {#1{#2}}$%
%:     \XINT_apply:x_loop {#1}$%
%: }$%
%)
%
% Some variant on 2020/02/24
% 
%( \long\def\xint_Bbye#1\xint_Bye{}$%
%: \long\def\xintApply:x #1#2$%
%: {$%
%:     \XINT_apply:x_loop {#1}#2$%
%:     {\xint_bye}{\xint_bye}{\xint_bye}{\xint_bye}$%
%:     {\xint_bye}{\xint_bye}{\xint_bye}{\xint_bye}\xint_bye
%: }$%
%: \long\def\XINT_apply:x_loop #1#2#3#4#5#6#7#8#9$%
%: {$%
%:     \xint_Bye #2\xint_bye {#1{#2}}$%
%:     \xint_Bye #3\xint_bye {#1{#3}}$%
%:     \xint_Bye #4\xint_bye {#1{#4}}$%
%:     \xint_Bye #5\xint_bye {#1{#5}}$%
%:     \xint_Bye #6\xint_bye {#1{#6}}$%
%:     \xint_Bye #7\xint_bye {#1{#7}}$%
%:     \xint_Bye #8\xint_bye {#1{#8}}$%
%:     \xint_Bye #9\xint_bye {#1{#9}}$%
%:     \XINT_apply:x_loop {#1}$%
%: }$%
%)
% |
% \subsection{\csh{xintApplyUnbraced}}
% \lverb|\xintApplyUnbraced {\macro}{{a}{b}...{z}} returns \macro{a}...\macro{z}
% where each instance of \macro is f-expanded using \romannumeral-`0. The second
% argument may be a macro as it is itself also f-expanded. No braces
% are added: this allows for example a non-expandable \def in \macro, without
% having to do \gdef. Introduced with release 1.06b.|
%    \begin{macrocode}
\def\xintApplyUnbraced {\romannumeral0\xintapplyunbraced }%
\def\xintApplyUnbracedNoExpand {\romannumeral0\xintapplyunbracednoexpand }%
\long\def\xintapplyunbraced #1#2%
{%
    \expandafter\XINT_applyunbr\expandafter {\romannumeral`&&@#2}%
    {#1}%
}%
\long\def\XINT_applyunbr #1#2{\XINT_applyunbr_loop_a {}{#2}#1\xint_bye }%
\long\def\xintapplyunbracednoexpand #1#2%
   {\XINT_applyunbr_loop_a {}{#1}#2\xint_bye }%
\long\def\XINT_applyunbr_loop_a #1#2#3%
{%
    \xint_bye #3\XINT_applyunbr_end\xint_bye
    \expandafter\XINT_applyunbr_loop_b
    \expandafter {\romannumeral`&&@#2{#3}}{#1}{#2}%
}%
\long\def\XINT_applyunbr_loop_b #1#2{\XINT_applyunbr_loop_a {#2#1}}%
\long\def\XINT_applyunbr_end\xint_bye\expandafter\XINT_applyunbr_loop_b
    \expandafter #1#2#3{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintApplyUnbraced:x} (WIP, commented-out)}
% \lverb|Done for 1.4, 2020/01/27. For usage in the NumPy-like slicing
% routines.
%
% The items should not contain \xint: and the applied macro should not contain
% \empty.
%
% Finally, xintexpr.sty 1.4 code did not use this macro but the f-expandable
% one \xintApplyUnbraced.
%
% For 1.4b I prefer leave the code commented out, and classify it as WIP.
%( \long\def\xintApplyUnbraced:x #1#2$%
%: {$%
%:     \XINT_applyunbraced:x_loop {#1}#2$%
%:     {\xint:\XINT_applyunbraced:x_loop_enda}{\xint:\XINT_applyunbraced:x_loop_endb}$%
%:     {\xint:\XINT_applyunbraced:x_loop_endc}{\xint:\XINT_applyunbraced:x_loop_endd}$%
%:     {\xint:\XINT_applyunbraced:x_loop_ende}{\xint:\XINT_applyunbraced:x_loop_endf}$%
%:     {\xint:\XINT_applyunbraced:x_loop_endg}{\xint:\XINT_applyunbraced:x_loop_endh}\xint_bye
%: }$%
%: \long\def\XINT_applyunbraced:x_loop #1#2#3#4#5#6#7#8#9$%
%: {$%
%:     \xint_gob_til_xint: #9\xint:
%:             #1{#2}$%
%:       \empty#1{#3}$%
%:       \empty#1{#4}$%
%:       \empty#1{#5}$%
%:       \empty#1{#6}$%
%:       \empty#1{#7}$%
%:       \empty#1{#8}$%
%:       \empty#1{#9}$%
%:     \XINT_applyunbraced:x_loop {#1}$%
%: }$%
%: \long\def\XINT_applyunbraced:x_loop_endh\xint: #1\xint_bye{}$%
%: \long\def\XINT_applyunbraced:x_loop_endg\xint: #1\empty#2\xint_bye{#1}$%
%: \long\def\XINT_applyunbraced:x_loop_endf\xint: #1\empty
%:                                                #2\empty#3\xint_bye{#1#2}$%
%: \long\def\XINT_applyunbraced:x_loop_ende\xint: #1\empty
%:                                                #2\empty
%:                                                #3\empty#4\xint_bye{#1#2#3}$%
%: \long\def\XINT_applyunbraced:x_loop_endd\xint: #1\empty
%:                                                #2\empty
%:                                                #3\empty
%:                                                #4\empty#5\xint_bye{#1#2#3#4}$%
%: \long\def\XINT_applyunbraced:x_loop_endc\xint: #1\empty
%:                                                #2\empty
%:                                                #3\empty
%:                                                #4\empty
%:                                                #5\empty#6\xint_bye{#1#2#3#4#5}$%
%: \long\def\XINT_applyunbraced:x_loop_endb\xint: #1\empty
%:                                                #2\empty
%:                                                #3\empty
%:                                                #4\empty
%:                                                #5\empty
%:                                                #6\empty#7\xint_bye{#1#2#3#4#5#6}$%
%: \long\def\XINT_applyunbraced:x_loop_enda\xint: #1\empty
%:                                                #2\empty
%:                                                #3\empty
%:                                                #4\empty
%:                                                #5\empty
%:                                                #6\empty
%:                                                #7\empty#8\xint_bye{#1#2#3#4#5#6#7}$%
%)
% |
% \subsection{\csh{xintZip} (WIP, not public)}
% \lverb|1.4b. (2020/02/25)
%
% Support for zip(). Requires \expanded.
%
% The implementation here thus considers the argument is already completely
% expanded and is a sequence of nut-ples. I will come back at later date for
% more generic macros.
%
% Consider even the name of the function zip() as WIP.
%
% As per what this does, it imitates the zip() function. See xint-manual.pdf.
%
% I use lame terminators. Will think again later on this. I have to be careful
% with the used terminators, in particular with the NE context in mind.
%
% Generally speaking I will think another day about efficiency else I will
% never start this.
%
% OK, done. More compact than I initially thought. Various things should be
% commented upon here. Well, actually not so compact in the end as I basically
% had to double the whole thing simply to avoid the overhead of having to grab
% the final result delimited by some
% \xint_bye\xint_bye\xint_bye\xint_bye\empty terminator. Now actually rather
% \xint_bye\xint_bye\xint_bye\xint_bye\xint: |
%    \begin{macrocode}
\def\xintZip #1{\expanded\XINT_zip_A#1\xint_bye\xint_bye}%
\def\XINT_zip_A#1%
{%
    \xint_bye#1{\expandafter}\xint_bye
    \expanded{\unexpanded{\XINT_ziptwo_A
        #1\xint_bye\xint_bye\xint_bye\xint_bye\xint:}\expandafter}%
    \expanded\XINT_zip_a
}%
\def\XINT_zip_a#1%
{%
    \xint_bye#1\XINT_zip_terminator\xint_bye
    \expanded{\unexpanded{\XINT_ziptwo_a
        #1\xint_bye\xint_bye\xint_bye\xint_bye\xint:}\expandafter}%
    \expanded\XINT_zip_a
}%
\def\XINT_zip_terminator\xint_bye#1\xint_bye{{}\empty\empty\empty\empty\xint:}%
\def\XINT_ziptwo_a #1#2#3#4#5\xint:#6#7#8#9%
{%
    \bgroup
    \xint_bye #1\XINT_ziptwo_e \xint_bye
    \xint_bye #6\XINT_ziptwo_e \xint_bye {{#1}#6}%
    \xint_bye #2\XINT_ziptwo_e \xint_bye
    \xint_bye #7\XINT_ziptwo_e \xint_bye {{#2}#7}%
    \xint_bye #3\XINT_ziptwo_e \xint_bye
    \xint_bye #8\XINT_ziptwo_e \xint_bye {{#3}#8}%
    \xint_bye #4\XINT_ziptwo_e \xint_bye
    \xint_bye #9\XINT_ziptwo_e \xint_bye {{#4}#9}%
%    \end{macrocode}
% \lverb|Attention here that #6 can very well deliver no tokens at all. But
% the \ifx will then do the expected thing. Only mentioning!
%
% By the way, the \xint_bye method means TeX needs to look into tokens
% but skipping braced groups. A conditional based method lets TeX look only
% at the start but then it has to find \else or \fi so here also it must looks
% at tokens, and actually goes into braced groups. But (written 2020/02/26) I
% never did serious testing comparing the two, and in xint I have usually
% preferred \xint_bye/\xint_gob_til_foo types of methods (they proved superior
% than \ifnum to check for 0000 in numerical core context for example, at the
% early days when xint used blocks of 4 digits, not 8), or usage of \if/\ifx
% only on single tokens, combined with some \xint_dothis/\xint_orthat syntax.
% |
%    \begin{macrocode}
    \ifx \empty#6\expandafter\XINT_zipone_a\fi
    \XINT_ziptwo_b #5\xint:
}%
\def\XINT_zipone_a\XINT_ziptwo_b{\XINT_zipone_b}%
\def\XINT_ziptwo_b #1#2#3#4#5\xint:#6#7#8#9%
{%
    \xint_bye #1\XINT_ziptwo_e \xint_bye
    \xint_bye #6\XINT_ziptwo_e \xint_bye {{#1}#6}%
    \xint_bye #2\XINT_ziptwo_e \xint_bye
    \xint_bye #7\XINT_ziptwo_e \xint_bye {{#2}#7}%
    \xint_bye #3\XINT_ziptwo_e \xint_bye
    \xint_bye #8\XINT_ziptwo_e \xint_bye {{#3}#8}%
    \xint_bye #4\XINT_ziptwo_e \xint_bye
    \xint_bye #9\XINT_ziptwo_e \xint_bye {{#4}#9}%
    \XINT_ziptwo_b #5\xint:
}%
\def\XINT_ziptwo_e #1\XINT_ziptwo_b #2\xint:#3\xint:
    {\iffalse{\fi}\xint_bye\xint_bye\xint_bye\xint_bye\xint:}%
\def\XINT_zipone_b #1#2#3#4%
{%
    \xint_bye #1\XINT_zipone_e \xint_bye {{#1}}%
    \xint_bye #2\XINT_zipone_e \xint_bye {{#2}}%
    \xint_bye #3\XINT_zipone_e \xint_bye {{#3}}%
    \xint_bye #4\XINT_zipone_e \xint_bye {{#4}}%
    \XINT_zipone_b
}%
\def\XINT_zipone_e #1\XINT_zipone_b #2\xint:
    {\iffalse{\fi}\xint_bye\xint_bye\xint_bye\xint_bye\empty}%
\def\XINT_ziptwo_A #1#2#3#4#5\xint:#6#7#8#9%
{%
    \bgroup
    \xint_bye #1\XINT_ziptwo_end \xint_bye
    \xint_bye #6\XINT_ziptwo_end \xint_bye {{#1}#6}%
    \xint_bye #2\XINT_ziptwo_end \xint_bye
    \xint_bye #7\XINT_ziptwo_end \xint_bye {{#2}#7}%
    \xint_bye #3\XINT_ziptwo_end \xint_bye
    \xint_bye #8\XINT_ziptwo_end \xint_bye {{#3}#8}%
    \xint_bye #4\XINT_ziptwo_end \xint_bye
    \xint_bye #9\XINT_ziptwo_end \xint_bye {{#4}#9}%
    \ifx \empty#6\expandafter\XINT_zipone_A\fi
    \XINT_ziptwo_B #5\xint:
}%
\def\XINT_zipone_A\XINT_ziptwo_B{\XINT_zipone_B}%
\def\XINT_ziptwo_B #1#2#3#4#5\xint:#6#7#8#9%
{%
    \xint_bye #1\XINT_ziptwo_end \xint_bye
    \xint_bye #6\XINT_ziptwo_end \xint_bye {{#1}#6}%
    \xint_bye #2\XINT_ziptwo_end \xint_bye
    \xint_bye #7\XINT_ziptwo_end \xint_bye {{#2}#7}%
    \xint_bye #3\XINT_ziptwo_end \xint_bye
    \xint_bye #8\XINT_ziptwo_end \xint_bye {{#3}#8}%
    \xint_bye #4\XINT_ziptwo_end \xint_bye
    \xint_bye #9\XINT_ziptwo_end \xint_bye {{#4}#9}%
    \XINT_ziptwo_B #5\xint:
}%
\def\XINT_ziptwo_end #1\XINT_ziptwo_B #2\xint:#3\xint:{\iffalse{\fi}}%
\def\XINT_zipone_B #1#2#3#4%
{%
    \xint_bye #1\XINT_zipone_end \xint_bye {{#1}}%
    \xint_bye #2\XINT_zipone_end \xint_bye {{#2}}%
    \xint_bye #3\XINT_zipone_end \xint_bye {{#3}}%
    \xint_bye #4\XINT_zipone_end \xint_bye {{#4}}%
    \XINT_zipone_B
}%
\def\XINT_zipone_end #1\XINT_zipone_B #2\xint:#3\xint:{\iffalse{\fi}}%
%    \end{macrocode}
% \subsection{\csh{xintSeq}}
% \lverb|1.09c. Without the optional argument puts stress on the input stack,
% should not be used to generated thousands of terms then.|
%
% \changed{1.4j}
% This venerable macro had a brace removal bug in case it produced a single
% number: |\xintSeq{10}{10}| expanded to |10| not |{10}|. When I looked at the
% code the bug looked almost deliberate to me, but reading the documentation
% (which I have not modified), the behaviour is really unexpected. And the
% variant with step parameter |\xintSeq[1]{10}{10}| did produce |{10}|, so
% yes, definitely it was a bug!
%
% I take this occasion to do some style (and perhaps efficiency) refactoring
% in the coding.  I feel there is room for improvement, no time this time.
% And I don't touch the variant with step parameter.
%
% Memo: \xintexprnameimp has some variants, a priori on ultra quick look
% they do not look like having similar bug as this one had.
%    \begin{macrocode}
\def\xintSeq {\romannumeral0\xintseq }%
\def\xintseq #1{\XINT_seq_chkopt  #1\xint_bye }%
\def\XINT_seq_chkopt #1%
{%
    \ifx [#1\expandafter\XINT_seq_opt
       \else\expandafter\XINT_seq_noopt
    \fi  #1%
}%
\def\XINT_seq_noopt #1\xint_bye #2%
{%
    \expandafter\XINT_seq
    \the\numexpr#1\expandafter.\the\numexpr #2.%
}%
\def\XINT_seq #1.#2.%
{%
   \ifnum #1=#2 \xint_dothis\XINT_seq_e\fi
   \ifnum #2>#1 \xint_dothis\XINT_seq_pa\fi
                \xint_orthat\XINT_seq_na
   #2.{#1}{#2}%
}%
\def\XINT_seq_e#1.#2{}%
\def\XINT_seq_pa {\expandafter\XINT_seq_p\the\numexpr-\xint_c_i+}%
\def\XINT_seq_na {\expandafter\XINT_seq_n\the\numexpr\xint_c_i+}%
\def\XINT_seq_p #1.#2%
{%
    \ifnum #1>#2
      \expandafter\XINT_seq_p\the
    \else
      \expandafter\XINT_seq_e
    \fi
    \numexpr #1-\xint_c_i.{#2}{#1}%
}%
\def\XINT_seq_n #1.#2%
{%
    \ifnum #1<#2
      \expandafter\XINT_seq_n\the
    \else
      \expandafter\XINT_seq_e
    \fi
    \numexpr #1+\xint_c_i.{#2}{#1}%
}%
%    \end{macrocode}
% Note at time of the |1.4j| bug fix : I definitely should improve this branch
% and diminish the number of expandafter's but no time this time.
%    \begin{macrocode}
\def\XINT_seq_opt [\xint_bye #1]#2#3%
{%
    \expandafter\XINT_seqo\expandafter
    {\the\numexpr #2\expandafter}\expandafter
    {\the\numexpr #3\expandafter}\expandafter
    {\the\numexpr #1}%
}%
\def\XINT_seqo #1#2%
{%
   \ifcase\ifnum #1=#2 0\else\ifnum #2>#1 1\else -1\fi\fi\space
      \expandafter\XINT_seqo_a
   \or
      \expandafter\XINT_seqo_pa
   \else
      \expandafter\XINT_seqo_na
   \fi
   {#1}{#2}%
}%
\def\XINT_seqo_a #1#2#3{ {#1}}%
\def\XINT_seqo_o #1#2#3#4{ #4}%
\def\XINT_seqo_pa #1#2#3%
{%
    \ifcase\ifnum #3=\xint_c_ 0\else\ifnum #3>\xint_c_ 1\else -1\fi\fi\space
           \expandafter\XINT_seqo_o
    \or
           \expandafter\XINT_seqo_pb
    \else
           \xint_afterfi{\expandafter\space\xint_gobble_iv}%
    \fi
    {#1}{#2}{#3}{{#1}}%
}%
\def\XINT_seqo_pb #1#2#3%
{%
    \expandafter\XINT_seqo_pc\expandafter{\the\numexpr #1+#3}{#2}{#3}%
}%
\def\XINT_seqo_pc #1#2%
{%
    \ifnum #1>#2
        \expandafter\XINT_seqo_o
    \else
        \expandafter\XINT_seqo_pd
    \fi
    {#1}{#2}%
}%
\def\XINT_seqo_pd #1#2#3#4{\XINT_seqo_pb {#1}{#2}{#3}{#4{#1}}}%
\def\XINT_seqo_na #1#2#3%
{%
    \ifcase\ifnum #3=\xint_c_ 0\else\ifnum #3>\xint_c_ 1\else -1\fi\fi\space
        \expandafter\XINT_seqo_o
    \or
        \xint_afterfi{\expandafter\space\xint_gobble_iv}%
    \else
        \expandafter\XINT_seqo_nb
    \fi
    {#1}{#2}{#3}{{#1}}%
}%
\def\XINT_seqo_nb #1#2#3%
{%
    \expandafter\XINT_seqo_nc\expandafter{\the\numexpr #1+#3}{#2}{#3}%
}%
\def\XINT_seqo_nc #1#2%
{%
    \ifnum #1<#2
        \expandafter\XINT_seqo_o
    \else
        \expandafter\XINT_seqo_nd
    \fi
    {#1}{#2}%
}%
\def\XINT_seqo_nd #1#2#3#4{\XINT_seqo_nb {#1}{#2}{#3}{#4{#1}}}%
%    \end{macrocode}
%\subsection{\csh{xintloop}, \csh{xintbreakloop}, \csh{xintbreakloopanddo},
% \csh{xintloopskiptonext}}
% \lverb|1.09g [2013/11/22]. Made long with 1.09h.|
%    \begin{macrocode}
\long\def\xintloop #1#2\repeat {#1#2\xintloop_again\fi\xint_gobble_i {#1#2}}%
\long\def\xintloop_again\fi\xint_gobble_i #1{\fi
                             #1\xintloop_again\fi\xint_gobble_i {#1}}%
\long\def\xintbreakloop #1\xintloop_again\fi\xint_gobble_i #2{}%
\long\def\xintbreakloopanddo #1#2\xintloop_again\fi\xint_gobble_i #3{#1}%
\long\def\xintloopskiptonext #1\xintloop_again\fi\xint_gobble_i #2{%
                        #2\xintloop_again\fi\xint_gobble_i {#2}}%
%    \end{macrocode}
% \subsection{\csh{xintiloop},
% \csh{xintiloopindex},
% \csh{xintbracediloopindex},
% \csh{xintouteriloopindex},
% \csh{xintbracedouteriloopindex},
% \csh{xintbreakiloop},
% \csh{xintbreakiloopanddo},
% \csh{xintiloopskiptonext},
% \csh{xintiloopskipandredo}}
%  \lverb|1.09g [2013/11/22]. Made long with 1.09h.
%
% «braced» variants added (2018/04/24) for 1.3b.|
%    \begin{macrocode}
\def\xintiloop [#1+#2]{%
    \expandafter\xintiloop_a\the\numexpr #1\expandafter.\the\numexpr #2.}%
\long\def\xintiloop_a #1.#2.#3#4\repeat{%
    #3#4\xintiloop_again\fi\xint_gobble_iii {#1}{#2}{#3#4}}%
\def\xintiloop_again\fi\xint_gobble_iii #1#2{%
    \fi\expandafter\xintiloop_again_b\the\numexpr#1+#2.#2.}%
\long\def\xintiloop_again_b #1.#2.#3{%
    #3\xintiloop_again\fi\xint_gobble_iii {#1}{#2}{#3}}%
\long\def\xintbreakiloop #1\xintiloop_again\fi\xint_gobble_iii #2#3#4{}%
\long\def\xintbreakiloopanddo
     #1.#2\xintiloop_again\fi\xint_gobble_iii #3#4#5{#1}%
\long\def\xintiloopindex #1\xintiloop_again\fi\xint_gobble_iii #2%
                {#2#1\xintiloop_again\fi\xint_gobble_iii {#2}}%
\long\def\xintbracediloopindex #1\xintiloop_again\fi\xint_gobble_iii #2%
     {{#2}#1\xintiloop_again\fi\xint_gobble_iii {#2}}%
\long\def\xintouteriloopindex #1\xintiloop_again
                         #2\xintiloop_again\fi\xint_gobble_iii #3%
   {#3#1\xintiloop_again #2\xintiloop_again\fi\xint_gobble_iii {#3}}%
\long\def\xintbracedouteriloopindex #1\xintiloop_again
                         #2\xintiloop_again\fi\xint_gobble_iii #3%
   {{#3}#1\xintiloop_again #2\xintiloop_again\fi\xint_gobble_iii {#3}}%
\long\def\xintiloopskiptonext #1\xintiloop_again\fi\xint_gobble_iii #2#3{%
    \expandafter\xintiloop_again_b \the\numexpr#2+#3.#3.}%
\long\def\xintiloopskipandredo #1\xintiloop_again\fi\xint_gobble_iii #2#3#4{%
    #4\xintiloop_again\fi\xint_gobble_iii {#2}{#3}{#4}}%
%    \end{macrocode}
% \subsection{\csh{XINT_xflet}}
% \lverb|1.09e [2013/10/29]: we f-expand unbraced tokens and swallow arising
% space tokens until the dust settles.|
%    \begin{macrocode}
\def\XINT_xflet #1%
{%
    \def\XINT_xflet_macro {#1}\XINT_xflet_zapsp
}%
\def\XINT_xflet_zapsp
{%
    \expandafter\futurelet\expandafter\XINT_token
    \expandafter\XINT_xflet_sp?\romannumeral`&&@%
}%
\def\XINT_xflet_sp?
{%
    \ifx\XINT_token\XINT_sptoken
         \expandafter\XINT_xflet_zapsp
    \else\expandafter\XINT_xflet_zapspB
    \fi
}%
\def\XINT_xflet_zapspB
{%
    \expandafter\futurelet\expandafter\XINT_tokenB
    \expandafter\XINT_xflet_spB?\romannumeral`&&@%
}%
\def\XINT_xflet_spB?
{%
    \ifx\XINT_tokenB\XINT_sptoken
         \expandafter\XINT_xflet_zapspB
    \else\expandafter\XINT_xflet_eq?
    \fi
}%
\def\XINT_xflet_eq?
{%
    \ifx\XINT_token\XINT_tokenB
         \expandafter\XINT_xflet_macro
    \else\expandafter\XINT_xflet_zapsp
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintApplyInline}}
% \lverb|1.09a: \xintApplyInline\macro{{a}{b}...{z}} has the same effect as
% executing \macro{a} and then applying again \xintApplyInline to the shortened
% list {{b}...{z}} until nothing is left. This is a non-expandable command
% which will result in quicker code than using \xintApplyUnbraced. It f-expands
% its second (list) argument first, which may thus be encapsulated in a macro.
%
% Rewritten in 1.09c. Nota bene: uses catcode 3 Z as privated list terminator.|
%    \begin{macrocode}
\catcode`Z 3
\long\def\xintApplyInline #1#2%
{%
  \long\expandafter\def\expandafter\XINT_inline_macro
  \expandafter ##\expandafter 1\expandafter {#1{##1}}%
  \XINT_xflet\XINT_inline_b #2Z% this Z has catcode 3
}%
\def\XINT_inline_b
{%
    \ifx\XINT_token Z\expandafter\xint_gobble_i
    \else\expandafter\XINT_inline_d\fi
}%
\long\def\XINT_inline_d #1%
{%
  \long\def\XINT_item{{#1}}\XINT_xflet\XINT_inline_e
}%
\def\XINT_inline_e
{%
    \ifx\XINT_token Z\expandafter\XINT_inline_w
    \else\expandafter\XINT_inline_f\fi
}%
\def\XINT_inline_f
{%
  \expandafter\XINT_inline_g\expandafter{\XINT_inline_macro {##1}}%
}%
\long\def\XINT_inline_g #1%
{%
   \expandafter\XINT_inline_macro\XINT_item
   \long\def\XINT_inline_macro ##1{#1}\XINT_inline_d
}%
\def\XINT_inline_w #1%
{%
   \expandafter\XINT_inline_macro\XINT_item
}%
%    \end{macrocode}
% \subsection{\csh{xintFor}, \csh{xintFor*}, \csh{xintBreakFor}, \csh{xintBreakForAndDo}}
% \lverb|1.09c [2013/10/09]: a new kind of loop which uses macro parameters
% #1, #2, #3, #4 rather than macros; while not expandable it survives executing
% code closing groups, like what happens in an alignment with the $& character.
% When inserted in a macro for later use, the # character must be doubled.
%
% The non-star variant works on a csv list, which it expands once, the
% star variant works on a token list, which it (repeatedly) f-expands.
%
% 1.09e adds \XINT_forever with \xintintegers, \xintdimensions, \xintrationals
% and \xintBreakFor, \xintBreakForAndDo, \xintifForFirst, \xintifForLast. On
% this occasion \xint_firstoftwo and \xint_secondoftwo are made long.
%
% 1.09f: rewrites large parts of \xintFor code in order to filter the comma
% separated list via \xintCSVtoList which gets rid of spaces. The #1 in
% \XINT_for_forever? has an initial space token which serves two purposes:
% preventing brace stripping, and stopping the expansion made by \xintcsvtolist.
% If the \XINT_forever branch is taken, the added space will not be a problem
% there.
%
% 1.09f rewrites (2013/11/03) the code which now allows all macro parameters
% from #1 to #9 in \xintFor, \xintFor*, and \XINT_forever.
% 1.2i: slightly more robust \xintifForFirst/Last in case of nesting.
% |
%    \begin{macrocode}
\def\XINT_tmpa #1#2{\ifnum #2<#1 \xint_afterfi {{#########2}}\fi}%
\def\XINT_tmpb #1#2{\ifnum #1<#2 \xint_afterfi {{#########2}}\fi}%
\def\XINT_tmpc #1%
{%
    \expandafter\edef \csname XINT_for_left#1\endcsname
               {\xintApplyUnbraced {\XINT_tmpa #1}{123456789}}%
    \expandafter\edef \csname XINT_for_right#1\endcsname
               {\xintApplyUnbraced {\XINT_tmpb #1}{123456789}}%
}%
\xintApplyInline \XINT_tmpc {123456789}%
\long\def\xintBreakFor      #1Z{}%
\long\def\xintBreakForAndDo #1#2Z{#1}%
\def\xintFor {\let\xintifForFirst\xint_firstoftwo
              \let\xintifForLast\xint_secondoftwo
              \futurelet\XINT_token\XINT_for_ifstar }%
\def\XINT_for_ifstar {\ifx\XINT_token*\expandafter\XINT_forx
                                 \else\expandafter\XINT_for \fi }%
\catcode`U 3 % with numexpr
\catcode`V 3 % with xintfrac.sty (xint.sty not enough)
\catcode`D 3 % with dimexpr
\def\XINT_flet_zapsp
{%
    \futurelet\XINT_token\XINT_flet_sp?
}%
\def\XINT_flet_sp?
{%
    \ifx\XINT_token\XINT_sptoken
         \xint_afterfi{\expandafter\XINT_flet_zapsp\romannumeral0}%
    \else\expandafter\XINT_flet_macro
    \fi
}%
\long\def\XINT_for #1#2in#3#4#5%
{%
    \expandafter\XINT_toks\expandafter
        {\expandafter\XINT_for_d\the\numexpr #2\relax {#5}}%
    \def\XINT_flet_macro {\expandafter\XINT_for_forever?\space}%
    \expandafter\XINT_flet_zapsp #3Z%
}%
\def\XINT_for_forever? #1Z%
{%
    \ifx\XINT_token U\XINT_to_forever\fi
    \ifx\XINT_token V\XINT_to_forever\fi
    \ifx\XINT_token D\XINT_to_forever\fi
    \expandafter\the\expandafter\XINT_toks\romannumeral0\xintcsvtolist {#1}Z%
}%
\def\XINT_to_forever\fi #1\xintcsvtolist #2{\fi \XINT_forever #2}%
\long\def\XINT_forx *#1#2in#3#4#5%
{%
    \expandafter\XINT_toks\expandafter
       {\expandafter\XINT_forx_d\the\numexpr #2\relax {#5}}%
    \XINT_xflet\XINT_forx_forever? #3Z%
}%
\def\XINT_forx_forever?
{%
    \ifx\XINT_token U\XINT_to_forxever\fi
    \ifx\XINT_token V\XINT_to_forxever\fi
    \ifx\XINT_token D\XINT_to_forxever\fi
    \XINT_forx_empty?
}%
\def\XINT_to_forxever\fi #1\XINT_forx_empty? {\fi \XINT_forever }%
\catcode`U 11
\catcode`D 11
\catcode`V 11
\def\XINT_forx_empty?
{%
    \ifx\XINT_token Z\expandafter\xintBreakFor\fi
    \the\XINT_toks
}%
\long\def\XINT_for_d #1#2#3%
{%
  \long\def\XINT_y ##1##2##3##4##5##6##7##8##9{#2}%
  \XINT_toks {{#3}}%
  \long\edef\XINT_x {\noexpand\XINT_y \csname XINT_for_left#1\endcsname
                     \the\XINT_toks   \csname XINT_for_right#1\endcsname }%
  \XINT_toks {\XINT_x\let\xintifForFirst\xint_secondoftwo
                     \let\xintifForLast\xint_secondoftwo\XINT_for_d #1{#2}}%
  \futurelet\XINT_token\XINT_for_last?
}%
\long\def\XINT_forx_d #1#2#3%
{%
  \long\def\XINT_y ##1##2##3##4##5##6##7##8##9{#2}%
  \XINT_toks {{#3}}%
  \long\edef\XINT_x {\noexpand\XINT_y \csname XINT_for_left#1\endcsname
                     \the\XINT_toks   \csname XINT_for_right#1\endcsname }%
  \XINT_toks {\XINT_x\let\xintifForFirst\xint_secondoftwo
                     \let\xintifForLast\xint_secondoftwo\XINT_forx_d #1{#2}}%
  \XINT_xflet\XINT_for_last?
}%
\def\XINT_for_last?
{%
  \ifx\XINT_token Z\expandafter\XINT_for_last?yes\fi
  \the\XINT_toks
}%
\def\XINT_for_last?yes
{%
 \let\xintifForLast\xint_firstoftwo
 \xintBreakForAndDo{\XINT_x\xint_gobble_i Z}%
}%
%    \end{macrocode}
% \subsection{\csh{XINT_forever}, \csh{xintintegers}, \csh{xintdimensions}, \csh{xintrationals}}
% \lverb|New with 1.09e. But this used inadvertently \xintiadd/\xintimul which
% have the unnecessary \xintnum overhead. Changed in 1.09f to use
% \xintiiadd/\xintiimul which do not have this overhead. Also 1.09f uses
% \xintZapSpacesB for the \xintrationals case to get rid of leading and ending
% spaces in the #4 and #5 delimited parameters of \XINT_forever_opt_a
% (for \xintintegers and \xintdimensions this is not necessary, due to the use
% of \numexpr resp. \dimexpr in \XINT_?expr_Ua, resp.\XINT_?expr_Da).|
%    \begin{macrocode}
\catcode`U 3
\catcode`D 3
\catcode`V 3
\let\xintegers      U%
\let\xintintegers   U%
\let\xintdimensions D%
\let\xintrationals  V%
\def\XINT_forever #1%
{%
  \expandafter\XINT_forever_a
  \csname XINT_?expr_\ifx#1UU\else\ifx#1DD\else V\fi\fi a\expandafter\endcsname
  \csname XINT_?expr_\ifx#1UU\else\ifx#1DD\else V\fi\fi i\expandafter\endcsname
  \csname XINT_?expr_\ifx#1UU\else\ifx#1DD\else V\fi\fi \endcsname
}%
\catcode`U 11
\catcode`D 11
\catcode`V 11
\def\XINT_?expr_Ua #1#2%
   {\expandafter{\expandafter\numexpr\the\numexpr #1\expandafter\relax
                              \expandafter\relax\expandafter}%
    \expandafter{\the\numexpr #2}}%
\def\XINT_?expr_Da #1#2%
   {\expandafter{\expandafter\dimexpr\number\dimexpr #1\expandafter\relax
                 \expandafter s\expandafter p\expandafter\relax\expandafter}%
    \expandafter{\number\dimexpr #2}}%
\catcode`Z 11
\def\XINT_?expr_Va #1#2%
{%
    \expandafter\XINT_?expr_Vb\expandafter
          {\romannumeral`&&@\xintrawwithzeros{\xintZapSpacesB{#2}}}%
          {\romannumeral`&&@\xintrawwithzeros{\xintZapSpacesB{#1}}}%
}%
\catcode`Z 3
\def\XINT_?expr_Vb #1#2{\expandafter\XINT_?expr_Vc #2.#1.}%
\def\XINT_?expr_Vc #1/#2.#3/#4.%
{%
     \xintifEq {#2}{#4}%
       {\XINT_?expr_Vf {#3}{#1}{#2}}%
       {\expandafter\XINT_?expr_Vd\expandafter
        {\romannumeral0\xintiimul {#2}{#4}}%
        {\romannumeral0\xintiimul {#1}{#4}}%
        {\romannumeral0\xintiimul {#2}{#3}}%
       }%
}%
\def\XINT_?expr_Vd #1#2#3{\expandafter\XINT_?expr_Ve\expandafter {#2}{#3}{#1}}%
\def\XINT_?expr_Ve #1#2{\expandafter\XINT_?expr_Vf\expandafter {#2}{#1}}%
\def\XINT_?expr_Vf #1#2#3{{#2/#3}{{0}{#1}{#2}{#3}}}%
\def\XINT_?expr_Ui {{\numexpr 1\relax}{1}}%
\def\XINT_?expr_Di {{\dimexpr 0pt\relax}{65536}}%
\def\XINT_?expr_Vi {{1/1}{0111}}%
\def\XINT_?expr_U #1#2%
   {\expandafter{\expandafter\numexpr\the\numexpr #1+#2\relax\relax}{#2}}%
\def\XINT_?expr_D #1#2%
   {\expandafter{\expandafter\dimexpr\the\numexpr #1+#2\relax sp\relax}{#2}}%
\def\XINT_?expr_V #1#2{\XINT_?expr_Vx #2}%
\def\XINT_?expr_Vx #1#2%
{%
     \expandafter\XINT_?expr_Vy\expandafter
        {\romannumeral0\xintiiadd {#1}{#2}}{#2}%
}%
\def\XINT_?expr_Vy #1#2#3#4%
{%
     \expandafter{\romannumeral0\xintiiadd {#3}{#1}/#4}{{#1}{#2}{#3}{#4}}%
}%
\def\XINT_forever_a #1#2#3#4%
{%
    \ifx #4[\expandafter\XINT_forever_opt_a
       \else\expandafter\XINT_forever_b
    \fi #1#2#3#4%
}%
\def\XINT_forever_b #1#2#3Z{\expandafter\XINT_forever_c\the\XINT_toks #2#3}%
\long\def\XINT_forever_c #1#2#3#4#5%
    {\expandafter\XINT_forever_d\expandafter #2#4#5{#3}Z}%
\def\XINT_forever_opt_a #1#2#3[#4+#5]#6Z%
{%
    \expandafter\expandafter\expandafter
    \XINT_forever_opt_c\expandafter\the\expandafter\XINT_toks
    \romannumeral`&&@#1{#4}{#5}#3%
}%
\long\def\XINT_forever_opt_c #1#2#3#4#5#6{\XINT_forever_d #2{#4}{#5}#6{#3}Z}%
\long\def\XINT_forever_d #1#2#3#4#5%
{%
  \long\def\XINT_y ##1##2##3##4##5##6##7##8##9{#5}%
  \XINT_toks {{#2}}%
  \long\edef\XINT_x {\noexpand\XINT_y \csname XINT_for_left#1\endcsname
                     \the\XINT_toks   \csname XINT_for_right#1\endcsname }%
  \XINT_x
  \let\xintifForFirst\xint_secondoftwo
  \let\xintifForLast\xint_secondoftwo
  \expandafter\XINT_forever_d\expandafter #1\romannumeral`&&@#4{#2}{#3}#4{#5}%
}%
%    \end{macrocode}
% \subsection{\csh{xintForpair}, \csh{xintForthree}, \csh{xintForfour}}
% \lverb|1.09c.
%
% [2013/11/02] 1.09f \xintForpair delegate to \xintCSVtoList and its
% \xintZapSpacesB the handling of spaces. Does not share code with \xintFor
% anymore.
%
% [2013/11/03] 1.09f: \xintForpair extended to accept #1#2, #2#3 etc... up to
% #8#9, \xintForthree, #1#2#3 up to #7#8#9, \xintForfour id.
%
% 1.2i: slightly more robust \xintifForFirst/Last in case of nesting.
% |
%    \begin{macrocode}
\catcode`j 3
\long\def\xintForpair #1#2#3in#4#5#6%
{%
    \let\xintifForFirst\xint_firstoftwo
    \let\xintifForLast\xint_secondoftwo
    \XINT_toks  {\XINT_forpair_d #2{#6}}%
    \expandafter\the\expandafter\XINT_toks #4jZ%
}%
\long\def\XINT_forpair_d #1#2#3(#4)#5%
{%
  \long\def\XINT_y ##1##2##3##4##5##6##7##8##9{#2}%
  \XINT_toks \expandafter{\romannumeral0\xintcsvtolist{ #4}}%
  \long\edef\XINT_x {\noexpand\XINT_y \csname XINT_for_left#1\endcsname
      \the\XINT_toks \csname XINT_for_right\the\numexpr#1+\xint_c_i\endcsname}%
  \ifx #5j\expandafter\XINT_for_last?yes\fi
  \XINT_x
  \let\xintifForFirst\xint_secondoftwo
  \let\xintifForLast\xint_secondoftwo
  \XINT_forpair_d #1{#2}%
}%
\long\def\xintForthree #1#2#3in#4#5#6%
{%
    \let\xintifForFirst\xint_firstoftwo
    \let\xintifForLast\xint_secondoftwo
    \XINT_toks  {\XINT_forthree_d #2{#6}}%
    \expandafter\the\expandafter\XINT_toks #4jZ%
}%
\long\def\XINT_forthree_d #1#2#3(#4)#5%
{%
  \long\def\XINT_y ##1##2##3##4##5##6##7##8##9{#2}%
  \XINT_toks \expandafter{\romannumeral0\xintcsvtolist{ #4}}%
  \long\edef\XINT_x {\noexpand\XINT_y \csname XINT_for_left#1\endcsname
      \the\XINT_toks \csname XINT_for_right\the\numexpr#1+\xint_c_ii\endcsname}%
  \ifx #5j\expandafter\XINT_for_last?yes\fi
  \XINT_x
  \let\xintifForFirst\xint_secondoftwo
  \let\xintifForLast\xint_secondoftwo
  \XINT_forthree_d #1{#2}%
}%
\long\def\xintForfour #1#2#3in#4#5#6%
{%
    \let\xintifForFirst\xint_firstoftwo
    \let\xintifForLast\xint_secondoftwo
    \XINT_toks  {\XINT_forfour_d #2{#6}}%
    \expandafter\the\expandafter\XINT_toks #4jZ%
}%
\long\def\XINT_forfour_d #1#2#3(#4)#5%
{%
  \long\def\XINT_y ##1##2##3##4##5##6##7##8##9{#2}%
  \XINT_toks \expandafter{\romannumeral0\xintcsvtolist{ #4}}%
  \long\edef\XINT_x {\noexpand\XINT_y \csname XINT_for_left#1\endcsname
     \the\XINT_toks \csname XINT_for_right\the\numexpr#1+\xint_c_iii\endcsname}%
  \ifx #5j\expandafter\XINT_for_last?yes\fi
  \XINT_x
  \let\xintifForFirst\xint_secondoftwo
  \let\xintifForLast\xint_secondoftwo
  \XINT_forfour_d #1{#2}%
}%
\catcode`Z 11
\catcode`j 11
%    \end{macrocode}
% \subsection{\csh{xintAssign}, \csh{xintAssignArray}, \csh{xintDigitsOf}}
% \lverb|\xintAssign {a}{b}..{z}\to\A\B...\Z resp. \xintAssignArray
% {a}{b}..{z}\to\U.
%
% \xintDigitsOf=\xintAssignArray.
%
% 1.1c 2015/09/12 has (belatedly) corrected some "features" of
% \xintAssign which didn't like the case of a space right before the "\to", or
% the case with the first token not an opening brace and the subsequent
% material containing brace groups. The new code handles gracefully these
% situations.|
%    \begin{macrocode}
\def\xintAssign{\def\XINT_flet_macro {\XINT_assign_fork}\XINT_flet_zapsp }%
\def\XINT_assign_fork
{%
    \let\XINT_assign_def\def
    \ifx\XINT_token[\expandafter\XINT_assign_opt
               \else\expandafter\XINT_assign_a
    \fi
}%
\def\XINT_assign_opt [#1]%
{%
    \ifcsname #1def\endcsname
      \expandafter\let\expandafter\XINT_assign_def \csname #1def\endcsname
    \else
      \expandafter\let\expandafter\XINT_assign_def \csname xint#1def\endcsname
    \fi
    \XINT_assign_a
}%
\long\def\XINT_assign_a #1\to
{%
    \def\XINT_flet_macro{\XINT_assign_b}%
    \expandafter\XINT_flet_zapsp\romannumeral`&&@#1\xint:\to
}%
\long\def\XINT_assign_b
{%
    \ifx\XINT_token\bgroup
         \expandafter\XINT_assign_c
    \else\expandafter\XINT_assign_f
    \fi
}%
\long\def\XINT_assign_f #1\xint:\to #2%
{%
    \XINT_assign_def #2{#1}%
}%
\long\def\XINT_assign_c #1%
{%
    \def\xint_temp {#1}%
    \ifx\xint_temp\xint_bracedstopper
        \expandafter\XINT_assign_e
    \else
        \expandafter\XINT_assign_d
    \fi
}%
\long\def\XINT_assign_d #1\to #2%
{%
    \expandafter\XINT_assign_def\expandafter #2\expandafter{\xint_temp}%
    \XINT_assign_c #1\to
}%
\def\XINT_assign_e #1\to {}%
\def\xintRelaxArray #1%
{%
    \edef\XINT_restoreescapechar {\escapechar\the\escapechar\relax}%
    \escapechar -1
    \expandafter\def\expandafter\xint_arrayname\expandafter {\string #1}%
    \XINT_restoreescapechar
    \xintiloop [\csname\xint_arrayname 0\endcsname+-1]
      \global
      \expandafter\let\csname\xint_arrayname\xintiloopindex\endcsname\relax
      \ifnum \xintiloopindex > \xint_c_
    \repeat
    \global\expandafter\let\csname\xint_arrayname 00\endcsname\relax
    \global\let #1\relax
}%
\def\xintAssignArray{\def\XINT_flet_macro {\XINT_assignarray_fork}%
                     \XINT_flet_zapsp }%
\def\XINT_assignarray_fork
{%
    \let\XINT_assignarray_def\def
    \ifx\XINT_token[\expandafter\XINT_assignarray_opt
               \else\expandafter\XINT_assignarray
    \fi
}%
\def\XINT_assignarray_opt [#1]%
{%
    \ifcsname #1def\endcsname
      \expandafter\let\expandafter\XINT_assignarray_def \csname #1def\endcsname
    \else
      \expandafter\let\expandafter\XINT_assignarray_def
                                  \csname xint#1def\endcsname
    \fi
    \XINT_assignarray
}%
\long\def\XINT_assignarray #1\to #2%
{%
    \edef\XINT_restoreescapechar {\escapechar\the\escapechar\relax }%
    \escapechar -1
    \expandafter\def\expandafter\xint_arrayname\expandafter {\string #2}%
    \XINT_restoreescapechar
    \def\xint_itemcount {0}%
    \expandafter\XINT_assignarray_loop \romannumeral`&&@#1\xint:
    \csname\xint_arrayname 00\expandafter\endcsname
    \csname\xint_arrayname 0\expandafter\endcsname
    \expandafter {\xint_arrayname}#2%
}%
\long\def\XINT_assignarray_loop #1%
{%
    \def\xint_temp {#1}%
    \ifx\xint_temp\xint_bracedstopper
       \expandafter\def\csname\xint_arrayname 0\expandafter\endcsname
                   \expandafter{\the\numexpr\xint_itemcount}%
       \expandafter\expandafter\expandafter\XINT_assignarray_end
    \else
       \expandafter\def\expandafter\xint_itemcount\expandafter
                   {\the\numexpr\xint_itemcount+\xint_c_i}%
       \expandafter\XINT_assignarray_def
          \csname\xint_arrayname\xint_itemcount\expandafter\endcsname
             \expandafter{\xint_temp }%
       \expandafter\XINT_assignarray_loop
    \fi
}%
\def\XINT_assignarray_end #1#2#3#4%
{%
    \def #4##1%
    {%
        \romannumeral0\expandafter #1\expandafter{\the\numexpr ##1}%
    }%
    \def #1##1%
    {%
        \ifnum ##1<\xint_c_
            \xint_afterfi{\XINT_expandableerror{Array index is negative: ##1.} }%
        \else
        \xint_afterfi {%
              \ifnum ##1>#2
                  \xint_afterfi
                  {\XINT_expandableerror{Array index is beyond range: ##1 > #2.} }%
              \else\xint_afterfi
       {\expandafter\expandafter\expandafter\space\csname #3##1\endcsname}%
              \fi}%
        \fi
     }%
}%
\let\xintDigitsOf\xintAssignArray
%    \end{macrocode}
%\subsection{CSV (non user documented) variants of Length, Keep, Trim, NthElt, Reverse}
%
% These routines are for use by |\xintListSel:x:csv| and |\xintListSel:f:csv|
% from \xintexprnameimp, and also for the |reversed| and |len| functions.
% Refactored for |1.2j| release, following |1.2i| updates to |\xintKeep|,
% |\xintTrim|, ...
% 
% These macros will remain undocumented in the user manual:
%
% -- they exist primarily for internal use by the \xintexprnameimp parsers,
% hence don't have to be general purpose; for example, they a priori need to
% handle only catcode 12 tokens (not true in |\xintNewExpr|, though)
% hence they are not really worried about
% controlling brace stripping (nevertheless |1.2j| has paid some secondary
% attention to it, see below.) They are not worried about normalizing leading
% spaces either, because none will be encountered when the macros are used as
% auxiliaries to the expression parsers.
%
% -- crucial design elements may change in future:
%
% 1. whether the handled lists must have or not have a final comma. Currently,
% the model is the one of comma separated lists with **no** final comma. But
% this means that there can not be a distinction of principle between a truly
% empty list and a list which contains one item which turns out to be empty.
% More importantly it makes the coding more complicated as it is needed to
% distinguish the empty list from the single-item list, both lacking commas.
%
% For the internal use of \xintexprnameimp, it would be ok to require all list
% items to be terminated by a comma, and this would bring quite some
% simplications here, but as initially I started with non-terminated lists, I
% have left it this way in the |1.2j| refactoring.
%
% 2. the way to represent the empty list. I was tempted for matter of
% optimization and synchronization with \xintexprnameimp context to require
% the empty list to be always represented by a space token and to not let the
% macros admit a completely empty input. But there were complications so for
% the time being |1.2j| does accept truly empty output (it is not
% distinguished from an input equal to a space token) and produces empty
% output for empty list. This means that the status of the «nil» object for
% the \xintexprnameimp parsers is not completely clarified (currently it is
% represented by a space token).
%
% The original Python slicing code in \xintexprnameimp |1.1| used
% |\xintCSVtoList| and |\xintListWithSep{,}| to convert back and forth to
% token lists and apply |\xintKeep/\xintTrim|. Release |1.2g| switched to
% devoted f-expandable macros added to \xinttoolsnameimp. Release |1.2j|
% refactored all these macros as a follow-up to |1.2i| improvements to
% |\xintKeep/\xintTrim|. They were made |\long| on this occasion and
% auxiliary |\xintLengthUpTo:f:csv| was added.
%
% Leading spaces in items are currently maintained as is by the |1.2j|
% macros, even by |\xintNthEltPy:f:csv|, with the exception of the first item,
% as the list is f-expanded. Perhaps |\xintNthEltPy:f:csv| should remove a
% leading space if present in the picked item; anyway, there are no spaces
% for the lists handled internally by the Python slicer of \xintexprnameimp,
% except the «nil» object currently represented by exactly one space.
%
% Kept items (with no leading spaces; but first item special as it will have
% lost a leading space due to f-expansion) will lose a brace pair under
% |\xintKeep:f:csv| if the first argument was positive and strictly less than
% the length of the list. This differs of course from |\xintKeep| (which
% always braces items it outputs when used with positive first argument) and
% also from |\xintKeepUnbraced| in the case when the whole list is kept.
% Actually the case of singleton list is special, and brace removal will
% happen then.
%
% This behaviour was otherwise for releases earlier than |1.2j| and may
% change again.
% 
% Directly usable names are provided, but these macros (and the behaviour as
% described above) are to be considered \emph{unstable} for the time being.
%
% \subsubsection{\csh{xintLength:f:csv}}
% \lverb|1.2g. Redone for 1.2j. Contrarily to \xintLength from xintkernel.sty,
% this one expands its argument.|
%    \begin{macrocode}
\def\xintLength:f:csv {\romannumeral0\xintlength:f:csv}%
\def\xintlength:f:csv #1%
{\long\def\xintlength:f:csv ##1{%
    \expandafter#1\the\numexpr\expandafter\XINT_length:f:csv_a
    \romannumeral`&&@##1\xint:,\xint:,\xint:,\xint:,%
      \xint:,\xint:,\xint:,\xint:,\xint:,%
      \xint_c_ix,\xint_c_viii,\xint_c_vii,\xint_c_vi,%
      \xint_c_v,\xint_c_iv,\xint_c_iii,\xint_c_ii,\xint_c_i,\xint_bye
    \relax
}}\xintlength:f:csv { }%
%    \end{macrocode}
% \lverb|Must first check if empty list.|
%    \begin{macrocode}
\long\def\XINT_length:f:csv_a #1%
{% 
    \xint_gob_til_xint: #1\xint_c_\xint_bye\xint:%
    \XINT_length:f:csv_loop #1%
}%
\long\def\XINT_length:f:csv_loop #1,#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \xint_gob_til_xint: #9\XINT_length:f:csv_finish\xint:%
    \xint_c_ix+\XINT_length:f:csv_loop
}%
\def\XINT_length:f:csv_finish\xint:\xint_c_ix+\XINT_length:f:csv_loop
    #1,#2,#3,#4,#5,#6,#7,#8,#9,{#9\xint_bye}%
%    \end{macrocode}
% \subsubsection{\csh{xintLengthUpTo:f:csv}}
% \lverb|1.2j. \xintLengthUpTo:f:csv{N}{comma-list}. No ending comma. Returns
% -0 if length>N, else returns difference N-length. **N must be non-negative!!**
% 
% Attention to the dot after \xint_bye for the loop interface.|
%    \begin{macrocode}
\def\xintLengthUpTo:f:csv {\romannumeral0\xintlengthupto:f:csv}%
\long\def\xintlengthupto:f:csv #1#2%
{%
    \expandafter\XINT_lengthupto:f:csv_a
    \the\numexpr#1\expandafter.%
    \romannumeral`&&@#2\xint:,\xint:,\xint:,\xint:,%
         \xint:,\xint:,\xint:,\xint:,%
         \xint_c_viii,\xint_c_vii,\xint_c_vi,\xint_c_v,%
         \xint_c_iv,\xint_c_iii,\xint_c_ii,\xint_c_i,\xint_bye.%
}%
%    \end{macrocode}
% \lverb|Must first recognize if empty list. If this is the case, return N.|
%    \begin{macrocode}
\long\def\XINT_lengthupto:f:csv_a #1.#2%
{%
    \xint_gob_til_xint: #2\XINT_lengthupto:f:csv_empty\xint:%
    \XINT_lengthupto:f:csv_loop_b #1.#2%
}%
\def\XINT_lengthupto:f:csv_empty\xint:%
    \XINT_lengthupto:f:csv_loop_b #1.#2\xint_bye.{ #1}%
\def\XINT_lengthupto:f:csv_loop_a #1%
{%
    \xint_UDsignfork
      #1\XINT_lengthupto:f:csv_gt
       -\XINT_lengthupto:f:csv_loop_b
    \krof #1%
}%
\long\def\XINT_lengthupto:f:csv_gt #1\xint_bye.{-0}%
\long\def\XINT_lengthupto:f:csv_loop_b #1.#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \xint_gob_til_xint: #9\XINT_lengthupto:f:csv_finish_a\xint:%
    \expandafter\XINT_lengthupto:f:csv_loop_a\the\numexpr #1-\xint_c_viii.%
}%
\def\XINT_lengthupto:f:csv_finish_a\xint:
    \expandafter\XINT_lengthupto:f:csv_loop_a
    \the\numexpr #1-\xint_c_viii.#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \expandafter\XINT_lengthupto:f:csv_finish_b\the\numexpr #1-#9\xint_bye
}%
\def\XINT_lengthupto:f:csv_finish_b #1#2.%
{%
    \xint_UDsignfork
       #1{-0}%
        -{ #1#2}%
    \krof
}%
%    \end{macrocode}
%\subsubsection{\csh{xintKeep:f:csv}}
% \lverb|1.2g 2016/03/17. Redone for 1.2j with use of \xintLengthUpTo:f:csv.
% Same code skeleton as \xintKeep but handling comma separated but non
% terminated lists has complications. The \xintKeep in case of a negative #1
% uses \xintgobble, we don't have that for comma delimited items, hence we do
% a special loop here (this style of loop is surely competitive with
% xintgobble for a few dozens items and even more). The loop knows before
% starting that it will not go too far.
%
%|
%    \begin{macrocode}
\def\xintKeep:f:csv {\romannumeral0\xintkeep:f:csv }%
\long\def\xintkeep:f:csv #1#2%
{%
    \expandafter\xint_stop_aftergobble
    \romannumeral0\expandafter\XINT_keep:f:csv_a
    \the\numexpr #1\expandafter.\expandafter{\romannumeral`&&@#2}%
}%
\def\XINT_keep:f:csv_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_keep:f:csv_keepnone
        0#1\XINT_keep:f:csv_neg
         0-{\XINT_keep:f:csv_pos #1}%
    \krof
}%
\long\def\XINT_keep:f:csv_keepnone .#1{,}%
\long\def\XINT_keep:f:csv_neg #1.#2%
{%
    \expandafter\XINT_keep:f:csv_neg_done\expandafter,%
    \romannumeral0%
    \expandafter\XINT_keep:f:csv_neg_a\the\numexpr
    #1-\numexpr\XINT_length:f:csv_a
    #2\xint:,\xint:,\xint:,\xint:,%
      \xint:,\xint:,\xint:,\xint:,\xint:,%
      \xint_c_ix,\xint_c_viii,\xint_c_vii,\xint_c_vi,%
      \xint_c_v,\xint_c_iv,\xint_c_iii,\xint_c_ii,\xint_c_i,\xint_bye
    .#2\xint_bye
}%
\def\XINT_keep:f:csv_neg_a #1%
{%
    \xint_UDsignfork
        #1{\expandafter\XINT_keep:f:csv_trimloop\the\numexpr-\xint_c_ix+}%
         -\XINT_keep:f:csv_keepall
    \krof
}%
\def\XINT_keep:f:csv_keepall #1.{ }%
\long\def\XINT_keep:f:csv_neg_done #1\xint_bye{#1}%
\def\XINT_keep:f:csv_trimloop #1#2.%
{%
    \xint_gob_til_minus#1\XINT_keep:f:csv_trimloop_finish-%
    \expandafter\XINT_keep:f:csv_trimloop
    \the\numexpr#1#2-\xint_c_ix\expandafter.\XINT_keep:f:csv_trimloop_trimnine
}%
\long\def\XINT_keep:f:csv_trimloop_trimnine #1,#2,#3,#4,#5,#6,#7,#8,#9,{}%
\def\XINT_keep:f:csv_trimloop_finish-%
    \expandafter\XINT_keep:f:csv_trimloop
    \the\numexpr-#1-\xint_c_ix\expandafter.\XINT_keep:f:csv_trimloop_trimnine
    {\csname XINT_trim:f:csv_finish#1\endcsname}%
\long\def\XINT_keep:f:csv_pos #1.#2%
{%
    \expandafter\XINT_keep:f:csv_pos_fork
    \romannumeral0\XINT_lengthupto:f:csv_a
    #1.#2\xint:,\xint:,\xint:,\xint:,%
         \xint:,\xint:,\xint:,\xint:,%
         \xint_c_viii,\xint_c_vii,\xint_c_vi,\xint_c_v,%
         \xint_c_iv,\xint_c_iii,\xint_c_ii,\xint_c_i,\xint_bye.%
    .#1.{}#2\xint_bye%
}%
\def\XINT_keep:f:csv_pos_fork #1#2.%
{%
    \xint_UDsignfork
      #1{\expandafter\XINT_keep:f:csv_loop\the\numexpr-\xint_c_viii+}%
       -\XINT_keep:f:csv_pos_keepall
    \krof
}%
\long\def\XINT_keep:f:csv_pos_keepall #1.#2#3\xint_bye{,#3}%
\def\XINT_keep:f:csv_loop #1#2.%
{%
    \xint_gob_til_minus#1\XINT_keep:f:csv_loop_end-%
    \expandafter\XINT_keep:f:csv_loop
    \the\numexpr#1#2-\xint_c_viii\expandafter.\XINT_keep:f:csv_loop_pickeight
}%
\long\def\XINT_keep:f:csv_loop_pickeight 
     #1#2,#3,#4,#5,#6,#7,#8,#9,{{#1,#2,#3,#4,#5,#6,#7,#8,#9}}%
\def\XINT_keep:f:csv_loop_end-\expandafter\XINT_keep:f:csv_loop
    \the\numexpr-#1-\xint_c_viii\expandafter.\XINT_keep:f:csv_loop_pickeight
    {\csname XINT_keep:f:csv_end#1\endcsname}%
\long\expandafter\def\csname XINT_keep:f:csv_end1\endcsname
   #1#2,#3,#4,#5,#6,#7,#8,#9\xint_bye {#1,#2,#3,#4,#5,#6,#7,#8}%
\long\expandafter\def\csname XINT_keep:f:csv_end2\endcsname
   #1#2,#3,#4,#5,#6,#7,#8\xint_bye {#1,#2,#3,#4,#5,#6,#7}%
\long\expandafter\def\csname XINT_keep:f:csv_end3\endcsname
   #1#2,#3,#4,#5,#6,#7\xint_bye {#1,#2,#3,#4,#5,#6}%
\long\expandafter\def\csname XINT_keep:f:csv_end4\endcsname
   #1#2,#3,#4,#5,#6\xint_bye {#1,#2,#3,#4,#5}%
\long\expandafter\def\csname XINT_keep:f:csv_end5\endcsname
   #1#2,#3,#4,#5\xint_bye {#1,#2,#3,#4}%
\long\expandafter\def\csname XINT_keep:f:csv_end6\endcsname
   #1#2,#3,#4\xint_bye {#1,#2,#3}%
\long\expandafter\def\csname XINT_keep:f:csv_end7\endcsname
   #1#2,#3\xint_bye {#1,#2}%
\long\expandafter\def\csname XINT_keep:f:csv_end8\endcsname
   #1#2\xint_bye {#1}%
%    \end{macrocode}
%\subsubsection{\csh{xintTrim:f:csv}}
% \lverb|1.2g 2016/03/17. Redone for 1.2j 2016/12/20 on the basis of new
% \xintTrim.|
%    \begin{macrocode}
\def\xintTrim:f:csv {\romannumeral0\xinttrim:f:csv }%
\long\def\xinttrim:f:csv #1#2%
{%
    \expandafter\xint_stop_aftergobble
    \romannumeral0\expandafter\XINT_trim:f:csv_a
    \the\numexpr #1\expandafter.\expandafter{\romannumeral`&&@#2}%
}%
\def\XINT_trim:f:csv_a #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_trim:f:csv_trimnone
        0#1\XINT_trim:f:csv_neg
         0-{\XINT_trim:f:csv_pos #1}%
    \krof
}%
\long\def\XINT_trim:f:csv_trimnone .#1{,#1}%
\long\def\XINT_trim:f:csv_neg #1.#2%
{%
    \expandafter\XINT_trim:f:csv_neg_a\the\numexpr
    #1-\numexpr\XINT_length:f:csv_a
    #2\xint:,\xint:,\xint:,\xint:,%
      \xint:,\xint:,\xint:,\xint:,\xint:,%
      \xint_c_ix,\xint_c_viii,\xint_c_vii,\xint_c_vi,%
      \xint_c_v,\xint_c_iv,\xint_c_iii,\xint_c_ii,\xint_c_i,\xint_bye
    .{}#2\xint_bye
}%
\def\XINT_trim:f:csv_neg_a #1%
{%
    \xint_UDsignfork
        #1{\expandafter\XINT_keep:f:csv_loop\the\numexpr-\xint_c_viii+}%
         -\XINT_trim:f:csv_trimall
    \krof
}%
\def\XINT_trim:f:csv_trimall {\expandafter,\xint_bye}%
\long\def\XINT_trim:f:csv_pos #1.#2%
{%
    \expandafter\XINT_trim:f:csv_pos_done\expandafter,%
    \romannumeral0%
    \expandafter\XINT_trim:f:csv_loop\the\numexpr#1-\xint_c_ix.%
     #2\xint:,\xint:,\xint:,\xint:,\xint:,%
       \xint:,\xint:,\xint:,\xint:,\xint:\xint_bye
}%
\def\XINT_trim:f:csv_loop #1#2.%
{%
    \xint_gob_til_minus#1\XINT_trim:f:csv_finish-%
    \expandafter\XINT_trim:f:csv_loop\the\numexpr#1#2\XINT_trim:f:csv_loop_trimnine
}%
\long\def\XINT_trim:f:csv_loop_trimnine #1,#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \xint_gob_til_xint: #9\XINT_trim:f:csv_toofew\xint:-\xint_c_ix.%
}%
\def\XINT_trim:f:csv_toofew\xint:{*\xint_c_}%
\def\XINT_trim:f:csv_finish-%
    \expandafter\XINT_trim:f:csv_loop\the\numexpr-#1\XINT_trim:f:csv_loop_trimnine
{%
    \csname XINT_trim:f:csv_finish#1\endcsname
}%
\long\expandafter\def\csname XINT_trim:f:csv_finish1\endcsname
  #1,#2,#3,#4,#5,#6,#7,#8,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish2\endcsname
  #1,#2,#3,#4,#5,#6,#7,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish3\endcsname
  #1,#2,#3,#4,#5,#6,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish4\endcsname
  #1,#2,#3,#4,#5,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish5\endcsname
  #1,#2,#3,#4,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish6\endcsname
  #1,#2,#3,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish7\endcsname
  #1,#2,{ }%
\long\expandafter\def\csname XINT_trim:f:csv_finish8\endcsname
  #1,{ }%
\expandafter\let\csname XINT_trim:f:csv_finish9\endcsname\space
\long\def\XINT_trim:f:csv_pos_done #1\xint:#2\xint_bye{#1}%
%    \end{macrocode}
% \subsubsection{\csh{xintNthEltPy:f:csv}}
% \lverb|Counts like Python starting at zero. Last refactored with 1.2j.
% Attention, makes currently no effort at removing leading spaces in the
% picked item.|
%    \begin{macrocode}
\def\xintNthEltPy:f:csv {\romannumeral0\xintntheltpy:f:csv }%
\long\def\xintntheltpy:f:csv #1#2%
{%
   \expandafter\XINT_nthelt:f:csv_a
   \the\numexpr #1\expandafter.\expandafter{\romannumeral`&&@#2}%
}%
\def\XINT_nthelt:f:csv_a #1%
{%
    \xint_UDsignfork
        #1\XINT_nthelt:f:csv_neg
         -\XINT_nthelt:f:csv_pos
    \krof #1%
}%
\long\def\XINT_nthelt:f:csv_neg -#1.#2%
{%
    \expandafter\XINT_nthelt:f:csv_neg_fork
    \the\numexpr\XINT_length:f:csv_a
    #2\xint:,\xint:,\xint:,\xint:,%
      \xint:,\xint:,\xint:,\xint:,\xint:,%
      \xint_c_ix,\xint_c_viii,\xint_c_vii,\xint_c_vi,%
      \xint_c_v,\xint_c_iv,\xint_c_iii,\xint_c_ii,\xint_c_i,\xint_bye
    -#1.#2,\xint_bye
}%
\def\XINT_nthelt:f:csv_neg_fork #1%
{%
    \if#1-\expandafter\xint_stop_afterbye\fi
    \expandafter\XINT_nthelt:f:csv_neg_done
    \romannumeral0%
    \expandafter\XINT_keep:f:csv_trimloop\the\numexpr-\xint_c_ix+#1%
}%
\long\def\XINT_nthelt:f:csv_neg_done#1,#2\xint_bye{ #1}%
\long\def\XINT_nthelt:f:csv_pos #1.#2%
{%
    \expandafter\XINT_nthelt:f:csv_pos_done
    \romannumeral0%
    \expandafter\XINT_trim:f:csv_loop\the\numexpr#1-\xint_c_ix.%
    #2\xint:,\xint:,\xint:,\xint:,\xint:,%
       \xint:,\xint:,\xint:,\xint:,\xint:,\xint_bye
}%
\def\XINT_nthelt:f:csv_pos_done #1{%
\long\def\XINT_nthelt:f:csv_pos_done ##1,##2\xint_bye{%
  \xint_gob_til_xint:##1\XINT_nthelt:f:csv_pos_cleanup\xint:#1##1}%
}\XINT_nthelt:f:csv_pos_done{ }%
%    \end{macrocode}
% \lverb|This strange thing is in case the picked item was the last one, hence
% there was an ending \xint: (we could not put a comma earlier for
% matters of not confusing empty list with a singleton list), and we do this
% here to activate brace-stripping of item as all other items may be
% brace-stripped if picked. This is done for coherence. Of course, in the
% context of the xintexpr.sty parsers, there are no braces in list items...|
%    \begin{macrocode}
\xint_firstofone{\long\def\XINT_nthelt:f:csv_pos_cleanup\xint:} %
   #1\xint:{ #1}%
%    \end{macrocode}
% \subsubsection{\csh{xintReverse:f:csv}}
% \lverb|1.2g. Contrarily to \xintReverseOrder from xintkernel.sty, this
% one expands its argument. Handles empty list too. 2016/03/17.
% Made \long for 1.2j.|
%    \begin{macrocode}
\def\xintReverse:f:csv {\romannumeral0\xintreverse:f:csv }%
\long\def\xintreverse:f:csv #1%
{%
    \expandafter\XINT_reverse:f:csv_loop
    \expandafter{\expandafter}\romannumeral`&&@#1,%
      \xint:,%
        \xint_bye,\xint_bye,\xint_bye,\xint_bye,%
        \xint_bye,\xint_bye,\xint_bye,\xint_bye,%
      \xint:
}%
\long\def\XINT_reverse:f:csv_loop #1#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \xint_bye #9\XINT_reverse:f:csv_cleanup\xint_bye
    \XINT_reverse:f:csv_loop {,#9,#8,#7,#6,#5,#4,#3,#2#1}%
}%
\long\def\XINT_reverse:f:csv_cleanup\xint_bye\XINT_reverse:f:csv_loop #1#2\xint:
{%
    \XINT_reverse:f:csv_finish #1%
}%
\long\def\XINT_reverse:f:csv_finish #1\xint:,{ }%
%    \end{macrocode}
% \subsubsection{\csh{xintFirstItem:f:csv}}
% \lverb|Added with 1.2k for use by first() in
% \xintexpr-essions, and some amount of compatibility with \xintNewExpr.|
%    \begin{macrocode}
\def\xintFirstItem:f:csv {\romannumeral0\xintfirstitem:f:csv}%
\long\def\xintfirstitem:f:csv #1%
{%
    \expandafter\XINT_first:f:csv_a\romannumeral`&&@#1,\xint_bye
}%
\long\def\XINT_first:f:csv_a #1,#2\xint_bye{ #1}%
%    \end{macrocode}
% \subsubsection{\csh{xintLastItem:f:csv}}
% \lverb|Added with 1.2k, based on and sharing code with xintkernel's
% \xintLastItem from 1.2i. Output empty if input empty. f-expands its argument
% (hence first item, if not protected.) For use by last() in
% \xintexpr-essions with to some extent \xintNewExpr compatibility.|
%    \begin{macrocode}
\def\xintLastItem:f:csv {\romannumeral0\xintlastitem:f:csv}%
\long\def\xintlastitem:f:csv #1%
{%
    \expandafter\XINT_last:f:csv_loop\expandafter{\expandafter}\expandafter.%
    \romannumeral`&&@#1,%
    \xint:\XINT_last_loop_enda,\xint:\XINT_last_loop_endb,%
    \xint:\XINT_last_loop_endc,\xint:\XINT_last_loop_endd,%
    \xint:\XINT_last_loop_ende,\xint:\XINT_last_loop_endf,%
    \xint:\XINT_last_loop_endg,\xint:\XINT_last_loop_endh,\xint_bye
}%
\long\def\XINT_last:f:csv_loop #1.#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    \xint_gob_til_xint: #9%
        {#8}{#7}{#6}{#5}{#4}{#3}{#2}{#1}\xint:
    \XINT_last:f:csv_loop {#9}.%
}%
%    \end{macrocode}
% \subsubsection{\csh{xintKeep:x:csv}}
% \lverb|Added to xintexpr at 1.2j.
%
% But data model changed at 1.4, this macro moved to xinttools, not part of
% publicly supported macros, may be removed at any time.
%
% This macro is used only with positive first argument.
% |
%    \begin{macrocode}
\def\xintKeep:x:csv #1#2%
{%
    \expandafter\xint_gobble_i
    \romannumeral0\expandafter\XINT_keep:x:csv_pos
    \the\numexpr #1\expandafter.\expandafter{\romannumeral`&&@#2}%
}%
\def\XINT_keep:x:csv_pos #1.#2%
{%
    \expandafter\XINT_keep:x:csv_loop\the\numexpr#1-\xint_c_viii.%
    #2\xint_Bye,\xint_Bye,\xint_Bye,\xint_Bye,%
       \xint_Bye,\xint_Bye,\xint_Bye,\xint_Bye,\xint_bye
}%
\def\XINT_keep:x:csv_loop #1%
{%
    \xint_gob_til_minus#1\XINT_keep:x:csv_finish-%
    \XINT_keep:x:csv_loop_pickeight #1%
}%
\def\XINT_keep:x:csv_loop_pickeight #1.#2,#3,#4,#5,#6,#7,#8,#9,%
{%
    ,#2,#3,#4,#5,#6,#7,#8,#9%
    \expandafter\XINT_keep:x:csv_loop\the\numexpr#1-\xint_c_viii.%
}%
\def\XINT_keep:x:csv_finish-\XINT_keep:x:csv_loop_pickeight -#1.%
{%
    \csname XINT_keep:x:csv_finish#1\endcsname
}%
\expandafter\def\csname XINT_keep:x:csv_finish1\endcsname
  #1,#2,#3,#4,#5,#6,#7,{,#1,#2,#3,#4,#5,#6,#7\xint_Bye}%
\expandafter\def\csname XINT_keep:x:csv_finish2\endcsname
  #1,#2,#3,#4,#5,#6,{,#1,#2,#3,#4,#5,#6\xint_Bye}%
\expandafter\def\csname XINT_keep:x:csv_finish3\endcsname
  #1,#2,#3,#4,#5,{,#1,#2,#3,#4,#5\xint_Bye}%
\expandafter\def\csname XINT_keep:x:csv_finish4\endcsname
  #1,#2,#3,#4,{,#1,#2,#3,#4\xint_Bye}%
\expandafter\def\csname XINT_keep:x:csv_finish5\endcsname
  #1,#2,#3,{,#1,#2,#3\xint_Bye}%
\expandafter\def\csname XINT_keep:x:csv_finish6\endcsname
  #1,#2,{,#1,#2\xint_Bye}%
\expandafter\def\csname XINT_keep:x:csv_finish7\endcsname
  #1,{,#1\xint_Bye}%
\expandafter\let\csname XINT_keep:x:csv_finish8\endcsname\xint_Bye
%    \end{macrocode}
% \subsubsection{Public names for the undocumented csv macros:
% \csh{xintCSVLength}, \csh{xintCSVKeep}, \csh{xintCSVKeepx}, \csh{xintCSVTrim},
% \csh{xintCSVNthEltPy}, \csh{xintCSVReverse},
% \csh{xintCSVFirstItem}, \csh{xintCSVLastItem}}
% 
% \lverb|Completely unstable macros: currently they expand the list argument
% and want no final comma. But for matters of xintexpr.sty I could as well
% decide to require a final comma, and then I could simplify implementation
% but of course this would break the macros if used with current
% functionalities.|
%    \begin{macrocode}
\let\xintCSVLength   \xintLength:f:csv
\let\xintCSVKeep     \xintKeep:f:csv
\let\xintCSVKeepx    \xintKeep:x:csv
\let\xintCSVTrim     \xintTrim:f:csv
\let\xintCSVNthEltPy \xintNthEltPy:f:csv
\let\xintCSVReverse  \xintReverse:f:csv
\let\xintCSVFirstItem\xintFirstItem:f:csv
\let\xintCSVLastItem \xintLastItem:f:csv
\let\XINT_tmpa\relax \let\XINT_tmpb\relax \let\XINT_tmpc\relax
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xinttools}
% \cleardoublepage\let\xinttoolsnameUp\undefined
%\gardesactifs
%\let</xinttools>\relax
%\let<*xintcore>\gardesinactifs
%</xinttools>^^A--------------------------------------------------
%<*xintcore>^^A---------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintcorenameUp\endcsname
% \section{Package \xintcorenameimp implementation}
% \RaisedLabel{sec:coreimp}
%
% \localtableofcontents
%
% Got split off from \xintnameimp with release |1.1|.
%
%   The core arithmetic routines have been entirely rewritten for release
%   |1.2|. The |1.2i| and |1.2l| brought again some improvements.
%
%   The commenting continues (\xintdocdate) to be very sparse: actually it got
%   worse than ever with release |1.2|. I will possibly add comments at a
%   later date, but for the time being the new routines are not commented at
%   all.
%
% |1.3| removes all macros which were deprecated at |1.2o|.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xintcore.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintkernel.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintcore}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintcore.sty
      \ifx\w\relax % but xintkernel.sty not yet loaded.
         \def\z{\endgroup\input xintkernel.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintkernel.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xintkernel}}%
          \fi
      \else
        \aftergroup\endinput % xintkernel already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintcore}%
  [2021/07/13 v1.4j Expandable arithmetic on big integers (JFB)]%
%    \end{macrocode}
% \subsection{(WIP!) Error conditions and exceptions}
% \lverb|As per the Mike Cowlishaw/IBM's General Decimal Arithmetic Specification
%
%    http://speleotrove.com/decimal/decarith.html
%
% and the Python3 implementation in its Decimal module.
%
% Clamped, ConversionSyntax, DivisionByZero, DivisionImpossible,
% DivisionUndefined, Inexact, InsufficientStorage, InvalidContext,
% InvalidOperation, Overflow, Inexact, Rounded, Subnormal,
% Underflow.
%
% X3.274 rajoute LostDigits
%
% Python rajoute FloatOperation (et n'inclut pas InsufficientStorage)
%
% quote de decarith.pdf:
% The Clamped, Inexact, Rounded, and Subnormal conditions can coincide with
% each other or with other conditions. In these cases then any trap enabled
% for another condition takes precedence over (is handled before) all of
% these, any Subnormal trap takes precedence over Inexact, any Inexact trap
% takes precedence over Rounded, and any Rounded trap takes precedence over
% Clamped.
%
% WORK IN PROGRESS ! (1.2l, 2017/07/26)
%
% I follow the Python terminology: a trapped signal means it raises an
% exception which for us means an expandable error message with some possible
% user interaction. In this WIP
% state, the interaction is commented out. A non-trapped signal or condition
% would activate a (presumably silent) handler.
%
% Here, no signal-raising condition is "ignored" and all are "trapped" which
% means that error handlers are never activated, thus left in garbage state in
% the code.
%
% Various conditions can raise the same signal.
%
% Only signals, not conditions, raise Flags.
%
% If a signal is ignored it does not raise a Flag, but it activates the signal
% handler (by default now no signal is ignored.)
%
% If a signal is not ignored it raises a Flag and then if it is not trapped it
% activates the handler of the _condition_.
%
% If trapped (which is default now) an «exception» is raised, which means an
% expandable error message (I copied over the LaTeX3 code for expandable error
% messages, basically)
% interrupts the TeX run. In future, user input could
% be solicited, but currently this is commented out.
%
% For now macros to reset flags are done but without public interface nor
% documentation.
%
% Only four conditions are currently possibly encountered:
%- InvalidOperation
%- DivisionByZero
%- DivisionUndefined (which signals InvalidOperation)
%- Underflow
%
% I did it quickly, anyhow this will become more palpable when some of the
% Decimal Specification is actually implemented. The plan is to first do the
% X3.274 norm, then more complete implementation will follow... perhaps...
% |
%    \begin{macrocode}
\csname XINT_Clamped_istrapped\endcsname
\csname XINT_ConversionSyntax_istrapped\endcsname
\csname XINT_DivisionByZero_istrapped\endcsname
\csname XINT_DivisionImpossible_istrapped\endcsname
\csname XINT_DivisionUndefined_istrapped\endcsname
\csname XINT_InvalidOperation_istrapped\endcsname
\csname XINT_Overflow_istrapped\endcsname
\csname XINT_Underflow_istrapped\endcsname
\catcode`- 11
\def\XINT_ConversionSyntax-signal  {{InvalidOperation}}%
\let\XINT_DivisionImpossible-signal\XINT_ConversionSyntax-signal
\let\XINT_DivisionUndefined-signal \XINT_ConversionSyntax-signal
\let\XINT_InvalidContext-signal    \XINT_ConversionSyntax-signal
\catcode`- 12
\def\XINT_signalcondition #1{\expandafter\XINT_signalcondition_a
    \romannumeral0\ifcsname XINT_#1-signal\endcsname
                    \xint_dothis{\csname XINT_#1-signal\endcsname}%
                  \fi\xint_orthat{{#1}}{#1}}%
\def\XINT_signalcondition_a #1#2#3#4#5{% copied over from Python Decimal module
% #1=signal, #2=condition, #3=explanation for user,
% #4=context for error handlers, #5=used 
    \ifcsname XINT_#1_isignoredflag\endcsname
       \xint_dothis{\csname XINT_#1.handler\endcsname {#4}}%
    \fi
    \expandafter\xint_gobble_i\csname XINT_#1Flag_ON\endcsname
    \unless\ifcsname XINT_#1_istrapped\endcsname
       \xint_dothis{\csname XINT_#2.handler\endcsname {#4}}%
    \fi
    \xint_orthat{%
       % the flag raised is named after the signal #1, but we show condition
       % #2
%    \end{macrocode}
% \lverb|&
%
% On 2021/05/19, 1.4g, I re-examined \XINT_expandableerror experimenting
% at first with an added ^^J to shift to next line the actual message.
%
% Previously I was calling it thrice (condition #2, user context #3, next
% tokens #5) here but it seems more reasonable to use it only once. As total
% size is so limited, I decided to only display #3 (information for user) and
% drop the #2 (condition, first argument of \XINT_signalcondition) and the
% display of the #5 (next tokens, fourth argument of \XINT_signalcondition).
%
% Besides, why was I doing here \xint_stop_atfirstofone{#5}, which adds
% limitations to usage?  Now inserting #5 directly so callers will have to
% insert a \romannumeral0 stopping space token if needed.  I thus have to
% update all usages across (mainly, I think) xintfrac. Done, but using here
% \xint_firstofone{#5}.  This looks silly, but allows some hypothetical future
% usage by user of I\xintUse{stuff} usage where \xintUse would be
% \xint_firstofthree.
%
% The problem is that this would have to be explained to user in the error
% context but space there is so extremely limited...
%
% After having reviewed existing usage of \XINT_signalcondition, I noticed
% there was free space in most cases and added here " (hit RET)" after #3.
%
% I experimented with ^^J here too (its effect in the "context" is
% independent of the \newlinechar setting, but it depends on the engine: works
% with TeXLive pdftex, requires -8bit with xetex)
%
% However, due to \errorcontextlines being 5 by default in etex (but
% xintsession 0.2b sets it to 0), I finally decided to not insert a ^^J
% ($&$&J) at all to separate the " (hit RET)" hint.
%
% On 2021/05/20 evening I found another completely different method for
% \XINT_expandableerror, which has some advantages. In particular it allows me
% to not use here "#3 (hit RET)" but simply "#3" as such information can be
% integrated in a non size limited generic message.
%
% The maximal size of #3 here was increased from 48 characters (method
% with \xint/ being badly delimited), to now 55 characters, longer
% messages being truncated at 56 characters with an appended "\ETC.".
% 
% |
%    \begin{macrocode}
       \XINT_expandableerror{#3}%
       % not for X3.274
       %\XINT_expandableerror{<RET>, or I\xintUse{...}<RET>, or I\xintCTRLC<RET>}%
       \xint_firstofone{#5}%
    }%
}%
%% \def\xintUse{\xint_firstofthree} % defined in xint.sty
\def\XINT_ifFlagRaised #1{%
    \ifcsname XINT_#1Flag_ON\endcsname
        \expandafter\xint_firstoftwo
    \else
        \expandafter\xint_secondoftwo
    \fi}%
\def\XINT_resetFlag #1%
    {\expandafter\let\csname XINT_#1Flag_ON\endcsname\XINT_undefined}%
\def\XINT_resetFlags {% WIP
    \XINT_resetFlag{InvalidOperation}% also from DivisionUndefined
    \XINT_resetFlag{DivisionByZero}%
    \XINT_resetFlag{Underflow}% (\xintiiPow with negative exponent)
    \XINT_resetFlag{Overflow}%   not encountered so far in xint code 1.2l
    % .. others .. 
}%
\def\XINT_RaiseFlag #1{\expandafter\xint_gobble_i\csname XINT_#1Flag_ON\endcsname}%
%    \end{macrocode}
% NOT IMPLEMENTED! WORK IN PROGRESS! (ALL SIGNALS TRAPPED, NO HANDLERS USED)
%    \begin{macrocode}
\catcode`. 11
\let\XINT_Clamped.handler\xint_firstofone % WIP
\def\XINT_InvalidOperation.handler#1{_NaN}% WIP
\def\XINT_ConversionSyntax.handler#1{_NaN}% WIP
\def\XINT_DivisionByZero.handler#1{_SignedInfinity(#1)}% WIP
\def\XINT_DivisionImpossible.handler#1{_NaN}% WIP
\def\XINT_DivisionUndefined.handler#1{_NaN}%  WIP
\let\XINT_Inexact.handler\xint_firstofone  %  WIP
\def\XINT_InvalidContext.handler#1{_NaN}%     WIP
\let\XINT_Rounded.handler\xint_firstofone  %  WIP
\let\XINT_Subnormal.handler\xint_firstofone%  WIP
\def\XINT_Overflow.handler#1{_NaN}%  WIP
\def\XINT_Underflow.handler#1{_NaN}% WIP
\catcode`. 12
%    \end{macrocode}
% \subsection{Counts for holding needed constants}
%    \begin{macrocode}
\ifdefined\m@ne\let\xint_c_mone\m@ne
          \else\csname newcount\endcsname\xint_c_mone \xint_c_mone -1 \fi
\ifdefined\xint_c_x^viii\else
\csname newcount\endcsname\xint_c_x^viii \xint_c_x^viii   100000000
\fi
\ifdefined\xint_c_x^ix\else
\csname newcount\endcsname\xint_c_x^ix   \xint_c_x^ix    1000000000
\fi
\newcount\xint_c_x^viii_mone        \xint_c_x^viii_mone    99999999
\newcount\xint_c_xii_e_viii          \xint_c_xii_e_viii  1200000000
\newcount\xint_c_xi_e_viii_mone  \xint_c_xi_e_viii_mone  1099999999
%    \end{macrocode}
% \subsection*{Routines handling integers as lists of token digits}
% \addcontentsline{toc}{subsection}{Routines handling integers as lists of token digits}
% \lverb|&
% Routines handling big integers which are lists of digit tokens with no
% special additional structure.
%
% Some
% routines do not accept non properly terminated inputs like "\the\numexpr1",
% or "\the\mathcode`\-", others do.
%
% These routines or their sub-routines are mainly for internal usage.
% |
%
% \subsection{\csh{XINT_cuz_small}}
% \lverb|&
% \XINT_cuz_small removes leading zeroes from the first eight digits. Expands
% following \romannumeral0. At least one digit is produced.|
%    \begin{macrocode}
\def\XINT_cuz_small#1{%
\def\XINT_cuz_small ##1##2##3##4##5##6##7##8%
{%
    \expandafter#1\the\numexpr ##1##2##3##4##5##6##7##8\relax
}}\XINT_cuz_small{ }%
%    \end{macrocode}
% \subsection{\csh{xintNum}, \csh{xintiNum}}
% \lverb|&
% For example \xintNum {----+-+++---+----000000000000003}
%
% Very old routine got completely rewritten at 1.2l.
%
% New code uses \numexpr governed expansion and fixes some issues of former
% version particularly regarding inputs of the \numexpr...\relax type without
% \the or \number prefix, and/or possibly no terminating \relax.
%
% \xintiNum{\numexpr 1}\foo in earlier versions caused premature expansion of
% \foo.
%
% \xintiNum{\the\numexpr 1} was ok, but a bit luckily so.
%
% Also, up to 1.2k inclusive, the macro fetched tokens eight by eight, and not
% nine by nine as is done now. I have no idea why.
%
% \xintNum gets redefined by $xintfracnameimp.
% |
%    \begin{macrocode}
\def\xintiNum {\romannumeral0\xintinum }%
\def\xintinum #1%
{%
    \expandafter\XINT_num_cleanup\the\numexpr\expandafter\XINT_num_loop
    \romannumeral`&&@#1\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z
}%
\def\xintNum {\romannumeral0\xintnum }%
\let\xintnum\xintinum
\def\XINT_num #1%
{%
    \expandafter\XINT_num_cleanup\the\numexpr\XINT_num_loop
    #1\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z
}%
\def\XINT_num_loop #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_xint: #9\XINT_num_end\xint:
    #1#2#3#4#5#6#7#8#9%
    \ifnum \numexpr #1#2#3#4#5#6#7#8#9+\xint_c_ = \xint_c_
%    \end{macrocode}
% \lverb|&
% means that so far only signs encountered, (if syntax is legal) then possibly
% zeroes
% or a terminated or not terminated \numexpr evaluating to zero
% In that latter case a correct zero will be produced in the end.
% |
%    \begin{macrocode}
      \expandafter\XINT_num_loop
    \else
%    \end{macrocode}
% \lverb|&
% non terminated \numexpr (with nine tokens total) are
% safe as after \fi, there is then \xint:
% |
%    \begin{macrocode}
     \expandafter\relax
    \fi
}%
\def\XINT_num_end\xint:#1\xint:{#1+\xint_c_\xint:}% empty input ok
\def\XINT_num_cleanup #1\xint:#2\Z { #1}%
%    \end{macrocode}
% \subsection{\csh{xintiiSgn}}
% \lverb|&
% 1.2l made \xintiiSgn robust against non terminated input.
%
% 1.2o deprecates here \xintSgn (it requires xintfrac.sty).
% |
%    \begin{macrocode}
\def\xintiiSgn {\romannumeral0\xintiisgn }%
\def\xintiisgn #1%
{%
    \expandafter\XINT_sgn \romannumeral`&&@#1\xint:
}%
\def\XINT_sgn #1#2\xint:
{%
    \xint_UDzerominusfork
      #1-{ 0}%
      0#1{-1}%
       0-{ 1}%
    \krof
}%
\def\XINT_Sgn #1#2\xint:
{%
    \xint_UDzerominusfork
      #1-{0}%
      0#1{-1}%
       0-{1}%
    \krof
}%
\def\XINT_cntSgn #1#2\xint:
{%
    \xint_UDzerominusfork
      #1-\xint_c_
      0#1\xint_c_mone
       0-\xint_c_i
    \krof
}%
%    \end{macrocode}
% \subsection{\csh{xintiiOpp}}
% \lverb|Attention, \xintiiOpp non robust against non terminated inputs.
% Reason is I don't want to have to grab a delimiter at the end, as everything
% happens "upfront".|
%    \begin{macrocode}
\def\xintiiOpp {\romannumeral0\xintiiopp }%
\def\xintiiopp #1%
{%
    \expandafter\XINT_opp \romannumeral`&&@#1%
}%
\def\XINT_Opp #1{\romannumeral0\XINT_opp #1}%
\def\XINT_opp #1%
{%
    \xint_UDzerominusfork
      #1-{ 0}%      zero
      0#1{ }%     negative
       0-{ -#1}%  positive
    \krof
}%
%    \end{macrocode}
% \subsection{\csh{xintiiAbs}}
% \lverb|&
%  Attention \xintiiAbs non robust against non terminated input.
%|
%    \begin{macrocode}
\def\xintiiAbs {\romannumeral0\xintiiabs }%
\def\xintiiabs #1%
{%
    \expandafter\XINT_abs \romannumeral`&&@#1%
}%
\def\XINT_abs #1%
{%
    \xint_UDsignfork
      #1{ }%
       -{ #1}%
    \krof
}%
\def\XINT_Abs #1%
{%
    \xint_UDsignfork
      #1{}%
       -{#1}%
    \krof
}%
%    \end{macrocode}
% \subsection{\csh{xintFDg}}
% \lverb|&
% FIRST DIGIT.
%
% 1.2l: \xintiiFDg made robust against non terminated input.
%
% 1.2o deprecates \xintiiFDg, gives to \xintFDg former meaning of \xintiiFDg.|
%    \begin{macrocode}
\def\xintFDg {\romannumeral0\xintfdg }%
\def\xintfdg #1{\expandafter\XINT_fdg \romannumeral`&&@#1\xint:\Z}%
\def\XINT_FDg #1%
   {\romannumeral0\expandafter\XINT_fdg\romannumeral`&&@\xintnum{#1}\xint:\Z }%
\def\XINT_fdg #1#2#3\Z
{%
    \xint_UDzerominusfork
      #1-{ 0}%   zero
      0#1{ #2}%  negative
       0-{ #1}%  positive
    \krof
}%
%    \end{macrocode}
% \subsection{\csh{xintLDg}}
% \lverb|&
% LAST DIGIT.
%
% Rewritten for 1.2i (2016/12/10). Surprisingly perhaps, it is faster than
% \xintLastItem from xintkernel.sty despite the \numexpr operations.
%
% 1.2o deprecates \xintiiLDg, gives to \xintLDg former meaning of \xintiiLDg.
%
% Attention \xintLDg non robust against non terminated input.
% |
%    \begin{macrocode}
\def\xintLDg {\romannumeral0\xintldg }%
\def\xintldg #1{\expandafter\XINT_ldg_fork\romannumeral`&&@#1%
    \XINT_ldg_c{}{}{}{}{}{}{}{}\xint_bye\relax}%
\def\XINT_ldg_fork #1%
{%
    \xint_UDsignfork
      #1\XINT_ldg
       -{\XINT_ldg#1}%
    \krof
}%
\def\XINT_ldg #1{%
\def\XINT_ldg ##1##2##3##4##5##6##7##8##9%
   {\expandafter#1%
    \the\numexpr##9##8##7##6##5##4##3##2##1*\xint_c_+\XINT_ldg_a##9}%
}\XINT_ldg{ }%
\def\XINT_ldg_a#1#2{\XINT_ldg_cbye#2\XINT_ldg_d#1\XINT_ldg_c\XINT_ldg_b#2}%
\def\XINT_ldg_b#1#2#3#4#5#6#7#8#9{#9#8#7#6#5#4#3#2#1*\xint_c_+\XINT_ldg_a#9}%
\def\XINT_ldg_c    #1#2\xint_bye{#1}%
\def\XINT_ldg_cbye #1\XINT_ldg_c{}%
\def\XINT_ldg_d#1#2\xint_bye{#1}%
%    \end{macrocode}
%
% \subsection{\csh{xintDouble}}
% \lverb|Attention \xintDouble non robust against non terminated input.|
%    \begin{macrocode}
\def\xintDouble {\romannumeral0\xintdouble}%
\def\xintdouble #1{\expandafter\XINT_dbl_fork\romannumeral`&&@#1%
                   \xint_bye2345678\xint_bye*\xint_c_ii\relax}%
\def\XINT_dbl_fork #1%
{%
    \xint_UDsignfork
     #1\XINT_dbl_neg
      -\XINT_dbl
    \krof #1%
}%
\def\XINT_dbl_neg-{\expandafter-\romannumeral0\XINT_dbl}%
\def\XINT_dbl #1{%
\def\XINT_dbl ##1##2##3##4##5##6##7##8%
   {\expandafter#1\the\numexpr##1##2##3##4##5##6##7##8\XINT_dbl_a}%
}\XINT_dbl{ }%
\def\XINT_dbl_a #1#2#3#4#5#6#7#8%
   {\expandafter\XINT_dbl_e\the\numexpr 1#1#2#3#4#5#6#7#8\XINT_dbl_a}%
\def\XINT_dbl_e#1{*\xint_c_ii\if#13+\xint_c_i\fi\relax}%
%    \end{macrocode}
% \subsection{\csh{xintHalf}}
% \lverb|Attention \xintHalf non robust against non terminated input.|
%    \begin{macrocode}
\def\xintHalf {\romannumeral0\xinthalf}%
\def\xinthalf #1{\expandafter\XINT_half_fork\romannumeral`&&@#1%
    \xint_bye\xint_Bye345678\xint_bye
    *\xint_c_v+\xint_c_v)/\xint_c_x-\xint_c_i\relax}%
\def\XINT_half_fork #1%
{%
    \xint_UDsignfork
     #1\XINT_half_neg
      -\XINT_half
    \krof #1%
}%
\def\XINT_half_neg-{\xintiiopp\XINT_half}%
\def\XINT_half #1{%
\def\XINT_half ##1##2##3##4##5##6##7##8%
   {\expandafter#1\the\numexpr(##1##2##3##4##5##6##7##8\XINT_half_a}%
}\XINT_half{ }%
\def\XINT_half_a#1{\xint_Bye#1\xint_bye\XINT_half_b#1}%
\def\XINT_half_b #1#2#3#4#5#6#7#8%
   {\expandafter\XINT_half_e\the\numexpr(1#1#2#3#4#5#6#7#8\XINT_half_a}%
\def\XINT_half_e#1{*\xint_c_v+#1-\xint_c_v)\relax}%
%    \end{macrocode}
% \subsection{\csh{xintInc}}
% \lverb|1.2i much delayed complete rewrite in 1.2 style.
%
% As we take 9 by 9 with the input save stack at 5000 this allows a bit less
% than 9 times 2500 = 22500 digits on input.
%
% Attention \xintInc non robust against non terminated input.|
%    \begin{macrocode}
\def\xintInc {\romannumeral0\xintinc}%
\def\xintinc #1{\expandafter\XINT_inc_fork\romannumeral`&&@#1%
                \xint_bye23456789\xint_bye+\xint_c_i\relax}%
\def\XINT_inc_fork #1%
{%
    \xint_UDsignfork
      #1\XINT_inc_neg
       -\XINT_inc
    \krof #1%
}%
\def\XINT_inc_neg-#1\xint_bye#2\relax
   {\xintiiopp\XINT_dec #1\XINT_dec_bye234567890\xint_bye}%
\def\XINT_inc #1{%
\def\XINT_inc ##1##2##3##4##5##6##7##8##9%
   {\expandafter#1\the\numexpr##1##2##3##4##5##6##7##8##9\XINT_inc_a}%
}\XINT_inc{ }%
\def\XINT_inc_a #1#2#3#4#5#6#7#8#9%
   {\expandafter\XINT_inc_e\the\numexpr 1#1#2#3#4#5#6#7#8#9\XINT_inc_a}%
\def\XINT_inc_e#1{\if#12+\xint_c_i\fi\relax}%
%    \end{macrocode}
% \subsection{\csh{xintDec}}
% \lverb|1.2i much delayed complete rewrite in the 1.2 style. Things are a
% bit more complicated than \xintInc because 2999999999 is too big for TeX.
%
% Attention \xintDec non robust against non terminated input.|
%    \begin{macrocode}
\def\xintDec {\romannumeral0\xintdec}%
\def\xintdec #1{\expandafter\XINT_dec_fork\romannumeral`&&@#1%
                \XINT_dec_bye234567890\xint_bye}%
\def\XINT_dec_fork #1%
{%
    \xint_UDsignfork
      #1\XINT_dec_neg
       -\XINT_dec
    \krof #1%
}%
\def\XINT_dec_neg-#1\XINT_dec_bye#2\xint_bye
   {\expandafter-%
    \romannumeral0\XINT_inc #1\xint_bye23456789\xint_bye+\xint_c_i\relax}%
\def\XINT_dec #1{%
\def\XINT_dec ##1##2##3##4##5##6##7##8##9%
   {\expandafter#1\the\numexpr##1##2##3##4##5##6##7##8##9\XINT_dec_a}%
}\XINT_dec{ }%
\def\XINT_dec_a #1#2#3#4#5#6#7#8#9%
   {\expandafter\XINT_dec_e\the\numexpr 1#1#2#3#4#5#6#7#8#9\XINT_dec_a}%
\def\XINT_dec_bye #1\XINT_dec_a#2#3\xint_bye
         {\if#20-\xint_c_ii\relax+\else-\fi\xint_c_i\relax}%
\def\XINT_dec_e#1{\unless\if#11\xint_dothis{-\xint_c_i#1}\fi\xint_orthat\relax}%
%    \end{macrocode}
% \subsection{\csh{xintDSL}}
% \lverb|DECIMAL SHIFT LEFT (=MULTIPLICATION PAR 10). Rewritten for 1.2i.
% This was very old code... I never came back to it, but I should have
% rewritten it long time ago.
%
% Attention \xintDSL non robust against non terminated input.|
%    \begin{macrocode}
\def\xintDSL {\romannumeral0\xintdsl }%
\def\xintdsl #1{\expandafter\XINT_dsl\romannumeral`&&@#10}%
\def\XINT_dsl#1{%
\def\XINT_dsl ##1{\xint_gob_til_zero ##1\xint_dsl_zero 0#1##1}%
}\XINT_dsl{ }%
\def\xint_dsl_zero 0 0{ }%
%    \end{macrocode}
% \subsection{\csh{xintDSR}}
% \lverb|Decimal shift right, truncates towards zero. Rewritten for 1.2i.
% Limited to 22483 digits on input.
%
% Attention \xintDSR non robust against non terminated input.|
%    \begin{macrocode}
\def\xintDSR{\romannumeral0\xintdsr}%
\def\xintdsr #1{\expandafter\XINT_dsr_fork\romannumeral`&&@#1%
    \xint_bye\xint_Bye3456789\xint_bye+\xint_c_v)/\xint_c_x-\xint_c_i\relax}%
\def\XINT_dsr_fork #1%
{%
    \xint_UDsignfork
      #1\XINT_dsr_neg
       -\XINT_dsr
    \krof #1%
}%
\def\XINT_dsr_neg-{\xintiiopp\XINT_dsr}%
\def\XINT_dsr #1{%
\def\XINT_dsr ##1##2##3##4##5##6##7##8##9%
   {\expandafter#1\the\numexpr(##1##2##3##4##5##6##7##8##9\XINT_dsr_a}%
}\XINT_dsr{ }%
\def\XINT_dsr_a#1{\xint_Bye#1\xint_bye\XINT_dsr_b#1}%
\def\XINT_dsr_b #1#2#3#4#5#6#7#8#9%
   {\expandafter\XINT_dsr_e\the\numexpr(1#1#2#3#4#5#6#7#8#9\XINT_dsr_a}%
\def\XINT_dsr_e #1{)\relax}%
%    \end{macrocode}
% \subsection{\csh{xintDSRr}}
% \lverb|New with 1.2i. Decimal shift right, rounds away from zero; done in
% the 1.2 spirit (with much delay, sorry). Used by \xintRound, \xintDivRound.
%
% This is about the first time I am happy that the division in \numexpr
% rounds!
%
% Attention \xintDSRr non robust against non terminated input.|
%    \begin{macrocode}
\def\xintDSRr{\romannumeral0\xintdsrr}%
\def\xintdsrr #1{\expandafter\XINT_dsrr_fork\romannumeral`&&@#1%
                \xint_bye\xint_Bye3456789\xint_bye/\xint_c_x\relax}%
\def\XINT_dsrr_fork #1%
{%
    \xint_UDsignfork
      #1\XINT_dsrr_neg
       -\XINT_dsrr
    \krof #1%
}%
\def\XINT_dsrr_neg-{\xintiiopp\XINT_dsrr}%
\def\XINT_dsrr #1{%
\def\XINT_dsrr ##1##2##3##4##5##6##7##8##9%
   {\expandafter#1\the\numexpr##1##2##3##4##5##6##7##8##9\XINT_dsrr_a}%
}\XINT_dsrr{ }%
\def\XINT_dsrr_a#1{\xint_Bye#1\xint_bye\XINT_dsrr_b#1}%
\def\XINT_dsrr_b #1#2#3#4#5#6#7#8#9%
   {\expandafter\XINT_dsrr_e\the\numexpr1#1#2#3#4#5#6#7#8#9\XINT_dsrr_a}%
\let\XINT_dsrr_e\XINT_inc_e
%    \end{macrocode}
% \subsection*{Blocks of eight digits}
% \addcontentsline{toc}{subsection}{Blocks of eight digits}
% \lverb|The lingua of release 1.2.|
%
% \subsection{\csh{XINT_cuz}}
% \lverb|This (launched by \romannumeral0) iterately removes all leading
% zeroes from a sequence of 8N digits ended by \R.
%
% Rewritten for 1.2l, now uses \numexpr governed expansion and \ifnum test
% rather than delimited gobbling macros.
%
% Note 2015/11/28: with only four digits the gob_til_fourzeroes had proved
% in some old testing faster than \ifnum test. But with eight digits, the
% execution times are much closer, as I tested back then.
% |
%    \begin{macrocode}
\def\XINT_cuz #1{%
\def\XINT_cuz {\expandafter#1\the\numexpr\XINT_cuz_loop}%
}\XINT_cuz{ }%
\def\XINT_cuz_loop #1#2#3#4#5#6#7#8#9%
{%
    #1#2#3#4#5#6#7#8%
       \xint_gob_til_R #9\XINT_cuz_hitend\R
       \ifnum #1#2#3#4#5#6#7#8>\xint_c_
             \expandafter\XINT_cuz_cleantoend
       \else\expandafter\XINT_cuz_loop
       \fi #9%
}%
\def\XINT_cuz_hitend\R #1\R{\relax}%
\def\XINT_cuz_cleantoend #1\R{\relax #1}%
%    \end{macrocode}
% \subsection{\csh{XINT_cuz_byviii}}
% \lverb|This removes eight by eight leading zeroes from a sequence of 8N digits
% ended by \R. Thus, we still have 8N digits on output. Expansion started by
% \romannumeral0 |
%    \begin{macrocode}
\def\XINT_cuz_byviii #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_R #9\XINT_cuz_byviii_e \R
    \xint_gob_til_eightzeroes #1#2#3#4#5#6#7#8\XINT_cuz_byviii_z 00000000%
    \XINT_cuz_byviii_done #1#2#3#4#5#6#7#8#9%
}%
\def\XINT_cuz_byviii_z 00000000\XINT_cuz_byviii_done 00000000{\XINT_cuz_byviii}%
\def\XINT_cuz_byviii_done #1\R { #1}%
\def\XINT_cuz_byviii_e\R #1\XINT_cuz_byviii_done #2\R{ #2}%
%    \end{macrocode}
% \subsection{\csh{XINT_unsep_loop}}
%
% \lverb|This is used as
%( \the\numexpr0\XINT_unsep_loop (blocks of 1<8digits>!)%
%:               \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax
%)
% It removes the 1's and !'s, and outputs the 8N digits with a 0 token as
% as prefix which will have to be cleaned out by caller.
%
% Actually it does not matter whether the blocks contain really 8 digits, all
% that matters is that they have 1 as first digit (and at most 9 digits after
% that to obey the TeX-\numexpr  bound).
%
% Done at 1.2l for usage by other macros. The similar code in earlier releases
% was strangely in O(N^2) style, apparently to avoid some memory constraints.
% But these memory constraints related to \numexpr chaining seems to be in
% many places in xint code base. The 1.2l version is written in the 1.2i style
% of \xintInc etc... and is compatible with some 1! block without digits
% among the treated blocks, they will disappear.|
%    \begin{macrocode}
\def\XINT_unsep_loop #1!#2!#3!#4!#5!#6!#7!#8!#9!%
{%
    \expandafter\XINT_unsep_clean
    \the\numexpr #1\expandafter\XINT_unsep_clean
    \the\numexpr #2\expandafter\XINT_unsep_clean
    \the\numexpr #3\expandafter\XINT_unsep_clean
    \the\numexpr #4\expandafter\XINT_unsep_clean
    \the\numexpr #5\expandafter\XINT_unsep_clean
    \the\numexpr #6\expandafter\XINT_unsep_clean
    \the\numexpr #7\expandafter\XINT_unsep_clean
    \the\numexpr #8\expandafter\XINT_unsep_clean
    \the\numexpr #9\XINT_unsep_loop
}%
\def\XINT_unsep_clean 1{\relax}%
%    \end{macrocode}
% \subsection{\csh{XINT_unsep_cuzsmall}}
%
% \lverb|This is used as
%( \romannumeral0\XINT_unsep_cuzsmall (blocks of 1<8d>!)%
%:               \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax
%)
% It removes the 1's and !'s, and removes the leading zeroes *of
% the first block*.
%
% Redone for 1.2l: the 1.2 variant was strangely in O(N^2) style.|
%    \begin{macrocode}
\def\XINT_unsep_cuzsmall
{%
    \expandafter\XINT_unsep_cuzsmall_x\the\numexpr0\XINT_unsep_loop
}%
\def\XINT_unsep_cuzsmall_x #1{%
\def\XINT_unsep_cuzsmall_x 0##1##2##3##4##5##6##7##8%
{%
    \expandafter#1\the\numexpr ##1##2##3##4##5##6##7##8\relax
}}\XINT_unsep_cuzsmall_x{ }%
%    \end{macrocode}
% \subsection{\csh{XINT_div_unsepQ}}
%
% \lverb|This is used by division to remove separators from the produced
% quotient. The quotient is produced in the correct order. The routine will
% also remove leading zeroes. An extra initial block of 8 zeroes is possible
% and thus if present must be removed. Then the next eight digits must be
% cleaned of leading zeroes. Attention that there might be a single
% block of 8 zeroes. Expansion launched by \romannumeral0.
%
% Rewritten for 1.2l in 1.2i style.|
%    \begin{macrocode}
\def\XINT_div_unsepQ_delim {\xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax\Z}%
\def\XINT_div_unsepQ
{%
    \expandafter\XINT_div_unsepQ_x\the\numexpr0\XINT_unsep_loop
}%
\def\XINT_div_unsepQ_x #1{%
\def\XINT_div_unsepQ_x 0##1##2##3##4##5##6##7##8##9%
{%
    \xint_gob_til_Z ##9\XINT_div_unsepQ_one\Z
    \xint_gob_til_eightzeroes ##1##2##3##4##5##6##7##8\XINT_div_unsepQ_y 00000000%
    \expandafter#1\the\numexpr ##1##2##3##4##5##6##7##8\relax ##9%
}}\XINT_div_unsepQ_x{ }%
\def\XINT_div_unsepQ_y #1{%
\def\XINT_div_unsepQ_y ##1\relax ##2##3##4##5##6##7##8##9%
{%
    \expandafter#1\the\numexpr ##2##3##4##5##6##7##8##9\relax
}}\XINT_div_unsepQ_y{ }%
\def\XINT_div_unsepQ_one#1\expandafter{\expandafter}%
%    \end{macrocode}
% \subsection{\csh{XINT_div_unsepR}}
%
% \lverb|This is used by division to remove separators from the produced
% remainder. The remainder is here in correct order. It must be cleaned of
% leading zeroes, possibly all the way.
%
% Also rewritten for 1.2l, the 1.2 version was O(N^2) style.
%
% Terminator \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax\R
%
% We have a need for something like \R because it is not guaranteed the thing
% is not actually zero.|
%    \begin{macrocode}
\def\XINT_div_unsepR
{%
    \expandafter\XINT_div_unsepR_x\the\numexpr0\XINT_unsep_loop
}%
\def\XINT_div_unsepR_x#1{%
\def\XINT_div_unsepR_x 0{\expandafter#1\the\numexpr\XINT_cuz_loop}%
}\XINT_div_unsepR_x{ }%
%    \end{macrocode}
% \subsection{\csh{XINT_zeroes_forviii}}
%
% \lverb|&
%( \romannumeral0\XINT_zeroes_forviii #1\R\R\R\R\R\R\R\R{10}0000001\W
%)
% produces a string of k 0's such that k+length(#1) is smallest bigger multiple
% of eight.|
%    \begin{macrocode}
\def\XINT_zeroes_forviii #1#2#3#4#5#6#7#8%
{%
    \xint_gob_til_R #8\XINT_zeroes_forviii_end\R\XINT_zeroes_forviii
}%
\def\XINT_zeroes_forviii_end#1{%
\def\XINT_zeroes_forviii_end\R\XINT_zeroes_forviii ##1##2##3##4##5##6##7##8##9\W
{%
    \expandafter#1\xint_gob_til_one ##2##3##4##5##6##7##8%
}}\XINT_zeroes_forviii_end{ }%
%    \end{macrocode}
% \subsection{\csh{XINT_sepbyviii_Z}}
%
% \lverb|This is used as
%( \the\numexpr\XINT_sepbyviii_Z <8Ndigits>\XINT_sepbyviii_Z_end 2345678\relax
%)
% It produces 1<8d>!...1<8d>!1;!
%
% Prior to 1.2l it used \Z as terminator not the semi-colon (hence the name).
% The switch to ; was done at a time I thought perhaps I would use an internal
% format maintaining such 8 digits blocks, and this has to be compatible with
% the \csname...\endcsname encapsulation in \xintexpr parsers.|
%    \begin{macrocode}
\def\XINT_sepbyviii_Z #1#2#3#4#5#6#7#8%
{%
    1#1#2#3#4#5#6#7#8\expandafter!\the\numexpr\XINT_sepbyviii_Z
}%
\def\XINT_sepbyviii_Z_end #1\relax {;!}%
%    \end{macrocode}
% \subsection{\csh{XINT_sepbyviii_andcount}}
%
% \lverb|This is used as
%( \the\numexpr\XINT_sepbyviii_andcount <8Ndigits>$%
%:     \XINT_sepbyviii_end 2345678\relax
%:     \xint_c_vii!\xint_c_vi!\xint_c_v!\xint_c_iv!$%
%:     \xint_c_iii!\xint_c_ii!\xint_c_i!\xint_c_\W
%)
% It will produce
%( 1<8d>!1<8d>!....1<8d>!1\xint:<count of blocks>\xint:
%)
% Used by
% \XINT_div_prepare_g for \XINT_div_prepare_h, and also by \xintiiCmp.|
%    \begin{macrocode}
\def\XINT_sepbyviii_andcount
{%
    \expandafter\XINT_sepbyviii_andcount_a\the\numexpr\XINT_sepbyviii
}%
\def\XINT_sepbyviii #1#2#3#4#5#6#7#8%
{%
    1#1#2#3#4#5#6#7#8\expandafter!\the\numexpr\XINT_sepbyviii
}%
\def\XINT_sepbyviii_end #1\relax {\relax\XINT_sepbyviii_andcount_end!}%
\def\XINT_sepbyviii_andcount_a {\XINT_sepbyviii_andcount_b \xint_c_\xint:}%
\def\XINT_sepbyviii_andcount_b #1\xint:#2!#3!#4!#5!#6!#7!#8!#9!%
{%
    #2\expandafter!\the\numexpr#3\expandafter!\the\numexpr#4\expandafter
    !\the\numexpr#5\expandafter!\the\numexpr#6\expandafter!\the\numexpr
    #7\expandafter!\the\numexpr#8\expandafter!\the\numexpr#9\expandafter!\the\numexpr
    \expandafter\XINT_sepbyviii_andcount_b\the\numexpr #1+\xint_c_viii\xint:%
}%
\def\XINT_sepbyviii_andcount_end #1\XINT_sepbyviii_andcount_b\the\numexpr
    #2+\xint_c_viii\xint:#3#4\W {\expandafter\xint:\the\numexpr #2+#3\xint:}%
%    \end{macrocode}
% \subsection{\csh{XINT_rsepbyviii}}
%
% \lverb|This is used as
%( \the\numexpr1\XINT_rsepbyviii <8Ndigits>$%
%:              \XINT_rsepbyviii_end_A 2345678$%
%:              \XINT_rsepbyviii_end_B 2345678\relax UV$%
%)
% and will produce
%( 1<8digits>!1<8digits>\xint:1<8digits>!...
%) 
% where the original
% digits are organized by eight, and the order inside successive pairs of
% blocks separated by \xint: has been reversed. Output ends either in 
% 1<8d>!1<8d>\xint:1U\xint: (even) or 1<8d>!1<8d>\xint:1V!1<8d>\xint: (odd)
%
% The U an V should be \numexpr1 stoppers (or will expand and be ended by !).
% This macro is currently (1.2..1.2l) exclusively used in combination with
% \XINT_sepandrev_andcount or \XINT_sepandrev.
% |
%    \begin{macrocode}
\def\XINT_rsepbyviii #1#2#3#4#5#6#7#8%
{%
    \XINT_rsepbyviii_b {#1#2#3#4#5#6#7#8}%
}%
\def\XINT_rsepbyviii_b #1#2#3#4#5#6#7#8#9%
{%
    #2#3#4#5#6#7#8#9\expandafter!\the\numexpr
    1#1\expandafter\xint:\the\numexpr 1\XINT_rsepbyviii
}%
\def\XINT_rsepbyviii_end_B #1\relax #2#3{#2\xint:}%
\def\XINT_rsepbyviii_end_A #11#2\expandafter #3\relax #4#5{#5!1#2\xint:}%
%    \end{macrocode}
% \subsection{\csh{XINT_sepandrev}}
% \lverb|This is used typically as
%( \romannumeral0\XINT_sepandrev <8Ndigits>$%
%:               \XINT_rsepbyviii_end_A 2345678$%
%:               \XINT_rsepbyviii_end_B 2345678\relax UV$%
%:               \R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\W
%)
% and will produce 
%( 1<8digits>!1<8digits>!1<8digits>!...
%)
% where the blocks have
% been globally reversed. The UV here are only place holders (must be \numexpr1
% stoppers) to share same
% syntax as \XINT_sepandrev_andcount, they are gobbled (#2 in \XINT_sepandrev_done).|
%    \begin{macrocode}
\def\XINT_sepandrev
{%
    \expandafter\XINT_sepandrev_a\the\numexpr 1\XINT_rsepbyviii
}%
\def\XINT_sepandrev_a {\XINT_sepandrev_b {}}%
\def\XINT_sepandrev_b #1#2\xint:#3\xint:#4\xint:#5\xint:#6\xint:#7\xint:#8\xint:#9\xint:%
{%
    \xint_gob_til_R #9\XINT_sepandrev_end\R
    \XINT_sepandrev_b {#9!#8!#7!#6!#5!#4!#3!#2!#1}%
}%
\def\XINT_sepandrev_end\R\XINT_sepandrev_b #1#2\W {\XINT_sepandrev_done #1}%
\def\XINT_sepandrev_done #11#2!{ }%
%    \end{macrocode}
% \subsection{\csh{XINT_sepandrev_andcount}}
% \lverb|This is used typically as
%( \romannumeral0\XINT_sepandrev_andcount <8Ndigits>$%
%:              \XINT_rsepbyviii_end_A 2345678$%
%:              \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
%:              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
%:              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
%)
% and will produce
%( <length>.1<8digits>!1<8digits>!1<8digits>!...
%)
% where the
% blocks have been globally reversed and <length> is the number of blocks.|
%    \begin{macrocode}
\def\XINT_sepandrev_andcount
{%
    \expandafter\XINT_sepandrev_andcount_a\the\numexpr 1\XINT_rsepbyviii
}%
\def\XINT_sepandrev_andcount_a {\XINT_sepandrev_andcount_b 0!{}}%
\def\XINT_sepandrev_andcount_b #1!#2#3\xint:#4\xint:#5\xint:#6\xint:#7\xint:#8\xint:#9\xint:%
{%
    \xint_gob_til_R #9\XINT_sepandrev_andcount_end\R
    \expandafter\XINT_sepandrev_andcount_b \the\numexpr #1+\xint_c_i!%
    {#9!#8!#7!#6!#5!#4!#3!#2}%
}%
\def\XINT_sepandrev_andcount_end\R
    \expandafter\XINT_sepandrev_andcount_b\the\numexpr #1+\xint_c_i!#2#3#4\W
{\expandafter\XINT_sepandrev_andcount_done\the\numexpr #3+\xint_c_xiv*#1!#2}%
\def\XINT_sepandrev_andcount_done#1{%
\def\XINT_sepandrev_andcount_done##1!##21##3!{\expandafter#1\the\numexpr##1-##3\xint:}%
}\XINT_sepandrev_andcount_done{ }%
%    \end{macrocode}
% \subsection{\csh{XINT_rev_nounsep}}
% \lverb|This is used as
%( \romannumeral0\XINT_rev_nounsep {}<blocks 1<8d>!>\R!\R!\R!\R!\R!\R!\R!\R!\W
%)
% It reverses the blocks, keeping the 1's and ! separators. Used multiple
% times in the division algorithm. The inserted {} here is not optional.|
%    \begin{macrocode}
\def\XINT_rev_nounsep #1#2!#3!#4!#5!#6!#7!#8!#9!%
{%
    \xint_gob_til_R #9\XINT_rev_nounsep_end\R
    \XINT_rev_nounsep {#9!#8!#7!#6!#5!#4!#3!#2!#1}%
}%
\def\XINT_rev_nounsep_end\R\XINT_rev_nounsep #1#2\W {\XINT_rev_nounsep_done #1}%
\def\XINT_rev_nounsep_done #11{ 1}%
%    \end{macrocode}
% \subsection{\csh{XINT_unrevbyviii}}
% \lverb|Used as \romannumeral0\XINT_unrevbyviii 1<8d>!....1<8d>! terminated
% by
%( 1;!1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
%)
% The \romannumeral in unrevbyviii_a is for special effects (expand some token
% which was put as 1<token>! at the end of the original blocks). This
% mechanism is used by 1.2 subtraction (still true for 1.2l).|
%    \begin{macrocode}
\def\XINT_unrevbyviii #11#2!1#3!1#4!1#5!1#6!1#7!1#8!1#9!%
{%
    \xint_gob_til_R #9\XINT_unrevbyviii_a\R
    \XINT_unrevbyviii {#9#8#7#6#5#4#3#2#1}%
}%
\def\XINT_unrevbyviii_a#1{%
\def\XINT_unrevbyviii_a\R\XINT_unrevbyviii ##1##2\W
    {\expandafter#1\romannumeral`&&@\xint_gob_til_sc ##1}%
}\XINT_unrevbyviii_a{ }%
%    \end{macrocode}
% \lverb|Can work with shorter ending pattern: 1;!1\R!1\R!1\R!1\R!1\R!1\R!\W
% but the longer one of unrevbyviii is ok here too. Used currently (1.2) only
% by addition, now (1.2c) with long ending pattern. Does the final clean up of
% leading zeroes contrarily to general \XINT_unrevbyviii.|
%    \begin{macrocode}
\def\XINT_smallunrevbyviii 1#1!1#2!1#3!1#4!1#5!1#6!1#7!1#8!#9\W%
{%
    \expandafter\XINT_cuz_small\xint_gob_til_sc #8#7#6#5#4#3#2#1%
}%
%    \end{macrocode}
% \subsection*{Core arithmetic}
% \addcontentsline{toc}{subsection}{Core arithmetic}
% \lverb|The four operations have been rewritten entirely for release 1.2.
% The new routines works with separated blocks of eight digits. They all measure
% first the lengths of the arguments, even addition and subtraction (this was
% not the case with xintcore.sty 1.1 or earlier.)
%
% The technique of chaining \the\numexpr induces a limitation on the
% maximal size depending on the size of the input save stack and the maximum
% expansion depth. For the current (TL2015) settings (5000, resp. 10000), the
% induced limit for addition of numbers is at 19968 and for multiplication
% it is observed to be 19959 (valid as of 2015/10/07).
%
% Side remark: I tested that \the\numexpr was more efficient than \number. But
% it reduced the allowable numbers for addition from 19976 digits to 19968
% digits.|
%
% \subsection{\csh{xintiiAdd}}
% \lverb|1.2l: \xintiiAdd made robust against non terminated input.|
%    \begin{macrocode}
\def\xintiiAdd   {\romannumeral0\xintiiadd }%
\def\xintiiadd #1{\expandafter\XINT_iiadd\romannumeral`&&@#1\xint:}%
\def\XINT_iiadd #1#2\xint:#3%
{%
    \expandafter\XINT_add_nfork\expandafter#1\romannumeral`&&@#3\xint:#2\xint:
}%
\def\XINT_add_fork #1#2\xint:#3\xint:{\XINT_add_nfork #1#3\xint:#2\xint:}%
\def\XINT_add_nfork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_add_firstiszero
      #2\XINT_add_secondiszero
       0{}%
    \krof
    \xint_UDsignsfork
          #1#2\XINT_add_minusminus
           #1-\XINT_add_minusplus
           #2-\XINT_add_plusminus
            --\XINT_add_plusplus
    \krof #1#2%
}%
\def\XINT_add_firstiszero  #1\krof 0#2#3\xint:#4\xint:{ #2#3}%
\def\XINT_add_secondiszero #1\krof #20#3\xint:#4\xint:{ #2#4}%
\def\XINT_add_minusminus   #1#2%
   {\expandafter-\romannumeral0\XINT_add_pp_a {}{}}%
\def\XINT_add_minusplus    #1#2{\XINT_sub_mm_a {}#2}%
\def\XINT_add_plusminus    #1#2%
   {\expandafter\XINT_opp\romannumeral0\XINT_sub_mm_a #1{}}%
\def\XINT_add_pp_a #1#2#3\xint:
{%
  \expandafter\XINT_add_pp_b
      \romannumeral0\expandafter\XINT_sepandrev_andcount
      \romannumeral0\XINT_zeroes_forviii #2#3\R\R\R\R\R\R\R\R{10}0000001\W
      #2#3\XINT_rsepbyviii_end_A 2345678%
          \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
   \X #1%
}%
\let\XINT_add_plusplus \XINT_add_pp_a
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_add_pp_b #1\xint:#2\X #3\xint:
{%
    \expandafter\XINT_add_checklengths
    \the\numexpr #1\expandafter\xint:%
    \romannumeral0\expandafter\XINT_sepandrev_andcount
    \romannumeral0\XINT_zeroes_forviii #3\R\R\R\R\R\R\R\R{10}0000001\W
    #3\XINT_rsepbyviii_end_A 2345678%
      \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
     1;!1;!1;!1;!\W #21;!1;!1;!1;!\W
     1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
%    \end{macrocode}
% \lverb|I keep #1.#2. to check if at most 6 + 6 base 10^8 digits which can be
% treated faster for final reverse. But is this overhead at all useful ? |
%    \begin{macrocode}
\def\XINT_add_checklengths #1\xint:#2\xint:%
{%
    \ifnum #2>#1
       \expandafter\XINT_add_exchange
    \else
       \expandafter\XINT_add_A
    \fi
    #1\xint:#2\xint:%
}%
\def\XINT_add_exchange #1\xint:#2\xint:#3\W #4\W
{%
    \XINT_add_A #2\xint:#1\xint:#4\W #3\W
}%
\def\XINT_add_A #1\xint:#2\xint:%
{%
    \ifnum #1>\xint_c_vi
          \expandafter\XINT_add_aa
    \else \expandafter\XINT_add_aa_small
    \fi
}%
\def\XINT_add_aa {\expandafter\XINT_add_out\the\numexpr\XINT_add_a \xint_c_ii}%
\def\XINT_add_out{\expandafter\XINT_cuz_small\romannumeral0\XINT_unrevbyviii {}}%
\def\XINT_add_aa_small
    {\expandafter\XINT_smallunrevbyviii\the\numexpr\XINT_add_a \xint_c_ii}%
%    \end{macrocode}
% \lverb|2 as first token of #1 stands for "no carry", 3 will mean a carry (we
% are adding 1<8digits> to 1<8digits>.) Version 1.2c has terminators of the
% shape 1;!, replacing the \Z! used in 1.2.
%
% Call: \the\numexpr\XINT_add_a 2#11;!1;!1;!1;!\W #21;!1;!1;!1;!\W
% where #1 and #2 are blocks of 1<8d>!, and #1 is at most as long as #2. This
% last requirement is a bit annoying (if one wants to do recursive algorithms
% but not have to check lengths), and I will probably remove it at some point.
%
% Output: blocks of 1<8d>! representing the addition, (least significant
% first), and a final 1;!. In recursive algorithm this 1;! terminator can
% thus conveniently be reused as part of input terminator (up to the length
% problem).
%
%|
%    \begin{macrocode}
\def\XINT_add_a #1!#2!#3!#4!#5\W
                #6!#7!#8!#9!%
{%
    \XINT_add_b 
        #1!#6!#2!#7!#3!#8!#4!#9!%
        #5\W
}%
\def\XINT_add_b #11#2#3!#4!%
{%
    \xint_gob_til_sc #2\XINT_add_bi ;%
    \expandafter\XINT_add_c\the\numexpr#1+1#2#3+#4-\xint_c_ii\xint:%
}%
\def\XINT_add_bi;\expandafter\XINT_add_c
    \the\numexpr#1+#2+#3-\xint_c_ii\xint:#4!#5!#6!#7!#8!#9!\W
{%
    \XINT_add_k #1#3!#5!#7!#9!%
}%
\def\XINT_add_c #1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_add_d #1%
}%
\def\XINT_add_d #11#2#3!#4!%
{%
    \xint_gob_til_sc #2\XINT_add_di ;%
    \expandafter\XINT_add_e\the\numexpr#1+1#2#3+#4-\xint_c_ii\xint:%
}%
\def\XINT_add_di;\expandafter\XINT_add_e
    \the\numexpr#1+#2+#3-\xint_c_ii\xint:#4!#5!#6!#7!#8\W
{%
    \XINT_add_k #1#3!#5!#7!%
}%
\def\XINT_add_e #1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_add_f #1%
}%
\def\XINT_add_f #11#2#3!#4!%
{%
    \xint_gob_til_sc #2\XINT_add_fi ;%
    \expandafter\XINT_add_g\the\numexpr#1+1#2#3+#4-\xint_c_ii\xint:%
}%
\def\XINT_add_fi;\expandafter\XINT_add_g
    \the\numexpr#1+#2+#3-\xint_c_ii\xint:#4!#5!#6\W
{%
    \XINT_add_k #1#3!#5!%
}%
\def\XINT_add_g #1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_add_h #1%
}%
\def\XINT_add_h #11#2#3!#4!%
{%
    \xint_gob_til_sc #2\XINT_add_hi ;%
    \expandafter\XINT_add_i\the\numexpr#1+1#2#3+#4-\xint_c_ii\xint:%
}%
\def\XINT_add_hi;%
    \expandafter\XINT_add_i\the\numexpr#1+#2+#3-\xint_c_ii\xint:#4\W
{%
    \XINT_add_k #1#3!%
}%
\def\XINT_add_i #1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_add_a #1%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_add_k #1{\if #12\expandafter\XINT_add_ke\else\expandafter\XINT_add_l \fi}%
\def\XINT_add_ke #11;#2\W {\XINT_add_kf #11;!}%
\def\XINT_add_kf 1{1\relax }%
\def\XINT_add_l 1#1#2{\xint_gob_til_sc #1\XINT_add_lf ;\XINT_add_m 1#1#2}%
\def\XINT_add_lf #1\W {1\relax 00000001!1;!}%
\def\XINT_add_m #1!{\expandafter\XINT_add_n\the\numexpr\xint_c_i+#1\xint:}%
\def\XINT_add_n #1#2\xint:{1#2\expandafter!\the\numexpr\XINT_add_o #1}%
%    \end{macrocode}
% \lverb|Here 2 stands for "carry", and 1 for "no carry" (we have been adding
% 1 to 1<8digits>.)|
%    \begin{macrocode}
\def\XINT_add_o #1{\if #12\expandafter\XINT_add_l\else\expandafter\XINT_add_ke \fi}%
%    \end{macrocode}
% \subsection{\csh{xintiiCmp}}
% \lverb|Moved from xint.sty to xintcore.sty and rewritten for 1.2l.
%
% 1.2l's \xintiiCmp is robust against non terminated input.
%
% 1.2o deprecates \xintCmp, with xintfrac loaded it will get overwritten anyhow.
%|
%    \begin{macrocode}
\def\xintiiCmp   {\romannumeral0\xintiicmp }%
\def\xintiicmp #1{\expandafter\XINT_iicmp\romannumeral`&&@#1\xint:}%
\def\XINT_iicmp #1#2\xint:#3%
{%
    \expandafter\XINT_cmp_nfork\expandafter #1\romannumeral`&&@#3\xint:#2\xint:
}%
\def\XINT_cmp_nfork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_cmp_firstiszero
      #2\XINT_cmp_secondiszero
       0{}%
    \krof
    \xint_UDsignsfork
          #1#2\XINT_cmp_minusminus
           #1-\XINT_cmp_minusplus
           #2-\XINT_cmp_plusminus
            --\XINT_cmp_plusplus
    \krof #1#2%
}%
\def\XINT_cmp_firstiszero  #1\krof 0#2#3\xint:#4\xint:
{%
    \xint_UDzerominusfork
      #2-{ 0}%
      0#2{ 1}%
       0-{ -1}%
    \krof
}%
\def\XINT_cmp_secondiszero #1\krof #20#3\xint:#4\xint:
{%
    \xint_UDzerominusfork
      #2-{ 0}%
      0#2{ -1}%
       0-{ 1}%
    \krof
}%
\def\XINT_cmp_plusminus    #1\xint:#2\xint:{ 1}%
\def\XINT_cmp_minusplus    #1\xint:#2\xint:{ -1}%
\def\XINT_cmp_minusminus
    --{\expandafter\XINT_opp\romannumeral0\XINT_cmp_plusplus {}{}}%
\def\XINT_cmp_plusplus  #1#2#3\xint:
{%
    \expandafter\XINT_cmp_pp
    \the\numexpr\expandafter\XINT_sepbyviii_andcount
    \romannumeral0\XINT_zeroes_forviii #2#3\R\R\R\R\R\R\R\R{10}0000001\W
    #2#3\XINT_sepbyviii_end 2345678\relax
        \xint_c_vii!\xint_c_vi!\xint_c_v!\xint_c_iv!%
        \xint_c_iii!\xint_c_ii!\xint_c_i!\xint_c_\W
    #1%
}%
\def\XINT_cmp_pp #1\xint:#2\xint:#3\xint:
{%
    \expandafter\XINT_cmp_checklengths
    \the\numexpr #2\expandafter\xint:%
    \the\numexpr\expandafter\XINT_sepbyviii_andcount
    \romannumeral0\XINT_zeroes_forviii #3\R\R\R\R\R\R\R\R{10}0000001\W
    #3\XINT_sepbyviii_end 2345678\relax
        \xint_c_vii!\xint_c_vi!\xint_c_v!\xint_c_iv!%
        \xint_c_iii!\xint_c_ii!\xint_c_i!\xint_c_\W
    #1;!1;!1;!1;!\W
}%
\def\XINT_cmp_checklengths #1\xint:#2\xint:#3\xint:
{%
    \ifnum #1=#3
       \expandafter\xint_firstoftwo
    \else
       \expandafter\xint_secondoftwo
    \fi
    \XINT_cmp_a {\XINT_cmp_distinctlengths {#1}{#3}}#2;!1;!1;!1;!\W
}%
\def\XINT_cmp_distinctlengths #1#2#3\W #4\W
{%
    \ifnum #1>#2
        \expandafter\xint_firstoftwo
    \else
        \expandafter\xint_secondoftwo
    \fi
    { -1}{ 1}%
}%
\def\XINT_cmp_a 1#1!1#2!1#3!1#4!#5\W 1#6!1#7!1#8!1#9!%
{%
    \xint_gob_til_sc #1\XINT_cmp_equal ;%
    \ifnum #1>#6 \XINT_cmp_gt\fi
    \ifnum #1<#6 \XINT_cmp_lt\fi
    \xint_gob_til_sc #2\XINT_cmp_equal ;%
    \ifnum #2>#7 \XINT_cmp_gt\fi
    \ifnum #2<#7 \XINT_cmp_lt\fi
    \xint_gob_til_sc #3\XINT_cmp_equal ;%
    \ifnum #3>#8 \XINT_cmp_gt\fi
    \ifnum #3<#8 \XINT_cmp_lt\fi
    \xint_gob_til_sc #4\XINT_cmp_equal ;%
    \ifnum #4>#9 \XINT_cmp_gt\fi
    \ifnum #4<#9 \XINT_cmp_lt\fi
    \XINT_cmp_a #5\W
}%
\def\XINT_cmp_lt#1{\def\XINT_cmp_lt\fi ##1\W ##2\W {\fi#1-1}}\XINT_cmp_lt{ }%
\def\XINT_cmp_gt#1{\def\XINT_cmp_gt\fi ##1\W ##2\W {\fi#11}}\XINT_cmp_gt{ }%
\def\XINT_cmp_equal #1\W #2\W { 0}%
%    \end{macrocode}
% \subsection{\csh{xintiiSub}}
% \lverb|Entirely rewritten for 1.2.
%
% Refactored at 1.2l. I was initially aiming at clinching some internal format
% of the type 1<8digits>!....1<8digits>! for chaining the arithmetic
% operations (as a preliminary step to deciding upon some internal format for
% $xintfracnameimp macros), thus I wanted to uniformize delimiters in
% particular and have some core macros inputting and outputting such formats.
% But the way division is implemented makes it currently very hard to obtain a
% satisfactory solution. For subtraction I got there almost, but there was
% added overhead and, as the core sub-routine still assumed the shorter number
% will be positioned first, one would need to record the length also in the
% basic internal format, or add the overhead to not make assumption on which
% one is shorter. I thus but back-tracked my steps but in passing I improved
% the efficiency (probably) in the worst case branch.
%
% Sadly this 1.2l refactoring left an extra ! in macro \XINT_sub_l_Ida. This
% bug shows only in rare circumstances which escaped out test suite :(
% Fixed at 1.2q.
%
% The other reason for backtracking was in relation with the decimal numbers.
% Having a core format in base 10^8 but ultimately the radix is actually 10
% leads to complications. I could use radix 10^8 for \xintiiexpr only, but
% then I need to make it compatible with sub-\xintiiexpr in \xintexpr, etc...
% there are many issues of this type.
%
% I considered also an approach like in the 1.2l \xintiiCmp, but decided to
% stick with the method here for now.|
%    \begin{macrocode}
\def\xintiiSub   {\romannumeral0\xintiisub }%
\def\xintiisub #1{\expandafter\XINT_iisub\romannumeral`&&@#1\xint:}%
\def\XINT_iisub #1#2\xint:#3%
{%
    \expandafter\XINT_sub_nfork\expandafter
    #1\romannumeral`&&@#3\xint:#2\xint:
}%
\def\XINT_sub_nfork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_sub_firstiszero
      #2\XINT_sub_secondiszero
       0{}%
    \krof
    \xint_UDsignsfork
          #1#2\XINT_sub_minusminus
           #1-\XINT_sub_minusplus
           #2-\XINT_sub_plusminus
            --\XINT_sub_plusplus
    \krof #1#2%
}%
\def\XINT_sub_firstiszero  #1\krof 0#2#3\xint:#4\xint:{\XINT_opp #2#3}%
\def\XINT_sub_secondiszero #1\krof #20#3\xint:#4\xint:{ #2#4}%
\def\XINT_sub_plusminus    #1#2{\XINT_add_pp_a #1{}}%
\def\XINT_sub_plusplus   #1#2%
   {\expandafter\XINT_opp\romannumeral0\XINT_sub_mm_a #1#2}%
\def\XINT_sub_minusplus    #1#2%
   {\expandafter-\romannumeral0\XINT_add_pp_a {}#2}%
\def\XINT_sub_minusminus #1#2{\XINT_sub_mm_a {}{}}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_sub_mm_a  #1#2#3\xint:
{%
  \expandafter\XINT_sub_mm_b
      \romannumeral0\expandafter\XINT_sepandrev_andcount
      \romannumeral0\XINT_zeroes_forviii #2#3\R\R\R\R\R\R\R\R{10}0000001\W
      #2#3\XINT_rsepbyviii_end_A 2345678%
          \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
  \X #1%
}%
\def\XINT_sub_mm_b #1\xint:#2\X #3\xint:
{%
    \expandafter\XINT_sub_checklengths
    \the\numexpr #1\expandafter\xint:%
    \romannumeral0\expandafter\XINT_sepandrev_andcount
    \romannumeral0\XINT_zeroes_forviii #3\R\R\R\R\R\R\R\R{10}0000001\W
    #3\XINT_rsepbyviii_end_A 2345678%
      \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
      1;!1;!1;!1;!\W
    #21;!1;!1;!1;!\W
    1;!1\R!1\R!1\R!1\R!%
    1\R!1\R!1\R!1\R!\W
}%
\def\XINT_sub_checklengths #1\xint:#2\xint:%
{%
    \ifnum #2>#1
       \expandafter\XINT_sub_exchange
    \else
       \expandafter\XINT_sub_aa
    \fi
}%
\def\XINT_sub_exchange #1\W #2\W
{%
    \expandafter\XINT_opp\romannumeral0\XINT_sub_aa #2\W #1\W
}%
\def\XINT_sub_aa
{%
    \expandafter\XINT_sub_out\the\numexpr\XINT_sub_a\xint_c_i
}%
%    \end{macrocode}
% \lverb|The post-processing (clean-up of zeros, or rescue of situation with
% A-B where actually B turns out bigger than A) will be done by a macro which
% depends on circumstances and will be initially last token before the
% reversion done by \XINT_unrevbyviii.|
%    \begin{macrocode}
\def\XINT_sub_out {\XINT_unrevbyviii{}}%
%    \end{macrocode}
% \lverb|1 as first token of #1 stands for "no carry", 0 will mean a carry.
%
%( Call: \the\numexpr
%:       \XINT_sub_a 1#11;!1;!1;!1;!\W
%:                    #21;!1;!1;!1;!\W
%)
% where #1 and #2
% are blocks of 1<8d>!, #1 (=B)  *must* be at most as long as #2 (=A),
% (in radix 10^8)
% and the routine wants to compute #2-#1 = A - B
%
% 1.2l uses 1;! delimiters to match those of addition (and multiplication).
% But in the end I reverted the code branch which made it possible to chain
% such operations keeping internal format in 8 digits blocks throughout.
%
% \numexpr governed expansion stops with various possibilities:
%
%- Type Ia:  #1 shorter than #2, no final carry
%- Type Ib:  #1 shorter than #2, a final carry but next block of #2 > 1
%- Type Ica: #1 shorter than #2, a final carry, next block of #2 is final and = 1
%- Type Icb: as Ica except that 00000001 block from #2 was not final
%- Type Id:  #1 shorter than #2, a final carry, next block of #2 = 0
%- Type IIa: #1 same length as #2, turns out it was <= #2.
%- Type IIb: #1 same length  as #2, but turned out > #2.
%
% Various type of post actions are then needed:
%
%- Ia: clean up of zeros in most significant block of 8 digits
%
%- Ib: as Ia
%
%- Ic: there may be significant blocks of 8 zeros to clean up from result.
% Only case Ica may have arbitrarily many of them, case Icb has only one such
% block.
%
%- Id: blocks of 99999999 may propagate and there might a be final zero block
% created which has to be cleaned up.
%
%- IIa: arbitrarily many zeros might have to be removed.
%
%- IIb: We wanted #2-#1 = - (#1-#2), but we got 10^{8N}+#2 -#1 = 10^{8N}-(#1-#2).
% We need to do the correction then we are as in IIa situation, except that
% final result can not be zero.
%
% The 1.2l method for this correction is (presumably, testing takes lots of
% time, which I do not have) more efficient than in 1.2 release. |
%    \begin{macrocode}
\def\XINT_sub_a #1!#2!#3!#4!#5\W #6!#7!#8!#9!%
{%
    \XINT_sub_b
    #1!#6!#2!#7!#3!#8!#4!#9!%
    #5\W
}%
%    \end{macrocode}
% \lverb|As 1.2l code uses 1<8digits>! blocks one has to be careful with
% the carry digit 1 or 0: A #11#2#3 pattern would result into an empty #1
% if the carry digit which is upfront is 1, rather than setting #1=1.|
%    \begin{macrocode}
\def\XINT_sub_b #1#2#3#4!#5!%
{%
    \xint_gob_til_sc #3\XINT_sub_bi ;%
    \expandafter\XINT_sub_c\the\numexpr#1+1#5-#3#4-\xint_c_i\xint:%
}%
\def\XINT_sub_c 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_sub_d #1%
}%
\def\XINT_sub_d #1#2#3#4!#5!%
{%
    \xint_gob_til_sc #3\XINT_sub_di ;%
    \expandafter\XINT_sub_e\the\numexpr#1+1#5-#3#4-\xint_c_i\xint:
}%
\def\XINT_sub_e 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_sub_f #1%
}%
\def\XINT_sub_f #1#2#3#4!#5!%
{%
    \xint_gob_til_sc #3\XINT_sub_fi ;%
    \expandafter\XINT_sub_g\the\numexpr#1+1#5-#3#4-\xint_c_i\xint:
}%
\def\XINT_sub_g 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_sub_h #1%
}%
\def\XINT_sub_h #1#2#3#4!#5!%
{%
    \xint_gob_til_sc #3\XINT_sub_hi ;%
    \expandafter\XINT_sub_i\the\numexpr#1+1#5-#3#4-\xint_c_i\xint:
}%
\def\XINT_sub_i 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_sub_a #1%
}%
\def\XINT_sub_bi;%
    \expandafter\XINT_sub_c\the\numexpr#1+1#2-#3\xint:
    #4!#5!#6!#7!#8!#9!\W
{%
    \XINT_sub_k #1#2!#5!#7!#9!%
}%
\def\XINT_sub_di;%
    \expandafter\XINT_sub_e\the\numexpr#1+1#2-#3\xint:
    #4!#5!#6!#7!#8\W
{%
    \XINT_sub_k #1#2!#5!#7!%
}%
\def\XINT_sub_fi;%
    \expandafter\XINT_sub_g\the\numexpr#1+1#2-#3\xint:
    #4!#5!#6\W
{%
    \XINT_sub_k #1#2!#5!%
}%
\def\XINT_sub_hi;%
    \expandafter\XINT_sub_i\the\numexpr#1+1#2-#3\xint:
    #4\W
{%
    \XINT_sub_k #1#2!%
}%
%    \end{macrocode}
% \lverb|B terminated. Have we reached the end of A (necessarily at least as
% long as B) ? (we are computing A-B, digits of B come first).
%
% If not, then we are certain that even if there is carry it will not
% propagate beyond the end of A. But it may propagate far transforming chains
% of 00000000 into 99999999, and if it does go to the final block which possibly is
% just 1<00000001>!, we will have those eight zeros to clean up.
%
% If A and B have the same length (in base 10^8) then arbitrarily many zeros
% might have to be cleaned up, and if A<B, the whole result will have to be
% complemented first.|
%    \begin{macrocode}
\def\XINT_sub_k #1#2#3%
{%
    \xint_gob_til_sc #3\XINT_sub_p;\XINT_sub_l #1#2#3%
}%
\def\XINT_sub_l #1%
   {\xint_UDzerofork #1\XINT_sub_l_carry 0\XINT_sub_l_Ia\krof}%
\def\XINT_sub_l_Ia 1#1;!#2\W{1\relax#1;!1\XINT_sub_fix_none!}%
%    \end{macrocode}
% \lverb|
%
% |
%    \begin{macrocode}
\def\XINT_sub_l_carry 1#1!{\ifcase #1
         \expandafter \XINT_sub_l_Id
    \or  \expandafter \XINT_sub_l_Ic
    \else\expandafter \XINT_sub_l_Ib\fi 1#1!}%
\def\XINT_sub_l_Ib #1;#2\W {-\xint_c_i+#1;!1\XINT_sub_fix_none!}%
\def\XINT_sub_l_Ic 1#1!1#2#3!#4;#5\W
{%
    \xint_gob_til_sc #2\XINT_sub_l_Ica;%
    1\relax 00000000!1#2#3!#4;!1\XINT_sub_fix_none!%
}%
%    \end{macrocode}
% \lverb|&
% We need to add some extra delimiters at the end for post-action by
% \XINT_num, so we first grab the material up to \W
% |
%    \begin{macrocode}
\def\XINT_sub_l_Ica#1\W
{%
    1;!1\XINT_sub_fix_cuz!%
    1;!1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
    \xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z
}%
\def\XINT_sub_l_Id 1#1!%
    {199999999\expandafter!\the\numexpr \XINT_sub_l_Id_a}%
\def\XINT_sub_l_Id_a 1#1!{\ifcase #1
         \expandafter \XINT_sub_l_Id
    \or  \expandafter \XINT_sub_l_Id_b
    \else\expandafter \XINT_sub_l_Ib\fi 1#1!}%
\def\XINT_sub_l_Id_b 1#1!1#2#3!#4;#5\W
{%
    \xint_gob_til_sc #2\XINT_sub_l_Ida;%
    1\relax 00000000!1#2#3!#4;!1\XINT_sub_fix_none!%
}%
\def\XINT_sub_l_Ida#1\XINT_sub_fix_none{1;!1\XINT_sub_fix_none}%
%    \end{macrocode}
% \lverb|&
% This is the case where both operands have same 10^8-base length.
%
% We were handling A-B but perhaps B>A. The situation with A=B is also
% annoying because we then have to clean up all zeros but don't know where to
% stop (if A>B the first non-zero 8 digits block would tell use when).
%
% Here again we need to grab #3\W to position the actually used terminating
% delimiters. 
% |
%    \begin{macrocode}
\def\XINT_sub_p;\XINT_sub_l #1#2\W #3\W
{%
    \xint_UDzerofork
       #1{1;!1\XINT_sub_fix_neg!%
          1;!1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
          \xint_bye2345678\xint_bye1099999988\relax}% A - B, B > A
        0{1;!1\XINT_sub_fix_cuz!%
          1;!1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W}%
    \krof
    \xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z
}%
%    \end{macrocode}
% \lverb|Routines for post-processing after reversal, and removal of
% separators. It is a matter of cleaning up zeros, and possibly in the bad
% case to take a complement before that.|
%    \begin{macrocode}
\def\XINT_sub_fix_none;{\XINT_cuz_small}%
\def\XINT_sub_fix_cuz ;{\expandafter\XINT_num_cleanup\the\numexpr\XINT_num_loop}%
%    \end{macrocode}
% \lverb|Case with A and B same number of digits in base 10^8 and B>A.
%
% 1.2l subtle chaining on the model of the 1.2i rewrite of \xintInc and
% similar routines. After taking complement, leading zeroes need to be
% cleaned up as in B<=A branch.|
%    \begin{macrocode}
\def\XINT_sub_fix_neg;%
{%
    \expandafter-\romannumeral0\expandafter
    \XINT_sub_comp_finish\the\numexpr\XINT_sub_comp_loop
}%
\def\XINT_sub_comp_finish 0{\XINT_sub_fix_cuz;}%
\def\XINT_sub_comp_loop #1#2#3#4#5#6#7#8%
{%
    \expandafter\XINT_sub_comp_clean
    \the\numexpr \xint_c_xi_e_viii_mone-#1#2#3#4#5#6#7#8\XINT_sub_comp_loop
}%
%    \end{macrocode}
% \lverb|#1 = 0 signifie une retenue, #1 = 1 pas de retenue, ce qui ne peut
% arriver que tant qu'il n'y a que des zéros du côté non significatif.
% Lorsqu'on est revenu au début on a forcément une retenue.|
%    \begin{macrocode}
\def\XINT_sub_comp_clean 1#1{+#1\relax}%
%    \end{macrocode}
% \subsection{\csh{xintiiMul}}
% \lverb|Completely rewritten for 1.2.
% 
% 1.2l: \xintiiMul made robust against non terminated input.|
%    \begin{macrocode}
\def\xintiiMul {\romannumeral0\xintiimul }%
\def\xintiimul #1%
{%
    \expandafter\XINT_iimul\romannumeral`&&@#1\xint:
}%
\def\XINT_iimul #1#2\xint:#3%
{%
    \expandafter\XINT_mul_nfork\expandafter #1\romannumeral`&&@#3\xint:#2\xint:
}%
%    \end{macrocode}
% \lverb|1.2 I have changed the fork, and it complicates matters elsewhere.
%
% ATTENTION for example that 1.4 \xintiiPrd uses \XINT_mul_nfork now.|
%    \begin{macrocode}
\def\XINT_mul_fork #1#2\xint:#3\xint:{\XINT_mul_nfork #1#3\xint:#2\xint:}%
\def\XINT_mul_nfork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_mul_zero
      #2\XINT_mul_zero
       0{}%
    \krof
    \xint_UDsignsfork
          #1#2\XINT_mul_minusminus
           #1-\XINT_mul_minusplus
           #2-\XINT_mul_plusminus
            --\XINT_mul_plusplus
    \krof #1#2%
}%
\def\XINT_mul_zero  #1\krof #2#3\xint:#4\xint:{ 0}%
\def\XINT_mul_minusminus   #1#2{\XINT_mul_plusplus {}{}}%
\def\XINT_mul_minusplus    #1#2%
   {\expandafter-\romannumeral0\XINT_mul_plusplus {}#2}%
\def\XINT_mul_plusminus    #1#2%
   {\expandafter-\romannumeral0\XINT_mul_plusplus #1{}}%
\def\XINT_mul_plusplus #1#2#3\xint:
{%
  \expandafter\XINT_mul_pre_b
      \romannumeral0\expandafter\XINT_sepandrev_andcount
      \romannumeral0\XINT_zeroes_forviii #2#3\R\R\R\R\R\R\R\R{10}0000001\W
      #2#3\XINT_rsepbyviii_end_A 2345678%
          \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
  \W #1%
}%
\def\XINT_mul_pre_b #1\xint:#2\W #3\xint:
{%
    \expandafter\XINT_mul_checklengths
    \the\numexpr #1\expandafter\xint:%
    \romannumeral0\expandafter\XINT_sepandrev_andcount
    \romannumeral0\XINT_zeroes_forviii #3\R\R\R\R\R\R\R\R{10}0000001\W
    #3\XINT_rsepbyviii_end_A 2345678%
      \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
     1;!\W #21;!%
    1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
%    \end{macrocode}
% \lverb|Cooking recipe, 2015/10/05.|
%    \begin{macrocode}
\def\XINT_mul_checklengths #1\xint:#2\xint:%
{%
    \ifnum #2=\xint_c_i\expandafter\XINT_mul_smallbyfirst\fi
    \ifnum #1=\xint_c_i\expandafter\XINT_mul_smallbysecond\fi
    \ifnum #2<#1
       \ifnum \numexpr (#2-\xint_c_i)*(#1-#2)<383
          \XINT_mul_exchange
       \fi
    \else
       \ifnum \numexpr (#1-\xint_c_i)*(#2-#1)>383
          \XINT_mul_exchange
       \fi
    \fi
    \XINT_mul_start
}%
\def\XINT_mul_smallbyfirst #1\XINT_mul_start 1#2!1;!\W
{%
    \ifnum#2=\xint_c_i\expandafter\XINT_mul_oneisone\fi
    \ifnum#2<\xint_c_xxii\expandafter\XINT_mul_verysmall\fi
    \expandafter\XINT_mul_out\the\numexpr\XINT_smallmul 1#2!%
}%
\def\XINT_mul_smallbysecond #1\XINT_mul_start #2\W 1#3!1;!%
{%
    \ifnum#3=\xint_c_i\expandafter\XINT_mul_oneisone\fi
    \ifnum#3<\xint_c_xxii\expandafter\XINT_mul_verysmall\fi
    \expandafter\XINT_mul_out\the\numexpr\XINT_smallmul 1#3!#2%
}%
\def\XINT_mul_oneisone #1!{\XINT_mul_out }%
\def\XINT_mul_verysmall\expandafter\XINT_mul_out
                       \the\numexpr\XINT_smallmul 1#1!%
    {\expandafter\XINT_mul_out\the\numexpr\XINT_verysmallmul 0\xint:#1!}%
\def\XINT_mul_exchange #1\XINT_mul_start #2\W #31;!%
   {\fi\fi\XINT_mul_start #31;!\W #2}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_mul_start
   {\expandafter\XINT_mul_out\the\numexpr\XINT_mul_loop 100000000!1;!\W}%
\def\XINT_mul_out
   {\expandafter\XINT_cuz_small\romannumeral0\XINT_unrevbyviii {}}%
%    \end{macrocode}
% \lverb|&
%
%( Call:
%: \the\numexpr \XINT_mul_loop 100000000!1;!\W #11;!\W #21;!
%)
% where #1 and #2 are (globally reversed) blocks 1<8d>!. Its is generally more
% efficient if #1 is the shorter one, but a better recipe is implemented in
% \XINT_mul_checklengths. One may call \XINT_mul_loop directly (but
% multiplication by zero will produce many 100000000! blocks on output).
%
% Ends after having produced: 1<8d>!....1<8d>!1;!. The last 8-digits block is
% significant one. It can not be 100000000! except if the loop was called with
% a zero operand.
%
% Thus \XINT_mul_loop can be conveniently called directly in recursive
% routines, as the output terminator can serve as input terminator, we can
% arrange to not have to grab the whole thing again.|
%    \begin{macrocode}
\def\XINT_mul_loop #1\W #2\W 1#3!%
{%
    \xint_gob_til_sc #3\XINT_mul_e ;%
    \expandafter\XINT_mul_a\the\numexpr \XINT_smallmul 1#3!#2\W
    #1\W #2\W
}%
%    \end{macrocode}
% \lverb|Each of #1 and #2 brings its 1;! for \XINT_add_a.|
%    \begin{macrocode}
\def\XINT_mul_a #1\W #2\W
{%
    \expandafter\XINT_mul_b\the\numexpr
    \XINT_add_a \xint_c_ii #21;!1;!1;!\W #11;!1;!1;!\W\W
}%
\def\XINT_mul_b 1#1!{1#1\expandafter!\the\numexpr\XINT_mul_loop }%
\def\XINT_mul_e;#1\W 1#2\W #3\W {1\relax #2}%
%    \end{macrocode}
% \lverb|1.2 small and mini multiplication in base 10^8 with carry. Used by
% the main multiplication routines. But division, float factorial, etc.. have
% their own variants as they need output with specific constraints.
%
% The minimulwc has 1<8digits carry>.<4 high digits>.<4 low digits!<8digits>.
%
% It produces a block 1<8d>! and then jump back into \XINT_smallmul_a with the
% new 8digits carry as argument. The \XINT_smallmul_a fetches a new 1<8d>!
% block to multiply, and calls back \XINT_minimul_wc having stored the
% multiplicand for re-use later. When the loop terminates, the final carry is
% checked for being nul, and in all cases the output is terminated by a 1;!
%
% Multiplication by zero will produce blocks of zeros.|
%    \begin{macrocode}
\def\XINT_minimulwc_a 1#1\xint:#2\xint:#3!#4#5#6#7#8\xint:%
{%
    \expandafter\XINT_minimulwc_b
    \the\numexpr \xint_c_x^ix+#1+#3*#8\xint:
                     #3*#4#5#6#7+#2*#8\xint:
                           #2*#4#5#6#7\xint:%
}%
\def\XINT_minimulwc_b 1#1#2#3#4#5#6\xint:#7\xint:%
{%
    \expandafter\XINT_minimulwc_c
    \the\numexpr \xint_c_x^ix+#1#2#3#4#5+#7\xint:#6\xint:%
}%
\def\XINT_minimulwc_c 1#1#2#3#4#5#6\xint:#7\xint:#8\xint:%
{%
    1#6#7\expandafter!%
    \the\numexpr\expandafter\XINT_smallmul_a
    \the\numexpr \xint_c_x^viii+#1#2#3#4#5+#8\xint:%
}%
\def\XINT_smallmul 1#1#2#3#4#5!{\XINT_smallmul_a 100000000\xint:#1#2#3#4\xint:#5!}%
\def\XINT_smallmul_a #1\xint:#2\xint:#3!1#4!%
{%
    \xint_gob_til_sc #4\XINT_smallmul_e;%
    \XINT_minimulwc_a #1\xint:#2\xint:#3!#4\xint:#2\xint:#3!%
}%
\def\XINT_smallmul_e;\XINT_minimulwc_a 1#1\xint:#2;#3!%
    {\xint_gob_til_eightzeroes #1\XINT_smallmul_f 000000001\relax #1!1;!}%
\def\XINT_smallmul_f 000000001\relax 00000000!1{1\relax}%
%    \end{macrocode}
%  \lverb|&
%  |
%    \begin{macrocode}
\def\XINT_verysmallmul #1\xint:#2!1#3!%
{%
    \xint_gob_til_sc #3\XINT_verysmallmul_e;%
    \expandafter\XINT_verysmallmul_a
    \the\numexpr #2*#3+#1\xint:#2!%
}%
\def\XINT_verysmallmul_e;\expandafter\XINT_verysmallmul_a\the\numexpr
    #1+#2#3\xint:#4!%
{\xint_gob_til_zero #2\XINT_verysmallmul_f 0\xint_c_x^viii+#2#3!1;!}%
\def\XINT_verysmallmul_f #1!1{1\relax}%
\def\XINT_verysmallmul_a #1#2\xint:%
{%
    \unless\ifnum #1#2<\xint_c_x^ix
    \expandafter\XINT_verysmallmul_bi\else
    \expandafter\XINT_verysmallmul_bj\fi
    \the\numexpr \xint_c_x^ix+#1#2\xint:%
}%
\def\XINT_verysmallmul_bj{\expandafter\XINT_verysmallmul_cj }%
\def\XINT_verysmallmul_cj 1#1#2\xint:%
    {1#2\expandafter!\the\numexpr\XINT_verysmallmul #1\xint:}%
\def\XINT_verysmallmul_bi\the\numexpr\xint_c_x^ix+#1#2#3\xint:%
    {1#3\expandafter!\the\numexpr\XINT_verysmallmul #1#2\xint:}%
%    \end{macrocode}
% \lverb|Used by division and by squaring, not by multiplication itself.
%
% This routine does not loop, it only does one mini multiplication with input
% format <4 high digits>.<4 low digits>!<8 digits>!, and on output
% 1<8d>!1<8d>!, with least significant block first.|
%    \begin{macrocode}
\def\XINT_minimul_a #1\xint:#2!#3#4#5#6#7!%
{%
    \expandafter\XINT_minimul_b
    \the\numexpr \xint_c_x^viii+#2*#7\xint:#2*#3#4#5#6+#1*#7\xint:#1*#3#4#5#6\xint:%
}%
\def\XINT_minimul_b 1#1#2#3#4#5\xint:#6\xint:%
{%
    \expandafter\XINT_minimul_c
    \the\numexpr \xint_c_x^ix+#1#2#3#4+#6\xint:#5\xint:%
}%
\def\XINT_minimul_c 1#1#2#3#4#5#6\xint:#7\xint:#8\xint:%
{%
    1#6#7\expandafter!\the\numexpr \xint_c_x^viii+#1#2#3#4#5+#8!%
}%
%    \end{macrocode}
% \subsection{\csh{xintiiDivision}}
% \lverb|Completely rewritten for 1.2.
%
% WARNING: some comments below try to describe the flow of tokens but they
% date back to xint 1.09j and I updated them on the fly while doing the 1.2
% version. As the routine now works in base 10^8, not 10^4 and "drops" the
% quotient digits,rather than store them upfront as the earlier code, I may
% well have not correctly converted all such comments. At the last minute some
% previously #1 became stuff like #1#2#3#4, then of course the old comments
% describing what the macro parameters stand for are necessarily wrong.
%
% Side remark: the way tokens are grouped was not essentially modified in
% 1.2, although the situation has changed. It was fine-tuned in xint
% 1.0/1.1 but the context has changed, and perhaps I should revisit this.
% As a corollary to the fact that quotient digits are now left behind thanks
% to the chains of \numexpr, some  macros which in 1.0/1.1 fetched up to 9
% parameters now need handle less such parameters. Thus, some rationale for
% the way the code was structured has disappeared.
%
%
% 1.2l: \xintiiDivision et al. made robust against non terminated input.
% |
% \lverb-#1 = A, #2 = B. On calcule le quotient et le reste dans la division
% euclidienne de A par B: A=BQ+R, 0<= R < |B|.-
%    \begin{macrocode}
\def\xintiiDivision   {\romannumeral0\xintiidivision }%
\def\xintiidivision  #1{\expandafter\XINT_iidivision \romannumeral`&&@#1\xint:}%
\def\XINT_iidivision #1#2\xint:#3{\expandafter\XINT_iidivision_a\expandafter #1%
                             \romannumeral`&&@#3\xint:#2\xint:}%
%    \end{macrocode}
% \lverb|On regarde les signes de A et de B.|
%    \begin{macrocode}
\def\XINT_iidivision_a #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_iidivision_divbyzero #1#2}\fi
    \if0#1\xint_dothis\XINT_iidivision_aiszero\fi
    \if-#2\xint_dothis{\expandafter\XINT_iidivision_bneg
                       \romannumeral0\XINT_iidivision_bpos #1}\fi
    \xint_orthat{\XINT_iidivision_bpos #1#2}%
}%
\def\XINT_iidivision_divbyzero#1#2#3\xint:#4\xint:
   {\if0#1\xint_dothis{\XINT_signalcondition{DivisionUndefined}}\fi
          \xint_orthat{\XINT_signalcondition{DivisionByZero}}%
    {Division by zero: #1#4/#2#3.}{}{{0}{0}}}%
\def\XINT_iidivision_aiszero #1\xint:#2\xint:{{0}{0}}%
\def\XINT_iidivision_bneg #1% q->-q, r unchanged
                          {\expandafter{\romannumeral0\XINT_opp #1}}%
\def\XINT_iidivision_bpos #1%
{%
    \xint_UDsignfork
            #1\XINT_iidivision_aneg
             -{\XINT_iidivision_apos #1}%
    \krof
}%
%    \end{macrocode}
% \lverb|Donc attention malgré son nom \XINT_div_prepare va jusqu'au bout.
% C'est donc en fait l'entrée principale (pour B>0, A>0) mais elle va
% regarder si B est < 10^8 et s'il vaut alors 1 ou 2, et si A < 10^8. Dans
% tous les cas le résultat est produit sous la forme {Q}{R}, avec Q et R sous
% leur forme final. On doit ensuite ajuster si le B ou le A initial était
% négatif. Je n'ai pas fait beaucoup d'efforts pour être un minimum efficace
% si A ou B n'est pas positif.|
%    \begin{macrocode}
\def\XINT_iidivision_apos #1#2\xint:#3\xint:{\XINT_div_prepare {#2}{#1#3}}%
\def\XINT_iidivision_aneg #1\xint:#2\xint:
   {\expandafter
    \XINT_iidivision_aneg_b\romannumeral0\XINT_div_prepare {#1}{#2}{#1}}%
\def\XINT_iidivision_aneg_b #1#2{\if0\XINT_Sgn #2\xint:
                              \expandafter\XINT_iidivision_aneg_rzero
                           \else
                              \expandafter\XINT_iidivision_aneg_rpos
                           \fi {#1}{#2}}%
\def\XINT_iidivision_aneg_rzero #1#2#3{{-#1}{0}}% necessarily q was >0
\def\XINT_iidivision_aneg_rpos #1%
{%
    \expandafter\XINT_iidivision_aneg_end\expandafter
               {\expandafter-\romannumeral0\xintinc {#1}}% q-> -(1+q)
}%
\def\XINT_iidivision_aneg_end #1#2#3%
{%
     \expandafter\xint_exchangetwo_keepbraces
     \expandafter{\romannumeral0\XINT_sub_mm_a {}{}#3\xint:#2\xint:}{#1}% r-> b-r
}%
%    \end{macrocode}
% \lverb|Le diviseur B va être étendu par des zéros pour que sa longueur soit
% multiple de huit. Les zéros seront mis du côté non significatif.|
%    \begin{macrocode}
\def\XINT_div_prepare #1%
{%
    \XINT_div_prepare_a #1\R\R\R\R\R\R\R\R {10}0000001\W !{#1}%
}%
\def\XINT_div_prepare_a #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_R #9\XINT_div_prepare_small\R
    \XINT_div_prepare_b #9%
}%
%    \end{macrocode}
% \lverb|B a au plus huit chiffres. On se débarrasse des trucs superflus. Si
% B>0 n'est ni 1 ni 2, le point d'entrée est \XINT_div_small_a {B}{A} (avec un
% A positif).|
%    \begin{macrocode}
\def\XINT_div_prepare_small\R #1!#2%
{%
    \ifcase #2
    \or\expandafter\XINT_div_BisOne
    \or\expandafter\XINT_div_BisTwo
    \else\expandafter\XINT_div_small_a
    \fi {#2}%
}%
\def\XINT_div_BisOne #1#2{{#2}{0}}%
\def\XINT_div_BisTwo #1#2%
{%
    \expandafter\expandafter\expandafter\XINT_div_BisTwo_a
    \ifodd\xintLDg{#2} \expandafter1\else \expandafter0\fi {#2}%
}%
\def\XINT_div_BisTwo_a #1#2%
{%
    \expandafter{\romannumeral0\XINT_half
     #2\xint_bye\xint_Bye345678\xint_bye
     *\xint_c_v+\xint_c_v)/\xint_c_x-\xint_c_i\relax}{#1}%
}%
%    \end{macrocode}
% \lverb|B a au plus huit chiffres et est au moins 3. On va l'utiliser
% directement, sans d'abord le multiplier par une puissance de 10 pour qu'il
% ait 8 chiffres.|
%    \begin{macrocode}
\def\XINT_div_small_a #1#2%
{%
    \expandafter\XINT_div_small_b
    \the\numexpr #1/\xint_c_ii\expandafter
    \xint:\the\numexpr \xint_c_x^viii+#1\expandafter!%
    \romannumeral0%
    \XINT_div_small_ba #2\R\R\R\R\R\R\R\R{10}0000001\W
       #2\XINT_sepbyviii_Z_end 2345678\relax
}%
%    \end{macrocode}
% \lverb|Le #2 poursuivra l'expansion par \XINT_div_dosmallsmall ou par
% \XINT_smalldivx_a suivi de \XINT_sdiv_out.|
%    \begin{macrocode}
\def\XINT_div_small_b #1!#2{#2#1!}%
%    \end{macrocode}
% \lverb|On ajoute des zéros avant A, puis on le prépare sous la forme de
% blocs 1<8d>! Au passage on repère le cas d'un A<10^8.|
%    \begin{macrocode}
\def\XINT_div_small_ba #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_R #9\XINT_div_smallsmall\R
    \expandafter\XINT_div_dosmalldiv
    \the\numexpr\expandafter\XINT_sepbyviii_Z
           \romannumeral0\XINT_zeroes_forviii
    #1#2#3#4#5#6#7#8#9%
}%
%    \end{macrocode}
% \lverb|Si A<10^8, on va poursuivre par \XINT_div_dosmallsmall
% round(B/2).10^8+B!{A}. On fait la division directe par \numexpr. Le résultat
% est produit sous la forme {Q}{R}.|
%    \begin{macrocode}
\def\XINT_div_smallsmall\R
    \expandafter\XINT_div_dosmalldiv
    \the\numexpr\expandafter\XINT_sepbyviii_Z
    \romannumeral0\XINT_zeroes_forviii #1\R #2\relax
   {{\XINT_div_dosmallsmall}{#1}}%
\def\XINT_div_dosmallsmall #1\xint:1#2!#3%
{%
    \expandafter\XINT_div_smallsmallend
    \the\numexpr (#3+#1)/#2-\xint_c_i\xint:#2\xint:#3\xint:%
}%
\def\XINT_div_smallsmallend #1\xint:#2\xint:#3\xint:{\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #3-#1*#2}}%
%    \end{macrocode}
% \lverb|Si A>=10^8, il est maintenant sous la forme 1<8d>!...1<8d>!1;! avec
% plus significatifs en premier. Donc on poursuit par$newline
% \expandafter\XINT_sdiv_out\the\numexpr\XINT_smalldivx_a
% x.1B!1<8d>!...1<8d>!1;! avec x =round(B/2), 1B=10^8+B.|
%    \begin{macrocode}
\def\XINT_div_dosmalldiv
    {{\expandafter\XINT_sdiv_out\the\numexpr\XINT_smalldivx_a}}%
%    \end{macrocode}
% \lverb|Ici B est au moins 10^8, on détermine combien de zéros lui adjoindre
% pour qu'il soit de longueur 8N.|
%    \begin{macrocode}
\def\XINT_div_prepare_b
   {\expandafter\XINT_div_prepare_c\romannumeral0\XINT_zeroes_forviii }%
\def\XINT_div_prepare_c #1!%
{%
     \XINT_div_prepare_d  #1.00000000!{#1}%
}%
\def\XINT_div_prepare_d #1#2#3#4#5#6#7#8#9%
{%
    \expandafter\XINT_div_prepare_e\xint_gob_til_dot #1#2#3#4#5#6#7#8#9!%
}%
\def\XINT_div_prepare_e #1!#2!#3#4%
{%
    \XINT_div_prepare_f #4#3\X {#1}{#3}%
}%
%    \end{macrocode}
% \lverb|attention qu'on calcule ici x'=x+1 (x = huit premiers chiffres du
% diviseur) et que si x=99999999, x' aura donc 9 chiffres, pas compatible avec
% div_mini (avant 1.2, x avait 4 chiffres, et on faisait la division avec x'
% dans un \numexpr). Bon, facile à dire après avoir laissé passer ce bug dans
% 1.2. C'est le problème lorsqu'au lieu de tout refaire à partir de zéro on
% recycle d'anciennes routines qui avaient un contexte différent.|
%    \begin{macrocode}
\def\XINT_div_prepare_f #1#2#3#4#5#6#7#8#9\X
{%
    \expandafter\XINT_div_prepare_g
     \the\numexpr  #1#2#3#4#5#6#7#8+\xint_c_i\expandafter
    \xint:\the\numexpr (#1#2#3#4#5#6#7#8+\xint_c_i)/\xint_c_ii\expandafter
    \xint:\the\numexpr #1#2#3#4#5#6#7#8\expandafter
    \xint:\romannumeral0\XINT_sepandrev_andcount
    #1#2#3#4#5#6#7#8#9\XINT_rsepbyviii_end_A 2345678%
                      \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
    \X
}%
\def\XINT_div_prepare_g #1\xint:#2\xint:#3\xint:#4\xint:#5\X #6#7#8%
{%
    \expandafter\XINT_div_prepare_h
    \the\numexpr\expandafter\XINT_sepbyviii_andcount
    \romannumeral0\XINT_zeroes_forviii #8#7\R\R\R\R\R\R\R\R{10}0000001\W
    #8#7\XINT_sepbyviii_end 2345678\relax
     \xint_c_vii!\xint_c_vi!\xint_c_v!\xint_c_iv!%
     \xint_c_iii!\xint_c_ii!\xint_c_i!\xint_c_\W
    {#1}{#2}{#3}{#4}{#5}{#6}%
}%
\def\XINT_div_prepare_h #11\xint:#2\xint:#3#4#5#6%#7#8%
{%
    \XINT_div_start_a {#2}{#6}{#1}{#3}{#4}{#5}%{#7}{#8}%
}%
%    \end{macrocode}
% \lverb|L, K, A, x',y,x, B, «c». Attention que K est diminué de 1 plus loin.
% Comme xint 1.2 a déjà repéré K=1, on a ici au minimum K=2. Attention B est à
% l'envers, A est à l'endroit et les deux avec séparateurs. Attention que ce
% n'est pas ici qu'on boucle mais en \XINT_div_I_a.|
%    \begin{macrocode}
\def\XINT_div_start_a #1#2%
{%
    \ifnum #1 < #2
      \expandafter\XINT_div_zeroQ
    \else
      \expandafter\XINT_div_start_b
    \fi
    {#1}{#2}%
}%
\def\XINT_div_zeroQ #1#2#3#4#5#6#7%
{%
    \expandafter\XINT_div_zeroQ_end
    \romannumeral0\XINT_unsep_cuzsmall
    #3\xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax\xint:
}%
\def\XINT_div_zeroQ_end #1\xint:#2%
    {\expandafter{\expandafter0\expandafter}\XINT_div_cleanR #1#2\xint:}%
%    \end{macrocode}
% \lverb|L, K, A, x',y,x, B, «c»->K.A.x{LK{x'y}x}B«c»|
%    \begin{macrocode}
\def\XINT_div_start_b #1#2#3#4#5#6%
{%
    \expandafter\XINT_div_finish\the\numexpr
    \XINT_div_start_c {#2}\xint:#3\xint:{#6}{{#1}{#2}{{#4}{#5}}{#6}}%
}%
\def\XINT_div_finish
{%
    \expandafter\XINT_div_finish_a \romannumeral`&&@\XINT_div_unsepQ
}%
\def\XINT_div_finish_a #1\Z #2\xint:{\XINT_div_finish_b #2\xint:{#1}}%
%    \end{macrocode}
% \lverb|Ici ce sont routines de fin. Le reste déjà nettoyé. R.Q«c».|
%    \begin{macrocode}
\def\XINT_div_finish_b #1%
{%
    \if0#1%
       \expandafter\XINT_div_finish_bRzero
    \else
       \expandafter\XINT_div_finish_bRpos
    \fi
    #1%
}%
\def\XINT_div_finish_bRzero 0\xint:#1#2{{#1}{0}}%
\def\XINT_div_finish_bRpos #1\xint:#2#3%
{%
    \expandafter\xint_exchangetwo_keepbraces\XINT_div_cleanR  #1#3\xint:{#2}%
}%
\def\XINT_div_cleanR #100000000\xint:{{#1}}%
%    \end{macrocode}
% \lverb|Kalpha.A.x{LK{x'y}x}, B, «c», au début #2=alpha est vide. On fait une
% boucle pour prendre K unités de A (on a au moins L égal à K) et les mettre
% dans alpha.|
%    \begin{macrocode}
\def\XINT_div_start_c #1%
{%
    \ifnum #1>\xint_c_vi
       \expandafter\XINT_div_start_ca
    \else
       \expandafter\XINT_div_start_cb
    \fi {#1}%
}%
\def\XINT_div_start_ca #1#2\xint:#3!#4!#5!#6!#7!#8!#9!%
{%
    \expandafter\XINT_div_start_c\expandafter
    {\the\numexpr #1-\xint_c_vii}#2#3!#4!#5!#6!#7!#8!#9!\xint:%
}%
\def\XINT_div_start_cb #1%
   {\csname XINT_div_start_c_\romannumeral\numexpr#1\endcsname}%
\def\XINT_div_start_c_i   #1\xint:#2!%
    {\XINT_div_start_c_   #1#2!\xint:}%
\def\XINT_div_start_c_ii  #1\xint:#2!#3!%
    {\XINT_div_start_c_   #1#2!#3!\xint:}%
\def\XINT_div_start_c_iii #1\xint:#2!#3!#4!%
    {\XINT_div_start_c_   #1#2!#3!#4!\xint:}%
\def\XINT_div_start_c_iv  #1\xint:#2!#3!#4!#5!%
    {\XINT_div_start_c_   #1#2!#3!#4!#5!\xint:}%
\def\XINT_div_start_c_v   #1\xint:#2!#3!#4!#5!#6!%
    {\XINT_div_start_c_   #1#2!#3!#4!#5!#6!\xint:}%
\def\XINT_div_start_c_vi  #1\xint:#2!#3!#4!#5!#6!#7!%
    {\XINT_div_start_c_   #1#2!#3!#4!#5!#6!#7!\xint:}%
%    \end{macrocode}
% \lverb|#1=a, #2=alpha (de longueur K, à l'endroit).#3=reste de A.#4=x,
% #5={LK{x'y}x},#6=B,«c» -> a, x, alpha, B, {00000000}, L, K, {x'y},x,
% alpha'=reste de A, B«c».|
%    \begin{macrocode}
\def\XINT_div_start_c_ 1#1!#2\xint:#3\xint:#4#5#6%
{%
    \XINT_div_I_a {#1}{#4}{1#1!#2}{#6}{00000000}#5{#3}{#6}%
}%
%    \end{macrocode}
% \lverb|Ceci est le point de retour de la boucle principale. a, x, alpha, B,
% q0, L, K, {x'y}, x, alpha', B«c» |
%    \begin{macrocode}
\def\XINT_div_I_a #1#2%
{%
    \expandafter\XINT_div_I_b\the\numexpr #1/#2\xint:{#1}{#2}%
}%
\def\XINT_div_I_b #1%
{%
    \xint_gob_til_zero #1\XINT_div_I_czero 0\XINT_div_I_c #1%
}%
%    \end{macrocode}
% \lverb|On intercepte petit quotient nul: #1=a, x, alpha, B, #5=q0, L, K,
%    {x'y}, x, alpha', B«c» -> on lâche un q puis {alpha} L, K, {x'y}, x,
%    alpha', B«c».|
%    \begin{macrocode}
\def\XINT_div_I_czero 0\XINT_div_I_c 0\xint:#1#2#3#4#5{1#5\XINT_div_I_g {#3}}%
\def\XINT_div_I_c #1\xint:#2#3%
{%
    \expandafter\XINT_div_I_da\the\numexpr #2-#1*#3\xint:#1\xint:{#2}{#3}%
}%
%    \end{macrocode}
% \lverb|r.q.alpha, B, q0, L, K, {x'y}, x, alpha', B«c»|
%    \begin{macrocode}
\def\XINT_div_I_da #1\xint:%
{%
    \ifnum #1>\xint_c_ix
       \expandafter\XINT_div_I_dP
    \else
       \ifnum #1<\xint_c_
        \expandafter\expandafter\expandafter\XINT_div_I_dN
       \else
        \expandafter\expandafter\expandafter\XINT_div_I_db
       \fi
    \fi
}%
%    \end{macrocode}
% \lverb|attention très mauvaises notations avec _b et _db.|
%    \begin{macrocode}
\def\XINT_div_I_dN #1\xint:%
{%
    \expandafter\XINT_div_I_b\the\numexpr #1-\xint_c_i\xint:%
}%
\def\XINT_div_I_db #1\xint:#2#3#4#5%
{%
    \expandafter\XINT_div_I_dc\expandafter #1%
    \romannumeral0\expandafter\XINT_div_sub\expandafter
       {\romannumeral0\XINT_rev_nounsep {}#4\R!\R!\R!\R!\R!\R!\R!\R!\W}%
       {\the\numexpr\XINT_div_verysmallmul #1!#51;!}%
    \Z {#4}{#5}%
}%
%    \end{macrocode}
% \lverb|La soustraction spéciale renvoie simplement - si le chiffre q est
% trop grand. On invoque dans ce cas I_dP.|
%    \begin{macrocode}
\def\XINT_div_I_dc #1#2%
{%
    \if-#2\expandafter\XINT_div_I_dd\else\expandafter\XINT_div_I_de\fi
     #1#2%
}%
\def\XINT_div_I_dd #1-\Z
{%
    \if #11\expandafter\XINT_div_I_dz\fi
    \expandafter\XINT_div_I_dP\the\numexpr #1-\xint_c_i\xint: XX%
}%
\def\XINT_div_I_dz #1XX#2#3#4%
{%
    1#4\XINT_div_I_g {#2}%
}%
\def\XINT_div_I_de #1#2\Z #3#4#5{1#5+#1\XINT_div_I_g {#2}}%
%    \end{macrocode}
% \lverb|q.alpha, B, q0, L, K, {x'y},x, alpha'B«c» (q=0 has been intercepted)
%        -> 1nouveauq.nouvel alpha, L, K, {x'y}, x, alpha',B«c»|
%    \begin{macrocode}
\def\XINT_div_I_dP #1\xint:#2#3#4#5#6%
{%
    1#6+#1\expandafter\XINT_div_I_g\expandafter
    {\romannumeral0\expandafter\XINT_div_sub\expandafter
      {\romannumeral0\XINT_rev_nounsep {}#4\R!\R!\R!\R!\R!\R!\R!\R!\W}%
      {\the\numexpr\XINT_div_verysmallmul #1!#51;!}%
    }%
}%
%    \end{macrocode}
% \lverb|1#1=nouveau q. nouvel alpha, L, K, {x'y},x,alpha', BQ«c»|
%    \begin{macrocode}
%    \end{macrocode}
% \lverb|#1=q,#2=nouvel alpha,#3=L, #4=K, #5={x'y}, #6=x, #7= alpha',#8=B,
% «c» -> on laisse q puis {x'y}alpha.alpha'.{{x'y}xKL}B«c»|
%    \begin{macrocode}
\def\XINT_div_I_g #1#2#3#4#5#6#7%
{%
     \expandafter !\the\numexpr
     \ifnum#2=#3
          \expandafter\XINT_div_exittofinish
     \else
          \expandafter\XINT_div_I_h
     \fi
     {#4}#1\xint:#6\xint:{{#4}{#5}{#3}{#2}}{#7}%
}%
%    \end{macrocode}
% \lverb|{x'y}alpha.alpha'.{{x'y}xKL}B«c» -> Attention retour à l'envoyeur ici
% par terminaison des \the\numexpr. On doit reprendre le Q déjà sorti, qui n'a
% plus de séparateurs, ni de leading 1. Ensuite R sans leading zeros.«c»|
%    \begin{macrocode}
\def\XINT_div_exittofinish #1#2\xint:#3\xint:#4#5%
{%
    1\expandafter\expandafter\expandafter!\expandafter\XINT_div_unsepQ_delim
    \romannumeral0\XINT_div_unsepR #2#3%
    \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax\R\xint:
}%
%    \end{macrocode}
% \lverb|ATTENTION DESCRIPTION OBSOLÈTE. #1={x'y}alpha.#2!#3=reste de A.
% #4={{x'y},x,K,L},#5=B,«c» devient {x'y},alpha sur K+4 chiffres.B,
% {{x'y},x,K,L}, #6= nouvel alpha',B,«c»|
%    \begin{macrocode}
\def\XINT_div_I_h #1\xint:#2!#3\xint:#4#5%
{%
    \XINT_div_II_b #1#2!\xint:{#5}{#4}{#3}{#5}%
}%
%    \end{macrocode}
% \lverb|{x'y}alpha.B, {{x'y},x,K,L}, nouveau alpha',B,«c»|
%    \begin{macrocode}
\def\XINT_div_II_b #11#2!#3!%
{%
    \xint_gob_til_eightzeroes #2\XINT_div_II_skipc 00000000%
    \XINT_div_II_c #1{1#2}{#3}%
}%
%    \end{macrocode}
% \lverb|x'y{100000000}{1<8>}reste de alpha.#6=B,#7={{x'y},x,K,L}, alpha',B,
% «c» -> {x'y}x,K,L (à diminuer de 4), {alpha sur
% K}B{q1=00000000}{alpha'}B,«c»|
%    \begin{macrocode}
\def\XINT_div_II_skipc 00000000\XINT_div_II_c #1#2#3#4#5\xint:#6#7%
{%
    \XINT_div_II_k #7{#4!#5}{#6}{00000000}%
}%
%    \end{macrocode}
% \lverb|x'ya->1qx'yalpha.B, {{x'y},x,K,L}, nouveau alpha',B, «c». En fait,
% attention, ici #3 et #4 sont les 16 premiers chiffres du numérateur,sous la
% forme blocs 1<8chiffres>.
% |
%    \begin{macrocode}
\def\XINT_div_II_c #1#2#3#4%
{%
     \expandafter\XINT_div_II_d\the\numexpr\XINT_div_xmini
     #1\xint:#2!#3!#4!{#1}{#2}#3!#4!%
}%
\def\XINT_div_xmini #1%
{%
    \xint_gob_til_one #1\XINT_div_xmini_a 1\XINT_div_mini #1%
}%
\def\XINT_div_xmini_a 1\XINT_div_mini 1#1%
{%
    \xint_gob_til_zero #1\XINT_div_xmini_b 0\XINT_div_mini 1#1%
}%
\def\XINT_div_xmini_b 0\XINT_div_mini 10#1#2#3#4#5#6#7%
{%
    \xint_gob_til_zero #7\XINT_div_xmini_c 0\XINT_div_mini 10#1#2#3#4#5#6#7%
}%
%    \end{macrocode}
% \lverb|x'=10^8 and we return #1=1<8digits>.|
%    \begin{macrocode}
\def\XINT_div_xmini_c 0\XINT_div_mini 100000000\xint:50000000!#1!#2!{#1!}%
%    \end{macrocode}
% \lverb|1 suivi de q1 sur huit chiffres! #2=x', #3=y, #4=alpha.#5=B,
% {{x'y},x,K,L}, alpha', B, «c» --> nouvel alpha.x',y,B,q1,{{x'y},x,K,L},
% alpha', B, «c» |
%    \begin{macrocode}
\def\XINT_div_II_d 1#1#2#3#4#5!#6#7#8\xint:#9%
{%
    \expandafter\XINT_div_II_e
    \romannumeral0\expandafter\XINT_div_sub\expandafter
      {\romannumeral0\XINT_rev_nounsep {}#8\R!\R!\R!\R!\R!\R!\R!\R!\W}%
      {\the\numexpr\XINT_div_smallmul_a 100000000\xint:#1#2#3#4\xint:#5!#91;!}%
    \xint:{#6}{#7}{#9}{#1#2#3#4#5}%
}%
%    \end{macrocode}
% \lverb|alpha.x',y,B,q1, {{x'y},x,K,L}, alpha', B, «c». Attention la
% soustraction spéciale doit maintenir les blocs 1<8>!|
%    \begin{macrocode}
\def\XINT_div_II_e 1#1!%
{%
    \xint_gob_til_eightzeroes #1\XINT_div_II_skipf 00000000%
    \XINT_div_II_f 1#1!%
}%
%    \end{macrocode}
% \lverb|100000000! alpha sur K chiffres.#2=x',#3=y,#4=B,#5=q1, #6={{x'y},x,K,L},
% #7=alpha',B«c» -> {x'y}x,K,L (à diminuer de 1),
% {alpha sur K}B{q1}{alpha'}B«c»|
%    \begin{macrocode}
\def\XINT_div_II_skipf 00000000\XINT_div_II_f 100000000!#1\xint:#2#3#4#5#6%
{%
    \XINT_div_II_k #6{#1}{#4}{#5}%
}%
%    \end{macrocode}
% \lverb|1<a1>!1<a2>!, alpha (sur K+1 blocs de 8). x', y, B, q1, {{x'y},x,K,L},
% alpha', B,«c».
%
% Here also we are dividing with x' which could be 10^8 in the exceptional
% case x=99999999. Must intercept it before sending to \XINT_div_mini.|
%    \begin{macrocode}
\def\XINT_div_II_f #1!#2!#3\xint:%
{%
    \XINT_div_II_fa {#1!#2!}{#1!#2!#3}%
}%
\def\XINT_div_II_fa #1#2#3#4%
{%
    \expandafter\XINT_div_II_g \the\numexpr\XINT_div_xmini #3\xint:#4!#1{#2}%
}%
%    \end{macrocode}
% \lverb|#1=q, #2=alpha (K+4), #3=B, #4=q1, {{x'y},x,K,L}, alpha', BQ«c»
%        -> 1 puis nouveau q sur 8 chiffres. nouvel alpha sur K blocs,
%        B, {{x'y},x,K,L}, alpha',B«c» |
%    \begin{macrocode}
\def\XINT_div_II_g 1#1#2#3#4#5!#6#7#8%
{%
    \expandafter \XINT_div_II_h
    \the\numexpr 1#1#2#3#4#5+#8\expandafter\expandafter\expandafter
    \xint:\expandafter\expandafter\expandafter
    {\expandafter\xint_gob_til_exclam
     \romannumeral0\expandafter\XINT_div_sub\expandafter
       {\romannumeral0\XINT_rev_nounsep {}#6\R!\R!\R!\R!\R!\R!\R!\R!\W}%
       {\the\numexpr\XINT_div_smallmul_a 100000000\xint:#1#2#3#4\xint:#5!#71;!}}%
    {#7}%
}%
%    \end{macrocode}
% \lverb|1 puis nouveau q sur 8 chiffres, #2=nouvel alpha sur K blocs,
% #3=B, #4={{x'y},x,K,L} avec L à ajuster,  alpha', BQ«c»
% -> {x'y}x,K,L à diminuer de 1, {alpha}B{q}, alpha', BQ«c»|
%    \begin{macrocode}
\def\XINT_div_II_h 1#1\xint:#2#3#4%
{%
    \XINT_div_II_k #4{#2}{#3}{#1}%
}%
%    \end{macrocode}
% \lverb|{x'y}x,K,L à diminuer de 1, alpha, B{q}alpha',B«c»
%        ->nouveau L.K,x',y,x,alpha.B,q,alpha',B,«c»
%        ->{LK{x'y}x},x,a,alpha.B,q,alpha',B,«c»|
%    \begin{macrocode}
\def\XINT_div_II_k #1#2#3#4#5%
{%
    \expandafter\XINT_div_II_l \the\numexpr #4-\xint_c_i\xint:{#3}#1{#2}#5\xint:%
}%
\def\XINT_div_II_l #1\xint:#2#3#4#51#6!%
{%
    \XINT_div_II_m {{#1}{#2}{{#3}{#4}}{#5}}{#5}{#6}1#6!%
}%
%    \end{macrocode}
% \lverb|{LK{x'y}x},x,a,alpha.B{q}alpha'B -> a, x, alpha, B, q,
% L, K, {x'y}, x, alpha', B«c» |
%    \begin{macrocode}
\def\XINT_div_II_m #1#2#3#4\xint:#5#6%
{%
     \XINT_div_I_a {#3}{#2}{#4}{#5}{#6}#1%
}%
%    \end{macrocode}
% \lverb|This multiplication is exactly like \XINT_smallmul -- apart from not
% inserting an ending 1;! --, but keeps ever a vanishing ending carry.|
%    \begin{macrocode}
\def\XINT_div_minimulwc_a 1#1\xint:#2\xint:#3!#4#5#6#7#8\xint:%
{%
    \expandafter\XINT_div_minimulwc_b
    \the\numexpr \xint_c_x^ix+#1+#3*#8\xint:#3*#4#5#6#7+#2*#8\xint:#2*#4#5#6#7\xint:%
}%
\def\XINT_div_minimulwc_b 1#1#2#3#4#5#6\xint:#7\xint:%
{%
    \expandafter\XINT_div_minimulwc_c
    \the\numexpr \xint_c_x^ix+#1#2#3#4#5+#7\xint:#6\xint:%
}%
\def\XINT_div_minimulwc_c 1#1#2#3#4#5#6\xint:#7\xint:#8\xint:%
{%
    1#6#7\expandafter!%
    \the\numexpr\expandafter\XINT_div_smallmul_a
    \the\numexpr \xint_c_x^viii+#1#2#3#4#5+#8\xint:%
}%
\def\XINT_div_smallmul_a #1\xint:#2\xint:#3!1#4!%
{%
    \xint_gob_til_sc #4\XINT_div_smallmul_e;%
    \XINT_div_minimulwc_a #1\xint:#2\xint:#3!#4\xint:#2\xint:#3!%
}%
\def\XINT_div_smallmul_e;\XINT_div_minimulwc_a 1#1\xint:#2;#3!{1\relax #1!}%
%    \end{macrocode}
% \lverb|Special very small multiplication for division. We only need to cater
% for multiplicands from 1 to 9. The ending is different from standard
% verysmallmul, a zero carry is not suppressed. And no final 1;! is added. If
% multiplicand is just 1 let's not forget to add the zero carry 100000000! at
% the end.|
%    \begin{macrocode}
\def\XINT_div_verysmallmul #1%
   {\xint_gob_til_one #1\XINT_div_verysmallisone 1\XINT_div_verysmallmul_a 0\xint:#1}%
\def\XINT_div_verysmallisone 1\XINT_div_verysmallmul_a 0\xint:1!1#11;!%
   {1\relax #1100000000!}%
\def\XINT_div_verysmallmul_a #1\xint:#2!1#3!%
{%
    \xint_gob_til_sc #3\XINT_div_verysmallmul_e;%
    \expandafter\XINT_div_verysmallmul_b
    \the\numexpr \xint_c_x^ix+#2*#3+#1\xint:#2!%
}%
\def\XINT_div_verysmallmul_b 1#1#2\xint:%
    {1#2\expandafter!\the\numexpr\XINT_div_verysmallmul_a #1\xint:}%
\def\XINT_div_verysmallmul_e;#1;+#2#3!{1\relax 0000000#2!}%
%    \end{macrocode}
% \lverb|Special subtraction for division purposes. If the subtracted thing
% turns out to be bigger, then just return a -. If not, then we must reverse
% the result, keeping the separators.|
%    \begin{macrocode}
\def\XINT_div_sub #1#2%
{%
    \expandafter\XINT_div_sub_clean
    \the\numexpr\expandafter\XINT_div_sub_a\expandafter
    1#2;!;!;!;!;!\W #1;!;!;!;!;!\W
}%
\def\XINT_div_sub_clean #1-#2#3\W
{%
    \if1#2\expandafter\XINT_rev_nounsep\else\expandafter\XINT_div_sub_neg\fi
    {}#1\R!\R!\R!\R!\R!\R!\R!\R!\W
}%
\def\XINT_div_sub_neg #1\W { -}%
\def\XINT_div_sub_a #1!#2!#3!#4!#5\W #6!#7!#8!#9!%
{%
    \XINT_div_sub_b #1!#6!#2!#7!#3!#8!#4!#9!#5\W
}%
\def\XINT_div_sub_b #1#2#3!#4!%
{%
    \xint_gob_til_sc #4\XINT_div_sub_bi ;%
    \expandafter\XINT_div_sub_c\the\numexpr#1-#3+1#4-\xint_c_i\xint:%
}%
\def\XINT_div_sub_c 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_div_sub_d #1%
}%
\def\XINT_div_sub_d #1#2#3!#4!%
{%
    \xint_gob_til_sc #4\XINT_div_sub_di ;%
    \expandafter\XINT_div_sub_e\the\numexpr#1-#3+1#4-\xint_c_i\xint:%
}%
\def\XINT_div_sub_e 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_div_sub_f #1%
}%
\def\XINT_div_sub_f #1#2#3!#4!%
{%
    \xint_gob_til_sc #4\XINT_div_sub_fi ;%
    \expandafter\XINT_div_sub_g\the\numexpr#1-#3+1#4-\xint_c_i\xint:%
}%
\def\XINT_div_sub_g 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_div_sub_h #1%
}%
\def\XINT_div_sub_h #1#2#3!#4!%
{%
    \xint_gob_til_sc #4\XINT_div_sub_hi ;%
    \expandafter\XINT_div_sub_i\the\numexpr#1-#3+1#4-\xint_c_i\xint:%
}%
\def\XINT_div_sub_i 1#1#2\xint:%
{%
    1#2\expandafter!\the\numexpr\XINT_div_sub_a #1%
}%
\def\XINT_div_sub_bi;%
    \expandafter\XINT_div_sub_c\the\numexpr#1-#2+#3\xint:#4!#5!#6!#7!#8!#9!;!\W
{%
    \XINT_div_sub_l #1#2!#5!#7!#9!%
}%
\def\XINT_div_sub_di;%
    \expandafter\XINT_div_sub_e\the\numexpr#1-#2+#3\xint:#4!#5!#6!#7!#8\W
{%
    \XINT_div_sub_l #1#2!#5!#7!%
}%
\def\XINT_div_sub_fi;%
    \expandafter\XINT_div_sub_g\the\numexpr#1-#2+#3\xint:#4!#5!#6\W
{%
    \XINT_div_sub_l #1#2!#5!%
}%
\def\XINT_div_sub_hi;%
    \expandafter\XINT_div_sub_i\the\numexpr#1-#2+#3\xint:#4\W
{%
    \XINT_div_sub_l #1#2!%
}%
\def\XINT_div_sub_l #1%
{%
   \xint_UDzerofork
      #1{-2\relax}%
       0\XINT_div_sub_r
   \krof
}%
\def\XINT_div_sub_r #1!%
{%
    -\ifnum 0#1=\xint_c_ 1\else2\fi\relax
}%
%    \end{macrocode}
% \lverb|Ici B<10^8 (et est >2). On
% exécute$newline
% \expandafter\XINT_sdiv_out\the\numexpr\XINT_smalldivx_a
%              x.1B!1<8d>!...1<8d>!1;!$newline
% avec x =round(B/2), 1B=10^8+B, et A déjà en
% blocs 1<8d>! (non renversés). Le \the\numexpr\XINT_smalldivx_a va produire
% Q\Z R\W avec un R<10^8, et un Q sous forme de blocs 1<8d>! terminé par 1!
% et nécessitant le nettoyage du premier bloc. Dans cette branche le B n'a pas
% été multiplié par une puissance de 10, il peut avoir moins de huit chiffres.
%
% |
%    \begin{macrocode}
\def\XINT_sdiv_out #1;!#2!%
    {\expandafter
     {\romannumeral0\XINT_unsep_cuzsmall
      #1\xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax}%
     {#2}}%
%    \end{macrocode}
% \lverb|La toute première étape fait la première division pour être sûr par
% la suite d'avoir un premier bloc pour A qui sera < B.|
%    \begin{macrocode}
\def\XINT_smalldivx_a #1\xint:1#2!1#3!%
{%
    \expandafter\XINT_smalldivx_b
    \the\numexpr (#3+#1)/#2-\xint_c_i!#1\xint:#2!#3!%
}%
\def\XINT_smalldivx_b #1#2!%
{%
    \if0#1\else
          \xint_c_x^viii+#1#2\xint_afterfi{\expandafter!\the\numexpr}\fi
    \XINT_smalldiv_c #1#2!%
}%
\def\XINT_smalldiv_c #1!#2\xint:#3!#4!%
{%
    \expandafter\XINT_smalldiv_d\the\numexpr #4-#1*#3!#2\xint:#3!%
}%
%    \end{macrocode}
% \lverb|On va boucler ici: #1 est un reste, #2 est x.B (avec B sans le 1 mais
% sur huit chiffres). #3#4 est le premier bloc qui reste de A. Si on a terminé
% avec A, alors #1 est le reste final. Le quotient lui est terminé par un 1!
% ce 1! disparaîtra dans le nettoyage par \XINT_unsep_cuzsmall.
% |
%    \begin{macrocode}
\def\XINT_smalldiv_d #1!#2!1#3#4!%
{%
    \xint_gob_til_sc #3\XINT_smalldiv_end ;%
    \XINT_smalldiv_e #1!#2!1#3#4!%
}%
\def\XINT_smalldiv_end;\XINT_smalldiv_e #1!#2!1;!{1!;!#1!}%
%    \end{macrocode}
% \lverb|Il est crucial que le reste #1 est < #3. J'ai documenté cette routine
% dans le fichier où j'ai préparé 1.2, il faudra transférer ici. Il n'est pas
% nécessaire pour cette routine que le diviseur B ait au moins 8 chiffres.
% Mais il doit être < 10^8.|
%    \begin{macrocode}
\def\XINT_smalldiv_e #1!#2\xint:#3!%
{%
    \expandafter\XINT_smalldiv_f\the\numexpr
    \xint_c_xi_e_viii_mone+#1*\xint_c_x^viii/#3!#2\xint:#3!#1!%
}%
\def\XINT_smalldiv_f 1#1#2#3#4#5#6!#7\xint:#8!%
{%
     \xint_gob_til_zero #1\XINT_smalldiv_fz 0%
     \expandafter\XINT_smalldiv_g
     \the\numexpr\XINT_minimul_a #2#3#4#5\xint:#6!#8!#2#3#4#5#6!#7\xint:#8!%
}%
\def\XINT_smalldiv_fz 0%
    \expandafter\XINT_smalldiv_g\the\numexpr\XINT_minimul_a
    9999\xint:9999!#1!99999999!#2!0!1#3!%
{%
    \XINT_smalldiv_i \xint:#3!\xint_c_!#2!%
}%
\def\XINT_smalldiv_g 1#1!1#2!#3!#4!#5!#6!%
{%
    \expandafter\XINT_smalldiv_h\the\numexpr 1#6-#1\xint:#2!#5!#3!#4!%
}%
\def\XINT_smalldiv_h 1#1#2\xint:#3!#4!%
{%
    \expandafter\XINT_smalldiv_i\the\numexpr #4-#3+#1-\xint_c_i\xint:#2!%
}%
\def\XINT_smalldiv_i #1\xint:#2!#3!#4\xint:#5!%
{%
    \expandafter\XINT_smalldiv_j\the\numexpr (#1#2+#4)/#5-\xint_c_i!#3!#1#2!#4\xint:#5!%
}%
\def\XINT_smalldiv_j #1!#2!%
{%
    \xint_c_x^viii+#1+#2\expandafter!\the\numexpr\XINT_smalldiv_k
    #1!%
}%
%    \end{macrocode}
% \lverb|On boucle vers \XINT_smalldiv_d.|
%    \begin{macrocode}
\def\XINT_smalldiv_k #1!#2!#3\xint:#4!%
{%
    \expandafter\XINT_smalldiv_d\the\numexpr #2-#1*#4!#3\xint:#4!%
}%
%    \end{macrocode}
% \lverb|Cette routine fait la division euclidienne d'un nombre de seize
% chiffres par #1 = C = diviseur sur huit chiffres >= 10^7, avec #2 = sa
% moitié utilisée dans \numexpr pour contrebalancer l'arrondi
% (ARRRRRRGGGGGHHHH) fait par /. Le nombre divisé XY = X*10^8+Y se présente
% sous la forme 1<8chiffres>!1<8chiffres>! avec plus significatif en premier.
%
% Seul le quotient est calculé, pas le reste. En effet la routine de division
% principale va utiliser ce quotient pour déterminer le "grand" reste, et le
% petit reste ici ne nous serait d'à peu près aucune utilité.
%
% ATTENTION UNIQUEMENT UTILISÉ POUR DES SITUATIONS OÙ IL EST GARANTI QUE X <
% C ! (et C au moins 10^7) le quotient euclidien de X*10^8+Y par C sera donc <
% 10^8. Il sera renvoyé sous la forme 1<8chiffres>.|
%    \begin{macrocode}
\def\XINT_div_mini #1\xint:#2!1#3!%
{%
    \expandafter\XINT_div_mini_a\the\numexpr
    \xint_c_xi_e_viii_mone+#3*\xint_c_x^viii/#1!#1\xint:#2!#3!%
}%
%    \end{macrocode}
% \lverb|Note (2015/10/08). Attention à la différence dans l'ordre des
% arguments avec ce que je vois en dans \XINT_smalldiv_f. Je ne me souviens
% plus du tout s'il y a une raison quelconque.|
%    \begin{macrocode}
\def\XINT_div_mini_a 1#1#2#3#4#5#6!#7\xint:#8!%
{%
     \xint_gob_til_zero #1\XINT_div_mini_w 0%
     \expandafter\XINT_div_mini_b
     \the\numexpr\XINT_minimul_a #2#3#4#5\xint:#6!#7!#2#3#4#5#6!#7\xint:#8!%
}%
\def\XINT_div_mini_w 0%
    \expandafter\XINT_div_mini_b\the\numexpr\XINT_minimul_a
    9999\xint:9999!#1!99999999!#2\xint:#3!00000000!#4!%
{%
    \xint_c_x^viii_mone+(#4+#3)/#2!%
}%
\def\XINT_div_mini_b 1#1!1#2!#3!#4!#5!#6!%
{%
    \expandafter\XINT_div_mini_c
    \the\numexpr 1#6-#1\xint:#2!#5!#3!#4!%
}%
\def\XINT_div_mini_c 1#1#2\xint:#3!#4!%
{%
    \expandafter\XINT_div_mini_d
    \the\numexpr #4-#3+#1-\xint_c_i\xint:#2!%
}%
\def\XINT_div_mini_d #1\xint:#2!#3!#4\xint:#5!%
{%
    \xint_c_x^viii_mone+#3+(#1#2+#5)/#4!%
}%
%    \end{macrocode}
% \subsection*{Derived arithmetic}
% \addcontentsline{toc}{subsection}{Derived arithmetic}
% \subsection{\csh{xintiiQuo}, \csh{xintiiRem}}
%    \begin{macrocode}
\def\xintiiQuo {\romannumeral0\xintiiquo }%
\def\xintiiRem {\romannumeral0\xintiirem }%
\def\xintiiquo
   {\expandafter\xint_stop_atfirstoftwo\romannumeral0\xintiidivision }%
\def\xintiirem
   {\expandafter\xint_stop_atsecondoftwo\romannumeral0\xintiidivision }%
%    \end{macrocode}
% \subsection{\csh{xintiiDivRound}}
% \lverb|1.1, transferred from first release of bnumexpr. Rewritten for 1.2.
% Ending rewritten for 1.2i. (new \xintDSRr).
%
% 1.2l: \xintiiDivRound made robust against non terminated input.|
%    \begin{macrocode}
\def\xintiiDivRound   {\romannumeral0\xintiidivround }%
\def\xintiidivround #1{\expandafter\XINT_iidivround\romannumeral`&&@#1\xint:}%
\def\XINT_iidivround #1#2\xint:#3%
    {\expandafter\XINT_iidivround_a\expandafter #1\romannumeral`&&@#3\xint:#2\xint:}%
\def\XINT_iidivround_a #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_iidivround_divbyzero#1#2}\fi
    \if0#1\xint_dothis\XINT_iidivround_aiszero\fi
    \if-#2\xint_dothis{\XINT_iidivround_bneg #1}\fi
          \xint_orthat{\XINT_iidivround_bpos #1#2}%
}%
\def\XINT_iidivround_divbyzero #1#2#3\xint:#4\xint:
   {\XINT_signalcondition{DivisionByZero}{Division by zero: #1#4/#2#3.}{}{ 0}}%
\def\XINT_iidivround_aiszero   #1\xint:#2\xint:{ 0}%
\def\XINT_iidivround_bpos #1%
{%
    \xint_UDsignfork
            #1{\xintiiopp\XINT_iidivround_pos {}}%
             -{\XINT_iidivround_pos #1}%
    \krof
}%
\def\XINT_iidivround_bneg #1%
{%
    \xint_UDsignfork
            #1{\XINT_iidivround_pos {}}%
             -{\xintiiopp\XINT_iidivround_pos #1}%
    \krof
}%
\def\XINT_iidivround_pos #1#2\xint:#3\xint:
{%
    \expandafter\expandafter\expandafter\XINT_dsrr
    \expandafter\xint_firstoftwo
    \romannumeral0\XINT_div_prepare {#2}{#1#30}%
    \xint_bye\xint_Bye3456789\xint_bye/\xint_c_x\relax
}%
%    \end{macrocode}
% \subsection{\csh{xintiiDivTrunc}}
% \lverb|1.2l: \xintiiDivTrunc made robust against non terminated input.|
%    \begin{macrocode}
\def\xintiiDivTrunc   {\romannumeral0\xintiidivtrunc }%
\def\xintiidivtrunc #1{\expandafter\XINT_iidivtrunc\romannumeral`&&@#1\xint:}%
\def\XINT_iidivtrunc #1#2\xint:#3{\expandafter\XINT_iidivtrunc_a\expandafter #1%
                             \romannumeral`&&@#3\xint:#2\xint:}%
\def\XINT_iidivtrunc_a #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_iidivtrunc_divbyzero#1#2}\fi
    \if0#1\xint_dothis\XINT_iidivtrunc_aiszero\fi
    \if-#2\xint_dothis{\XINT_iidivtrunc_bneg #1}\fi
          \xint_orthat{\XINT_iidivtrunc_bpos #1#2}%
}%
%    \end{macrocode}
% \lverb|Attention to not move DivRound code beyond that point.|
%    \begin{macrocode}
\let\XINT_iidivtrunc_divbyzero\XINT_iidivround_divbyzero
\let\XINT_iidivtrunc_aiszero  \XINT_iidivround_aiszero
\def\XINT_iidivtrunc_bpos #1%
{%
    \xint_UDsignfork
            #1{\xintiiopp\XINT_iidivtrunc_pos {}}%
             -{\XINT_iidivtrunc_pos #1}%
    \krof
}%
\def\XINT_iidivtrunc_bneg #1%
{%
    \xint_UDsignfork
            #1{\XINT_iidivtrunc_pos {}}%
             -{\xintiiopp\XINT_iidivtrunc_pos #1}%
    \krof
}%
\def\XINT_iidivtrunc_pos #1#2\xint:#3\xint:
    {\expandafter\xint_stop_atfirstoftwo
     \romannumeral0\XINT_div_prepare {#2}{#1#3}}%
%    \end{macrocode}
% \subsection{\csh{xintiiModTrunc}}
% \lverb|Renamed from \xintiiMod to \xintiiModTrunc at 1.2p.|
%    \begin{macrocode}
\def\xintiiModTrunc {\romannumeral0\xintiimodtrunc }%
\def\xintiimodtrunc #1{\expandafter\XINT_iimodtrunc\romannumeral`&&@#1\xint:}%
\def\XINT_iimodtrunc #1#2\xint:#3{\expandafter\XINT_iimodtrunc_a\expandafter #1%
                             \romannumeral`&&@#3\xint:#2\xint:}%
\def\XINT_iimodtrunc_a #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_iimodtrunc_divbyzero#1#2}\fi
    \if0#1\xint_dothis\XINT_iimodtrunc_aiszero\fi
    \if-#2\xint_dothis{\XINT_iimodtrunc_bneg #1}\fi
          \xint_orthat{\XINT_iimodtrunc_bpos #1#2}%
}%
%    \end{macrocode}
% \lverb|Attention to not move DivRound code beyond that point. A bit of abuse
% here for divbyzero defaulted-to value, which happily works in both.|
%    \begin{macrocode}
\let\XINT_iimodtrunc_divbyzero\XINT_iidivround_divbyzero
\let\XINT_iimodtrunc_aiszero  \XINT_iidivround_aiszero
\def\XINT_iimodtrunc_bpos #1%
{%
    \xint_UDsignfork
            #1{\xintiiopp\XINT_iimodtrunc_pos {}}%
             -{\XINT_iimodtrunc_pos #1}%
    \krof
}%
\def\XINT_iimodtrunc_bneg #1%
{%
    \xint_UDsignfork
            #1{\xintiiopp\XINT_iimodtrunc_pos {}}%
             -{\XINT_iimodtrunc_pos #1}%
    \krof
}%
\def\XINT_iimodtrunc_pos #1#2\xint:#3\xint:
    {\expandafter\xint_stop_atsecondoftwo\romannumeral0\XINT_div_prepare
      {#2}{#1#3}}%
%    \end{macrocode}
% \subsection{\csh{xintiiDivMod}}
% \changed{1.2p}
% It is associated with floored division (like Python divmod
% function), and with the |//| operator in \csbxint{iiexpr}.
%    \begin{macrocode}
\def\xintiiDivMod   {\romannumeral0\xintiidivmod }%
\def\xintiidivmod #1{\expandafter\XINT_iidivmod\romannumeral`&&@#1\xint:}%
\def\XINT_iidivmod #1#2\xint:#3{\expandafter\XINT_iidivmod_a\expandafter #1%
                             \romannumeral`&&@#3\xint:#2\xint:}%
\def\XINT_iidivmod_a #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_iidivmod_divbyzero#1#2}\fi
    \if0#1\xint_dothis\XINT_iidivmod_aiszero\fi
    \if-#2\xint_dothis{\XINT_iidivmod_bneg #1}\fi
          \xint_orthat{\XINT_iidivmod_bpos #1#2}%
}%
\def\XINT_iidivmod_divbyzero #1#2\xint:#3\xint:
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #1#3/#2.}{}%
    {{0}{0}}% à revoir...
}%
\def\XINT_iidivmod_aiszero #1\xint:#2\xint:{{0}{0}}%
\def\XINT_iidivmod_bneg #1%
{%
    \expandafter\XINT_iidivmod_bneg_finish
    \romannumeral0\xint_UDsignfork
            #1{\XINT_iidivmod_bpos {}}%
             -{\XINT_iidivmod_bpos {-#1}}%
    \krof
}%
\def\XINT_iidivmod_bneg_finish#1#2%
{%
    \expandafter\xint_exchangetwo_keepbraces\expandafter
    {\romannumeral0\xintiiopp#2}{#1}%
}%
\def\XINT_iidivmod_bpos #1#2\xint:#3\xint:{\xintiidivision{#1#3}{#2}}%
%    \end{macrocode}
% \subsection{\csh{xintiiDivFloor}}
% \lverb|1.2p. For bnumexpr actually, because \xintiiexpr could use
% \xintDivFloor which also outputs an integer in strict format.|
%    \begin{macrocode}
\def\xintiiDivFloor {\romannumeral0\xintiidivfloor}%
\def\xintiidivfloor {\expandafter\xint_stop_atfirstoftwo
                \romannumeral0\xintiidivmod}%
%    \end{macrocode}
% \subsection{\csh{xintiiMod}}
% \lverb|Associated with floored division at 1.2p. Formerly was associated with
% truncated division.|
%    \begin{macrocode}
\def\xintiiMod {\romannumeral0\xintiimod}%
\def\xintiimod {\expandafter\xint_stop_atsecondoftwo
                \romannumeral0\xintiidivmod}%
%    \end{macrocode}
% \subsection{\csh{xintiiSqr}}
% \lverb|1.2l: \xintiiSqr made robust against non terminated input.|
%    \begin{macrocode}
\def\xintiiSqr {\romannumeral0\xintiisqr }%
\def\xintiisqr #1%
{%
    \expandafter\XINT_sqr\romannumeral0\xintiiabs{#1}\xint:
}%
\def\XINT_sqr #1\xint:
{%
    \expandafter\XINT_sqr_a
      \romannumeral0\expandafter\XINT_sepandrev_andcount
      \romannumeral0\XINT_zeroes_forviii #1\R\R\R\R\R\R\R\R{10}0000001\W
      #1\XINT_rsepbyviii_end_A 2345678%
        \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
      \xint:
}%
%    \end{macrocode}
% \lverb|1.2c \XINT_mul_loop can now be called directly even with small
% arguments, thus the following check is not anymore a necessity.|
%    \begin{macrocode}
\def\XINT_sqr_a #1\xint:
{%
    \ifnum #1=\xint_c_i \expandafter\XINT_sqr_small
                   \else\expandafter\XINT_sqr_start\fi
}%
\def\XINT_sqr_small 1#1#2#3#4#5!\xint:
{%
    \ifnum #1#2#3#4#5<46341 \expandafter\XINT_sqr_verysmall\fi
    \expandafter\XINT_sqr_small_out
    \the\numexpr\XINT_minimul_a #1#2#3#4\xint:#5!#1#2#3#4#5!%
}%
\def\XINT_sqr_verysmall#1{%
\def\XINT_sqr_verysmall
    \expandafter\XINT_sqr_small_out\the\numexpr\XINT_minimul_a ##1!##2!%
    {\expandafter#1\the\numexpr ##2*##2\relax}%
}\XINT_sqr_verysmall{ }%
\def\XINT_sqr_small_out 1#1!1#2!%
{%
    \XINT_cuz #2#1\R
}%
%    \end{macrocode}
% \lverb|An ending 1;! is produced on output for \XINT_mul_loop and gets
% incorporated to the delimiter needed by the \XINT_unrevbyviii done by
% \XINT_mul_out.|
%    \begin{macrocode}
\def\XINT_sqr_start #1\xint:
{%
    \expandafter\XINT_mul_out
    \the\numexpr\XINT_mul_loop
                100000000!1;!\W #11;!\W #11;!%
    1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
%    \end{macrocode}
% \subsection{\csh{xintiiPow}}
% \lverb|&
% The exponent is not limited but with current default settings of tex memory,
% with xint 1.2, the maximal exponent for 2^N is N = 2^17 = 131072.
%
% 1.2f Modifies the initial steps: 1) in order to be able to let more easily
% \xintiPow use \xintNum on the exponent once xintfrac.sty is loaded; 2) also
% because I noticed it was not very well coded. And it did only a \numexpr on
% the exponent, contradicting the documentation related to the "i" convention
% in names.
% 
% 1.2l: \xintiiPow made robust against non terminated input.|
%    \begin{macrocode}
\def\xintiiPow {\romannumeral0\xintiipow }%
\def\xintiipow #1#2%
{%
    \expandafter\xint_pow\the\numexpr #2\expandafter
    .\romannumeral`&&@#1\xint:
}%
\def\xint_pow #1.#2%#3\xint:
{%
    \xint_UDzerominusfork
      #2-\XINT_pow_AisZero
      0#2\XINT_pow_Aneg
      0-{\XINT_pow_Apos #2}%
    \krof {#1}%
}%
\def\XINT_pow_AisZero #1#2\xint:
{%
     \ifcase\XINT_cntSgn #1\xint:
         \xint_afterfi { 1}%
     \or
         \xint_afterfi { 0}%
     \else
         \xint_afterfi
        {\XINT_signalcondition{DivisionByZero}{0 raised to power #1.}{}{ 0}}%
     \fi
}%
\def\XINT_pow_Aneg #1%
{%
   \ifodd #1
       \expandafter\XINT_opp\romannumeral0%
   \fi
   \XINT_pow_Apos {}{#1}%
}%
\def\XINT_pow_Apos #1#2{\XINT_pow_Apos_a {#2}#1}%
\def\XINT_pow_Apos_a #1#2#3%
{%
    \xint_gob_til_xint: #3\XINT_pow_Apos_short\xint:
    \XINT_pow_AatleastTwo {#1}#2#3%
}%
\def\XINT_pow_Apos_short\xint:\XINT_pow_AatleastTwo #1#2\xint:
{%
    \ifcase #2
         \xintError:thiscannothappen
    \or  \expandafter\XINT_pow_AisOne
    \else\expandafter\XINT_pow_AatleastTwo
    \fi {#1}#2\xint:
}%
\def\XINT_pow_AisOne #1\xint:{ 1}%
\def\XINT_pow_AatleastTwo #1%
{%
    \ifcase\XINT_cntSgn #1\xint:
        \expandafter\XINT_pow_BisZero
    \or
        \expandafter\XINT_pow_I_in
    \else
        \expandafter\XINT_pow_BisNegative
    \fi
    {#1}%
}%
\def\XINT_pow_BisNegative #1\xint:{\XINT_signalcondition{Underflow}%
    {Inverse power is not an integer.}{}{ 0}}%
\def\XINT_pow_BisZero #1\xint:{ 1}%
%    \end{macrocode}
% \lverb|B = #1 > 0, A = #2 > 1. Earlier code checked if size of B did not
% exceed a given limit (for example 131000).|
%    \begin{macrocode}
\def\XINT_pow_I_in #1#2\xint:
{%
    \expandafter\XINT_pow_I_loop
    \the\numexpr #1\expandafter\xint:%
    \romannumeral0\expandafter\XINT_sepandrev
    \romannumeral0\XINT_zeroes_forviii #2\R\R\R\R\R\R\R\R{10}0000001\W
    #2\XINT_rsepbyviii_end_A 2345678%
      \XINT_rsepbyviii_end_B 2345678\relax XX%
    \R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\R\xint:\W
    1;!\W
    1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
\def\XINT_pow_I_loop #1\xint:%
{%
    \ifnum #1 = \xint_c_i\expandafter\XINT_pow_I_exit\fi
    \ifodd #1
       \expandafter\XINT_pow_II_in
    \else
       \expandafter\XINT_pow_I_squareit
    \fi #1\xint:%
}%
\def\XINT_pow_I_exit \ifodd #1\fi #2\xint:#3\W {\XINT_mul_out #3}%
%    \end{macrocode}
% \lverb|The 1.2c \XINT_mul_loop can be called directly even with small
% arguments, hence the "butcheckifsmall" is not a necessity as it was earlier
% with 1.2. On 2^30, it does bring roughly a 40$char37 $space time gain
% though, and 30$char37 $space gain for 2^60. The overhead on big computations
% should be negligible.|
%    \begin{macrocode}
\def\XINT_pow_I_squareit #1\xint:#2\W%
{%
    \expandafter\XINT_pow_I_loop
    \the\numexpr #1/\xint_c_ii\expandafter\xint:%
    \the\numexpr\XINT_pow_mulbutcheckifsmall #2\W #2\W
}%
\def\XINT_pow_mulbutcheckifsmall #1!1#2%
{%
    \xint_gob_til_sc #2\XINT_pow_mul_small;%
    \XINT_mul_loop 100000000!1;!\W #1!1#2%
}%
\def\XINT_pow_mul_small;\XINT_mul_loop
    100000000!1;!\W 1#1!1;!\W
{%
    \XINT_smallmul 1#1!%
}%
\def\XINT_pow_II_in #1\xint:#2\W
{%
    \expandafter\XINT_pow_II_loop
    \the\numexpr #1/\xint_c_ii-\xint_c_i\expandafter\xint:%
    \the\numexpr\XINT_pow_mulbutcheckifsmall #2\W #2\W #2\W
}%
\def\XINT_pow_II_loop #1\xint:%
{%
    \ifnum #1 = \xint_c_i\expandafter\XINT_pow_II_exit\fi
    \ifodd #1
       \expandafter\XINT_pow_II_odda
    \else
       \expandafter\XINT_pow_II_even
    \fi #1\xint:%
}%
\def\XINT_pow_II_exit\ifodd #1\fi #2\xint:#3\W #4\W
{%
    \expandafter\XINT_mul_out
    \the\numexpr\XINT_pow_mulbutcheckifsmall #4\W #3%
}%
\def\XINT_pow_II_even #1\xint:#2\W
{%
    \expandafter\XINT_pow_II_loop
    \the\numexpr #1/\xint_c_ii\expandafter\xint:%
    \the\numexpr\XINT_pow_mulbutcheckifsmall #2\W #2\W
}%
\def\XINT_pow_II_odda #1\xint:#2\W #3\W
{%
    \expandafter\XINT_pow_II_oddb
    \the\numexpr #1/\xint_c_ii-\xint_c_i\expandafter\xint:%
    \the\numexpr\XINT_pow_mulbutcheckifsmall #3\W #2\W #2\W
}%
\def\XINT_pow_II_oddb #1\xint:#2\W #3\W
{%
    \expandafter\XINT_pow_II_loop
    \the\numexpr #1\expandafter\xint:%
    \the\numexpr\XINT_pow_mulbutcheckifsmall #3\W #3\W #2\W
}%
%    \end{macrocode}
% \subsection{\csh{xintiiFac}}
% \lverb|Moved here from xint.sty with release 1.2 (to be usable by \bnumexpr).
%
% An \xintiFac is needed by xintexpr.sty. Prior to 1.2o it was defined here
% as an alias to \xintiiFac, then redefined by xintfrac to use \xintNum. This
% was incoherent. Contrarily to other similarly named macros,
% \xintiiFac uses \numexpr on its input. This is also incoherent with the
% naming scheme, alas.
%
% Partially rewritten with release 1.2 to benefit from the inner format of the
% 1.2 multiplication.
% 
% With current default settings of the etex memory and a.t.t.o.w (11/2015) the
% maximal possible computation is 5971! (which has 19956 digits).
%
%
%
% Note (end november 2015): I also tried out a quickly written recursive
% (binary split) implementation
%
%( \catcode`_ 11
%: \catcode`^ 11
%: \long\def\xint_firstofthree  #1#2#3{#1}$%
%: \long\def\xint_secondofthree #1#2#3{#2}$%
%: \long\def\xint_thirdofthree  #1#2#3{#3}$%
%: $% quickly written factorial using binary split recursive method
%: \def\tFac   {\romannumeral-`0\tfac }$%
%: \def\tfac #1{\expandafter\XINT_mul_out
%:              \romannumeral-`0\ufac {1}{#1}1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W}$%
%: \def\ufac #1#2{\ifcase\numexpr#2-#1\relax
%:                  \expandafter\xint_firstofthree
%:                \or
%:                  \expandafter\xint_secondofthree
%:                \else
%:                  \expandafter\xint_thirdofthree
%:                \fi
%:                {\the\numexpr\xint_c_x^viii+#1!1;!}$%
%:                {\the\numexpr\xint_c_x^viii+#1*#2!1;!}$%
%:                {\expandafter\vfac\the\numexpr (#1+#2)/\xint_c_ii.#1.#2.}$%
%: }$%
%: \def\vfac #1.#2.#3.$%
%: {$%
%:     \expandafter
%:     \wfac\expandafter
%:         {\romannumeral-`0\expandafter
%:          \ufac\expandafter{\the\numexpr #1+\xint_c_i}{#3}}$%
%:         {\ufac {#2}{#1}}$%
%: }$%
%: \def\wfac #1#2{\expandafter\zfac\romannumeral-`0#2\W #1}$%
%: \def\zfac {\the\numexpr\XINT_mul_loop 100000000!1;!\W }$% core multiplication...
%: \catcode`_ 8
%: \catcode`^ 7
%)
% and I was quite surprised that it was only about 1.6x--2x slower in the range
% N=200 to 2000 than the \xintiiFac here which attempts to be smarter...
%
% Note (2017, 1.2l): I found out some code comment of mine that the code here
% should be more in the style of \xintiiBinomial, but I left matters
% untouched.
%
%
%
% 1.2o modifies \xintiFac to be coherent with \xintiBinomial: only with
% xintfrac.sty loaded does it use \xintNum. It is documented only as macro of
% xintfrac.sty, not as macro of xint.sty.
% |
%    \begin{macrocode}
\def\xintiiFac {\romannumeral0\xintiifac }%
\def\xintiifac #1{\expandafter\XINT_fac_fork\the\numexpr#1.}%
\def\XINT_fac_fork #1#2.%
{%
    \xint_UDzerominusfork
     #1-\XINT_fac_zero
     0#1\XINT_fac_neg
      0-\XINT_fac_checksize
    \krof #1#2.%
}%
\def\XINT_fac_zero #1.{ 1}%
\def\XINT_fac_neg  #1.{\XINT_signalcondition{InvalidOperation}{Factorial of
    negative argument: #1.}{}{ 0}}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_fac_checksize #1.%
{%
    \ifnum #1>\xint_c_x^iv \xint_dothis{\XINT_fac_toobig #1.}\fi
    \ifnum #1>465 \xint_dothis{\XINT_fac_bigloop_a   #1.}\fi
    \ifnum #1>101 \xint_dothis{\XINT_fac_medloop_a   #1.\XINT_mul_out}\fi
                  \xint_orthat{\XINT_fac_smallloop_a #1.\XINT_mul_out}%
    1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
\def\XINT_fac_toobig
#1.#2\W{\XINT_signalcondition{InvalidOperation}{Factorial
    argument is too large: #1 > 10^4.}{}{ 0}}%
\def\XINT_fac_bigloop_a #1.%
{%
    \expandafter\XINT_fac_bigloop_b \the\numexpr
    #1+\xint_c_i-\xint_c_ii*((#1-464)/\xint_c_ii).#1.%
}%
\def\XINT_fac_bigloop_b #1.#2.%
{%
    \expandafter\XINT_fac_medloop_a
        \the\numexpr #1-\xint_c_i.{\XINT_fac_bigloop_loop #1.#2.}%
}%
\def\XINT_fac_bigloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_fac_bigloop_exit\fi
    \expandafter\XINT_fac_bigloop_loop
    \the\numexpr #1+\xint_c_ii\expandafter.%
    \the\numexpr #2\expandafter.\the\numexpr\XINT_fac_bigloop_mul #1!%
}%
\def\XINT_fac_bigloop_exit #1!{\XINT_mul_out}%
\def\XINT_fac_bigloop_mul #1!%
{%
    \expandafter\XINT_smallmul
        \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)!%
}%
\def\XINT_fac_medloop_a #1.%
{%
    \expandafter\XINT_fac_medloop_b
        \the\numexpr #1+\xint_c_i-\xint_c_iii*((#1-100)/\xint_c_iii).#1.%
}%
\def\XINT_fac_medloop_b #1.#2.%
{%
    \expandafter\XINT_fac_smallloop_a
        \the\numexpr #1-\xint_c_i.{\XINT_fac_medloop_loop #1.#2.}%
}%
\def\XINT_fac_medloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_fac_loop_exit\fi
    \expandafter\XINT_fac_medloop_loop
    \the\numexpr #1+\xint_c_iii\expandafter.%
    \the\numexpr #2\expandafter.\the\numexpr\XINT_fac_medloop_mul #1!%
}%
\def\XINT_fac_medloop_mul #1!%
{%
    \expandafter\XINT_smallmul
    \the\numexpr
        \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_fac_smallloop_a #1.%
{%
    \csname
       XINT_fac_smallloop_\the\numexpr #1-\xint_c_iv*(#1/\xint_c_iv)\relax
    \endcsname #1.%
}%
\expandafter\def\csname XINT_fac_smallloop_1\endcsname #1.%
{%
    \XINT_fac_smallloop_loop 2.#1.100000001!1;!%
}%
\expandafter\def\csname XINT_fac_smallloop_-2\endcsname #1.%
{%
    \XINT_fac_smallloop_loop 3.#1.100000002!1;!%
}%
\expandafter\def\csname XINT_fac_smallloop_-1\endcsname #1.%
{%
    \XINT_fac_smallloop_loop 4.#1.100000006!1;!%
}%
\expandafter\def\csname XINT_fac_smallloop_0\endcsname #1.%
{%
    \XINT_fac_smallloop_loop 5.#1.1000000024!1;!%
}%
\def\XINT_fac_smallloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_fac_loop_exit\fi
    \expandafter\XINT_fac_smallloop_loop
    \the\numexpr #1+\xint_c_iv\expandafter.%
    \the\numexpr #2\expandafter.\the\numexpr\XINT_fac_smallloop_mul #1!%
}%
\def\XINT_fac_smallloop_mul #1!%
{%
    \expandafter\XINT_smallmul
    \the\numexpr
        \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_fac_loop_exit #1!#2;!#3{#3#2;!}%
%    \end{macrocode}
% \subsection{\csh{XINT_useiimessage}}
% \lverb|1.2o|
%    \begin{macrocode}
\def\XINT_useiimessage #1% used in LaTeX only
{%
    \XINT_ifFlagRaised {#1}%
    {\@backslashchar#1
     (load xintfrac or use \@backslashchar xintii\xint_gobble_iv#1!)\MessageBreak}%
    {}%
}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintcore}
% \cleardoublepage\let\xintcorenameUp\undefined
%\gardesactifs
%\let</xintcore>\relax
%\let<*xint>\gardesinactifs
%</xintcore>^^A---------------------------------------------------
%<*xint>^^A-------------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintnameUp\endcsname
% \section{Package \xintnameimp implementation}
% \RaisedLabel{sec:xintimp}
%
% \localtableofcontents
%
% With release |1.1| the core arithmetic routines |\xintiiAdd|,
% |\xintiiSub|, |\xintiiMul|, |\xintiiQuo|, |\xintiiPow| were separated to be
% the main component of the then new
% \xintcorenameimp.
%
% At |1.3| the macros deprecated at |1.2o| got all removed.
%
% |1.3b| adds randomness related macros.
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xint.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintcore.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xint}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintcore.sty
      \ifx\w\relax % but xintkernel.sty not yet loaded.
         \def\z{\endgroup\input xintcore.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintcore.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xintcore}}%
          \fi
      \else
        \aftergroup\endinput % xint already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty (loaded by xintcore.sty)
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xint}%
  [2021/07/13 v1.4j Expandable operations on big integers (JFB)]%
%    \end{macrocode}
% \subsection{More token management}
%    \begin{macrocode}
\long\def\xint_firstofthree  #1#2#3{#1}%
\long\def\xint_secondofthree #1#2#3{#2}%
\long\def\xint_thirdofthree  #1#2#3{#3}%
\long\def\xint_stop_atfirstofthree  #1#2#3{ #1}%
\long\def\xint_stop_atsecondofthree #1#2#3{ #2}%
\long\def\xint_stop_atthirdofthree  #1#2#3{ #3}%
%    \end{macrocode}
% \subsection{(WIP) A constant needed by \cshnolabel{xintRandomDigits} et al.}
%    \begin{macrocode}
\ifdefined\xint_texuniformdeviate
  \unless\ifdefined\xint_c_nine_x^viii
      \csname newcount\endcsname\xint_c_nine_x^viii
      \xint_c_nine_x^viii 900000000
  \fi
\fi
%    \end{macrocode}
% \subsection{\csh{xintLen}, \csh{xintiLen}}
% \lverb|\xintLen gets extended to fractions by xintfrac.sty: A/B is given
% length len(A)+len(B)-1 (somewhat arbitrary). It applies \xintNum to its
% argument. A minus sign is accepted and ignored.
%
%
% For parallelism with \xintiNum/\xintNum, 1.2o defines \xintiLen.
%
% \xintLen gets redefined by $xintfracnameimp.
% |
%    \begin{macrocode}
\def\xintiLen {\romannumeral0\xintilen }%
\def\xintilen #1{\def\xintilen ##1%
{%
    \expandafter#1\the\numexpr
    \expandafter\XINT_len_fork\romannumeral0\xintinum{##1}%
      \xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye\relax
}}\xintilen{ }%
\def\xintLen {\romannumeral0\xintlen }%
\let\xintlen\xintilen
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_len_fork #1%
{%
    \expandafter\XINT_length_loop\xint_UDsignfork#1{}-#1\krof
}%
%    \end{macrocode}
% \subsection{\csh{xintiiLogTen}}
% \lverb|1.3e. Support for ilog10() function in \xintiiexpr. See \XINTiLogTen
% in xintfrac.sty which also currently uses -"7FFF8000 as value if input is
% zero.|
%    \begin{macrocode}
\def\xintiiLogTen {\the\numexpr\xintiilogten }%
\def\xintiilogten #1%
{%
    \expandafter\XINT_iilogten\romannumeral`&&@#1%
      \xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    \relax
}%
\def\XINT_iilogten #1{\if#10-"7FFF8000\fi -1+%
                      \expandafter\XINT_length_loop\xint_UDsignfork#1{}-#1\krof}%
%    \end{macrocode}
% \subsection{\csh{xintReverseDigits}}
% \lverb|&
% 1.2.
%
% This puts digits in reverse order, not suppressing leading zeros
% after reverse. Despite lacking the "ii" in its name, it does not apply
% \xintNum to its argument (contrarily to \xintLen, this is not very coherent).
%
% 1.2l variant is robust against non terminated \the\numexpr input.
%
% This macro is currently not used elsewhere in xint code.
% |
%    \begin{macrocode}
\def\xintReverseDigits {\romannumeral0\xintreversedigits }%
\def\xintreversedigits #1%
{%
    \expandafter\XINT_revdigits\romannumeral`&&@#1%
     {\XINT_microrevsep_end\W}\XINT_microrevsep_end
      \XINT_microrevsep_end\XINT_microrevsep_end
      \XINT_microrevsep_end\XINT_microrevsep_end
      \XINT_microrevsep_end\XINT_microrevsep_end\XINT_microrevsep_end\Z
    1\Z!1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
\def\XINT_revdigits #1%
{%
    \xint_UDsignfork
      #1{\expandafter-\romannumeral0\XINT_revdigits_a}%
       -{\XINT_revdigits_a #1}%
    \krof
}%
\def\XINT_revdigits_a
{%
    \expandafter\XINT_revdigits_b\expandafter{\expandafter}%
    \the\numexpr\XINT_microrevsep
}%
\def\XINT_microrevsep #1#2#3#4#5#6#7#8#9%
{%
    1#9#8#7#6#5#4#3#2#1\expandafter!\the\numexpr\XINT_microrevsep
}%
\def\XINT_microrevsep_end #1\W #2\expandafter #3\Z{\relax#2!}%
\def\XINT_revdigits_b #11#2!1#3!1#4!1#5!1#6!1#7!1#8!1#9!%
{%
    \xint_gob_til_R #9\XINT_revdigits_end\R
                      \XINT_revdigits_b {#9#8#7#6#5#4#3#2#1}%
}%
\def\XINT_revdigits_end#1{%
\def\XINT_revdigits_end\R\XINT_revdigits_b ##1##2\W
   {\expandafter#1\xint_gob_til_Z ##1}%
}\XINT_revdigits_end{ }%
\let\xintRev\xintReverseDigits
%    \end{macrocode}
% \subsection{\csh{xintiiE}}
% \lverb|Originally was used in \xintiiexpr. Transferred from xintfrac for
% 1.1.
% Code rewritten for 1.2i.
% \xintiiE{x}{e} extends x with e zeroes if e is positive and simply outputs
% x if e is zero or negative. Attention, le comportement pour e < 0 ne doit
% pas être modifié car \xintMod et autres macros en dépendent.
% |
%    \begin{macrocode}
\def\xintiiE {\romannumeral0\xintiie }%
\def\xintiie #1#2%
   {\expandafter\XINT_iie_fork\the\numexpr #2\expandafter.\romannumeral`&&@#1;}%
\def\XINT_iie_fork #1%
{%
    \xint_UDsignfork
      #1\XINT_iie_neg
       -\XINT_iie_a
    \krof #1%
}%
%    \end{macrocode}
% \lverb|&
% le #2 a le bon pattern terminé par ; #1=0 est OK pour \XINT_rep.
% |
%    \begin{macrocode}
\def\XINT_iie_a #1.%
 {\expandafter\XINT_dsx_append\romannumeral\XINT_rep #1\endcsname 0.}%
\def\XINT_iie_neg #1.#2;{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintDecSplit}}
% \lverb@DECIMAL SPLIT
%
% The macro \xintDecSplit {x}{A} cuts A which is composed of digits (leading
% zeroes ok, but no sign) (*) into two (each possibly empty) pieces L and R.
% The concatenation LR always reproduces A.
%
% The position of the cut is specified by the first argument x. If x is zero
% or positive the cut location is x slots to the left of the right end of the
% number. If x becomes equal to or larger than the length of the number then L
% becomes empty. If x is negative the location of the cut is |x| slots to the
% right of the left end of the number.
%
% (*) versions earlier than 1.2i first replaced A with its absolute value.
% This is not the case anymore. This macro should NOT be used for A with a
% leading sign (+ or -).
%
% Entirely rewritten for 1.2i (2016/12/11).
%
% Attention: \xintDecSplit not robust against non terminated second argument.
% @
%    \begin{macrocode}
\def\xintDecSplit {\romannumeral0\xintdecsplit }%
\def\xintdecsplit #1#2%
{%
    \expandafter\XINT_split_finish
    \romannumeral0\expandafter\XINT_split_xfork
    \the\numexpr #1\expandafter.\romannumeral`&&@#2%
    \xint_bye2345678\xint_bye..%
}%
\def\XINT_split_finish  #1.#2.{{#1}{#2}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_split_xfork #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_split_zerosplit
      0#1\XINT_split_fromleft
       0-{\XINT_split_fromright #1}%
    \krof
}%
\def\XINT_split_zerosplit .#1\xint_bye#2\xint_bye..{ #1..}%
\def\XINT_split_fromleft
    {\expandafter\XINT_split_fromleft_a\the\numexpr\xint_c_viii-}%
\def\XINT_split_fromleft_a #1%
{%
    \xint_UDsignfork
      #1\XINT_split_fromleft_b
       -{\XINT_split_fromleft_end_a #1}%
    \krof
}%
\def\XINT_split_fromleft_b #1.#2#3#4#5#6#7#8#9%
{%
    \expandafter\XINT_split_fromleft_clean
    \the\numexpr1#2#3#4#5#6#7#8#9\expandafter
    \XINT_split_fromleft_a\the\numexpr\xint_c_viii-#1.%
}%
\def\XINT_split_fromleft_end_a #1.%
{%
    \expandafter\XINT_split_fromleft_clean
    \the\numexpr1\csname XINT_split_fromleft_end#1\endcsname
}%
\def\XINT_split_fromleft_clean 1{ }%
\expandafter\def\csname XINT_split_fromleft_end7\endcsname #1%
   {#1\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end6\endcsname #1#2%
   {#1#2\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end5\endcsname #1#2#3%
   {#1#2#3\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end4\endcsname #1#2#3#4%
   {#1#2#3#4\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end3\endcsname #1#2#3#4#5%
   {#1#2#3#4#5\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end2\endcsname #1#2#3#4#5#6%
   {#1#2#3#4#5#6\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end1\endcsname #1#2#3#4#5#6#7%
   {#1#2#3#4#5#6#7\XINT_split_fromleft_end_b}%
\expandafter\def\csname XINT_split_fromleft_end0\endcsname #1#2#3#4#5#6#7#8%
   {#1#2#3#4#5#6#7#8\XINT_split_fromleft_end_b}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_split_fromleft_end_b #1\xint_bye#2\xint_bye.{.#1}% puis .
\def\XINT_split_fromright #1.#2\xint_bye
{%
    \expandafter\XINT_split_fromright_a
    \the\numexpr#1-\numexpr\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    .#2\xint_bye
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_split_fromright_a #1%
{%
    \xint_UDsignfork
      #1\XINT_split_fromleft
       -\XINT_split_fromright_Lempty
    \krof
}%
\def\XINT_split_fromright_Lempty #1.#2\xint_bye#3..{.#2.}%
%    \end{macrocode}
% \subsection{\csh{xintDecSplitL}}
%    \begin{macrocode}
\def\xintDecSplitL {\romannumeral0\xintdecsplitl }%
\def\xintdecsplitl #1#2%
{%
    \expandafter\XINT_splitl_finish
    \romannumeral0\expandafter\XINT_split_xfork
    \the\numexpr #1\expandafter.\romannumeral`&&@#2%
    \xint_bye2345678\xint_bye..%
}%
\def\XINT_splitl_finish #1.#2.{ #1}%
%    \end{macrocode}
% \subsection{\csh{xintDecSplitR}}
%    \begin{macrocode}
\def\xintDecSplitR {\romannumeral0\xintdecsplitr }%
\def\xintdecsplitr #1#2%
{%
    \expandafter\XINT_splitr_finish
    \romannumeral0\expandafter\XINT_split_xfork
    \the\numexpr #1\expandafter.\romannumeral`&&@#2%
    \xint_bye2345678\xint_bye..%
}%
\def\XINT_splitr_finish #1.#2.{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintDSHr}}
% \lverb@DECIMAL SHIFTS \xintDSH {x}{A}$\
% si x <= 0, fait A -> A.10^(|x|).
% si x >  0, et A >=0, fait A -> quo(A,10^(x))$\
% si x >  0, et A < 0, fait A -> -quo(-A,10^(x))$\
% (donc pour x > 0 c'est comme DSR itéré x fois)$\
% \xintDSHr donne le `reste' (si x<=0 donne zéro).
%
% Badly named macros.
% 
% Rewritten for 1.2i, this was old code and \xintDSx has changed interface.
% @
%    \begin{macrocode}
\def\xintDSHr {\romannumeral0\xintdshr }%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\xintdshr #1#2%
{%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \expandafter\XINT_dshr_fork\the\numexpr#1\expandafter.\romannumeral`&&@#2;%
}%
\def\XINT_dshr_fork #1%
{%
    \xint_UDzerominusfork
      0#1\XINT_dshr_xzeroorneg
      #1-\XINT_dshr_xzeroorneg
       0-\XINT_dshr_xpositive
    \krof #1%
}%
\def\XINT_dshr_xzeroorneg #1;{ 0}%
\def\XINT_dshr_xpositive
{%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \expandafter\xint_stop_atsecondoftwo\romannumeral0\XINT_dsx_xisPos
}%
%    \end{macrocode}
% \subsection{\csh{xintDSH}}
%    \begin{macrocode}
\def\xintDSH {\romannumeral0\xintdsh }%
\def\xintdsh #1#2%
{%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \expandafter\XINT_dsh_fork\the\numexpr#1\expandafter.\romannumeral`&&@#2;%
}%
\def\XINT_dsh_fork #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_dsh_xiszero
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
      0#1\XINT_dsx_xisNeg_checkA
       0-{\XINT_dsh_xisPos #1}%
    \krof
}%
\def\XINT_dsh_xiszero #1.#2;{ #2}%
\def\XINT_dsh_xisPos
{%
%    \end{macrocode}
% \lverb|&
    \expandafter\xint_stop_atfirstoftwo\romannumeral0\XINT_dsx_xisPos
% |
%    \begin{macrocode}
}%
%    \end{macrocode}
% \subsection{\csh{xintDSx}}
% \lverb@&
% --> Attention le cas x=0 est traité dans la même catégorie que x > 0 <--
%
%( si x < 0, fait A -> A.10^(|x|)
%: si x >=  0, et A >=0, fait A -> {quo(A,10^(x))}{rem(A,10^(x))}
%: si x >=  0, et A < 0, d'abord on calcule {quo(-A,10^(x))}{rem(-A,10^(x))}
%:    puis, si le premier n'est pas nul on lui donne le signe - 
%:          si le premier est nul on donne le signe - au second.
%)
% On peut donc toujours reconstituer l'original A par 10^x Q \pm R
% où il faut prendre le signe plus si Q est positif ou nul et le signe moins si
% Q est strictement négatif.
%
% Rewritten for 1.2i, this was old code.
%
% @
%    \begin{macrocode}
\def\xintDSx {\romannumeral0\xintdsx }%
\def\xintdsx #1#2%
{%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \expandafter\XINT_dsx_fork\the\numexpr#1\expandafter.\romannumeral`&&@#2;%
}%
\def\XINT_dsx_fork #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_dsx_xisZero
      0#1\XINT_dsx_xisNeg_checkA
       0-{\XINT_dsx_xisPos #1}%
    \krof
}%
\def\XINT_dsx_xisZero #1.#2;{{#2}{0}}%
\def\XINT_dsx_xisNeg_checkA #1.#2%
{%
    \xint_gob_til_zero #2\XINT_dsx_xisNeg_Azero 0%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \expandafter\XINT_dsx_append\romannumeral\XINT_rep #1\endcsname 0.#2%
}%
\def\XINT_dsx_xisNeg_Azero #1;{ 0}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_dsx_addzeros #1%
   {\expandafter\XINT_dsx_append\romannumeral\XINT_rep#1\endcsname0.}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_dsx_addzerosnofuss #1%
   {\expandafter\XINT_dsx_append\romannumeral\xintreplicate{#1}0.}%
\def\XINT_dsx_append #1.#2;{ #2#1}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_dsx_xisPos #1.#2%
{%
    \xint_UDzerominusfork
      #2-\XINT_dsx_AisZero
      0#2\XINT_dsx_AisNeg
       0-\XINT_dsx_AisPos
    \krof #1.#2%
}%
\def\XINT_dsx_AisZero #1;{{0}{0}}%
\def\XINT_dsx_AisNeg #1.-#2;%
{%
    \expandafter\XINT_dsx_AisNeg_checkiffirstempty
    \romannumeral0\XINT_split_xfork #1.#2\xint_bye2345678\xint_bye..%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_dsx_AisNeg_checkiffirstempty #1%
{%
    \xint_gob_til_dot #1\XINT_dsx_AisNeg_finish_zero.%
    \XINT_dsx_AisNeg_finish_notzero #1%
}%
\def\XINT_dsx_AisNeg_finish_zero.\XINT_dsx_AisNeg_finish_notzero.#1.%
{%
    \expandafter\XINT_dsx_end
    \expandafter {\romannumeral0\XINT_num {-#1}}{0}%
}%
\def\XINT_dsx_AisNeg_finish_notzero #1.#2.%
{%
    \expandafter\XINT_dsx_end
    \expandafter {\romannumeral0\XINT_num {#2}}{-#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_dsx_AisPos #1.#2;%
{%
    \expandafter\XINT_dsx_AisPos_finish
    \romannumeral0\XINT_split_xfork #1.#2\xint_bye2345678\xint_bye..%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_dsx_AisPos_finish #1.#2.%
{%
    \expandafter\XINT_dsx_end
    \expandafter {\romannumeral0\XINT_num {#2}}%
                 {\romannumeral0\XINT_num {#1}}%
}%
\def\XINT_dsx_end #1#2{\expandafter{#2}{#1}}%
%    \end{macrocode}
% \subsection{\csh{xintiiEq}}
% \lverb|no \xintiieq.|
%    \begin{macrocode}
\def\xintiiEq #1#2{\romannumeral0\xintiiifeq{#1}{#2}{1}{0}}%
%    \end{macrocode}
% \subsection{\csh{xintiiNotEq}}
% \lverb|Pour xintexpr. Pas de version en lowercase.|
%    \begin{macrocode}
\def\xintiiNotEq #1#2{\romannumeral0\xintiiifeq {#1}{#2}{0}{1}}%
%    \end{macrocode}
% \subsection{\csh{xintiiGeq}}
% \lverb|&
% PLUS GRAND OU ÉGAL
% attention compare les **valeurs absolues**
%
% 1.2l made \xintiiGeq robust against non terminated items.
%
% 1.2l rewrote \xintiiCmp, but forgot to handle \xintiiGeq too. Done at 1.2m.
%
% This macro should have been called \xintGEq for example.
% |
%    \begin{macrocode}
\def\xintiiGeq   {\romannumeral0\xintiigeq }%
\def\xintiigeq #1{\expandafter\XINT_iigeq\romannumeral`&&@#1\xint:}%
\def\XINT_iigeq #1#2\xint:#3%
{%
    \expandafter\XINT_geq_fork\expandafter #1\romannumeral`&&@#3\xint:#2\xint:
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_geq #1#2\xint:#3%
{%
    \expandafter\XINT_geq_fork\expandafter #1\romannumeral0\xintnum{#3}\xint:#2\xint:
}%
\def\XINT_geq_fork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_geq_firstiszero
      #2\XINT_geq_secondiszero
       0{}%
    \krof
    \xint_UDsignsfork
          #1#2\XINT_geq_minusminus
           #1-\XINT_geq_minusplus
           #2-\XINT_geq_plusminus
            --\XINT_geq_plusplus
    \krof #1#2%
}%
\def\XINT_geq_firstiszero  #1\krof 0#2#3\xint:#4\xint:
                              {\xint_UDzerofork #2{ 1}0{ 0}\krof }%
\def\XINT_geq_secondiszero #1\krof #20#3\xint:#4\xint:{ 1}%
\def\XINT_geq_plusminus    #1-{\XINT_geq_plusplus #1{}}%
\def\XINT_geq_minusplus    -#1{\XINT_geq_plusplus {}#1}%
\def\XINT_geq_minusminus    --{\XINT_geq_plusplus {}{}}%
\def\XINT_geq_plusplus
   {\expandafter\XINT_geq_finish\romannumeral0\XINT_cmp_plusplus}%
\def\XINT_geq_finish #1{\if-#1\expandafter\XINT_geq_no
                         \else\expandafter\XINT_geq_yes\fi}%
\def\XINT_geq_no 1{ 0}%
\def\XINT_geq_yes { 1}%
%    \end{macrocode}
% \subsection{\csh{xintiiGt}}
%    \begin{macrocode}
\def\xintiiGt #1#2{\romannumeral0\xintiiifgt{#1}{#2}{1}{0}}%
%    \end{macrocode}
% \subsection{\csh{xintiiLt}}
%    \begin{macrocode}
\def\xintiiLt #1#2{\romannumeral0\xintiiiflt{#1}{#2}{1}{0}}%
%    \end{macrocode}
% \subsection{\csh{xintiiGtorEq}}
%    \begin{macrocode}
\def\xintiiGtorEq #1#2{\romannumeral0\xintiiiflt {#1}{#2}{0}{1}}%
%    \end{macrocode}
% \subsection{\csh{xintiiLtorEq}}
%    \begin{macrocode}
\def\xintiiLtorEq #1#2{\romannumeral0\xintiiifgt {#1}{#2}{0}{1}}%
%    \end{macrocode}
% \subsection{\csh{xintiiIsZero}}
% \lverb|1.09a. restyled in 1.09i. 1.1 adds \xintiiIsZero, etc... for
% optimization in \xintexpr|
%    \begin{macrocode}
\def\xintiiIsZero {\romannumeral0\xintiiiszero }%
\def\xintiiiszero #1{\if0\xintiiSgn{#1}\xint_afterfi{ 1}\else\xint_afterfi{ 0}\fi}%
%    \end{macrocode}
% \subsection{\csh{xintiiIsNotZero}}
% \lverb|1.09a. restyled in 1.09i. 1.1 adds \xintiiIsZero, etc... for
% optimization in \xintexpr|
%    \begin{macrocode}
\def\xintiiIsNotZero {\romannumeral0\xintiiisnotzero }%
\def\xintiiisnotzero
          #1{\if0\xintiiSgn{#1}\xint_afterfi{ 0}\else\xint_afterfi{ 1}\fi}%
%    \end{macrocode}
% \subsection{\csh{xintiiIsOne}}
% \lverb|Added in 1.03. 1.09a defines \xintIsOne. 1.1a adds \xintiiIsOne.
%
% \XINT_isOne rewritten for 1.2g. Works with expanded strict integers,
% positive or negative.
%
%
%
%|
%    \begin{macrocode}
\def\xintiiIsOne {\romannumeral0\xintiiisone }%
\def\xintiiisone #1{\expandafter\XINT_isone\romannumeral`&&@#1XY}%
\def\XINT_isone #1#2#3Y%
{%
    \unless\if#2X\xint_dothis{ 0}\fi
    \unless\if#11\xint_dothis{ 0}\fi
    \xint_orthat{ 1}%
}%
\def\XINT_isOne #1{\XINT_is_One#1XY}%
\def\XINT_is_One #1#2#3Y%
{%
    \unless\if#2X\xint_dothis0\fi
    \unless\if#11\xint_dothis0\fi
    \xint_orthat1%
}%
%    \end{macrocode}
% \subsection{\csh{xintiiOdd}}
% \lverb|\xintOdd is needed for the xintexpr-essions even() and odd()
% functions (and also by \xintNewExpr).|
%    \begin{macrocode}
\def\xintiiOdd {\romannumeral0\xintiiodd }%
\def\xintiiodd #1%
{%
    \ifodd\xintLDg{#1} %<- intentional space
        \xint_afterfi{ 1}%
    \else
        \xint_afterfi{ 0}%
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiEven}}
%    \begin{macrocode}
\def\xintiiEven {\romannumeral0\xintiieven }%
\def\xintiieven #1%
{%
    \ifodd\xintLDg{#1} %<- intentional space
        \xint_afterfi{ 0}%
    \else
        \xint_afterfi{ 1}%
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiMON}}
% \lverb|MINUS ONE TO THE POWER N|
%    \begin{macrocode}
\def\xintiiMON {\romannumeral0\xintiimon }%
\def\xintiimon #1%
{%
    \ifodd\xintLDg {#1} %<- intentional space
        \xint_afterfi{ -1}%
    \else
        \xint_afterfi{ 1}%
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiMMON}}
% \lverb|MINUS ONE TO THE POWER N-1|
%    \begin{macrocode}
\def\xintiiMMON {\romannumeral0\xintiimmon }%
\def\xintiimmon #1%
{%
    \ifodd\xintLDg {#1} %<- intentional space
        \xint_afterfi{ 1}%
    \else
        \xint_afterfi{ -1}%
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintSgnFork}}
% \lverb|Expandable three-way fork added in 1.07. The argument #1 must expand
% to non-self-ending -1,0 or 1. 1.09i with _thenstop (now _stop_at...).|
%    \begin{macrocode}
\def\xintSgnFork {\romannumeral0\xintsgnfork }%
\def\xintsgnfork #1%
{%
    \ifcase #1 \expandafter\xint_stop_atsecondofthree
            \or\expandafter\xint_stop_atthirdofthree
          \else\expandafter\xint_stop_atfirstofthree
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifSgn}}
% \lverb|Expandable three-way fork added in 1.09a. Branches expandably
% depending on whether <0, =0, >0. Choice of branch guaranteed in two steps.
%
% 1.09i has \xint_firstofthreeafterstop (now \xint_stop_atfirstofthree) etc
% for faster expansion.
%
% 1.1 adds \xintiiifSgn for optimization in xintexpr-essions. Should I move
% them to xintcore? (for bnumexpr)|
%    \begin{macrocode}
\def\xintiiifSgn {\romannumeral0\xintiiifsgn }%
\def\xintiiifsgn #1%
{%
    \ifcase \xintiiSgn{#1}
               \expandafter\xint_stop_atsecondofthree
            \or\expandafter\xint_stop_atthirdofthree
          \else\expandafter\xint_stop_atfirstofthree
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifCmp}}
% \lverb|1.09e
% \xintifCmp {n}{m}{if n<m}{if n=m}{if n>m}. 1.1a adds ii variant|
%    \begin{macrocode}
\def\xintiiifCmp {\romannumeral0\xintiiifcmp }%
\def\xintiiifcmp #1#2%
{%
    \ifcase\xintiiCmp {#1}{#2}
               \expandafter\xint_stop_atsecondofthree
            \or\expandafter\xint_stop_atthirdofthree
          \else\expandafter\xint_stop_atfirstofthree
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifEq}}
% \lverb|1.09a \xintifEq {n}{m}{YES if n=m}{NO if n<>m}. 1.1a adds ii variant|
%    \begin{macrocode}
\def\xintiiifEq {\romannumeral0\xintiiifeq }%
\def\xintiiifeq #1#2%
{%
    \if0\xintiiCmp{#1}{#2}%
               \expandafter\xint_stop_atfirstoftwo
          \else\expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifGt}}
% \lverb|1.09a \xintifGt {n}{m}{YES if n>m}{NO if n<=m}. 1.1a adds ii variant|
%    \begin{macrocode}
\def\xintiiifGt {\romannumeral0\xintiiifgt }%
\def\xintiiifgt #1#2%
{%
    \if1\xintiiCmp{#1}{#2}%
               \expandafter\xint_stop_atfirstoftwo
          \else\expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifLt}}
% \lverb|1.09a \xintifLt {n}{m}{YES if n<m}{NO if n>=m}. Restyled in 1.09i.
% 1.1a adds ii variant|
%    \begin{macrocode}
\def\xintiiifLt {\romannumeral0\xintiiiflt }%
\def\xintiiiflt #1#2%
{%
    \ifnum\xintiiCmp{#1}{#2}<\xint_c_
          \expandafter\xint_stop_atfirstoftwo
    \else \expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifZero}}
% \lverb|Expandable two-way fork added in 1.09a. Branches expandably depending on
% whether the argument is zero (branch A) or not (branch B). 1.09i restyling. By
% the way it appears (not thoroughly tested, though) that \if tests are faster
% than \ifnum tests. 1.1 adds ii  versions.
%
% 1.2o deprecates \xintifZero.|
%    \begin{macrocode}
\def\xintiiifZero {\romannumeral0\xintiiifzero }%
\def\xintiiifzero #1%
{%
    \if0\xintiiSgn{#1}%
       \expandafter\xint_stop_atfirstoftwo
    \else
       \expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifNotZero}}
%    \begin{macrocode}
\def\xintiiifNotZero {\romannumeral0\xintiiifnotzero }%
\def\xintiiifnotzero #1%
{%
    \if0\xintiiSgn{#1}%
       \expandafter\xint_stop_atsecondoftwo
    \else
       \expandafter\xint_stop_atfirstoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifOne}}
% \lverb|added in 1.09i. 1.1a adds \xintiiifOne.|
%    \begin{macrocode}
\def\xintiiifOne {\romannumeral0\xintiiifone }%
\def\xintiiifone #1%
{%
    \if1\xintiiIsOne{#1}%
       \expandafter\xint_stop_atfirstoftwo
    \else
       \expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintiiifOdd}}
% \lverb|1.09e. Restyled in 1.09i. 1.1a adds \xintiiifOdd.|
%    \begin{macrocode}
\def\xintiiifOdd {\romannumeral0\xintiiifodd }%
\def\xintiiifodd #1%
{%
    \if\xintiiOdd{#1}1%
       \expandafter\xint_stop_atfirstoftwo
    \else
       \expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintifTrueAelseB}, \csh{xintifFalseAelseB}}
% \lverb|1.09i. 1.2i has removed deprecated \xintifTrueFalse, \xintifTrue.
%
% 1.2o uses \xintiiifNotZero, see comments to \xintAND etc... This will work
% fine with arguments being nested xintfrac.sty macros, without the overhead
% of \xintNum or \xintRaw parsing.|
%    \begin{macrocode}
\def\xintifTrueAelseB {\romannumeral0\xintiiifnotzero}%
\def\xintifFalseAelseB{\romannumeral0\xintiiifzero}%
%    \end{macrocode}
% \subsection{\csh{xintIsTrue}, \csh{xintIsFalse}}
% \lverb|1.09c. Suppressed at 1.2o. They seem not to have been documented, fortunately.|
%    \begin{macrocode}
%\let\xintIsTrue \xintIsNotZero
%\let\xintIsFalse\xintIsZero
%    \end{macrocode}
% \subsection{\csh{xintNOT}}
% \lverb|1.09c. But it should have been called \xintNOT, not \xintNot. Former
% denomination deprecated at 1.2o. Besides, the macro is now defined as ii-type.
% |
%    \begin{macrocode}
\def\xintNOT{\romannumeral0\xintiiiszero}%
%    \end{macrocode}
% \subsection{\csh{xintAND}, \csh{xintOR}, \csh{xintXOR}}
% \lverb|Added with 1.09a. But they used \xintSgn, etc... rather than
% \xintiiSgn. This brings \xintNum overhead which is not really desired, and
% which is not needed for use by xintexpr.sty. At 1.2o I modify them to use
% only ii macros. This is enough for sign or zeroness even for xintfrac
% format, as manipulated inside the \xintexpr. Big hesitation whether there
% should be however \xintiiAND outputting 1 or 0 versus an \xintAND outputting
% 1[0] versus 0[0] for example.|
%    \begin{macrocode}
\def\xintAND {\romannumeral0\xintand }%
\def\xintand #1#2{\if0\xintiiSgn{#1}\expandafter\xint_firstoftwo
                             \else\expandafter\xint_secondoftwo\fi
                  { 0}{\xintiiisnotzero{#2}}}%
\def\xintOR {\romannumeral0\xintor }%
\def\xintor #1#2{\if0\xintiiSgn{#1}\expandafter\xint_firstoftwo
                            \else\expandafter\xint_secondoftwo\fi
                 {\xintiiisnotzero{#2}}{ 1}}%
\def\xintXOR {\romannumeral0\xintxor }%
\def\xintxor #1#2{\if\xintiiIsZero{#1}\xintiiIsZero{#2}%
                     \xint_afterfi{ 0}\else\xint_afterfi{ 1}\fi }%
%    \end{macrocode}
% \subsection{\csh{xintANDof}}
% \lverb|New with 1.09a. \xintANDof works also with an empty list. Empty items
% however are not accepted.
%
% 1.2l made \xintANDof robust against non terminated items.
%
% 1.2o's \xintifTrueAelseB is now an ii macro, actually.
%
% 1.4. This macro as well as ORof and XORof were formally not used by
% xintexpr, which uses comma separated items, but at 1.4 xintexpr uses braced
% items. And the macros here got slightly refactored and \XINT_ANDof added for
% usage by xintexpr and the NewExpr hook. For some random reason I decided to
% use ^ as delimiter this has to do that other macros in xintfrac in same
% family (such as \xintGCDof, \xintSum) also use \xint: internally and
% although not strictly needed having two separate ones clarifies.
%  
% |
%    \begin{macrocode}
\def\xintANDof  {\romannumeral0\xintandof }%
\def\xintandof  #1{\expandafter\XINT_andof\romannumeral`&&@#1^}%
\def\XINT_ANDof {\romannumeral0\XINT_andof}%
\def\XINT_andof #1%
{%
    \xint_gob_til_^ #1\XINT_andof_yes ^%
    \xintiiifNotZero{#1}\XINT_andof\XINT_andof_no
}%
\def\XINT_andof_no  #1^{ 0}%
\def\XINT_andof_yes ^#1\XINT_andof_no{ 1}%
%    \end{macrocode}
% \subsection{\csh{xintORof}}
% \lverb|New with 1.09a. Works also with an empty list. Empty items
% however are not accepted.
%
% 1.2l made \xintORof robust against non terminated items.
%
% Refactored at 1.4.|
%    \begin{macrocode}
\def\xintORof  {\romannumeral0\xintorof }%
\def\xintorof  #1{\expandafter\XINT_orof\romannumeral`&&@#1^}%
\def\XINT_ORof {\romannumeral0\XINT_orof}%
\def\XINT_orof #1%
{%
    \xint_gob_til_^ #1\XINT_orof_no ^%
    \xintiiifNotZero{#1}\XINT_orof_yes\XINT_orof
}%
\def\XINT_orof_yes#1^{ 1}%
\def\XINT_orof_no ^#1\XINT_orof{ 0}%
%    \end{macrocode}
% \subsection{\csh{xintXORof}}
% \lverb|New with 1.09a. Works with an empty list, too.  Empty items
% however are not accepted. \XINT_xorof_c more
% efficient in 1.09i.
%
% 1.2l made \xintXORof robust against non terminated items.
%
% Refactored at 1.4 to use \numexpr (or an \ifnum). I have not tested if
% more efficient or not or if one can do better without \the.
% \XINT_XORof for xintexpr matters.
% |
%    \begin{macrocode}
\def\xintXORof  {\romannumeral0\xintxorof }%
\def\xintxorof  #1{\expandafter\XINT_xorof\romannumeral`&&@#1^}%
\def\XINT_XORof {\romannumeral0\XINT_xorof}%
\def\XINT_xorof {\if1\the\numexpr\XINT_xorof_a}%
\def\XINT_xorof_a #1%
{%
    \xint_gob_til_^ #1\XINT_xorof_e ^%
    \xintiiifNotZero{#1}{-}{}\XINT_xorof_a
}%
\def\XINT_xorof_e ^#1\XINT_xorof_a
   {1\relax\xint_afterfi{ 0}\else\xint_afterfi{ 1}\fi}%
%    \end{macrocode}
% \subsection{\csh{xintiiMax}}
% \lverb|&
% At 1.2m, a long-standing bug was fixed: \xintiiMax had the overhead of
% applying \xintNum to its arguments due to use of a sub-macro of \xintGeq
% code to which this overhead was added at some point.
%
% And on this occasion I reduced even more number of times input is grabbed.
% |
%    \begin{macrocode}
\def\xintiiMax {\romannumeral0\xintiimax }%
\def\xintiimax #1%
{%
    \expandafter\xint_iimax \romannumeral`&&@#1\xint:
}%
\def\xint_iimax #1\xint:#2%
{%
    \expandafter\XINT_max_fork\romannumeral`&&@#2\xint:#1\xint:
}%
%    \end{macrocode}
% \lverb|&
% #3#4 vient du *premier*,
% #1#2 vient du *second*. I have renamed the sub-macros at 1.2m because the
% terminology was quite counter-intuitive; there was no bug, but still.|
%    \begin{macrocode}
\def\XINT_max_fork #1#2\xint:#3#4\xint:
{%
    \xint_UDsignsfork
          #1#3\XINT_max_minusminus  % A < 0, B < 0
           #1-\XINT_max_plusminus   % B < 0, A >= 0
           #3-\XINT_max_minusplus   % A < 0, B >= 0
            --{\xint_UDzerosfork
                      #1#3\XINT_max_zerozero % A = B = 0
                       #10\XINT_max_pluszero % B = 0, A > 0
                       #30\XINT_max_zeroplus % A = 0, B > 0
                        00\XINT_max_plusplus % A, B > 0
                      \krof }%
    \krof
    #3#1#2\xint:#4\xint:
      \expandafter\xint_stop_atfirstoftwo
    \else
      \expandafter\xint_stop_atsecondoftwo
    \fi
    {#3#4}{#1#2}%
}%
%    \end{macrocode}
% \lverb|&
% Refactored at 1.2m for avoiding grabbing arguments. Position of inputs
% shared with iiCmp and iiGeq code.|
%    \begin{macrocode}
\def\XINT_max_zerozero  #1\fi{\xint_stop_atfirstoftwo }%
\def\XINT_max_zeroplus  #1\fi{\xint_stop_atsecondoftwo }%
\def\XINT_max_pluszero  #1\fi{\xint_stop_atfirstoftwo }%
\def\XINT_max_minusplus #1\fi{\xint_stop_atsecondoftwo }%
\def\XINT_max_plusminus #1\fi{\xint_stop_atfirstoftwo }%
\def\XINT_max_plusplus
{%
    \if1\romannumeral0\XINT_geq_plusplus
}%
%    \end{macrocode}
% \lverb+Premier des testés |A|=-A, second est |B|=-B. On veut le max(A,B),
% c'est donc A si |A|<|B| (ou |A|=|B|, mais peu importe alors). Donc on peut
% faire cela avec \unless. Simple.+
%    \begin{macrocode}
\def\XINT_max_minusminus --%
{%
    \unless\if1\romannumeral0\XINT_geq_plusplus{}{}%
}%
%    \end{macrocode}
% \subsection{\csh{xintiiMin}}
% \lverb|\xintnum added New with 1.09a. I add \xintiiMin in 1.1 and mark as
% deprecated \xintMin, renamed \xintiMin. \xintMin NOW REMOVED (1.2, as
% \xintMax, \xintMaxof), only provided by \xintfracnameimp.
%
% At 1.2m, a long-standing bug was fixed: \xintiiMin had the overhead of
% applying \xintNum to its arguments due to use of a sub-macro of \xintGeq
% code to which this overhead was added at some point.
%
% And on this occasion I reduced even more number of times input is grabbed.
% |
%    \begin{macrocode}
\def\xintiiMin {\romannumeral0\xintiimin }%
\def\xintiimin #1%
{%
    \expandafter\xint_iimin \romannumeral`&&@#1\xint:
}%
\def\xint_iimin #1\xint:#2%
{%
    \expandafter\XINT_min_fork\romannumeral`&&@#2\xint:#1\xint:
}%
\def\XINT_min_fork #1#2\xint:#3#4\xint:
{%
    \xint_UDsignsfork
          #1#3\XINT_min_minusminus  % A < 0, B < 0
           #1-\XINT_min_plusminus   % B < 0, A >= 0
           #3-\XINT_min_minusplus   % A < 0, B >= 0
            --{\xint_UDzerosfork
                      #1#3\XINT_min_zerozero % A = B = 0
                       #10\XINT_min_pluszero % B = 0, A > 0
                       #30\XINT_min_zeroplus % A = 0, B > 0
                        00\XINT_min_plusplus % A, B > 0
                      \krof }%
    \krof
    #3#1#2\xint:#4\xint:
      \expandafter\xint_stop_atsecondoftwo
    \else
      \expandafter\xint_stop_atfirstoftwo
    \fi
    {#3#4}{#1#2}%
}%
\def\XINT_min_zerozero  #1\fi{\xint_stop_atfirstoftwo }%
\def\XINT_min_zeroplus  #1\fi{\xint_stop_atfirstoftwo }%
\def\XINT_min_pluszero  #1\fi{\xint_stop_atsecondoftwo }%
\def\XINT_min_minusplus #1\fi{\xint_stop_atfirstoftwo }%
\def\XINT_min_plusminus #1\fi{\xint_stop_atsecondoftwo }%
\def\XINT_min_plusplus
{%
    \if1\romannumeral0\XINT_geq_plusplus
}%
\def\XINT_min_minusminus --%
{%
    \unless\if1\romannumeral0\XINT_geq_plusplus{}{}%
}%
%    \end{macrocode}
% \subsection{\csh{xintiiMaxof}}
% \lverb|New with 1.09a. 1.2 has NO MORE \xintMaxof, requires \xintfracname.
% 1.2a adds \xintiiMaxof, as \xintiiMaxof:csv is not public.
%
% NOT compatible with empty list.
%
% 1.2l made \xintiiMaxof robust against non terminated items.
%
% 1.4 refactors code to allow empty argument. For usage by \xintiiexpr.
% Slight deterioration, will come back.
% |
%    \begin{macrocode}
\def\xintiiMaxof {\romannumeral0\xintiimaxof }%
\def\xintiimaxof #1{\expandafter\XINT_iimaxof\romannumeral`&&@#1^}%
\def\XINT_iiMaxof{\romannumeral0\XINT_iimaxof}%
\def\XINT_iimaxof#1%
{%
    \xint_gob_til_^ #1\XINT_iimaxof_empty ^%
    \expandafter\XINT_iimaxof_loop\romannumeral`&&@#1\xint:
}%
\def\XINT_iimaxof_empty ^#1\xint:{ 0}%
\def\XINT_iimaxof_loop #1\xint:#2%
{%
    \xint_gob_til_^ #2\XINT_iimaxof_e ^%
    \expandafter\XINT_iimaxof_loop\romannumeral0\xintiimax{#1}{#2}\xint:
}%
\def\XINT_iimaxof_e ^#1\xintiimax #2#3\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintiiMinof}}
% \lverb|1.09a. 1.2a adds \xintiiMinof which was lacking.
%
% 1.4 refactoring for \xintiiexpr matters.|
%    \begin{macrocode}
\def\xintiiMinof {\romannumeral0\xintiiminof }%
\def\xintiiminof #1{\expandafter\XINT_iiminof\romannumeral`&&@#1^}%
\def\XINT_iiMinof{\romannumeral0\XINT_iiminof}%
\def\XINT_iiminof#1%
{%
    \xint_gob_til_^ #1\XINT_iiminof_empty ^%
    \expandafter\XINT_iiminof_loop\romannumeral`&&@#1\xint:
}%
\def\XINT_iiminof_empty ^#1\xint:{ 0}%
\def\XINT_iiminof_loop #1\xint:#2%
{%
    \xint_gob_til_^ #2\XINT_iiminof_e ^%
    \expandafter\XINT_iiminof_loop\romannumeral0\xintiimin{#1}{#2}\xint:
}%
\def\XINT_iiminof_e ^#1\xintiimin #2#3\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintiiSum}}
% \lverb|\xintiiSum {{a}{b}...{z}}
%  Refactored at 1.4 for matters initially related to xintexpr delimiter
%  choice.
%
%
% |
%    \begin{macrocode}
\def\xintiiSum {\romannumeral0\xintiisum }%
\def\xintiisum #1{\expandafter\XINT_iisum\romannumeral`&&@#1^}%
\def\XINT_iiSum{\romannumeral0\XINT_iisum}%
\def\XINT_iisum #1%
{%
    \expandafter\XINT_iisum_a\romannumeral`&&@#1\xint:
}%
\def\XINT_iisum_a #1%
{%
    \xint_gob_til_^ #1\XINT_iisum_empty ^%
    \XINT_iisum_loop #1%
}%
\def\XINT_iisum_empty ^#1\xint:{ 0}%
%    \end{macrocode}
% \lverb|bad coding as it depends on internal conventions of \XINT_add_nfork|
%    \begin{macrocode}
\def\XINT_iisum_loop #1#2\xint:#3%
{%
    \expandafter\XINT_iisum_loop_a
    \expandafter#1\romannumeral`&&@#3\xint:#2\xint:\xint:
}%
\def\XINT_iisum_loop_a #1#2%
{%
    \xint_gob_til_^ #2\XINT_iisum_loop_end ^%
    \expandafter\XINT_iisum_loop\romannumeral0\XINT_add_nfork #1#2%
}%
%    \end{macrocode}
% \lverb|see previous comment!|
%    \begin{macrocode}
\def\XINT_iisum_loop_end ^#1\XINT_add_nfork #2#3\xint:#4\xint:\xint:{ #2#4}%
%    \end{macrocode}
% \subsection{\csh{xintiiPrd}}
% \lverb|\xintiiPrd {{a}...{z}}
%
%
% Macros renamed and refactored (slightly more macros here to supposedly bring
% micro-gain) at 1.4 to match changes in xintfrac of delimiter, in sync with
% some usage in xintexpr.
%
% Contrarily to the xintfrac version \xintPrd, this one aborts as soon as it
% hits a zero value.
% 
%
%|
%    \begin{macrocode}
\def\xintiiPrd {\romannumeral0\xintiiprd }%
\def\xintiiprd #1{\expandafter\XINT_iiprd\romannumeral`&&@#1^}%
\def\XINT_iiPrd{\romannumeral0\XINT_iiprd}%
%    \end{macrocode}
% \lverb|The above romannumeral caused f-expansion of the list argument.
% We f-expand below the first item and each successive items because
% we do not use \xintiiMul but jump directly into \XINT_mul_nfork.
%
% |
%    \begin{macrocode}
\def\XINT_iiprd #1%
{%
    \expandafter\XINT_iiprd_a\romannumeral`&&@#1\xint:
}%
\def\XINT_iiprd_a #1%
{%
    \xint_gob_til_^ #1\XINT_iiprd_empty ^%
    \xint_gob_til_zero #1\XINT_iiprd_zero 0%
    \XINT_iiprd_loop #1%
}%
\def\XINT_iiprd_empty ^#1\xint:{ 1}%
\def\XINT_iiprd_zero  0#1^{ 0}%
%    \end{macrocode}
% \lverb|bad coding as it depends on internal conventions of \XINT_mul_nfork|
%    \begin{macrocode}
\def\XINT_iiprd_loop #1#2\xint:#3%
{%
    \expandafter\XINT_iiprd_loop_a
    \expandafter#1\romannumeral`&&@#3\xint:#2\xint:\xint:
}%
\def\XINT_iiprd_loop_a #1#2%
{%
    \xint_gob_til_^    #2\XINT_iiprd_loop_end ^%
    \xint_gob_til_zero #2\XINT_iiprd_zero 0%
    \expandafter\XINT_iiprd_loop\romannumeral0\XINT_mul_nfork #1#2%
}%
%    \end{macrocode}
% \lverb|see previous comment!|
%    \begin{macrocode}
\def\XINT_iiprd_loop_end ^#1\XINT_mul_nfork #2#3\xint:#4\xint:\xint:{ #2#4}%
%    \end{macrocode}
% \subsection{\csh{xintiiSquareRoot}}
% \lverb|First done with 1.08.
%
% 1.1 added \xintiiSquareRoot.
% 
% 1.1a added \xintiiSqrtR.
%
% 1.2f (2016/03/01-02-03) has rewritten the implementation, the underlying
% mathematics remaining about the same. The routine is much faster for inputs
% having up to 16 digits (because it does it all with \numexpr directly now),
% and also much faster for very long inputs (because it now fetches only the
% needed new digits after the first 16 (or 17) ones, via the geometric
% sequence 16, then 32, then 64, etc...; earlier version did the computations
% with all remaining digits after a suitable starting point with correct 4 or
% 5 leading digits). Note however that the fetching of tokens is via
% intrinsically O(N^2) macros, hence inevitably inputs with thousands of
% digits start being treated less well.
%
% Actually there is some room for improvements, one could prepare better
% input X for the upcoming treatment of fetching its digits by 16, then 32,
% then 64, etc...
%
% Incidently, as \xintiiSqrt uses subtraction and subtraction was broken from
% 1.2 to 1.2c, then for another reason from 1.2c to 1.2f, it could
% get wrong in certain (relatively rare) cases. There was also a bug that
% made it unneedlessly slow for odd number of digits on input.
%
% 1.2f also modifies \xintFloatSqrt in xintfrac.sty which now has more
% code in common with here and benefits from the same speed improvements.
%
% 1.2k belatedly corrects the output to {1}{1} and not 11 when input is zero.
% As braces are used in all other cases they should have been used here too.
%
% Also, 1.2k adds an \xintiSqrtR macro, for coherence as \xintiSqrt is
% defined (and mentioned in user manual.)
%
% |
%
%    \begin{macrocode}
\def\xintiiSquareRoot {\romannumeral0\xintiisquareroot }%
\def\xintiisquareroot #1{\expandafter\XINT_sqrt_checkin\romannumeral`&&@#1\xint:}%
\def\XINT_sqrt_checkin #1%
{%
    \xint_UDzerominusfork
     #1-\XINT_sqrt_iszero
     0#1\XINT_sqrt_isneg
      0-\XINT_sqrt
    \krof #1%
}%
\def\XINT_sqrt_iszero #1\xint:{{1}{1}}%
\def\XINT_sqrt_isneg  #1\xint:
   {\XINT_signalcondition{InvalidOperation}%
                         {Square root of negative: #1.}{}{{0}{0}}}%
\def\XINT_sqrt #1\xint:
{%
    \expandafter\XINT_sqrt_start\romannumeral0\xintlength {#1}.#1.%
}%
\def\XINT_sqrt_start #1.%
{%
    \ifnum #1<\xint_c_x\xint_dothis\XINT_sqrt_small_a\fi
    \xint_orthat\XINT_sqrt_big_a #1.%
}%
\def\XINT_sqrt_small_a #1.{\XINT_sqrt_a #1.\XINT_sqrt_small_d }%
\def\XINT_sqrt_big_a   #1.{\XINT_sqrt_a #1.\XINT_sqrt_big_d   }%
\def\XINT_sqrt_a #1.%
{%
   \ifodd #1
     \expandafter\XINT_sqrt_bO
   \else
     \expandafter\XINT_sqrt_bE
   \fi
   #1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_bE #1.#2#3#4%
{%
    \XINT_sqrt_c {#3#4}#2{#1}#3#4%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_bO #1.#2#3%
{%
    \XINT_sqrt_c #3#2{#1}#3%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_c #1#2%
{%
    \expandafter #2%
    \the\numexpr \ifnum #1>\xint_c_ii
                 \ifnum #1>\xint_c_vi
                 \ifnum #1>12 \ifnum #1>20 \ifnum #1>30
                 \ifnum #1>42 \ifnum #1>56 \ifnum #1>72
                 \ifnum #1>90
      10\else 9\fi \else 8\fi \else 7\fi \else 6\fi \else 5\fi
        \else 4\fi \else 3\fi \else 2\fi \else 1\fi .%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_d #1.#2%
{%
   \expandafter\XINT_sqrt_small_e
   \the\numexpr #1\ifcase \numexpr #2/\xint_c_ii-\xint_c_i\relax
                   \or 0\or 00\or 000\or 0000\fi .%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_e #1.#2.%
{%
   \expandafter\XINT_sqrt_small_ea\the\numexpr #1*#1-#2.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_ea #1%
{%
    \if0#1\xint_dothis\XINT_sqrt_small_ez\fi
    \if-#1\xint_dothis\XINT_sqrt_small_eb\fi
    \xint_orthat\XINT_sqrt_small_f #1%
}%
\def\XINT_sqrt_small_ez 0.#1.{\expandafter{\the\numexpr#1+\xint_c_i
         \expandafter}\expandafter{\the\numexpr #1*\xint_c_ii+\xint_c_i}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_eb -#1.#2.%
{%
    \expandafter\XINT_sqrt_small_ec \the\numexpr
    (#1-\xint_c_i+#2)/(\xint_c_ii*#2).#1.#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_ec #1.#2.#3.%
{%
    \expandafter\XINT_sqrt_small_f \the\numexpr
      -#2+\xint_c_ii*#3*#1+#1*#1\expandafter.\the\numexpr #3+#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_f #1.#2.%
{%
   \expandafter\XINT_sqrt_small_g
   \the\numexpr (#1+#2)/(\xint_c_ii*#2)-\xint_c_i.#1.#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_g #1#2.%
{%
    \if 0#1%
       \expandafter\XINT_sqrt_small_end
    \else
       \expandafter\XINT_sqrt_small_h
    \fi
    #1#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_small_h #1.#2.#3.%
{%
    \expandafter\XINT_sqrt_small_f
    \the\numexpr #2-\xint_c_ii*#1*#3+#1*#1\expandafter.%
    \the\numexpr #3-#1.%
}%
\def\XINT_sqrt_small_end #1.#2.#3.{{#3}{#2}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_d #1.#2%
{%
   \ifodd #2 \xint_dothis{\expandafter\XINT_sqrt_big_eO}\fi
   \xint_orthat{\expandafter\XINT_sqrt_big_eE}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
   \the\numexpr (#2-\xint_c_i)/\xint_c_ii.#1;%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_eE  #1;#2#3#4#5#6#7#8#9%
{%
    \XINT_sqrt_big_eE_a #1;{#2#3#4#5#6#7#8#9}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_eE_a #1.#2;#3%
{%
    \expandafter\XINT_sqrt_bigormed_f
    \romannumeral0\XINT_sqrt_small_e #2000.#3.#1;%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_eO #1;#2#3#4#5#6#7#8#9%
{%
    \XINT_sqrt_big_eO_a #1;{#2#3#4#5#6#7#8#9}%
}%
\def\XINT_sqrt_big_eO_a #1.#2;#3#4%
{%
    \expandafter\XINT_sqrt_bigormed_f
    \romannumeral0\XINT_sqrt_small_e #20000.#3#4.#1;%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_bigormed_f #1#2#3;%
{%
    \ifnum#3<\xint_c_ix 
          \xint_dothis {\csname XINT_sqrt_med_f\romannumeral#3\endcsname}%
    \fi
    \xint_orthat\XINT_sqrt_big_f #1.#2.#3;%
}%
\def\XINT_sqrt_med_fv   {\XINT_sqrt_med_fa .}%
\def\XINT_sqrt_med_fvi  {\XINT_sqrt_med_fa 0.}%
\def\XINT_sqrt_med_fvii {\XINT_sqrt_med_fa 00.}%
\def\XINT_sqrt_med_fviii{\XINT_sqrt_med_fa 000.}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_med_fa #1.#2.#3.#4;%
{%
    \expandafter\XINT_sqrt_med_fb
    \the\numexpr (#30#1-5#1)/(\xint_c_ii*#2).#1.#2.#3.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_med_fb #1.#2.#3.#4.#5.%
{%
    \expandafter\XINT_sqrt_small_ea
    \the\numexpr (#40#2-\xint_c_ii*#3*#1)*10#2+(#1*#1-#5)\expandafter.%
    \the\numexpr #30#2-#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_f #1;#2#3#4#5#6#7#8#9%
{%
    \XINT_sqrt_big_fa #1;{#2#3#4#5#6#7#8#9}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_fa #1.#2.#3;#4%
{%
    \expandafter\XINT_sqrt_big_ga 
    \the\numexpr #3-\xint_c_viii\expandafter.%
    \romannumeral0\XINT_sqrt_med_fa 000.#1.#2.;#4.%
}%
%    \end{macrocode}
% \lverb|&
%
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_ga #1.#2#3%
{%
    \ifnum #1>\xint_c_viii
      \expandafter\XINT_sqrt_big_gb\else
      \expandafter\XINT_sqrt_big_ka 
    \fi #1.#3.#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gb #1.#2.#3.%
{%
    \expandafter\XINT_sqrt_big_gc 
    \the\numexpr (\xint_c_ii*#2-\xint_c_i)*\xint_c_x^viii/(\xint_c_iv*#3).%
    #3.#2.#1;%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gc #1.#2.#3.%
{%
    \expandafter\XINT_sqrt_big_gd
    \romannumeral0\xintiiadd
        {\xintiiSub {#300000000}{\xintDouble{\xintiiMul{#2}{#1}}}00000000}%
        {\xintiiSqr {#1}}.%
    \romannumeral0\xintiisub{#200000000}{#1}.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gd #1.#2.%
{%
    \expandafter\XINT_sqrt_big_ge #2.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_ge #1;#2#3#4#5#6#7#8#9%
   {\XINT_sqrt_big_gf #1.#2#3#4#5#6#7#8#9;}%
\def\XINT_sqrt_big_gf #1;#2#3#4#5#6#7#8#9%
   {\XINT_sqrt_big_gg #1#2#3#4#5#6#7#8#9.}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gg #1.#2.#3.#4.%
{%
    \expandafter\XINT_sqrt_big_gloop
    \expandafter\xint_c_xvi\expandafter.%
    \the\numexpr #3-\xint_c_viii\expandafter.%
    \romannumeral0\xintiisub {#2}{\xintiNum{#4}}.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gloop #1.#2.%
{%
    \unless\ifnum #1<#2 \xint_dothis\XINT_sqrt_big_ka \fi
    \xint_orthat{\XINT_sqrt_big_gi #1.}#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gi #1.%
{%
    \expandafter\XINT_sqrt_big_gj\romannumeral\xintreplicate{#1}0.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gj #1.#2.#3.#4.#5.%
{%
    \expandafter\XINT_sqrt_big_gk
    \romannumeral0\xintiidivision {#4#1}%
                  {\XINT_dbl #5\xint_bye2345678\xint_bye*\xint_c_ii\relax}.%
    #1.#5.#2.#3.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gk #1#2.#3.#4.%
{%
    \expandafter\XINT_sqrt_big_gl
    \romannumeral0\xintiiadd {#2#3}{\xintiiSqr{#1}}.%
    \romannumeral0\xintiisub {#4#3}{#1}.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gl #1.#2.%
{%
    \expandafter\XINT_sqrt_big_gm #2.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gm #1.#2.#3.#4.#5.%
{%
    \expandafter\XINT_sqrt_big_gn
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \romannumeral0\XINT_split_fromleft\xint_c_ii*#3.#5\xint_bye2345678\xint_bye..%
    #1.#2.#3.#4.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_gn #1.#2.#3.#4.#5.#6.%
{%
    \expandafter\XINT_sqrt_big_gloop
    \the\numexpr \xint_c_ii*#5\expandafter.%
    \the\numexpr #6-#5\expandafter.%
    \romannumeral0\xintiisub{#4}{\xintiNum{#1}}.#3.#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_ka #1.#2.#3.#4.%
{%
    \expandafter\XINT_sqrt_big_kb
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \romannumeral0\XINT_dsx_addzeros {#1}#3;.%
    \romannumeral0\xintiisub
      {\XINT_dsx_addzerosnofuss {\xint_c_ii*#1}#2;}%
      {\xintiNum{#4}}.%
}%
\def\XINT_sqrt_big_kb #1.#2.%
{%
    \expandafter\XINT_sqrt_big_kc #2.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_kc #1%
{%
    \if0#1\xint_dothis\XINT_sqrt_big_kz\fi
    \xint_orthat\XINT_sqrt_big_kloop #1%
}%
\def\XINT_sqrt_big_kz 0.#1.%
{%
    \expandafter\XINT_sqrt_big_kend
    \romannumeral0%
    \xintinc{\XINT_dbl#1\xint_bye2345678\xint_bye*\xint_c_ii\relax}.#1.%
}%
\def\XINT_sqrt_big_kend #1.#2.%
{%
    \expandafter{\romannumeral0\xintinc{#2}}{#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_kloop #1.#2.%
{%
    \expandafter\XINT_sqrt_big_ke
    \romannumeral0\xintiidivision{#1}%
     {\romannumeral0\XINT_dbl #2\xint_bye2345678\xint_bye*\xint_c_ii\relax}{#2}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_ke #1%
{%
    \if0\XINT_Sgn #1\xint:
          \expandafter \XINT_sqrt_big_end
    \else \expandafter \XINT_sqrt_big_kf
    \fi {#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_kf #1#2#3%
{%
    \expandafter\XINT_sqrt_big_kg
    \romannumeral0\xintiisub {#3}{#1}.%
    \romannumeral0\xintiiadd {#2}{\xintiiSqr {#1}}.%
}%
\def\XINT_sqrt_big_kg #1.#2.%
{%
   \expandafter\XINT_sqrt_big_kloop #2.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_sqrt_big_end #1#2#3{{#3}{#2}}%
%    \end{macrocode}
% \subsection{\csh{xintiiSqrt}, \csh{xintiiSqrtR}}
%    \begin{macrocode}
\def\xintiiSqrt  {\romannumeral0\xintiisqrt  }%
\def\xintiisqrt  {\expandafter\XINT_sqrt_post\romannumeral0\xintiisquareroot  }%
\def\XINT_sqrt_post #1#2{\XINT_dec #1\XINT_dec_bye234567890\xint_bye}%
\def\xintiiSqrtR {\romannumeral0\xintiisqrtr }%
\def\xintiisqrtr {\expandafter\XINT_sqrtr_post\romannumeral0\xintiisquareroot }%
%    \end{macrocode}
% \lverb|N = (#1)^2 - #2 avec #1 le plus petit possible et #2>0 (hence #2<2*#1).
% (#1-.5)^2=#1^2-#1+.25=N+#2-#1+.25. Si 0<#2<#1, <= N-0.75<N, donc rounded->#1
% si #2>=#1, (#1-.5)^2>=N+.25>N, donc rounded->#1-1.|
%    \begin{macrocode}
\def\XINT_sqrtr_post #1#2%
   {\xintiiifLt {#2}{#1}{ #1}{\XINT_dec #1\XINT_dec_bye234567890\xint_bye}}%
%    \end{macrocode}
% \subsection{\csh{xintiiBinomial}}
% \lverb|2015/11/28-29 for 1.2f.
%
% 2016/11/19 for 1.2h: I truly can't understand why I hard-coded last
% year an error-message for arguments outside of the range for binomial
% formula. Naturally there should be no error but a rather a 0 return
% value for binomial(x,y), if y<0 or x<y !
%
% I really lack some kind of infinity or NaN value.
%
% 1.2o deprecates \xintiBinomial. (which xintfrac.sty redefined to use
% \xintNum)
% |
%    \begin{macrocode}
\def\xintiiBinomial {\romannumeral0\xintiibinomial }%
\def\xintiibinomial #1#2%
{%
    \expandafter\XINT_binom_pre\the\numexpr #1\expandafter.\the\numexpr #2.%
}%
\def\XINT_binom_pre #1.#2.%
{%
    \expandafter\XINT_binom_fork \the\numexpr#1-#2.#2.#1.%
}%
%    \end{macrocode}
% \lverb|k.x-k.x. I hesitated to restrict maximal allowed value of x to 10000.
% Finally I don't. But due to using small multiplication and small division, x
% must have at most eight digits. If x>=2^31 an arithmetic overflow error will
% have happened already.|
%    \begin{macrocode}
\def\XINT_binom_fork #1#2.#3#4.#5#6.%
{%
    \if-#5\xint_dothis{\XINT_signalcondition{InvalidOperation}%
                       {Binomial with negative first argument: #5#6.}{}{ 0}}\fi
    \if-#1\xint_dothis{ 0}\fi
    \if-#3\xint_dothis{ 0}\fi
    \if0#1\xint_dothis{ 1}\fi
    \if0#3\xint_dothis{ 1}\fi
    \ifnum #5#6>\xint_c_x^viii_mone\xint_dothis
       {\XINT_signalcondition{InvalidOperation}%
            {Binomial with too large argument: #5#6 >= 10^8.}{}{ 0}}\fi
    \ifnum #1#2>#3#4  \xint_dothis{\XINT_binom_a #1#2.#3#4.}\fi
                      \xint_orthat{\XINT_binom_a #3#4.#1#2.}%
}%
%    \end{macrocode}
% \lverb|x-k.k. avec 0<k<x, k<=x-k. Les divisions produiront en extra après le
% quotient un terminateur 1!\Z!0!. On va procéder par petite multiplication
% suivie par petite division. Donc ici on met le 1!\Z!0! pour amorcer.
%
% Le \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax est le terminateur pour le
% \XINT_unsep_cuzsmall final.|
%    \begin{macrocode}
\def\XINT_binom_a #1.#2.%
{%
    \expandafter\XINT_binom_b\the\numexpr \xint_c_i+#1.1.#2.100000001!1!;!0!%
}%
%    \end{macrocode}
% \lverb|y=x-k+1.j=1.k. On va évaluer par y/1*(y+1)/2*(y+2)/3 etc... On essaie
% de regrouper de manière à utiliser au mieux \numexpr. On peut aller jusqu'à
% x=10000 car 9999*10000<10^8. 463*464*465=99896880, 98*99*100*101=97990200.
% On va vérifier à chaque étape si on dépasse un seuil. Le style de
% l'implémentation diffère de celui que j'avais utilisé pour \xintiiFac. On
% pourrait tout-à-fait avoir une verybigloop, mais bon. Je rajoute aussi un
% verysmall. Le traitement est un peu différent pour elle afin d'aller jusqu'à
% x=29 (et pas seulement 26 si je suivais le modèle des autres, mais je veux
% pouvoir faire binomial(29,1), binomial(29,2), ... en vsmall).|
%    \begin{macrocode}
\def\XINT_binom_b #1.%
{%
    \ifnum #1>9999 \xint_dothis\XINT_binom_vbigloop \fi
    \ifnum #1>463  \xint_dothis\XINT_binom_bigloop   \fi
    \ifnum #1>98   \xint_dothis\XINT_binom_medloop   \fi
    \ifnum #1>29   \xint_dothis\XINT_binom_smallloop \fi
                   \xint_orthat\XINT_binom_vsmallloop #1.%
}%
%    \end{macrocode}
% \lverb|y.j.k. Au départ on avait x-k+1.1.k. Ensuite on a des blocs 1<8d>!
% donnant le résultat intermédiaire, dans l'ordre, et à la fin on a 1!1;!0!.
% Dans smallloop on peut prendre 4 par 4.|
%    \begin{macrocode}
\def\XINT_binom_smallloop #1.#2.#3.%
{%
    \ifcase\numexpr #3-#2\relax
        \expandafter\XINT_binom_end_
    \or \expandafter\XINT_binom_end_i
    \or \expandafter\XINT_binom_end_ii
    \or \expandafter\XINT_binom_end_iii
    \else\expandafter\XINT_binom_smallloop_a
    \fi #1.#2.#3.%
}%
%    \end{macrocode}
% \lverb|Ça m'ennuie un peu de reprendre les #1, #2, #3 ici. On a besoin de
% \numexpr pour \XINT_binom_div, mais de \romannumeral0 pour le unsep après
% \XINT_binom_mul.|
%    \begin{macrocode}
\def\XINT_binom_smallloop_a #1.#2.#3.%
{%
    \expandafter\XINT_binom_smallloop_b
    \the\numexpr #1+\xint_c_iv\expandafter.%
    \the\numexpr #2+\xint_c_iv\expandafter.%
    \the\numexpr #3\expandafter.%
    \the\numexpr\expandafter\XINT_binom_div
    \the\numexpr #2*(#2+\xint_c_i)*(#2+\xint_c_ii)*(#2+\xint_c_iii)\expandafter
    !\romannumeral0\expandafter\XINT_binom_mul
    \the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_binom_smallloop_b #1.%
{%
    \ifnum #1>98  \expandafter\XINT_binom_medloop   \else
                  \expandafter\XINT_binom_smallloop \fi #1.%
}%
%    \end{macrocode}
% \lverb|Ici on prend trois par trois.|
%    \begin{macrocode}
\def\XINT_binom_medloop #1.#2.#3.%
{%
    \ifcase\numexpr #3-#2\relax
        \expandafter\XINT_binom_end_
    \or \expandafter\XINT_binom_end_i
    \or \expandafter\XINT_binom_end_ii
    \else\expandafter\XINT_binom_medloop_a
    \fi #1.#2.#3.%
}%
\def\XINT_binom_medloop_a #1.#2.#3.%
{%
    \expandafter\XINT_binom_medloop_b
    \the\numexpr #1+\xint_c_iii\expandafter.%
    \the\numexpr #2+\xint_c_iii\expandafter.%
    \the\numexpr #3\expandafter.%
    \the\numexpr\expandafter\XINT_binom_div
        \the\numexpr #2*(#2+\xint_c_i)*(#2+\xint_c_ii)\expandafter
    !\romannumeral0\expandafter\XINT_binom_mul
        \the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_binom_medloop_b #1.%
{%
    \ifnum #1>463 \expandafter\XINT_binom_bigloop   \else
                  \expandafter\XINT_binom_medloop   \fi #1.%
}%
%    \end{macrocode}
% \lverb|Ici on prend deux par deux.|
%    \begin{macrocode}
\def\XINT_binom_bigloop #1.#2.#3.%
{%
    \ifcase\numexpr #3-#2\relax
        \expandafter\XINT_binom_end_
    \or \expandafter\XINT_binom_end_i
    \else\expandafter\XINT_binom_bigloop_a
    \fi #1.#2.#3.%
}%
\def\XINT_binom_bigloop_a #1.#2.#3.%
{%
    \expandafter\XINT_binom_bigloop_b
    \the\numexpr #1+\xint_c_ii\expandafter.%
    \the\numexpr #2+\xint_c_ii\expandafter.%
    \the\numexpr #3\expandafter.%
    \the\numexpr\expandafter\XINT_binom_div
        \the\numexpr #2*(#2+\xint_c_i)\expandafter
    !\romannumeral0\expandafter\XINT_binom_mul
        \the\numexpr #1*(#1+\xint_c_i)!%
}%
\def\XINT_binom_bigloop_b #1.%
{%
    \ifnum #1>9999 \expandafter\XINT_binom_vbigloop  \else
                   \expandafter\XINT_binom_bigloop   \fi #1.%
}%
%    \end{macrocode}
% \lverb|Et finalement un par un.|
%    \begin{macrocode}
\def\XINT_binom_vbigloop #1.#2.#3.%
{%
    \ifnum #3=#2
         \expandafter\XINT_binom_end_
    \else\expandafter\XINT_binom_vbigloop_a
    \fi #1.#2.#3.%
}%
\def\XINT_binom_vbigloop_a #1.#2.#3.%
{%
    \expandafter\XINT_binom_vbigloop
    \the\numexpr #1+\xint_c_i\expandafter.%
    \the\numexpr #2+\xint_c_i\expandafter.%
    \the\numexpr #3\expandafter.%
    \the\numexpr\expandafter\XINT_binom_div\the\numexpr #2\expandafter
    !\romannumeral0\XINT_binom_mul #1!%
}%
%    \end{macrocode}
% \lverb|y.j.k. La partie very small. y est au plus 26 (non 29 mais retesté
% dans \XINT_binom_vsmallloop_a), et tous les binomial(29,n) sont <10^8. On
% peut donc faire y(y+1)(y+2)(y+3) et aussi il y a le fait que etex fait a*b/c
% en double precision. Pour ne pas bifurquer à la fin sur smallloop, si n=27,
% 27, ou 29 on procède un peu différemment des autres boucles. Si je testais
% aussi #1 après #3-#2 pour les autres il faudrait des terminaisons
% différentes.|
%    \begin{macrocode}
\def\XINT_binom_vsmallloop #1.#2.#3.%
{%
    \ifcase\numexpr #3-#2\relax
        \expandafter\XINT_binom_vsmallend_
    \or \expandafter\XINT_binom_vsmallend_i
    \or \expandafter\XINT_binom_vsmallend_ii
    \or \expandafter\XINT_binom_vsmallend_iii
    \else\expandafter\XINT_binom_vsmallloop_a
    \fi #1.#2.#3.%
}%
\def\XINT_binom_vsmallloop_a #1.%
{%
    \ifnum #1>26  \expandafter\XINT_binom_smallloop_a  \else
                  \expandafter\XINT_binom_vsmallloop_b \fi #1.%
}%
\def\XINT_binom_vsmallloop_b #1.#2.#3.%
{%
    \expandafter\XINT_binom_vsmallloop
    \the\numexpr #1+\xint_c_iv\expandafter.%
    \the\numexpr #2+\xint_c_iv\expandafter.%
    \the\numexpr #3\expandafter.%
    \the\numexpr \expandafter\XINT_binom_vsmallmuldiv
    \the\numexpr  #2*(#2+\xint_c_i)*(#2+\xint_c_ii)*(#2+\xint_c_iii)\expandafter
    !\the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_binom_mul #1!#21!;!0!%
{%
    \expandafter\XINT_rev_nounsep\expandafter{\expandafter}%
    \the\numexpr\expandafter\XINT_smallmul
    \the\numexpr\xint_c_x^viii+#1\expandafter
    !\romannumeral0\XINT_rev_nounsep {}1;!#2%
    \R!\R!\R!\R!\R!\R!\R!\R!\W
    \R!\R!\R!\R!\R!\R!\R!\R!\W
    1;!%
}%
\def\XINT_binom_div #1!1;!%
{%
    \expandafter\XINT_smalldivx_a
    \the\numexpr #1/\xint_c_ii\expandafter\xint:
    \the\numexpr \xint_c_x^viii+#1!%
}%
%    \end{macrocode}
% \lverb|Vaguement envisagé d'éviter le 10^8+ mais bon.|
%    \begin{macrocode}
\def\XINT_binom_vsmallmuldiv #1!#2!1#3!{\xint_c_x^viii+#2*#3/#1!}%
%    \end{macrocode}
% \lverb|On a des terminaisons communes aux trois situations small, med, big,
% et on est sûr de pouvoir faire les multiplications dans \numexpr, car on
% vient ici *après* avoir comparé à 9999 ou 463 ou 98.|
%    \begin{macrocode}
\def\XINT_binom_end_iii #1.#2.#3.%
{%
    \expandafter\XINT_binom_finish
    \the\numexpr\expandafter\XINT_binom_div
        \the\numexpr #2*(#2+\xint_c_i)*(#2+\xint_c_ii)*(#2+\xint_c_iii)\expandafter
    !\romannumeral0\expandafter\XINT_binom_mul
        \the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_binom_end_ii #1.#2.#3.%
{%
    \expandafter\XINT_binom_finish
    \the\numexpr\expandafter\XINT_binom_div
        \the\numexpr #2*(#2+\xint_c_i)*(#2+\xint_c_ii)\expandafter
    !\romannumeral0\expandafter\XINT_binom_mul
        \the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_binom_end_i #1.#2.#3.%
{%
    \expandafter\XINT_binom_finish
    \the\numexpr\expandafter\XINT_binom_div
        \the\numexpr #2*(#2+\xint_c_i)\expandafter
    !\romannumeral0\expandafter\XINT_binom_mul
        \the\numexpr #1*(#1+\xint_c_i)!%
}%
\def\XINT_binom_end_ #1.#2.#3.%
{%
    \expandafter\XINT_binom_finish
    \the\numexpr\expandafter\XINT_binom_div\the\numexpr #2\expandafter
    !\romannumeral0\XINT_binom_mul #1!%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_binom_finish #1;!0!%
   {\XINT_unsep_cuzsmall #1\xint_bye!2!3!4!5!6!7!8!9!\xint_bye\xint_c_i\relax}%
%    \end{macrocode}
% \lverb|Duplication de code seulement pour la boucle avec très
% petits coeffs, mais en plus on fait au maximum des possibilités. (on
% pourrait tester plus le résultat déjà obtenu).|
%    \begin{macrocode}
\def\XINT_binom_vsmallend_iii #1.%
{%
    \ifnum #1>26  \expandafter\XINT_binom_end_iii \else
                  \expandafter\XINT_binom_vsmallend_iiib \fi #1.%
}%
\def\XINT_binom_vsmallend_iiib #1.#2.#3.%
{%
    \expandafter\XINT_binom_vsmallfinish
    \the\numexpr \expandafter\XINT_binom_vsmallmuldiv
    \the\numexpr  #2*(#2+\xint_c_i)*(#2+\xint_c_ii)*(#2+\xint_c_iii)\expandafter
    !\the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_binom_vsmallend_ii #1.%
{%
    \ifnum #1>27  \expandafter\XINT_binom_end_ii \else
                  \expandafter\XINT_binom_vsmallend_iib \fi #1.%
}%
\def\XINT_binom_vsmallend_iib #1.#2.#3.%
{%
    \expandafter\XINT_binom_vsmallfinish
    \the\numexpr \expandafter\XINT_binom_vsmallmuldiv
    \the\numexpr  #2*(#2+\xint_c_i)*(#2+\xint_c_ii)\expandafter
    !\the\numexpr #1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_binom_vsmallend_i #1.%
{%
    \ifnum #1>28  \expandafter\XINT_binom_end_i \else
                  \expandafter\XINT_binom_vsmallend_ib \fi #1.%
}%
\def\XINT_binom_vsmallend_ib #1.#2.#3.%
{%
    \expandafter\XINT_binom_vsmallfinish
    \the\numexpr \expandafter\XINT_binom_vsmallmuldiv
    \the\numexpr  #2*(#2+\xint_c_i)\expandafter
    !\the\numexpr #1*(#1+\xint_c_i)!%
}%
\def\XINT_binom_vsmallend_ #1.%
{%
    \ifnum #1>29  \expandafter\XINT_binom_end_ \else
                  \expandafter\XINT_binom_vsmallend_b \fi #1.%
}%
\def\XINT_binom_vsmallend_b #1.#2.#3.%
{%
    \expandafter\XINT_binom_vsmallfinish
    \the\numexpr\XINT_binom_vsmallmuldiv #2!#1!%
}%
\def\XINT_binom_vsmallfinish#1{%
\def\XINT_binom_vsmallfinish1##1!1!;!0!{\expandafter#1\the\numexpr##1\relax}%
}\XINT_binom_vsmallfinish{ }%
%    \end{macrocode}
% \subsection{\csh{xintiiPFactorial}}
% \lverb?2015/11/29 for 1.2f. Partial factorial pfac(a,b)=(a+1)...b, only for
% non-negative integers with a<=b<10^8.
%
% 1.2h (2016/11/20) removes the non-negativity condition. It was a bit
% unfortunate that the code raised \xintError:OutOfRangePFac if 0<=a<=b<10^8
% was violated. The rule now applied is to interpret pfac(a,b) as the product
% for a<j<=b (not as a ratio of Gamma function), hence if a>=b, return 1
% because of an empty product. If a<b: if a<0, return 0 for b>=0 and
% (-1)^(b-a) times |b|...(|a|-1) for b<0. But only for the range 0<=
% a <= b < 10^8 is the macro result to be considered as stable.?
%    \begin{macrocode}
\def\xintiiPFactorial {\romannumeral0\xintiipfactorial }%
\def\xintiipfactorial #1#2%
{%
    \expandafter\XINT_pfac_fork\the\numexpr#1\expandafter.\the\numexpr #2.%
}%
\def\xintPFactorial{\romannumeral0\xintpfactorial}%
\let\xintpfactorial\xintiipfactorial
%    \end{macrocode}
% \lverb|Code is a simplified version of the one for \xintiiBinomial, with no
% attempt at implementing a "very small" branch.|
%    \begin{macrocode}
\def\XINT_pfac_fork #1#2.#3#4.%
{%
    \unless\ifnum #1#2<#3#4 \xint_dothis\XINT_pfac_one\fi
    \if-#3\xint_dothis\XINT_pfac_neg\fi
    \if-#1\xint_dothis\XINT_pfac_zero\fi
    \ifnum #3#4>\xint_c_x^viii_mone\xint_dothis\XINT_pfac_outofrange\fi
    \xint_orthat \XINT_pfac_a #1#2.#3#4.%
}%
\def\XINT_pfac_outofrange #1.#2.%
   {\XINT_signalcondition{InvalidOperation}%
         {pFactorial with too large argument: #2 >= 10^8.}{}{ 0}}%
\def\XINT_pfac_one        #1.#2.{ 1}%
\def\XINT_pfac_zero       #1.#2.{ 0}%
\def\XINT_pfac_neg -#1.-#2.%
{%
    \ifnum #1>\xint_c_x^viii\xint_dothis\XINT_pfac_outofrange\fi
    \xint_orthat
   {\ifodd\numexpr#2-#1\relax\xint_afterfi{\expandafter-\romannumeral`&&@}\fi
    \expandafter\XINT_pfac_a }%
    \the\numexpr #2-\xint_c_i\expandafter.\the\numexpr#1-\xint_c_i.%
}%
\def\XINT_pfac_a #1.#2.%
{%
    \expandafter\XINT_pfac_b\the\numexpr \xint_c_i+#1.#2.100000001!1;!%
    1\R!1\R!1\R!1\R!1\R!1\R!1\R!1\R!\W
}%
\def\XINT_pfac_b #1.%
{%
    \ifnum #1>9999 \xint_dothis\XINT_pfac_vbigloop \fi
    \ifnum #1>463  \xint_dothis\XINT_pfac_bigloop   \fi
    \ifnum #1>98   \xint_dothis\XINT_pfac_medloop   \fi
                   \xint_orthat\XINT_pfac_smallloop #1.%
}%
\def\XINT_pfac_smallloop #1.#2.%
{%
    \ifcase\numexpr #2-#1\relax
        \expandafter\XINT_pfac_end_
    \or \expandafter\XINT_pfac_end_i
    \or \expandafter\XINT_pfac_end_ii
    \or \expandafter\XINT_pfac_end_iii
    \else\expandafter\XINT_pfac_smallloop_a
    \fi #1.#2.%
}%
\def\XINT_pfac_smallloop_a #1.#2.%
{%
    \expandafter\XINT_pfac_smallloop_b
    \the\numexpr #1+\xint_c_iv\expandafter.%
    \the\numexpr #2\expandafter.%
    \the\numexpr\expandafter\XINT_smallmul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_pfac_smallloop_b #1.%
{%
    \ifnum #1>98  \expandafter\XINT_pfac_medloop   \else
                  \expandafter\XINT_pfac_smallloop \fi #1.%
}%
\def\XINT_pfac_medloop #1.#2.%
{%
    \ifcase\numexpr #2-#1\relax
        \expandafter\XINT_pfac_end_
    \or \expandafter\XINT_pfac_end_i
    \or \expandafter\XINT_pfac_end_ii
    \else\expandafter\XINT_pfac_medloop_a
    \fi #1.#2.%
}%
\def\XINT_pfac_medloop_a #1.#2.%
{%
    \expandafter\XINT_pfac_medloop_b
    \the\numexpr #1+\xint_c_iii\expandafter.%
    \the\numexpr #2\expandafter.%
    \the\numexpr\expandafter\XINT_smallmul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_pfac_medloop_b #1.%
{%
    \ifnum #1>463 \expandafter\XINT_pfac_bigloop   \else
                  \expandafter\XINT_pfac_medloop   \fi #1.%
}%
\def\XINT_pfac_bigloop #1.#2.%
{%
    \ifcase\numexpr #2-#1\relax
        \expandafter\XINT_pfac_end_
    \or \expandafter\XINT_pfac_end_i
    \else\expandafter\XINT_pfac_bigloop_a
    \fi #1.#2.%
}%
\def\XINT_pfac_bigloop_a #1.#2.%
{%
    \expandafter\XINT_pfac_bigloop_b
    \the\numexpr #1+\xint_c_ii\expandafter.%
    \the\numexpr #2\expandafter.%
    \the\numexpr\expandafter
    \XINT_smallmul\the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)!%
}%
\def\XINT_pfac_bigloop_b #1.%
{%
    \ifnum #1>9999 \expandafter\XINT_pfac_vbigloop  \else
                   \expandafter\XINT_pfac_bigloop   \fi #1.%
}%
\def\XINT_pfac_vbigloop #1.#2.%
{%
    \ifnum #2=#1
         \expandafter\XINT_pfac_end_
    \else\expandafter\XINT_pfac_vbigloop_a
    \fi #1.#2.%
}%
\def\XINT_pfac_vbigloop_a #1.#2.%
{%
    \expandafter\XINT_pfac_vbigloop
    \the\numexpr #1+\xint_c_i\expandafter.%
    \the\numexpr #2\expandafter.%
    \the\numexpr\expandafter\XINT_smallmul\the\numexpr\xint_c_x^viii+#1!%
}%
\def\XINT_pfac_end_iii #1.#2.%
{%
    \expandafter\XINT_mul_out
    \the\numexpr\expandafter\XINT_smallmul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_pfac_end_ii #1.#2.%
{%
    \expandafter\XINT_mul_out
    \the\numexpr\expandafter\XINT_smallmul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_pfac_end_i #1.#2.%
{%
    \expandafter\XINT_mul_out
    \the\numexpr\expandafter\XINT_smallmul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)!%
}%
\def\XINT_pfac_end_ #1.#2.%
{%
    \expandafter\XINT_mul_out
    \the\numexpr\expandafter\XINT_smallmul\the\numexpr \xint_c_x^viii+#1!%
}%
%    \end{macrocode}
% \subsection{\csh{xintBool}, \csh{xintToggle}}
% \lverb|1.09c|
%    \begin{macrocode}
\def\xintBool #1{\romannumeral`&&@%
                 \csname if#1\endcsname\expandafter1\else\expandafter0\fi }%
\def\xintToggle #1{\romannumeral`&&@\iftoggle{#1}{1}{0}}%
%    \end{macrocode}
% \subsection{\csh{xintiiGCD}}
% |1.3d|: |\xintiiGCD| code from \xintgcdnameimp is copied here to
% support |gcd()| function in \csbxint{iiexpr}.
%
% |1.4|: removed from \xintgcdnameimp the original caode as now
% \xintgcdnameimp loads \xintnameimp.
%
% \changed[2021/03/22]{1.4d} Damn'ed! Since |1.3d| (2019/01/06) the code was
% broken if one of the arguments vanished due to a typo in macro names:
% "AisZero" at one location and "Aiszero" at next, and same for B...
%
% How could this not be detected by my tests !?!
%
% This caused |\xintiiGCDof| hence the |gcd()| function in |\xintiiexpr| to
% break as soon as one argument was zero.
%    \begin{macrocode}
\def\xintiiGCD {\romannumeral0\xintiigcd }%
\def\xintiigcd #1{\expandafter\XINT_iigcd\romannumeral0\xintiiabs#1\xint:}%
\def\XINT_iigcd #1#2\xint:#3%
{%
    \expandafter\XINT_gcd_fork\expandafter#1%
                \romannumeral0\xintiiabs#3\xint:#1#2\xint:
}%
\def\XINT_gcd_fork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_gcd_Aiszero
      #2\XINT_gcd_Biszero
       0\XINT_gcd_loop
    \krof
    #2%
}%
\def\XINT_gcd_Aiszero #1\xint:#2\xint:{ #1}%
\def\XINT_gcd_Biszero #1\xint:#2\xint:{ #2}%
\def\XINT_gcd_loop #1\xint:#2\xint:
{%
    \expandafter\expandafter\expandafter\XINT_gcd_CheckRem
    \expandafter\xint_secondoftwo
    \romannumeral0\XINT_div_prepare {#1}{#2}\xint:#1\xint:
}%
\def\XINT_gcd_CheckRem #1%
{%
    \xint_gob_til_zero #1\XINT_gcd_end0\XINT_gcd_loop #1%
}%
\def\XINT_gcd_end0\XINT_gcd_loop #1\xint:#2\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintiiGCDof}}
% \lverb|New with 1.09a (was located in xintgcd.sty).
%
% 1.2l adds protection against items being non-terminated \the\numexpr.
%
% 1.4 renames the macro into \xintiiGCDof and moves it here.
% Terminator modified to ^ for direct call by \xintiiexpr function.
%
% 1.4d fixes breakage inherited since 1.3d rom \xintiiGCD, in case
% any argument vanished.
%
% Currently does not support empty list of arguments.
% |
%    \begin{macrocode}
\def\xintiiGCDof    {\romannumeral0\xintiigcdof }%
\def\xintiigcdof    #1{\expandafter\XINT_iigcdof_a\romannumeral`&&@#1^}%
\def\XINT_iiGCDof   {\romannumeral0\XINT_iigcdof_a}%
\def\XINT_iigcdof_a #1{\expandafter\XINT_iigcdof_b\romannumeral`&&@#1!}%
\def\XINT_iigcdof_b #1!#2{\expandafter\XINT_iigcdof_c\romannumeral`&&@#2!{#1}!}%
\def\XINT_iigcdof_c #1{\xint_gob_til_^ #1\XINT_iigcdof_e ^\XINT_iigcdof_d #1}%
\def\XINT_iigcdof_d #1!{\expandafter\XINT_iigcdof_b\romannumeral0\xintiigcd {#1}}%
\def\XINT_iigcdof_e #1!#2!{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintiiLCM}}
% Copied over |\xintiiLCM| code from \xintgcdnameimp at |1.3d| in order to
% support |lcm()| function in \csbxint{iiexpr}.
%
% At |1.4| original code removed from \xintgcdnameimp as the latter now requires
% \xintnameimp.
%    \begin{macrocode}
\def\xintiiLCM {\romannumeral0\xintiilcm}%
\def\xintiilcm #1{\expandafter\XINT_iilcm\romannumeral0\xintiiabs#1\xint:}%
\def\XINT_iilcm #1#2\xint:#3%
{%
    \expandafter\XINT_lcm_fork\expandafter#1%
                \romannumeral0\xintiiabs#3\xint:#1#2\xint:
}%
\def\XINT_lcm_fork #1#2%
{%
    \xint_UDzerofork
      #1\XINT_lcm_iszero
      #2\XINT_lcm_iszero
       0\XINT_lcm_notzero
    \krof
    #2%
}%
\def\XINT_lcm_iszero #1\xint:#2\xint:{ 0}%
\def\XINT_lcm_notzero #1\xint:#2\xint:
{%
    \expandafter\XINT_lcm_end\romannumeral0%
      \expandafter\expandafter\expandafter\XINT_gcd_CheckRem
      \expandafter\xint_secondoftwo
      \romannumeral0\XINT_div_prepare {#1}{#2}\xint:#1\xint:
    \xint:#1\xint:#2\xint:
}%
\def\XINT_lcm_end #1\xint:#2\xint:#3\xint:{\xintiimul {#2}{\xintiiQuo{#3}{#1}}}%
%    \end{macrocode}
% \subsection{\csh{xintiiLCMof}}
% \lverb|See comments of \xintiiGCDof|.
%    \begin{macrocode}
\def\xintiiLCMof      {\romannumeral0\xintiilcmof }%
\def\xintiilcmof    #1{\expandafter\XINT_iilcmof_a\romannumeral`&&@#1^}%
\def\XINT_iiLCMof   {\romannumeral0\XINT_iilcmof_a}%
\def\XINT_iilcmof_a #1{\expandafter\XINT_iilcmof_b\romannumeral`&&@#1!}%
\def\XINT_iilcmof_b #1!#2{\expandafter\XINT_iilcmof_c\romannumeral`&&@#2!{#1}!}%
\def\XINT_iilcmof_c #1{\xint_gob_til_^ #1\XINT_iilcmof_e ^\XINT_iilcmof_d #1}%
\def\XINT_iilcmof_d #1!{\expandafter\XINT_iilcmof_b\romannumeral0\xintiilcm {#1}}%
\def\XINT_iilcmof_e #1!#2!{ #2}%
%    \end{macrocode}
% \subsection{(WIP) \csh{xintRandomDigits}}
% \lverb|1.3b. See user manual. Whether this will be part of xintkernel,
% xintcore, or xint is yet to be decided.|
%    \begin{macrocode}
\def\xintRandomDigits{\romannumeral0\xintrandomdigits}%
\def\xintrandomdigits#1%
{%
    \csname xint_gob_andstop_\expandafter\XINT_randomdigits\the\numexpr#1\xint:
}%
\def\XINT_randomdigits#1\xint:
{%
    \expandafter\XINT_randomdigits_a
    \the\numexpr(#1+\xint_c_iii)/\xint_c_viii\xint:#1\xint:
}%
\def\XINT_randomdigits_a#1\xint:#2\xint:
{%
    \romannumeral\numexpr\xint_c_viii*#1-#2\csname XINT_%
      \romannumeral\XINT_replicate #1\endcsname \csname
    XINT_rdg\endcsname
}%
\def\XINT_rdg
{%
    \expandafter\XINT_rdg_aux\the\numexpr%
                 \xint_c_nine_x^viii%
                                -\xint_texuniformdeviate\xint_c_ii^vii%
                 -\xint_c_ii^vii*\xint_texuniformdeviate\xint_c_ii^vii%
                 -\xint_c_ii^xiv*\xint_texuniformdeviate\xint_c_ii^vii%
                 -\xint_c_ii^xxi*\xint_texuniformdeviate\xint_c_ii^vii%
                 +\xint_texuniformdeviate\xint_c_x^viii%
                 \relax%
}%
\def\XINT_rdg_aux#1{XINT_rdg\endcsname}%
\let\XINT_XINT_rdg\endcsname
%    \end{macrocode}
% \subsection{(WIP) \csh{XINT_eightrandomdigits}, \csh{xintEightRandomDigits}}
% \lverb|1.3b. 1.4 adds some public alias...|
%    \begin{macrocode}
\def\XINT_eightrandomdigits
{%
     \expandafter\xint_gobble_i\the\numexpr%
                 \xint_c_nine_x^viii%
                                -\xint_texuniformdeviate\xint_c_ii^vii%
                 -\xint_c_ii^vii*\xint_texuniformdeviate\xint_c_ii^vii%
                 -\xint_c_ii^xiv*\xint_texuniformdeviate\xint_c_ii^vii%
                 -\xint_c_ii^xxi*\xint_texuniformdeviate\xint_c_ii^vii%
                 +\xint_texuniformdeviate\xint_c_x^viii%
                 \relax%
}%
\let\xintEightRandomDigits\XINT_eightrandomdigits
\def\xintRandBit{\xint_texuniformdeviate\xint_c_ii}%
%    \end{macrocode}
% \subsection{(WIP) \csh{xintRandBit}}
% \lverb|1.4 And let's add also \xintRandBit while we are at it.|
%    \begin{macrocode}
\def\xintRandBit{\xint_texuniformdeviate\xint_c_ii}%
%    \end{macrocode}
% \subsection{(WIP) \csh{xintXRandomDigits}}
% \lverb|1.3b.|
%    \begin{macrocode}
\def\xintXRandomDigits#1%
{%
    \csname xint_gobble_\expandafter\XINT_xrandomdigits\the\numexpr#1\xint:
}%
\def\XINT_xrandomdigits#1\xint:
{%
    \expandafter\XINT_xrandomdigits_a
    \the\numexpr(#1+\xint_c_iii)/\xint_c_viii\xint:#1\xint:
}%
\def\XINT_xrandomdigits_a#1\xint:#2\xint:
{%
    \romannumeral\numexpr\xint_c_viii*#1-#2\expandafter\endcsname
    \romannumeral`&&@\romannumeral
                     \XINT_replicate #1\endcsname\XINT_eightrandomdigits
}%
%    \end{macrocode}
% \subsection{(WIP) \csh{xintiiRandRangeAtoB}}
% \lverb|1.3b. Support for randrange() function.
%
% We do it f-expandably for matters of \xintNewExpr etc... The \xintexpr will
% add \xintNum wrapper to possible fractional input. But \xintiiexpr will call
% as is.
%
% TODO: ? implement third argument (STEP)
% TODO: \xintNum wrapper (which truncates) not so good in floatexpr. Use round?
%
% It is an error if b<=a, as in Python.|
%    \begin{macrocode}
\def\xintiiRandRangeAtoB{\romannumeral`&&@\xintiirandrangeAtoB}%
\def\xintiirandrangeAtoB#1%
{%
    \expandafter\XINT_randrangeAtoB_a\romannumeral`&&@#1\xint:
}%
\def\XINT_randrangeAtoB_a#1\xint:#2%
{%
    \xintiiadd{\expandafter\XINT_randrange
               \romannumeral0\xintiisub{#2}{#1}\xint:}%
              {#1}%
}%
%    \end{macrocode}
% \subsection{(WIP) \csh{xintiiRandRange}}
% \lverb|1.3b. Support for randrange().|
%    \begin{macrocode}
\def\xintiiRandRange{\romannumeral`&&@\xintiirandrange}%
\def\xintiirandrange#1%
{%
    \expandafter\XINT_randrange\romannumeral`&&@#1\xint:
}%
\def\XINT_randrange #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_randrange_err:empty
      0#1\XINT_randrange_err:empty
       0-\XINT_randrange_a
    \krof #1%
}%
\def\XINT_randrange_err:empty#1\xint:
{%
    \XINT_expandableerror{Empty range for randrange.} 0%
}%
\def\XINT_randrange_a #1\xint:
{%
    \expandafter\XINT_randrange_b\romannumeral0\xintlength{#1}.#1\xint:
}%
\def\XINT_randrange_b #1.%
{%
    \ifnum#1<\xint_c_x\xint_dothis{\the\numexpr\XINT_uniformdeviate{}}\fi
    \xint_orthat{\XINT_randrange_c #1.}%
}%
\def\XINT_randrange_c #1.#2#3#4#5#6#7#8#9%
{%
    \expandafter\XINT_randrange_d
    \the\numexpr\expandafter\XINT_uniformdeviate\expandafter
       {\expandafter}\the\numexpr\xint_c_i+#2#3#4#5#6#7#8#9\xint:\xint:
    #2#3#4#5#6#7#8#9\xint:#1\xint:
}%
%    \end{macrocode}
% \lverb|This raises following annex question: immediately after setting the
% seed is it possible for \xintUniformDeviate{N} where N>0 has exactly eight
% digits to return either 0 or N-1 ? It could be that this is never the case,
% then there is a bias in randrange(). Of course there are anyhow only 2^28
% seeds so randrange(10^X) is by necessity biased when executed immediately
% after setting the seed, if X is at least 9.|
%    \begin{macrocode}
\def\XINT_randrange_d #1\xint:#2\xint:
{%
    \ifnum#1=\xint_c_\xint_dothis\XINT_randrange_Z\fi
    \ifnum#1=#2 \xint_dothis\XINT_randrange_A\fi
    \xint_orthat\XINT_randrange_e #1\xint:
}%
\def\XINT_randrange_e #1\xint:#2\xint:#3\xint:
{%
    \the\numexpr#1\expandafter\relax
    \romannumeral0\xintrandomdigits{#2-\xint_c_viii}%
}%
%    \end{macrocode}
% \lverb|This is quite unlikely to get executed but if it does it must
% pay attention to leading zeros, hence the \xintinum.
% We don't have to be
% overly obstinate about removing overheads...|
%    \begin{macrocode}
\def\XINT_randrange_Z 0\xint:#1\xint:#2\xint:
{%
    \xintinum{\xintRandomDigits{#1-\xint_c_viii}}%
}%
%    \end{macrocode}
% \lverb|Here too, overhead is not such a problem. The idea is that we got by
% extraordinary same first 8 digits as upper range bound so we pick at random
% the remaining needed digits in one go and compare with the upper bound. If too
% big, we start again with another random 8 leading digits in given range. No
% need to aim at any kind of efficiency for the check and loop back.|
%    \begin{macrocode}
\def\XINT_randrange_A #1\xint:#2\xint:#3\xint:
{%
    \expandafter\XINT_randrange_B
    \romannumeral0\xintrandomdigits{#2-\xint_c_viii}\xint:
    #3\xint:#2.#1\xint:
}%
\def\XINT_randrange_B #1\xint:#2\xint:#3.#4\xint:
{%
    \xintiiifLt{#1}{#2}{\XINT_randrange_E}{\XINT_randrange_again}%
    #4#1\xint:#3.#4#2\xint:
}%
\def\XINT_randrange_E #1\xint:#2\xint:{ #1}%
\def\XINT_randrange_again #1\xint:{\XINT_randrange_c}%
%    \end{macrocode}
% \subsection{(WIP) Adjustments for engines without uniformdeviate primitive}
% \lverb|1.3b.|
%    \begin{macrocode}
\ifdefined\xint_texuniformdeviate
\else
  \def\xintrandomdigits#1%
  {%
      \XINT_expandableerror
      {No uniformdeviate at engine level.} 0%
  }%
  \let\xintXRandomDigits\xintRandomDigits
  \def\XINT_randrange#1\xint:
  {%
      \XINT_expandableerror
      {No uniformdeviate at engine level.} 0%
  }%
\fi
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xint}
% \cleardoublepage\let\xintnameUp\undefined
%\gardesactifs
%\let</xint>\relax
%\let<*xintbinhex>\gardesinactifs
%</xint>^^A-------------------------------------------------------
%<*xintbinhex>^^A-------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintbinhexnameUp\endcsname
% \section{Package \xintbinhexnameimp implementation}
% \RaisedLabel{sec:binheximp}
%
% \localtableofcontents
%
% The commenting is currently (\xintdocdate) very sparse.
%
% The macros from |1.08| (|2013/06/07|) remained unchanged
% until their complete rewrite at |1.2m| (|2017/07/31|).
%
% At |1.2n| dependencies on \xintcorenameimp were removed, so now the package
% loads only \xintkernelnameimp (this could have been done earlier).
%
% Also at |1.2n|, macros evolved again, the main improvements being in the
% increased allowable sizes of the input for |\xintDecToHex|, |\xintDecToBin|,
% |\xintBinToHex|. Use of |\csname| governed expansion at some places rather
% than |\numexpr| with some clean-up after it.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xintbinhex.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintkernel.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintbinhex}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintbinhex.sty
      \ifx\w\relax % but xintkernel.sty not yet loaded.
         \def\z{\endgroup\input xintkernel.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintkernel.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xintkernel}}%
          \fi
      \else
        \aftergroup\endinput % xintbinhex already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintbinhex}%
  [2021/07/13 v1.4j Expandable binary and hexadecimal conversions (JFB)]%
%    \end{macrocode}
% \subsection{Constants, etc...}
% \lverb|1.2n switches to \csname-governed expansion at various places.|
%    \begin{macrocode}
\newcount\xint_c_ii^xv  \xint_c_ii^xv   32768
\newcount\xint_c_ii^xvi \xint_c_ii^xvi  65536
\def\XINT_tmpa #1{\ifx\relax#1\else
  \expandafter\edef\csname XINT_csdth_#1\endcsname
  {\endcsname\ifcase #1 0\or 1\or 2\or 3\or 4\or 5\or 6\or 7\or
              8\or 9\or A\or B\or C\or D\or E\or F\fi}%
  \expandafter\XINT_tmpa\fi }%
\XINT_tmpa {0}{1}{2}{3}{4}{5}{6}{7}{8}{9}{10}{11}{12}{13}{14}{15}\relax
\def\XINT_tmpa #1{\ifx\relax#1\else
  \expandafter\edef\csname XINT_csdtb_#1\endcsname
  {\endcsname\ifcase #1
   0000\or 0001\or 0010\or 0011\or 0100\or 0101\or 0110\or 0111\or
   1000\or 1001\or 1010\or 1011\or 1100\or 1101\or 1110\or 1111\fi}%
  \expandafter\XINT_tmpa\fi }%
\XINT_tmpa {0}{1}{2}{3}{4}{5}{6}{7}{8}{9}{10}{11}{12}{13}{14}{15}\relax
\let\XINT_tmpa\relax
\expandafter\def\csname XINT_csbth_0000\endcsname {\endcsname0}%
\expandafter\def\csname XINT_csbth_0001\endcsname {\endcsname1}%
\expandafter\def\csname XINT_csbth_0010\endcsname {\endcsname2}%
\expandafter\def\csname XINT_csbth_0011\endcsname {\endcsname3}%
\expandafter\def\csname XINT_csbth_0100\endcsname {\endcsname4}%
\expandafter\def\csname XINT_csbth_0101\endcsname {\endcsname5}%
\expandafter\def\csname XINT_csbth_0110\endcsname {\endcsname6}%
\expandafter\def\csname XINT_csbth_0111\endcsname {\endcsname7}%
\expandafter\def\csname XINT_csbth_1000\endcsname {\endcsname8}%
\expandafter\def\csname XINT_csbth_1001\endcsname {\endcsname9}%
\expandafter\def\csname XINT_csbth_1010\endcsname {\endcsname A}%
\expandafter\def\csname XINT_csbth_1011\endcsname {\endcsname B}%
\expandafter\def\csname XINT_csbth_1100\endcsname {\endcsname C}%
\expandafter\def\csname XINT_csbth_1101\endcsname {\endcsname D}%
\expandafter\def\csname XINT_csbth_1110\endcsname {\endcsname E}%
\expandafter\def\csname XINT_csbth_1111\endcsname {\endcsname F}%
\let\XINT_csbth_none \endcsname
\expandafter\def\csname XINT_cshtb_0\endcsname {\endcsname0000}%
\expandafter\def\csname XINT_cshtb_1\endcsname {\endcsname0001}%
\expandafter\def\csname XINT_cshtb_2\endcsname {\endcsname0010}%
\expandafter\def\csname XINT_cshtb_3\endcsname {\endcsname0011}%
\expandafter\def\csname XINT_cshtb_4\endcsname {\endcsname0100}%
\expandafter\def\csname XINT_cshtb_5\endcsname {\endcsname0101}%
\expandafter\def\csname XINT_cshtb_6\endcsname {\endcsname0110}%
\expandafter\def\csname XINT_cshtb_7\endcsname {\endcsname0111}%
\expandafter\def\csname XINT_cshtb_8\endcsname {\endcsname1000}%
\expandafter\def\csname XINT_cshtb_9\endcsname {\endcsname1001}%
\def\XINT_cshtb_A {\endcsname1010}%
\def\XINT_cshtb_B {\endcsname1011}%
\def\XINT_cshtb_C {\endcsname1100}%
\def\XINT_cshtb_D {\endcsname1101}%
\def\XINT_cshtb_E {\endcsname1110}%
\def\XINT_cshtb_F {\endcsname1111}%
\let\XINT_cshtb_none \endcsname
%    \end{macrocode}
% \subsection{Helper macros}
% \subsubsection{\csh{XINT_zeroes_foriv}}
% \lverb|&
%( \romannumeral0\XINT_zeroes_foriv #1\R{0\R}{00\R}{000\R}$%
%:                                    \R{0\R}{00\R}{000\R}\R\W
%)
% expands to the <empty> or 0 or 00 or 000 needed which when adjoined to #1
% extend it to length 4N.|
%    \begin{macrocode}
\def\XINT_zeroes_foriv #1#2#3#4#5#6#7#8%
{%
    \xint_gob_til_R #8\XINT_zeroes_foriv_end\R\XINT_zeroes_foriv
}%
\def\XINT_zeroes_foriv_end\R\XINT_zeroes_foriv #1#2\W
   {\XINT_zeroes_foriv_done #1}%
\def\XINT_zeroes_foriv_done #1\R{ #1}%
%    \end{macrocode}
% \subsection{\csh{xintDecToHex}}
% \lverb|Complete rewrite at 1.2m in the 1.2 style. Also, 1.2m is robust
% against non terminated inputs.
%
% Improvements of coding at 1.2n, increased maximal size. Again some coding
% improvement at 1.2o, about 6$% speed gain.
%
% An input without leading zeroes gives an output without leading zeroes.|
%    \begin{macrocode}
\def\xintDecToHex {\romannumeral0\xintdectohex }%
\def\xintdectohex #1%
{%
    \expandafter\XINT_dth_checkin\romannumeral`&&@#1\xint:
}%
\def\XINT_dth_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_dth_neg
        -{\XINT_dth_main #1}%
     \krof
}%
\def\XINT_dth_neg {\expandafter-\romannumeral0\XINT_dth_main}%
\def\XINT_dth_main #1\xint:
{%
    \expandafter\XINT_dth_finish
    \romannumeral`&&@\expandafter\XINT_dthb_start
    \romannumeral0\XINT_zeroes_foriv
       #1\R{0\R}{00\R}{000\R}\R{0\R}{00\R}{000\R}\R\W
    #1\xint_bye\XINT_dth_tohex
}%
\def\XINT_dthb_start #1#2#3#4#5%
{%
    \xint_bye#5\XINT_dthb_small\xint_bye\XINT_dthb_start_a #1#2#3#4#5%
}%
\def\XINT_dthb_small\xint_bye\XINT_dthb_start_a #1\xint_bye#2{#2#1!}%
\def\XINT_dthb_start_a #1#2#3#4#5#6#7#8#9%
{%
    \expandafter\XINT_dthb_again\the\numexpr\expandafter\XINT_dthb_update
    \the\numexpr#1#2#3#4%
    \xint_bye#9\XINT_dthb_lastpass\xint_bye
    #5#6#7#8!\XINT_dthb_exclam\relax\XINT_dthb_nextfour #9%
}%
%    \end{macrocode}
% \lverb|The 1.2n inserted
% exclamations marks, which when bumping back from \XINT_dthb_again gave rise
% to a \numexpr-loop which gathered the ! delimited arguments and inserted
% \expandafter\XINT_dthb_update\the\numexpr dynamically. The 1.2o trick is to
% insert it here immediately. Then at \XINT_dthb_again the \numexpr will
% trigger an already prepared chain.
%
% The crux of the thing is handling of #3 at \XINT_dthb_update_a.
% |
%    \begin{macrocode}
\def\XINT_dthb_exclam {!\XINT_dthb_exclam\relax
                       \expandafter\XINT_dthb_update\the\numexpr}%
\def\XINT_dthb_update #1!%
{%
    \expandafter\XINT_dthb_update_a
    \the\numexpr (#1+\xint_c_ii^xv)/\xint_c_ii^xvi-\xint_c_i\xint:
    #1\xint:%
}%
\def\XINT_dthb_update_a #1\xint:#2\xint:#3%
{%
    0000+#1\expandafter#3\the\numexpr#2-#1*\xint_c_ii^xvi
}%
%    \end{macrocode}
% \lverb|1.2m and 1.2n had some unduly complicated ending pattern for
% \XINT_dthb_nextfour as inheritance of a loop needing ! separators which was
% pruned out at 1.2o (see previous comment).
% |
%    \begin{macrocode}
\def\XINT_dthb_nextfour #1#2#3#4#5%
{%
    \xint_bye#5\XINT_dthb_lastpass\xint_bye
    #1#2#3#4!\XINT_dthb_exclam\relax\XINT_dthb_nextfour#5%
}%
\def\XINT_dthb_lastpass\xint_bye #1!#2\xint_bye#3{#1!#3!}%
\def\XINT_dth_tohex
{%
    \expandafter\expandafter\expandafter\XINT_dth_tohex_a\csname\XINT_tofourhex
}%
\def\XINT_dth_tohex_a\endcsname{!\XINT_dth_tohex!}%
\def\XINT_dthb_again #1!#2#3%
{%
    \ifx#3\relax
          \expandafter\xint_firstoftwo
      \else
          \expandafter\xint_secondoftwo
    \fi
    {\expandafter\XINT_dthb_again
     \the\numexpr
     \ifnum #1>\xint_c_
        \xint_afterfi{\expandafter\XINT_dthb_update\the\numexpr#1}%
     \fi}%
    {\ifnum #1>\xint_c_ \xint_dothis{#2#1!}\fi\xint_orthat{!#2!}}%
}%
\def\XINT_tofourhex #1!%
{%
    \expandafter\XINT_tofourhex_a
    \the\numexpr (#1+\xint_c_ii^vii)/\xint_c_ii^viii-\xint_c_i\xint:
    #1\xint:
}%
\def\XINT_tofourhex_a #1\xint:#2\xint:
{%
    \expandafter\XINT_tofourhex_c
    \the\numexpr (#1+\xint_c_viii)/\xint_c_xvi-\xint_c_i\xint:
    #1\xint:
    \the\numexpr #2-\xint_c_ii^viii*#1!%
}%
\def\XINT_tofourhex_c #1\xint:#2\xint:
{%
    XINT_csdth_#1%
    \csname XINT_csdth_\the\numexpr #2-\xint_c_xvi*#1\relax
    \csname \expandafter\XINT_tofourhex_d
}%
\def\XINT_tofourhex_d #1!%
{%
    \expandafter\XINT_tofourhex_e
    \the\numexpr (#1+\xint_c_viii)/\xint_c_xvi-\xint_c_i\xint:
    #1\xint:
}%
\def\XINT_tofourhex_e #1\xint:#2\xint:
{%
    XINT_csdth_#1%
    \csname XINT_csdth_\the\numexpr #2-\xint_c_xvi*#1\endcsname
}%
%    \end{macrocode}
% \lverb|We only clean-up up to 3 zero hexadecimal digits, as output was
% produced in chunks of 4 hex digits. If input had no leading zero, output
% will have none either. If input had many leading zeroes, output will have
% some number (unspecified, but a recipe can be given...) of leading zeroes...
%
% The coding is for varying a bit, I did not check if efficient, it does not
% matter.|
%    \begin{macrocode}
\def\XINT_dth_finish !\XINT_dth_tohex!#1#2#3%
{%
    \unless\if#10\xint_dothis{ #1#2#3}\fi
    \unless\if#20\xint_dothis{ #2#3}\fi
    \unless\if#30\xint_dothis{ #3}\fi
    \xint_orthat{ }%
}%
%    \end{macrocode}
% \subsection{\csh{xintDecToBin}}
% \lverb|Complete rewrite at 1.2m in the 1.2 style. Also, 1.2m is robust
% against non terminated inputs.
%
% Revisited at 1.2n like in \xintDecToHex: increased maximal size.
%
% An input without leading zeroes gives an output without leading zeroes.
%
% Most of the code canvas is shared with \xintDecToHex.
% |
%    \begin{macrocode}
\def\xintDecToBin {\romannumeral0\xintdectobin }%
\def\xintdectobin #1%
{%
    \expandafter\XINT_dtb_checkin\romannumeral`&&@#1\xint:
}%
\def\XINT_dtb_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_dtb_neg
        -{\XINT_dtb_main #1}%
     \krof
}%
\def\XINT_dtb_neg {\expandafter-\romannumeral0\XINT_dtb_main}%
\def\XINT_dtb_main #1\xint:
{%
    \expandafter\XINT_dtb_finish
    \romannumeral`&&@\expandafter\XINT_dthb_start
    \romannumeral0\XINT_zeroes_foriv
       #1\R{0\R}{00\R}{000\R}\R{0\R}{00\R}{000\R}\R\W
    #1\xint_bye\XINT_dtb_tobin
}%
\def\XINT_dtb_tobin
{%
    \expandafter\expandafter\expandafter\XINT_dtb_tobin_a\csname\XINT_tosixteenbits
}%
\def\XINT_dtb_tobin_a\endcsname{!\XINT_dtb_tobin!}%
\def\XINT_tosixteenbits #1!%
{%
    \expandafter\XINT_tosixteenbits_a
    \the\numexpr (#1+\xint_c_ii^vii)/\xint_c_ii^viii-\xint_c_i\xint:
    #1\xint:
}%
\def\XINT_tosixteenbits_a #1\xint:#2\xint:
{%
    \expandafter\XINT_tosixteenbits_c
    \the\numexpr (#1+\xint_c_viii)/\xint_c_xvi-\xint_c_i\xint:
    #1\xint:
    \the\numexpr #2-\xint_c_ii^viii*#1!%
}%
\def\XINT_tosixteenbits_c #1\xint:#2\xint:
{%
    XINT_csdtb_#1%
    \csname XINT_csdtb_\the\numexpr #2-\xint_c_xvi*#1\relax
    \csname \expandafter\XINT_tosixteenbits_d
}%
\def\XINT_tosixteenbits_d #1!%
{%
    \expandafter\XINT_tosixteenbits_e
    \the\numexpr (#1+\xint_c_viii)/\xint_c_xvi-\xint_c_i\xint:
    #1\xint:
}%
\def\XINT_tosixteenbits_e #1\xint:#2\xint:
{%
    XINT_csdtb_#1%
    \csname XINT_csdtb_\the\numexpr #2-\xint_c_xvi*#1\endcsname
}%
\def\XINT_dtb_finish !\XINT_dtb_tobin!#1#2#3#4#5#6#7#8%
{%
    \expandafter\XINT_dtb_finish_a\the\numexpr #1#2#3#4#5#6#7#8\relax
}%
\def\XINT_dtb_finish_a #1{%
\def\XINT_dtb_finish_a ##1##2##3##4##5##6##7##8##9%
{%
    \expandafter#1\the\numexpr ##1##2##3##4##5##6##7##8##9\relax
}}\XINT_dtb_finish_a { }%
%    \end{macrocode}
% \subsection{\csh{xintHexToDec}}
% \lverb|Completely (and belatedly) rewritten at 1.2m in the 1.2 style.
%
% 1.2m version robust against non terminated inputs, but there is no primitive
% from TeX which may generate hexadecimal digits and provoke expansion ahead,
% afaik, except of course if decimal digits are treated as hexadecimal. This
% robustness is not on purpose but from need to expand argument and then grab
% it again. So we do it safely.
%
% Increased maximal size at 1.2n.
%
% 1.2m version robust against non terminated inputs.
%
% An input without leading zeroes gives an output without leading zeroes.
% |
%    \begin{macrocode}
\def\xintHexToDec {\romannumeral0\xinthextodec }%
\def\xinthextodec #1%
{%
    \expandafter\XINT_htd_checkin\romannumeral`&&@#1\xint:
}%
\def\XINT_htd_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_htd_neg
        -{\XINT_htd_main #1}%
     \krof
}%
\def\XINT_htd_neg {\expandafter-\romannumeral0\XINT_htd_main}%
\def\XINT_htd_main #1\xint:
{%
    \expandafter\XINT_htd_startb
    \the\numexpr\expandafter\XINT_htd_starta
    \romannumeral0\XINT_zeroes_foriv
       #1\R{0\R}{00\R}{000\R}\R{0\R}{00\R}{000\R}\R\W
    #1\xint_bye!2!3!4!5!6!7!8!9!\xint_bye\relax
}%
\def\XINT_htd_starta #1#2#3#4{"#1#2#3#4+100000!}%
\def\XINT_htd_startb 1#1%
{%
    \if#10\expandafter\XINT_htd_startba\else
          \expandafter\XINT_htd_startbb
    \fi 1#1%
}%
\def\XINT_htd_startba 10#1!{\XINT_htd_again #1%
    \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\XINT_htd_nextfour}%
\def\XINT_htd_startbb 1#1#2!{\XINT_htd_again #1!#2%
    \xint_bye!2!3!4!5!6!7!8!9!\xint_bye\XINT_htd_nextfour}%
%    \end{macrocode}
% \lverb|It is a bit annoying to grab all to the end here. I have a version,
% modeled on the 1.2n variant of \xintDecToHex which solved that problem
% there, but it did not prove enough if at all faster in my brief testing and
% it had the defect of a reduced maximal allowed size of the input. |
%    \begin{macrocode}
\def\XINT_htd_again #1\XINT_htd_nextfour #2%
{%
    \xint_bye #2\XINT_htd_finish\xint_bye
    \expandafter\XINT_htd_A\the\numexpr
    \XINT_htd_a #1\XINT_htd_nextfour #2%
}%
\def\XINT_htd_a #1!#2!#3!#4!#5!#6!#7!#8!#9!%
{%
    #1\expandafter\XINT_htd_update
    \the\numexpr #2\expandafter\XINT_htd_update
    \the\numexpr #3\expandafter\XINT_htd_update
    \the\numexpr #4\expandafter\XINT_htd_update
    \the\numexpr #5\expandafter\XINT_htd_update
    \the\numexpr #6\expandafter\XINT_htd_update
    \the\numexpr #7\expandafter\XINT_htd_update
    \the\numexpr #8\expandafter\XINT_htd_update
    \the\numexpr #9\expandafter\XINT_htd_update
    \the\numexpr \XINT_htd_a
}% 
\def\XINT_htd_nextfour #1#2#3#4%
{%
    *\xint_c_ii^xvi+"#1#2#3#4+1000000000\relax\xint_bye!%
    2!3!4!5!6!7!8!9!\xint_bye\XINT_htd_nextfour
}%
%    \end{macrocode}
% \lverb|If the innocent looking commented out $#6 is left in the pattern as
% was the case at 1.2m, the maximal size becomes limited at 5538 digits, not
% 8298! (with parameter stack size = 10000.) |
%    \begin{macrocode}
\def\XINT_htd_update 1#1#2#3#4#5%#6!%
{%
    *\xint_c_ii^xvi+10000#1#2#3#4#5!%#6!%
}%
\def\XINT_htd_A 1#1%
{%
    \if#10\expandafter\XINT_htd_Aa\else
          \expandafter\XINT_htd_Ab
    \fi 1#1%
}%
\def\XINT_htd_Aa 10#1#2#3#4{\XINT_htd_again #1#2#3#4!}%
\def\XINT_htd_Ab 1#1#2#3#4#5{\XINT_htd_again #1!#2#3#4#5!}%
\def\XINT_htd_finish\xint_bye
    \expandafter\XINT_htd_A\the\numexpr \XINT_htd_a #1\XINT_htd_nextfour
{%
    \expandafter\XINT_htd_finish_cuz\the\numexpr0\XINT_htd_unsep_loop #1% 
}%
\def\XINT_htd_unsep_loop #1!#2!#3!#4!#5!#6!#7!#8!#9!%
{%
    \expandafter\XINT_unsep_clean
    \the\numexpr 1#1#2\expandafter\XINT_unsep_clean
    \the\numexpr 1#3#4\expandafter\XINT_unsep_clean
    \the\numexpr 1#5#6\expandafter\XINT_unsep_clean
    \the\numexpr 1#7#8\expandafter\XINT_unsep_clean
    \the\numexpr 1#9\XINT_htd_unsep_loop_a
}%
\def\XINT_htd_unsep_loop_a #1!#2!#3!#4!#5!#6!#7!#8!#9!%
{%
    #1\expandafter\XINT_unsep_clean
    \the\numexpr 1#2#3\expandafter\XINT_unsep_clean
    \the\numexpr 1#4#5\expandafter\XINT_unsep_clean
    \the\numexpr 1#6#7\expandafter\XINT_unsep_clean
    \the\numexpr 1#8#9\XINT_htd_unsep_loop
}%
\def\XINT_unsep_clean 1{\relax}% also in xintcore
\def\XINT_htd_finish_cuz #1{%
\def\XINT_htd_finish_cuz ##1##2##3##4##5%
   {\expandafter#1\the\numexpr ##1##2##3##4##5\relax}%
}\XINT_htd_finish_cuz{ }%
%    \end{macrocode}
% \subsection{\csh{xintBinToDec}}
% \lverb|Redone entirely for 1.2m. Starts by converting to hexadecimal
% first.
%
% Increased maximal size at 1.2n.
%
% An input without leading zeroes gives an output without leading zeroes.
%
% Robust against non-terminated input.|
%    \begin{macrocode}
\def\xintBinToDec {\romannumeral0\xintbintodec }%
\def\xintbintodec #1%
{%
    \expandafter\XINT_btd_checkin\romannumeral`&&@#1\xint:
}%
\def\XINT_btd_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_btd_N
        -{\XINT_btd_main #1}%
     \krof
}%
\def\XINT_btd_N {\expandafter-\romannumeral0\XINT_btd_main }%
\def\XINT_btd_main #1\xint:
{%
    \csname XINT_btd_htd\csname\expandafter\XINT_bth_loop
    \romannumeral0\XINT_zeroes_foriv
       #1\R{0\R}{00\R}{000\R}\R{0\R}{00\R}{000\R}\R\W
    #1\xint_bye2345678\xint_bye none\endcsname\xint:
}%
\def\XINT_btd_htd #1\xint:
{%
    \expandafter\XINT_htd_startb
    \the\numexpr\expandafter\XINT_htd_starta
    \romannumeral0\XINT_zeroes_foriv
       #1\R{0\R}{00\R}{000\R}\R{0\R}{00\R}{000\R}\R\W
    #1\xint_bye!2!3!4!5!6!7!8!9!\xint_bye\relax
}%
%    \end{macrocode}
% \subsection{\csh{xintBinToHex}}
% \lverb|Complete rewrite for 1.2m.
% But input for 1.2m version limited to about 13320 binary digits (expansion
% depth=10000).
%
% Again redone for 1.2n for \csname governed expansion: increased maximal size.
%
% Size of output is ceil(size(input)/4), leading zeroes in output (inherited
% from the input) are not trimmed.
%
% An input without leading zeroes gives an output without leading zeroes.
%
% Robust against non-terminated input.
% |
%    \begin{macrocode}
\def\xintBinToHex {\romannumeral0\xintbintohex }%
\def\xintbintohex #1%
{%
    \expandafter\XINT_bth_checkin\romannumeral`&&@#1\xint:
}%
\def\XINT_bth_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_bth_N
        -{\XINT_bth_main #1}%
     \krof
}%
\def\XINT_bth_N {\expandafter-\romannumeral0\XINT_bth_main }%
\def\XINT_bth_main #1\xint:
{%
    \csname space\csname\expandafter\XINT_bth_loop
    \romannumeral0\XINT_zeroes_foriv
       #1\R{0\R}{00\R}{000\R}\R{0\R}{00\R}{000\R}\R\W
    #1\xint_bye2345678\xint_bye none\endcsname
}%
\def\XINT_bth_loop #1#2#3#4#5#6#7#8%
{%
            XINT_csbth_#1#2#3#4%
    \csname XINT_csbth_#5#6#7#8%
    \csname\XINT_bth_loop
}%
%    \end{macrocode}
% \subsection{\csh{xintHexToBin}}
% \lverb|Completely rewritten for 1.2m.
%
% Attention this macro is not robust against arguments expanding after
% themselves.
%
% Only up to three zeros are removed on front of output: if the input had a
% leading zero, there will be a leading zero (and then possibly 4n of them if
% inputs had more leading zeroes) on output.
%
% Rewritten again at 1.2n for \csname governed expansion.|
%    \begin{macrocode}
\def\xintHexToBin {\romannumeral0\xinthextobin }%
\def\xinthextobin #1%
{%
    \expandafter\XINT_htb_checkin\romannumeral`&&@#1%
    \xint_bye 23456789\xint_bye none\endcsname
}%
\def\XINT_htb_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_htb_N
        -{\XINT_htb_main #1}%
     \krof
}%
\def\XINT_htb_N {\expandafter-\romannumeral0\XINT_htb_main }%
\def\XINT_htb_main {\csname XINT_htb_cuz\csname\XINT_htb_loop}%
\def\XINT_htb_loop #1#2#3#4#5#6#7#8#9%
{%
             XINT_cshtb_#1%
     \csname XINT_cshtb_#2%
     \csname XINT_cshtb_#3%
     \csname XINT_cshtb_#4%
     \csname XINT_cshtb_#5%
     \csname XINT_cshtb_#6%
     \csname XINT_cshtb_#7%
     \csname XINT_cshtb_#8%
     \csname XINT_cshtb_#9%
     \csname \XINT_htb_loop
}%
\def\XINT_htb_cuz #1{%
\def\XINT_htb_cuz ##1##2##3##4%
   {\expandafter#1\the\numexpr##1##2##3##4\relax}%
}\XINT_htb_cuz { }%
%    \end{macrocode}
% \subsection{\csh{xintCHexToBin}}
% \lverb|The 1.08 macro had same functionality as \xintHexToBin, and slightly
% different code, the 1.2m version has the same code as \xintHexToBin except
% that it does not remove leading zeros from output: if the input had N
% hexadecimal digits, the output will have exactly 4N binary digits.
%
% Rewritten again at 1.2n for \csname governed expansion.|
%    \begin{macrocode}
\def\xintCHexToBin {\romannumeral0\xintchextobin }%
\def\xintchextobin #1%
{%
    \expandafter\XINT_chtb_checkin\romannumeral`&&@#1%
    \xint_bye 23456789\xint_bye none\endcsname
}%
\def\XINT_chtb_checkin #1%
{%
    \xint_UDsignfork
       #1\XINT_chtb_N
        -{\XINT_chtb_main #1}%
     \krof
}%
\def\XINT_chtb_N {\expandafter-\romannumeral0\XINT_chtb_main }%
\def\XINT_chtb_main {\csname space\csname\XINT_htb_loop}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintbinhex}
% \cleardoublepage\let\xintbinhexnameUp\undefined
%\gardesactifs
%\let</xintbinhex>\relax
%\let<*xintgcd>\gardesinactifs
%</xintbinhex>^^A-------------------------------------------------
%<*xintgcd>^^A----------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintgcdnameUp\endcsname
% \section{Package \xintgcdnameimp implementation}
% \RaisedLabel{sec:gcdimp}
%
% \localtableofcontents
%
% The commenting is currently (\xintdocdate) very sparse.
%
% Release |1.09h| has
% modified a bit the |\xintTypesetEuclideAlgorithm| and
% |\xintTypesetBezoutAlgorithm| layout with respect to line indentation in
% particular. And they use the \xinttoolsnameimp |\xintloop| rather than the
% Plain \TeX{} or \LaTeX{}'s |\loop|.
%
%
% Breaking change at |1.2p|: |\xintBezout{A}{B}| formerly had output
% |{A}{B}{U}{V}{D}| with |AU-BV=D|, now it is |{U}{V}{D}| with |AU+BV=D|.
%
% From |1.1| to |1.3f| the package loaded only \xintcorenameimp. At |1.4| it
% now automatically loads both of \xintnameimp and \xinttoolsnameimp (the
% latter being in fact a requirement of \csbxint{TypesetEuclideAlgorithm} and
% \csbxint{TypesetBezoutAlgorithm} since |1.09h|).
%
% \begin{framed}
% At |1.4| \csbxint{GCD}, \csbxint{LCM}, \csbxint{GCDof}, and \csbxint{LCMof}
% are \emph{removed} from the package:\IMPORTANTf they are provided only by
% \xintfracnameimp and they handle general fractions, not only integers.
%
% The original\CHANGEDf{1.4} integer-only macros have been renamed into respectively
% \csbxint{iiGCD}, \csbxint{iiLCM}, \csbxint{iiGCDof}, and \csbxint{iiLCMof}
% and got relocated into \xintnameimp package.
% \end{framed}
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \def\z{\endgroup}%
  \expandafter\let\expandafter\x\csname ver@xintgcd.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xint.sty\endcsname
  \expandafter\let\expandafter\t\csname ver@xinttools.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintgcd}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintgcd.sty
      \ifx\w\relax % but xint.sty not yet loaded.
         \expandafter\def\expandafter\z\expandafter{\z\input xint.sty\relax}%
      \fi
      \ifx\t\relax % but xinttools.sty not yet loaded.
         \expandafter\def\expandafter\z\expandafter{\z\input xinttools.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xint.sty not yet loaded.
            \expandafter\def\expandafter\z\expandafter{\z\RequirePackage{xint}}%
          \fi
          \ifx\t\relax % xinttools.sty not yet loaded.
            \expandafter\def\expandafter\z\expandafter{\z\RequirePackage{xinttools}}%
          \fi
      \else
        \aftergroup\endinput % xintgcd already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintgcd}%
  [2021/07/13 v1.4j Euclide algorithm with xint package (JFB)]%
%    \end{macrocode}
% \subsection{\csh{xintBezout}}
% \lverb|&
% \xintBezout{#1}{#2}
% produces {U}{V}{D} with UA+VB=D, D = PGCD(A,B) (non-positive),
% where #1 and #2 f-expand to big integers A and B.
%
% I had not checked this macro for about three years when I realized in
% January 2017 that \xintBezout{A}{B} was buggy for the cases A = 0 or B = 0.
% I fixed that blemish in 1.2l but overlooked the other blemish that
% \xintBezout{A}{B} with A multiple of B produced a coefficient U as -0 in
% place of 0.
%
% Hence I rewrote again for 1.2p. On this occasion I  modified the output
% of the macro to be {U}{V}{D} with AU+BV=D, formerly it was
% {A}{B}{U}{V}{D} with  AU - BV = D. This is quite breaking change!
%
% Note in particular change of sign of V.
%
% I don't know why I had designed this macro to contain {A}{B} in its output.
% Perhaps I initially intended to output {A//D}{B//D} (but forgot), as this is
% actually possible from outcome of the last iteration, with no need of
% actually dividing. Current code however arranges to skip this last update,
% as U and V are already furnished by the iteration prior to realizing that
% the last non-zero remainder was found.
%
% Also 1.2l raised InvalidOperation if both A and B vanished, but I removed
% this behaviour at 1.2p.
%|
%    \begin{macrocode}
\def\xintBezout {\romannumeral0\xintbezout }%
\def\xintbezout #1%
{%
    \expandafter\XINT_bezout\expandafter {\romannumeral0\xintnum{#1}}%
}%
\def\XINT_bezout #1#2%
{%
    \expandafter\XINT_bezout_fork \romannumeral0\xintnum{#2}\Z #1\Z
}%
%    \end{macrocode}
% \lverb|#3#4 = A, #1#2=B. Micro improvement for 1.2l.|
%    \begin{macrocode}
\def\XINT_bezout_fork #1#2\Z #3#4\Z
{%
    \xint_UDzerosfork
     #1#3\XINT_bezout_botharezero
      #10\XINT_bezout_secondiszero
      #30\XINT_bezout_firstiszero
       00\xint_UDsignsfork
    \krof
          #1#3\XINT_bezout_minusminus % A < 0, B < 0
           #1-\XINT_bezout_minusplus  % A > 0, B < 0
           #3-\XINT_bezout_plusminus  % A < 0, B > 0
            --\XINT_bezout_plusplus   % A > 0, B > 0
    \krof
    {#2}{#4}#1#3% #1#2=B, #3#4=A
}%
\def\XINT_bezout_botharezero #1\krof#2#300{{0}{0}{0}}%
\def\XINT_bezout_firstiszero #1\krof#2#3#4#5%
{%
    \xint_UDsignfork
      #4{{0}{-1}{#2}}%
       -{{0}{1}{#4#2}}%
    \krof
}%
\def\XINT_bezout_secondiszero #1\krof#2#3#4#5%
{%
    \xint_UDsignfork
       #5{{-1}{0}{#3}}%
        -{{1}{0}{#5#3}}%
    \krof
}%
%    \end{macrocode}
% \lverb|#4#2= A < 0, #3#1 = B < 0|
%    \begin{macrocode}
\def\XINT_bezout_minusminus #1#2#3#4%
{%
    \expandafter\XINT_bezout_mm_post
    \romannumeral0\expandafter\XINT_bezout_preloop_a
    \romannumeral0\XINT_div_prepare {#1}{#2}{#1}%
}%
\def\XINT_bezout_mm_post #1#2%
{%
    \expandafter\XINT_bezout_mm_postb\expandafter
    {\romannumeral0\xintiiopp{#2}}{\romannumeral0\xintiiopp{#1}}%
}%
\def\XINT_bezout_mm_postb #1#2{\expandafter{#2}{#1}}%
%    \end{macrocode}
% \lverb|minusplus  #4#2= A > 0, B < 0|
%    \begin{macrocode}
\def\XINT_bezout_minusplus #1#2#3#4%
{%
    \expandafter\XINT_bezout_mp_post
    \romannumeral0\expandafter\XINT_bezout_preloop_a
    \romannumeral0\XINT_div_prepare {#1}{#4#2}{#1}%
}%
\def\XINT_bezout_mp_post #1#2%
{%
    \expandafter\xint_exchangetwo_keepbraces\expandafter
    {\romannumeral0\xintiiopp {#2}}{#1}%
}%
%    \end{macrocode}
% \lverb|plusminus  A < 0, B > 0|
%    \begin{macrocode}
\def\XINT_bezout_plusminus #1#2#3#4%
{%
    \expandafter\XINT_bezout_pm_post
    \romannumeral0\expandafter\XINT_bezout_preloop_a
    \romannumeral0\XINT_div_prepare {#3#1}{#2}{#3#1}%
}%
\def\XINT_bezout_pm_post #1{\expandafter{\romannumeral0\xintiiopp{#1}}}%
%    \end{macrocode}
% \lverb|plusplus, B = #3#1 > 0, A = #4#2 > 0|
%    \begin{macrocode}
\def\XINT_bezout_plusplus #1#2#3#4%
{%
    \expandafter\XINT_bezout_preloop_a
    \romannumeral0\XINT_div_prepare {#3#1}{#4#2}{#3#1}%
}%
%    \end{macrocode}
% \lverb|&
%( n = 0: BA1001 (B, A, e=1, vv, uu, v, u)
%: r(1)=B, r(0)=A, après n étapes {r(n+1)}{r(n)}{vv}{uu}{v}{u}
%: q(n) quotient de r(n-1) par r(n)
%: si reste nul, exit et renvoie U = -e*uu, V = e*vv, A*U+B*V=D
%: sinon mise à jour
%:   vv, v = q * vv + v, vv
%:   uu, u = q * uu + u, uu
%:   e = -e
%: puis calcul quotient reste et itération
%)
%
% We arrange for \xintiiMul sub-routine to be called only with positive
% arguments, thus skipping some un-needed sign parsing there. For that though
% we have to screen out the special cases A divides B, or B divides A. And we
% first want to exchange A and B if A < B. These special cases are the only
% one possibly leading to U or V zero (for A and B positive which is the case
% here.) Thus the general case always leads to non-zero U and V's and assigning
% a final sign is done simply adding a - to one of them, with no fear of
% producing -0. |
%    \begin{macrocode}
\def\XINT_bezout_preloop_a #1#2#3%
{%
    \if0#1\xint_dothis\XINT_bezout_preloop_exchange\fi
    \if0#2\xint_dothis\XINT_bezout_preloop_exit\fi
    \xint_orthat{\expandafter\XINT_bezout_loop_B}%
    \romannumeral0\XINT_div_prepare {#2}{#3}{#2}{#1}110%
}%
\def\XINT_bezout_preloop_exit
    \romannumeral0\XINT_div_prepare #1#2#3#4#5#6#7%
{%
    {0}{1}{#2}%
}%
\def\XINT_bezout_preloop_exchange
{%
    \expandafter\xint_exchangetwo_keepbraces
    \romannumeral0\expandafter\XINT_bezout_preloop_A
}%
\def\XINT_bezout_preloop_A #1#2#3#4%
{%
    \if0#2\xint_dothis\XINT_bezout_preloop_exit\fi
    \xint_orthat{\expandafter\XINT_bezout_loop_B}%
    \romannumeral0\XINT_div_prepare {#2}{#3}{#2}{#1}%
}%
\def\XINT_bezout_loop_B #1#2%
{%  
    \if0#2\expandafter\XINT_bezout_exitA
     \else\expandafter\XINT_bezout_loop_C
    \fi {#1}{#2}%
}%
%    \end{macrocode}
% \lverb|&
% We use the fact that the \romannumeral-`0 (or equivalent) done by \xintiiadd
% will absorb the initial space token left by \XINT_mul_plusplus in its
% output.
%
% We arranged for operands here to be always positive which is needed for
% \XINT_mul_plusplus entry point (last time I checked...). Admittedly this
% kind of optimization is not good for maintenance of code, but I can't resist
% temptation of limiting the shuffling around of tokens...
% |
%    \begin{macrocode}
\def\XINT_bezout_loop_C #1#2#3#4#5#6#7%
{%
    \expandafter\XINT_bezout_loop_D\expandafter
        {\romannumeral0\xintiiadd{\XINT_mul_plusplus{}{}#1\xint:#4\xint:}{#6}}%
        {\romannumeral0\xintiiadd{\XINT_mul_plusplus{}{}#1\xint:#5\xint:}{#7}}%
    {#2}{#3}{#4}{#5}%
}%
\def\XINT_bezout_loop_D #1#2%
{%
    \expandafter\XINT_bezout_loop_E\expandafter{#2}{#1}%
}%
\def\XINT_bezout_loop_E #1#2#3#4%
{%
    \expandafter\XINT_bezout_loop_b
    \romannumeral0\XINT_div_prepare {#3}{#4}{#3}{#2}{#1}%
}%
\def\XINT_bezout_loop_b #1#2%
{%  
    \if0#2\expandafter\XINT_bezout_exita
     \else\expandafter\XINT_bezout_loop_c
    \fi {#1}{#2}%
}%
\def\XINT_bezout_loop_c #1#2#3#4#5#6#7%
{%
    \expandafter\XINT_bezout_loop_d\expandafter
        {\romannumeral0\xintiiadd{\XINT_mul_plusplus{}{}#1\xint:#4\xint:}{#6}}%
        {\romannumeral0\xintiiadd{\XINT_mul_plusplus{}{}#1\xint:#5\xint:}{#7}}%
    {#2}{#3}{#4}{#5}%
}%
\def\XINT_bezout_loop_d #1#2%
{%
    \expandafter\XINT_bezout_loop_e\expandafter{#2}{#1}%
}%
\def\XINT_bezout_loop_e #1#2#3#4%
{%
    \expandafter\XINT_bezout_loop_B
    \romannumeral0\XINT_div_prepare {#3}{#4}{#3}{#2}{#1}%
}%
%    \end{macrocode}
% \lverb|&
% sortir U, V, D mais on a travaillé avec vv, uu, v, u dans cet ordre.$\
% The code is structured so that #4 and #5 are guaranteed non-zero
% if we exit here, hence we can not create a -0 in output.|
%    \begin{macrocode}
\def\XINT_bezout_exita #1#2#3#4#5#6#7{{-#5}{#4}{#3}}%
\def\XINT_bezout_exitA #1#2#3#4#5#6#7{{#5}{-#4}{#3}}%
%    \end{macrocode}
% \subsection{\csh{xintEuclideAlgorithm}}
% \lverb|&
% Pour Euclide:
% {N}{A}{D=r(n)}{B}{q1}{r1}{q2}{r2}{q3}{r3}....{qN}{rN=0}$\
% u<2n> = u<2n+3>u<2n+2> + u<2n+4> à la n ième étape.
%
% Formerly, used \xintiabs, but got deprecated at 1.2o.|
%    \begin{macrocode}
\def\xintEuclideAlgorithm {\romannumeral0\xinteuclidealgorithm }%
\def\xinteuclidealgorithm #1%
{%
    \expandafter\XINT_euc\expandafter{\romannumeral0\xintiiabs{\xintNum{#1}}}%
}%
\def\XINT_euc #1#2%
{%
    \expandafter\XINT_euc_fork\romannumeral0\xintiiabs{\xintNum{#2}}\Z #1\Z
}%
%    \end{macrocode}
% \lverb|Ici #3#4=A, #1#2=B|
%    \begin{macrocode}
\def\XINT_euc_fork #1#2\Z #3#4\Z
{%
    \xint_UDzerofork
      #1\XINT_euc_BisZero
      #3\XINT_euc_AisZero
       0\XINT_euc_a
    \krof
    {0}{#1#2}{#3#4}{{#3#4}{#1#2}}{}\Z
}%
%    \end{macrocode}
% \lverb|&
% Le {} pour protéger {{A}{B}} si on s'arrête après une étape (B divise
% A).
% On va renvoyer:$\
% {N}{A}{D=r(n)}{B}{q1}{r1}{q2}{r2}{q3}{r3}....{qN}{rN=0}|
%    \begin{macrocode}
\def\XINT_euc_AisZero #1#2#3#4#5#6{{1}{0}{#2}{#2}{0}{0}}%
\def\XINT_euc_BisZero #1#2#3#4#5#6{{1}{0}{#3}{#3}{0}{0}}%
%    \end{macrocode}
% \lverb|&
% {n}{rn}{an}{{qn}{rn}}...{{A}{B}}{}\Z$\
%  a(n) = r(n-1). Pour n=0 on a juste {0}{B}{A}{{A}{B}}{}\Z$\
% \XINT_div_prepare {u}{v} divise v par u|
%    \begin{macrocode}
\def\XINT_euc_a #1#2#3%
{%
    \expandafter\XINT_euc_b\the\numexpr #1+\xint_c_i\expandafter.%
    \romannumeral0\XINT_div_prepare {#2}{#3}{#2}%
}%
%    \end{macrocode}
% \lverb|{n+1}{q(n+1)}{r(n+1)}{rn}{{qn}{rn}}...|
%    \begin{macrocode}
\def\XINT_euc_b #1.#2#3#4%
{%
    \XINT_euc_c #3\Z {#1}{#3}{#4}{{#2}{#3}}%
}%
%    \end{macrocode}
% \lverb|r(n+1)\Z {n+1}{r(n+1)}{r(n)}{{q(n+1)}{r(n+1)}}{{qn}{rn}}...$\
% Test si r(n+1) est nul.|
%    \begin{macrocode}
\def\XINT_euc_c #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_euc_end0\XINT_euc_a
}%
%    \end{macrocode}
% \lverb|&
% {n+1}{r(n+1)}{r(n)}{{q(n+1)}{r(n+1)}}...{}\Z
% Ici r(n+1) = 0. On arrête on se prépare à inverser
% {n+1}{0}{r(n)}{{q(n+1)}{r(n+1)}}.....{{q1}{r1}}{{A}{B}}{}\Z$\
% On veut renvoyer: {N=n+1}{A}{D=r(n)}{B}{q1}{r1}{q2}{r2}{q3}{r3}....{qN}{rN=0}|
%    \begin{macrocode}
\def\XINT_euc_end0\XINT_euc_a #1#2#3#4\Z%
{%
    \expandafter\XINT_euc_end_a
    \romannumeral0%
    \XINT_rord_main {}#4{{#1}{#3}}%
    \xint:
      \xint_bye\xint_bye\xint_bye\xint_bye
      \xint_bye\xint_bye\xint_bye\xint_bye
    \xint:
}%
\def\XINT_euc_end_a #1#2#3{{#1}{#3}{#2}}%
%    \end{macrocode}
% \subsection{\csh{xintBezoutAlgorithm}}
% \lverb|&
% Pour Bezout: objectif, renvoyer$\
% {N}{A}{0}{1}{D=r(n)}{B}{1}{0}{q1}{r1}{alpha1=q1}{beta1=1}$\
%       {q2}{r2}{alpha2}{beta2}....{qN}{rN=0}{alphaN=A/D}{betaN=B/D}$\
% alpha0=1, beta0=0, alpha(-1)=0, beta(-1)=1|
%    \begin{macrocode}
\def\xintBezoutAlgorithm {\romannumeral0\xintbezoutalgorithm }%
\def\xintbezoutalgorithm #1%
{%
    \expandafter \XINT_bezalg
    \expandafter{\romannumeral0\xintiiabs{\xintNum{#1}}}%
}%
\def\XINT_bezalg #1#2%
{%
    \expandafter\XINT_bezalg_fork\romannumeral0\xintiiabs{\xintNum{#2}}\Z #1\Z
}%
%    \end{macrocode}
% \lverb|Ici #3#4=A, #1#2=B|
%    \begin{macrocode}
\def\XINT_bezalg_fork #1#2\Z #3#4\Z
{%
    \xint_UDzerofork
      #1\XINT_bezalg_BisZero
      #3\XINT_bezalg_AisZero
       0\XINT_bezalg_a
    \krof
    0{#1#2}{#3#4}1001{{#3#4}{#1#2}}{}\Z
}%
\def\XINT_bezalg_AisZero #1#2#3\Z{{1}{0}{0}{1}{#2}{#2}{1}{0}{0}{0}{0}{1}}%
\def\XINT_bezalg_BisZero #1#2#3#4\Z{{1}{0}{0}{1}{#3}{#3}{1}{0}{0}{0}{0}{1}}%
%    \end{macrocode}
% \lverb|&
% pour préparer l'étape n+1 il faut
% {n}{r(n)}{r(n-1)}{alpha(n)}{beta(n)}{alpha(n-1)}{beta(n-1)}&
%                    {{q(n)}{r(n)}{alpha(n)}{beta(n)}}...
% division de #3 par #2|
%    \begin{macrocode}
\def\XINT_bezalg_a #1#2#3%
{%
    \expandafter\XINT_bezalg_b\the\numexpr #1+\xint_c_i\expandafter.%
    \romannumeral0\XINT_div_prepare {#2}{#3}{#2}%
}%
%    \end{macrocode}
% \lverb|&
% {n+1}{q(n+1)}{r(n+1)}{r(n)}{alpha(n)}{beta(n)}{alpha(n-1)}{beta(n-1)}...|
%    \begin{macrocode}
\def\XINT_bezalg_b #1.#2#3#4#5#6#7#8%
{%
    \expandafter\XINT_bezalg_c\expandafter
     {\romannumeral0\xintiiadd {\xintiiMul {#6}{#2}}{#8}}%
     {\romannumeral0\xintiiadd {\xintiiMul {#5}{#2}}{#7}}%
     {#1}{#2}{#3}{#4}{#5}{#6}%
}%
%    \end{macrocode}
% \lverb|&
% {beta(n+1)}{alpha(n+1)}{n+1}{q(n+1)}{r(n+1)}{r(n)}{alpha(n)}{beta(n}}|
%    \begin{macrocode}
\def\XINT_bezalg_c #1#2#3#4#5#6%
{%
    \expandafter\XINT_bezalg_d\expandafter {#2}{#3}{#4}{#5}{#6}{#1}%
}%
%    \end{macrocode}
% \lverb|{alpha(n+1)}{n+1}{q(n+1)}{r(n+1)}{r(n)}{beta(n+1)}|
%    \begin{macrocode}
\def\XINT_bezalg_d #1#2#3#4#5#6#7#8%
{%
    \XINT_bezalg_e #4\Z {#2}{#4}{#5}{#1}{#6}{#7}{#8}{{#3}{#4}{#1}{#6}}%
}%
%    \end{macrocode}
% \lverb|r(n+1)\Z {n+1}{r(n+1)}{r(n)}{alpha(n+1)}{beta(n+1)}$\
%                              {alpha(n)}{beta(n)}{q,r,alpha,beta(n+1)}$\
% Test si r(n+1) est nul.|
%    \begin{macrocode}
\def\XINT_bezalg_e #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_bezalg_end0\XINT_bezalg_a
}%
%    \end{macrocode}
% \lverb|&
% Ici r(n+1) = 0. On arrête on se prépare à inverser.$\
% {n+1}{r(n+1)}{r(n)}{alpha(n+1)}{beta(n+1)}{alpha(n)}{beta(n)}$\
%                     {q,r,alpha,beta(n+1)}...{{A}{B}}{}\Z$\
% On veut renvoyer$\
% {N}{A}{0}{1}{D=r(n)}{B}{1}{0}{q1}{r1}{alpha1=q1}{beta1=1}$\
%       {q2}{r2}{alpha2}{beta2}....{qN}{rN=0}{alphaN=A/D}{betaN=B/D}|
%    \begin{macrocode}
\def\XINT_bezalg_end0\XINT_bezalg_a #1#2#3#4#5#6#7#8\Z
{%
    \expandafter\XINT_bezalg_end_a
    \romannumeral0%
    \XINT_rord_main {}#8{{#1}{#3}}%
    \xint:
      \xint_bye\xint_bye\xint_bye\xint_bye
      \xint_bye\xint_bye\xint_bye\xint_bye
    \xint:
}%
%    \end{macrocode}
% \lverb|&
% {N}{D}{A}{B}{q1}{r1}{alpha1=q1}{beta1=1}{q2}{r2}{alpha2}{beta2}$\
%      ....{qN}{rN=0}{alphaN=A/D}{betaN=B/D}$\
% On veut renvoyer$\
% {N}{A}{0}{1}{D=r(n)}{B}{1}{0}{q1}{r1}{alpha1=q1}{beta1=1}$\
%        {q2}{r2}{alpha2}{beta2}....{qN}{rN=0}{alphaN=A/D}{betaN=B/D}|
%    \begin{macrocode}
\def\XINT_bezalg_end_a #1#2#3#4{{#1}{#3}{0}{1}{#2}{#4}{1}{0}}%
%    \end{macrocode}
% \subsection{\csh{xintTypesetEuclideAlgorithm}}
% \lverb|&
% TYPESETTING
%
% Organisation:
%
% {N}{A}{D}{B}{q1}{r1}{q2}{r2}{q3}{r3}....{qN}{rN=0}$\
% \U1 = N = nombre d'étapes, \U3 = PGCD, \U2 = A, \U4=B
% q1 = \U5, q2 = \U7 --> qn = \U<2n+3>, rn = \U<2n+4>
% bn = rn. B = r0. A=r(-1)
%
% r(n-2) = q(n)r(n-1)+r(n) (n e étape)
%
% \U{2n} = \U{2n+3} \times \U{2n+2} + \U{2n+4}, n e étape.
% (avec n entre 1 et N)
%
% 1.09h uses \xintloop, and \par rather than \endgraf; and \par rather than
% \hfill\break|
%    \begin{macrocode}
\def\xintTypesetEuclideAlgorithm {%
    \unless\ifdefined\xintAssignArray
       \errmessage
        {xintgcd: package xinttools is required for \string\xintTypesetEuclideAlgorithm}%
       \expandafter\xint_gobble_iii
    \fi
    \XINT_TypesetEuclideAlgorithm
}%
\def\XINT_TypesetEuclideAlgorithm #1#2%
{% l'algo remplace #1 et #2 par |#1| et |#2|
  \par
  \begingroup
    \xintAssignArray\xintEuclideAlgorithm {#1}{#2}\to\U
    \edef\A{\U2}\edef\B{\U4}\edef\N{\U1}%
    \setbox 0 \vbox{\halign {$##$\cr \A\cr \B \cr}}%
    \count 255 1
    \xintloop
      \indent\hbox to \wd 0 {\hfil$\U{\numexpr 2*\count255\relax}$}%
      ${} =  \U{\numexpr 2*\count255 + 3\relax}
      \times \U{\numexpr 2*\count255 + 2\relax}
          +  \U{\numexpr 2*\count255 + 4\relax}$%
    \ifnum \count255 < \N
      \par
      \advance \count255 1
    \repeat
  \endgroup
}%
%    \end{macrocode}
% \subsection{\csh{xintTypesetBezoutAlgorithm}}
% \lverb|&
% Pour Bezout on a:
% {N}{A}{0}{1}{D=r(n)}{B}{1}{0}{q1}{r1}{alpha1=q1}{beta1=1}$\
%             {q2}{r2}{alpha2}{beta2}....{qN}{rN=0}{alphaN=A/D}{betaN=B/D}%
% Donc 4N+8 termes:
% U1 = N, U2= A, U5=D, U6=B, q1 = U9, qn = U{4n+5}, n au moins 1$\
% rn = U{4n+6}, n au moins -1$\
% alpha(n) = U{4n+7}, n au moins -1$\
% beta(n)  = U{4n+8}, n au moins -1
%
% 1.09h uses \xintloop, and \par rather than \endgraf; and no more \parindent0pt
% |
%    \begin{macrocode}
\def\xintTypesetBezoutAlgorithm {%
    \unless\ifdefined\xintAssignArray
       \errmessage
        {xintgcd: package xinttools is required for \string\xintTypesetBezoutAlgorithm}%
       \expandafter\xint_gobble_iii
    \fi
    \XINT_TypesetBezoutAlgorithm
}%
\def\XINT_TypesetBezoutAlgorithm #1#2%
{%
  \par
  \begingroup
    \xintAssignArray\xintBezoutAlgorithm {#1}{#2}\to\BEZ
    \edef\A{\BEZ2}\edef\B{\BEZ6}\edef\N{\BEZ1}% A = |#1|, B = |#2|
    \setbox 0 \vbox{\halign {$##$\cr \A\cr \B \cr}}%
    \count255 1
    \xintloop
      \indent\hbox to \wd 0 {\hfil$\BEZ{4*\count255 - 2}$}%
      ${} =  \BEZ{4*\count255 + 5}
      \times \BEZ{4*\count255 + 2}
          +  \BEZ{4*\count255 + 6}$\hfill\break
      \hbox to \wd 0 {\hfil$\BEZ{4*\count255 +7}$}%
      ${} = \BEZ{4*\count255 + 5}
      \times \BEZ{4*\count255 + 3}
          +  \BEZ{4*\count255 - 1}$\hfill\break
      \hbox to \wd 0 {\hfil$\BEZ{4*\count255 +8}$}%
      ${} =  \BEZ{4*\count255 + 5}
      \times \BEZ{4*\count255 + 4}
          +  \BEZ{4*\count255 }$
      \par
    \ifnum \count255 < \N
    \advance \count255 1
  \repeat
    \edef\U{\BEZ{4*\N + 4}}%
    \edef\V{\BEZ{4*\N + 3}}%
    \edef\D{\BEZ5}%
    \ifodd\N
       $\U\times\A  - \V\times \B = -\D$%
    \else
       $\U\times\A - \V\times\B = \D$%
    \fi
    \par
  \endgroup
}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintgcd}
% \cleardoublepage\let\xintgcdnameUp\undefined
%\gardesactifs
%\let</xintgcd>\relax
%\let<*xintfrac>\gardesinactifs
%</xintgcd>^^A----------------------------------------------------
%<*xintfrac>^^A---------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintfracnameUp\endcsname
% \section{Package \xintfracnameimp implementation}
% \RaisedLabel{sec:fracimp}
%
% \localtableofcontents
%
% The commenting is currently (\xintdocdate) very sparse.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xintfrac.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xint.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintfrac}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintfrac.sty
      \ifx\w\relax % but xint.sty not yet loaded.
         \def\z{\endgroup\input xint.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xint.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xint}}%
          \fi
      \else
        \aftergroup\endinput % xintfrac already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintfrac}%
  [2021/07/13 v1.4j Expandable operations on fractions (JFB)]%
%    \end{macrocode}
% \subsection{\csh{XINT_cntSgnFork}}
% \lverb|1.09i. Used internally, #1 must expand to \m@ne, \z@, or \@ne or
% equivalent. \XINT_cntSgnFork does not insert a romannumeral stopper.|
%    \begin{macrocode}
\def\XINT_cntSgnFork #1%
{%
    \ifcase #1\expandafter\xint_secondofthree
            \or\expandafter\xint_thirdofthree
          \else\expandafter\xint_firstofthree
    \fi
}%
%    \end{macrocode}
% \subsection{\cshnolabel{xintLen}}
% \lverb|The used formula is disputable, the idea is that A/1 and A should have
% same length. Venerable code rewritten for 1.2i, following updates to
% \xintLength in xintkernel.sty. And sadly, I forgot on this
% occasion that this macro is not supposed to count the sign... Fixed in 1.2k.|
%    \begin{macrocode}
\def\xintLen {\romannumeral0\xintlen }%
\def\xintlen #1%
{%
    \expandafter\XINT_flen\romannumeral0\XINT_infrac {#1}%
}%
\def\XINT_flen#1{\def\XINT_flen ##1##2##3%
{%
    \expandafter#1%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \the\numexpr \XINT_abs##1+%
    \XINT_len_fork ##2##3\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye-\xint_c_i
    \relax
}}\XINT_flen{ }%
%    \end{macrocode}
% \subsection{\csh{XINT_outfrac}}
% \lverb|&
% Months later (2014/10/22): perhaps I should document what this macro does
% before I forget?  from {e}{N}{D} it outputs N/D[e], checking in passing if
% D=0 or if N=0. It also makes sure D is not < 0. I am not sure but I don't
% think there is any place in the code which could call \XINT_outfrac with a D
% < 0, but I should check.|
%    \begin{macrocode}
\def\XINT_outfrac #1#2#3%
{%
    \ifcase\XINT_cntSgn #3\xint:
        \expandafter \XINT_outfrac_divisionbyzero
    \or
        \expandafter \XINT_outfrac_P
    \else
        \expandafter \XINT_outfrac_N
    \fi
    {#2}{#3}[#1]%
}%
\def\XINT_outfrac_divisionbyzero #1#2[#3]%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #1/#2.}{}{ 0/1[0]}%
}%
\def\XINT_outfrac_P#1{%
\def\XINT_outfrac_P ##1##2%
   {\if0\XINT_Sgn ##1\xint:\expandafter\XINT_outfrac_Zero\fi#1##1/##2}%
}\XINT_outfrac_P{ }%
\def\XINT_outfrac_Zero #1[#2]{ 0/1[0]}%
\def\XINT_outfrac_N #1#2%
{%
    \expandafter\XINT_outfrac_N_a\expandafter
    {\romannumeral0\XINT_opp #2}{\romannumeral0\XINT_opp #1}%
}%
\def\XINT_outfrac_N_a #1#2%
{%
    \expandafter\XINT_outfrac_P\expandafter {#2}{#1}%
}%
%    \end{macrocode}
% \subsection{\csh{XINT_inFrac}}\label{src-XINT_infrac}
% \lverb|&
% Parses fraction, scientific notation, etc... and produces {n}{A}{B}
% corresponding to A/B times 10^n. No reduction to smallest terms.
%
% Extended in 1.07 to accept scientific notation on input. With lowercase
% e only. The \xintexpr parser does accept uppercase E also. Ah, by the way,
% perhaps I should at least say what this macro does? (belated addition
% 2014/10/22...), before I forget! It prepares the fraction in the internal
% format {exponent}{Numerator}{Denominator} where Denominator is at least 1.
%
% 2015/10/09: this venerable macro from the very early days (1.03, 2013/04/14)
% has gotten a lifting for release 1.2. There were two kinds of issues:
%
% 1) use of \W, \Z, \T delimiters was very poor choice as this could clash with
% user input,
%
% 2) the new \XINT_frac_gen handles macros (possibly empty) in the input as
% general as \A.\Be\C/\D.\Ee\F. The earlier version would not have expanded
% the \B or \E: digits after decimal mark were constrained to arise from
% expansion of the first token. Thus the 1.03 original code would have
% expanded only \A, \D, \C, and \F for this input.
%
% This reminded me think I should revisit the remaining earlier
% portions of code, as I was still learning TeX coding when I wrote them.
%
% Also I thought about parsing even faster the A/B[N] input, not expanding B,
% but this turned out to clash with some established uses in the documentation
% such as 1/\xintiiSqr{...}[0]. For the implementation, careful here about
% potential brace removals with parameter patterns such as like #1/#2#3[#4]for
% example.
%
% While I was at it 1.2 added \numexpr parsing of the N, which earlier was
% restricted to be only explicit digits. I allowed [] with empty N, but the
% way I did it in 1.2 with \the\numexpr 0#1 was buggy, as it did not allow #1
% to be a \count for example or itself a \numexpr (although such inputs were
% not previously allowed, I later turned out to use them in the code itself,
% e.g. the float factorial of version 1.2f). The better way would be
% \the\numexpr#1+\xint_c_ but 1.2f finally does only \the\numexpr #1 and #1 is
% not allowed to be empty.
%
% The 1.2 \XINT_frac_gen had two locations with such a problematic \numexpr
% 0#1 which I replaced for 1.2f with \numexpr#1+\xint_c_.
%
% Regarding calling the macro with an argument A[<expression>], a / inthe
% expression must be suitably hidden for example in \firstofone type
% constructs.
%
% Note: when the numerator is found to be zero \XINT_inFrac *always* returns
% {0}{0}{1}. This behaviour must not change because 1.2g \xintFloat and
% XINTinFloat (for example) rely upon it: if the denominator on output is not
% 1, then \xintFloat assumes that the numerator is not zero.
%
% As described in the manual, if the input contains a (final) [N] part, it is
% assumed that it is in the shape A[N] or A/B[N] with A (and B) not containing
% neither decimal mark nor scientific part, moreover B must be positive and A
% have at most one minus sign (and no plus sign). Else there will be errors,
% for example -0/2[0] would not be recognized as being zero at this stage and
% this could cause issues afterwards. When there is no ending [N] part, both
% numerator and denominator will be parsed for the more general format
% allowing decimal digits and scientific part and possibly multiple leading
% signs.
%
% 1.2l fixes frailty of \XINT_infrac (hence basically of all xintfrac macros)
% respective to non terminated \numexpr input: \xintRaw{\the\numexpr1} for
% example. The issue was that \numexpr sees the / and expands what's next.
% But even \numexpr 1// for example creates an error, and to my mind this is
% a defect of \numexpr. It should be able to trace back and see that / was
% used as delimiter not as operator. Anyway, I thus fixed this problem
% belatedly here regarding \XINT_infrac.
% |
%    \begin{macrocode}
\def\XINT_inFrac {\romannumeral0\XINT_infrac }%
\def\XINT_infrac #1%
{%
    \expandafter\XINT_infrac_fork\romannumeral`&&@#1\xint:/\XINT_W[\XINT_W\XINT_T
}%
\def\XINT_infrac_fork #1[#2%
{%
    \xint_UDXINTWfork
      #2\XINT_frac_gen          % input has no brackets [N]
      \XINT_W\XINT_infrac_res_a % there is some [N], must be strict A[N] or A/B[N] input
    \krof
    #1[#2%
}%
\def\XINT_infrac_res_a #1%
{%
    \xint_gob_til_zero #1\XINT_infrac_res_zero 0\XINT_infrac_res_b #1%
}%
%    \end{macrocode}
% \lverb|Note that input exponent is here ignored and forced to be zero.|
%    \begin{macrocode}
\def\XINT_infrac_res_zero 0\XINT_infrac_res_b #1\XINT_T {{0}{0}{1}}%
\def\XINT_infrac_res_b #1/#2%
{%
    \xint_UDXINTWfork
     #2\XINT_infrac_res_ca      % it was A[N] input
     \XINT_W\XINT_infrac_res_cb % it was A/B[N] input
    \krof
    #1/#2%
}%
%    \end{macrocode}
% \lverb|An empty [] is not allowed. (this was authorized in 1.2, removed in
% 1.2f). As nobody reads xint documentation, no one will have noticed the
% fleeting possibility.|
%    \begin{macrocode}
\def\XINT_infrac_res_ca #1[#2]\xint:/\XINT_W[\XINT_W\XINT_T
   {\expandafter{\the\numexpr #2}{#1}{1}}%
\def\XINT_infrac_res_cb #1/#2[%
   {\expandafter\XINT_infrac_res_cc\romannumeral`&&@#2~#1[}%
\def\XINT_infrac_res_cc #1~#2[#3]\xint:/\XINT_W[\XINT_W\XINT_T
   {\expandafter{\the\numexpr #3}{#2}{#1}}%
%    \end{macrocode}
% \subsection{\csh{XINT_frac_gen}}
% \lverb|Extended in 1.07 to recognize and accept scientific notation both at
% the numerator and (possible) denominator. Only a lowercase e will do here,
% but uppercase E is possible within an \xintexpr..\relax
%
% Completely rewritten for 1.2 2015/10/10. The parsing handles inputs such as
% \A.\Be\C/\D.\Ee\F where each of \A, \B, \D, and \E may need f-expansion and
% \C and \F will end up in \numexpr.
%
% 1.2f corrects an issue to allow \C and \F to be \count variable (or
% expressions with \numexpr): 1.2 did a bad \numexpr0#1 which allowed only
% explicit digits for expanded #1.|
%    \begin{macrocode}
\def\XINT_frac_gen #1/#2%
{%
    \xint_UDXINTWfork
      #2\XINT_frac_gen_A      % there was no /
      \XINT_W\XINT_frac_gen_B % there was a /
    \krof
    #1/#2%
}%
%    \end{macrocode}
% \lverb|Note that #1 is only expanded so far up to decimal mark or "e".|
%    \begin{macrocode}
\def\XINT_frac_gen_A #1\xint:/\XINT_W [\XINT_W {\XINT_frac_gen_C 0~1!#1ee.\XINT_W }%
\def\XINT_frac_gen_B #1/#2\xint:/\XINT_W[%\XINT_W
{%
    \expandafter\XINT_frac_gen_Ba
    \romannumeral`&&@#2ee.\XINT_W\XINT_Z #1ee.%\XINT_W
}%
\def\XINT_frac_gen_Ba #1.#2%
{%
    \xint_UDXINTWfork
      #2\XINT_frac_gen_Bb
      \XINT_W\XINT_frac_gen_Bc
    \krof
    #1.#2%
}%
\def\XINT_frac_gen_Bb #1e#2e#3\XINT_Z
                {\expandafter\XINT_frac_gen_C\the\numexpr #2+\xint_c_~#1!}%
\def\XINT_frac_gen_Bc #1.#2e%
{%
    \expandafter\XINT_frac_gen_Bd\romannumeral`&&@#2.#1e%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_frac_gen_Bd #1.#2e#3e#4\XINT_Z
{%
    \expandafter\XINT_frac_gen_C\the\numexpr #3-%
    \numexpr\XINT_length_loop
    #1\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    ~#2#1!%
}%
\def\XINT_frac_gen_C #1!#2.#3%
{%
    \xint_UDXINTWfork
      #3\XINT_frac_gen_Ca
      \XINT_W\XINT_frac_gen_Cb
    \krof
    #1!#2.#3%
}%
\def\XINT_frac_gen_Ca #1~#2!#3e#4e#5\XINT_T
{%
    \expandafter\XINT_frac_gen_F\the\numexpr #4-#1\expandafter
    ~\romannumeral0\expandafter\XINT_num_cleanup\the\numexpr\XINT_num_loop
     #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z~#3~%
}%
\def\XINT_frac_gen_Cb #1.#2e%
{%
    \expandafter\XINT_frac_gen_Cc\romannumeral`&&@#2.#1e%
}%
\def\XINT_frac_gen_Cc #1.#2~#3!#4e#5e#6\XINT_T
{%
    \expandafter\XINT_frac_gen_F\the\numexpr #5-#2-%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \numexpr\XINT_length_loop
    #1\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \relax\expandafter~%
    \romannumeral0\expandafter\XINT_num_cleanup\the\numexpr\XINT_num_loop
    #3\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z
    ~#4#1~%
}%
\def\XINT_frac_gen_F #1~#2%
{%
    \xint_UDzerominusfork
      #2-\XINT_frac_gen_Gdivbyzero
      0#2{\XINT_frac_gen_G  -{}}%
       0-{\XINT_frac_gen_G {}#2}%
    \krof  #1~%
}%
\def\XINT_frac_gen_Gdivbyzero #1~~#2~%
{%
   \expandafter\XINT_frac_gen_Gdivbyzero_a
   \romannumeral0\expandafter\XINT_num_cleanup\the\numexpr\XINT_num_loop
   #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z~#1~%
}%
\def\XINT_frac_gen_Gdivbyzero_a #1~#2~%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #1/0.}{}{{#2}{#1}{0}}%
}%
\def\XINT_frac_gen_G #1#2#3~#4~#5~%
{%
    \expandafter\XINT_frac_gen_Ga
    \romannumeral0\expandafter\XINT_num_cleanup\the\numexpr\XINT_num_loop
    #1#5\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\Z~#3~{#2#4}%
}%
\def\XINT_frac_gen_Ga #1#2~#3~%
{%
    \xint_gob_til_zero #1\XINT_frac_gen_zero 0%
    {#3}{#1#2}%
}%
\def\XINT_frac_gen_zero 0#1#2#3{{0}{0}{1}}%
%    \end{macrocode}
% \subsection{\csh{XINT_factortens}}
% \lverb|This is the core macro for \xintREZ. To be used as
% \romannumeral0\XINT_factortens{...}. Output is A.N. (formerly {A}{N}) where
% A is the integer stripped from trailing zeroes and N is the number of
% removed zeroes. Only for positive strict integers!
%
% Completely rewritten at 1.3a to replace a double \xintReverseOrder by a
% direct \numexpr governed expansion to the end and back, à la 1.2. I should
% comment more... and perhaps improve again in future.
%
% Testing shows significant gain at 100 digits or more.|
%    \begin{macrocode}
\def\XINT_factortens #1{\expandafter\XINT_factortens_z
                        \romannumeral0\XINT_factortens_a#1%
                        \XINT_factortens_b123456789.}%
\def\XINT_factortens_z.\XINT_factortens_y{ }%
\def\XINT_factortens_a #1#2#3#4#5#6#7#8#9%
   {\expandafter\XINT_factortens_x
    \the\numexpr 1#1#2#3#4#5#6#7#8#9\XINT_factortens_a}%
\def\XINT_factortens_b#1\XINT_factortens_a#2#3.%
   {.\XINT_factortens_cc 000000000-#2.}%
\def\XINT_factortens_x1#1.#2{#2#1}%
\def\XINT_factortens_y{.\XINT_factortens_y}%
\def\XINT_factortens_cc #1#2#3#4#5#6#7#8#9%
   {\if#90\xint_dothis
     {\expandafter\XINT_factortens_d\the\numexpr #8#7#6#5#4#3#2#1\relax
      \xint_c_i 2345678.}\fi
    \xint_orthat{\XINT_factortens_yy{#1#2#3#4#5#6#7#8#9}}}%
\def\XINT_factortens_yy #1#2.{.\XINT_factortens_y#1.0.}%
\def\XINT_factortens_c #1#2#3#4#5#6#7#8#9%
   {\if#90\xint_dothis
     {\expandafter\XINT_factortens_d\the\numexpr #8#7#6#5#4#3#2#1\relax
      \xint_c_i 2345678.}\fi
    \xint_orthat{.\XINT_factortens_y #1#2#3#4#5#6#7#8#9.}}%
\def\XINT_factortens_d #1#2#3#4#5#6#7#8#9%
   {\if#10\expandafter\XINT_factortens_e\fi
    \XINT_factortens_f #9#9#8#7#6#5#4#3#2#1.}%
\def\XINT_factortens_f #1#2\xint_c_i#3.#4.#5.%
   {\expandafter\XINT_factortens_g\the\numexpr#1+#5.#3.}%
\def\XINT_factortens_g #1.#2.{.\XINT_factortens_y#2.#1.}%
\def\XINT_factortens_e #1..#2.%
   {\expandafter.\expandafter\XINT_factortens_c
    \the\numexpr\xint_c_ix+#2.}%
%    \end{macrocode}
% \subsection{\xintListWithSep{, }
%             {\xintApply{ \csh}{{xintEq}{xintNotEq}{xintGt}{xintLt}{xintGtorEq}
%             {xintLtorEq}{xintIsZero}{xintIsNotZero}{xintOdd}
%             {xintEven}{xintifSgn}{xintifCmp}{xintifEq}{xintifGt}{xintifLt}
%             {xintifZero}{xintifNotZero}{xintifOne}{xintifOdd}}}}
%
% \lverb|Moved here at 1.3. Formerly these macros were already defined in
% xint.sty or even xintcore.sty. They are slim wrappers of macros defined
% elsewhere in xintfrac.
% |
%    \begin{macrocode}
\def\xintEq   {\romannumeral0\xinteq }%
\def\xinteq   #1#2{\xintifeq{#1}{#2}{1}{0}}%
\def\xintNotEq#1#2{\romannumeral0\xintifeq {#1}{#2}{0}{1}}%
\def\xintGt {\romannumeral0\xintgt }%
\def\xintgt #1#2{\xintifgt{#1}{#2}{1}{0}}%
\def\xintLt   {\romannumeral0\xintlt }%
\def\xintlt #1#2{\xintiflt{#1}{#2}{1}{0}}%
\def\xintGtorEq #1#2{\romannumeral0\xintiflt {#1}{#2}{0}{1}}%
\def\xintLtorEq #1#2{\romannumeral0\xintifgt {#1}{#2}{0}{1}}%
\def\xintIsZero   {\romannumeral0\xintiszero }%
\def\xintiszero #1{\if0\xintSgn{#1}\xint_afterfi{ 1}\else\xint_afterfi{ 0}\fi}%
\def\xintIsNotZero{\romannumeral0\xintisnotzero }%
\def\xintisnotzero
          #1{\if0\xintSgn{#1}\xint_afterfi{ 0}\else\xint_afterfi{ 1}\fi}%
\def\xintOdd      {\romannumeral0\xintodd }%
\def\xintodd #1%
{%
    \ifodd\xintLDg{\xintNum{#1}} %<- intentional space
        \xint_afterfi{ 1}%
    \else
        \xint_afterfi{ 0}%
    \fi
}%
\def\xintEven     {\romannumeral0\xinteven }%
\def\xinteven #1%
{%
    \ifodd\xintLDg{\xintNum{#1}} %<- intentional space
        \xint_afterfi{ 0}%
    \else
        \xint_afterfi{ 1}%
    \fi
}%
\def\xintifSgn{\romannumeral0\xintifsgn }%
\def\xintifsgn #1%
{%
    \ifcase \xintSgn{#1}
               \expandafter\xint_stop_atsecondofthree
            \or\expandafter\xint_stop_atthirdofthree
          \else\expandafter\xint_stop_atfirstofthree
    \fi
}%
\def\xintifCmp{\romannumeral0\xintifcmp }%
\def\xintifcmp #1#2%
{%
    \ifcase\xintCmp {#1}{#2}
               \expandafter\xint_stop_atsecondofthree
            \or\expandafter\xint_stop_atthirdofthree
          \else\expandafter\xint_stop_atfirstofthree
    \fi
}%
\def\xintifEq {\romannumeral0\xintifeq }%
\def\xintifeq #1#2%
{%
    \if0\xintCmp{#1}{#2}%
               \expandafter\xint_stop_atfirstoftwo
          \else\expandafter\xint_stop_atsecondoftwo
    \fi
}%
\def\xintifGt {\romannumeral0\xintifgt }%
\def\xintifgt #1#2%
{%
    \if1\xintCmp{#1}{#2}%
               \expandafter\xint_stop_atfirstoftwo
          \else\expandafter\xint_stop_atsecondoftwo
    \fi
}%
\def\xintifLt {\romannumeral0\xintiflt }%
\def\xintiflt #1#2%
{%
    \ifnum\xintCmp{#1}{#2}<\xint_c_
          \expandafter\xint_stop_atfirstoftwo
    \else \expandafter\xint_stop_atsecondoftwo
    \fi
}%
\def\xintifZero   {\romannumeral0\xintifzero }%
\def\xintifzero #1%
{%
    \if0\xintSgn{#1}%
       \expandafter\xint_stop_atfirstoftwo
    \else
       \expandafter\xint_stop_atsecondoftwo
    \fi
}%
\def\xintifNotZero{\romannumeral0\xintifnotzero }%
\def\xintifnotzero #1%
{%
    \if0\xintSgn{#1}%
       \expandafter\xint_stop_atsecondoftwo
    \else
       \expandafter\xint_stop_atfirstoftwo
    \fi
}%
\def\xintifOne {\romannumeral0\xintifone }%
\def\xintifone #1%
{%
    \if1\xintIsOne{#1}%
       \expandafter\xint_stop_atfirstoftwo
    \else
       \expandafter\xint_stop_atsecondoftwo
    \fi
}%
\def\xintifOdd {\romannumeral0\xintifodd }%
\def\xintifodd #1%
{%
    \if\xintOdd{#1}1%
       \expandafter\xint_stop_atfirstoftwo
    \else
       \expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintRaw}}
% \lverb|&
% 1.07: this macro simply prints in a user readable form the fraction after its
% initial scanning. Useful when put inside braces in an \xintexpr, when the
% input is not yet in the A/B[n] form.|
%    \begin{macrocode}
\def\xintRaw {\romannumeral0\xintraw }%
\def\xintraw
{%
    \expandafter\XINT_raw\romannumeral0\XINT_infrac
}%
\def\XINT_raw #1#2#3{ #2/#3[#1]}%
%    \end{macrocode}
% \subsection{\csh{xintiLogTen}}
% \lverb|&
% New at 1.3e. The exponent a, such that 10^a<= abs(x) < 10^(a+1).
% |
%    \begin{macrocode}
\def\xintiLogTen {\the\numexpr\xintilogten}%
\def\xintilogten
{%
    \expandafter\XINT_ilogten\romannumeral0\xintraw
}%
\def\XINT_ilogten #1%
{%
    \xint_UDzerominusfork
      0#1\XINT_ilogten_p
      #1-\XINT_ilogten_z
      0-{\XINT_ilogten_p#1}%
    \krof
}%
\def\XINT_ilogten_z #1[#2]{-"7FFF8000\relax}%
\def\XINT_ilogten_p #1/#2[#3]%
{%
    #3+\expandafter\XINT_ilogten_a
       \the\numexpr\xintLength{#1}\expandafter.\the\numexpr\xintLength{#2}.#1.#2.%
}%
\def\XINT_ilogten_a #1.#2.%
{%
    #1-#2\ifnum#1>#2
      \expandafter\XINT_ilogten_aa
    \else
      \expandafter\XINT_ilogten_ab
    \fi #1.#2.%
}%
\def\XINT_ilogten_aa #1.#2.#3.#4.%
{%
    \xintiiifLt{#3}{\XINT_dsx_addzerosnofuss{#1-#2}#4;}{-1}{}\relax
}%
\def\XINT_ilogten_ab #1.#2.#3.#4.%
{%
    \xintiiifLt{\XINT_dsx_addzerosnofuss{#2-#1}#3;}{#4}{-1}{}\relax
}%
%    \end{macrocode}
% \subsection{\csh{xintPRaw}}
% \lverb|1.09b|
%    \begin{macrocode}
\def\xintPRaw {\romannumeral0\xintpraw }%
\def\xintpraw
{%
    \expandafter\XINT_praw\romannumeral0\XINT_infrac
}%
\def\XINT_praw #1%
{%
    \ifnum #1=\xint_c_ \expandafter\XINT_praw_a\fi \XINT_praw_A {#1}%
}%
\def\XINT_praw_A #1#2#3%
{%
    \if\XINT_isOne{#3}1\expandafter\xint_firstoftwo
                  \else\expandafter\xint_secondoftwo
    \fi { #2[#1]}{ #2/#3[#1]}%
}%
\def\XINT_praw_a\XINT_praw_A #1#2#3%
{%
    \if\XINT_isOne{#3}1\expandafter\xint_firstoftwo
                  \else\expandafter\xint_secondoftwo
    \fi { #2}{ #2/#3}%
}%
%    \end{macrocode}
% \subsection{\csh{xintSPRaw}}
% \lverb|This private macro was for internal usage by \xinttheexpr.
% It got moved here
% at 1.4 and is not used anymore by the package.
%
% It checks if input has a [N] part, if yes uses \xintPRaw, else
% simply lets the input pass through as is.
% |
%    \begin{macrocode}
\def\xintSPRaw    {\romannumeral0\xintspraw }%
\def\xintspraw  #1{\expandafter\XINT_spraw\romannumeral`&&@#1[\W]}%
\def\XINT_spraw #1[#2#3]{\xint_gob_til_W #2\XINT_spraw_a\W\XINT_spraw_p #1[#2#3]}%
\def\XINT_spraw_a\W\XINT_spraw_p #1[\W]{ #1}%
\def\XINT_spraw_p #1[\W]{\xintpraw {#1}}%
%    \end{macrocode}
% \subsection{\csh{xintFracToSci}}
% \lverb|1.4, refactored and much simplified at 1.4e.
%
% It only needs to be x-expandable, and indeed the implementation here is only
% x-expandable.
%
% (2021/04/13) the user documentation was really deplorable, I have
% tried to improve it and in the process tried to remember what this macro was
% supposed to do, and improved comments here, also lamentable.
%
% At 1.4e-dev this became provisorily basically like defunct \xintSPRaw, but
% doing less parsing at it does not go to \xintPRaw with its \XINT_infrac
% induced overhead.  Previous 1.4b \xintFracToSci was much complicated from
% having to allow fixed point notation on input and scientific notation with a
% catcode 12 "e".  Refactoring of \xintiexpr has removed these constraints.
% Now:
%
%( Input:  A, A/B, A[N], A/B[N]
%: Output: AeN/B with special cases:
%:         0 if input gives a zero value
%:         /B is skipped in output if B=1 in input
%:         eN is skipped in output if N=0 in input
%)
%
% 0[N] when N not zero is possible as input, but 0/B currently not I think,
% and -0 for example never arises as one is guaranteed that A is in strict
% integer format.
%
% (2021/05/05) Finally for 1.4e release I  modify. This is breaking
% change for all \xinteval output in case of scientific notation: it will not
% be with an integer mantissa, but in regular scientific notation, using the
% same rules as \xintPFloat.
%
% Of course there will be no float rounding applied! Also, as [0] will always
% or almost always be present from an \xinteval, we want then to use integer
% not scientific notation. But expression contained decimal fixed point input,
% or uses scientific functions, then probably the N will not be zero and this
% will trigger usage of scientific notation in output.
%
% Implementing these changes sort of ruin our previous efforts to minimize
% grabbing the argument, but well. So the rules now are
%
%( Input:  A, A/B, A[N], A/B[N]
%: Output: A, A/B, A if N=0, A/B if N=0
%:         If N is not zero, scientific notation like \xintPFloat,&
%          i.e. behaviour like \xintfloateval apart from the rounding&
%          to Digits. In particular trailing zeros are trimmed.
%: The zero gives 0, except in A[N] and A/B[N] cases, it may give&
%  0.0
%)
%
% As a result of these last minute 1.4e changes, the \xintFracToSciE is
% removed.
% |
%
%    \begin{macrocode}
\def\xintFracToSci #1{\expandafter\XINT_FracToSci\romannumeral`&&@#1/\W[\R}%
\def\XINT_FracToSci #1/#2#3[#4%
{%
    \xint_gob_til_W #2\XINT_FracToSci_noslash\W
    \xint_gob_til_R #4\XINT_FracToSci_slash_noN\R
    \XINT_FracToSci_slash_N #1/#2#3[#4%
}%
\def\XINT_FracToSci_noslash#1\XINT_FracToSci_slash_N #2[#3%
{%
    \xint_gob_til_R #3\XINT_FracToSci_noslash_noN\R
    \XINT_FracToSci_noslash_N #2[#3%
}%
\def\XINT_FracToSci_noslash_noN\R\XINT_FracToSci_noslash_N #1/\W[\R{#1}%
\def\XINT_FracToSci_noslash_N #1[#2]/\W[\R%
{%
    \ifnum#2=\xint_c_ #1\else
      \romannumeral0\expandafter\XINT_pfloat_fork\romannumeral0\xintrez{#1[#2]}%
    \fi
}%
\def\XINT_FracToSci_slash_noN\R\XINT_FracToSci_slash_N #1#2/#3/\W[\R%
{%
    #1\xint_gob_til_zero#1\expandafter\iffalse\xint_gobble_ii0\iftrue
    #2\if\XINT_isOne{#3}1\else/#3\fi\fi
}%
\def\XINT_FracToSci_slash_N #1#2/#3[#4]/\W[\R%
{%
    \ifnum#4=\xint_c_ #1#2\else
      \romannumeral0\expandafter\XINT_pfloat_fork\romannumeral0\xintrez{#1#2[#4]}%
    \fi
    \if\XINT_isOne{#3}1\else\if#10\else/#3\fi\fi
}%
%    \end{macrocode}
% \subsection{\csh{xintRawWithZeros}}
% \lverb|&
% This was called \xintRaw in versions earlier than 1.07|
%    \begin{macrocode}
\def\xintRawWithZeros {\romannumeral0\xintrawwithzeros }%
\def\xintrawwithzeros
{%
    \expandafter\XINT_rawz_fork\romannumeral0\XINT_infrac
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_rawz_fork #1%
{%
    \ifnum#1<\xint_c_
      \expandafter\XINT_rawz_Ba
    \else
      \expandafter\XINT_rawz_A
    \fi
    #1.%
}%
\def\XINT_rawz_A  #1.#2#3{\XINT_dsx_addzeros{#1}#2;/#3}%
\def\XINT_rawz_Ba -#1.#2#3{\expandafter\XINT_rawz_Bb
    \expandafter{\romannumeral0\XINT_dsx_addzeros{#1}#3;}{#2}}%
\def\XINT_rawz_Bb #1#2{ #2/#1}%
%    \end{macrocode}
% \subsection{\csh{xintDecToString}}
% \lverb|1.3. This is a backport from polexpr 0.4. It is definitely not in
% final form, consider it to be an unstable macro.|
%    \begin{macrocode}
\def\xintDecToString{\romannumeral0\xintdectostring}%
\def\xintdectostring#1{\expandafter\XINT_dectostr\romannumeral0\xintraw{#1}}%
\def\XINT_dectostr #1/#2[#3]{\xintiiifZero {#1}%
       \XINT_dectostr_z
       {\if1\XINT_isOne{#2}\expandafter\XINT_dectostr_a
                      \else\expandafter\XINT_dectostr_b
        \fi}%
  #1/#2[#3]%
}%
\def\XINT_dectostr_z#1[#2]{ 0}%
\def\XINT_dectostr_a#1/#2[#3]{%
    \ifnum#3<\xint_c_\xint_dothis{\xinttrunc{-#3}{#1[#3]}}\fi
    \xint_orthat{\xintiie{#1}{#3}}%
}%
\def\XINT_dectostr_b#1/#2[#3]{% just to handle this somehow
    \ifnum#3<\xint_c_\xint_dothis{\xinttrunc{-#3}{#1[#3]}/#2}\fi
    \xint_orthat{\xintiie{#1}{#3}/#2}%
}%
%    \end{macrocode}
% \subsection{\csh{xintDecToStringREZ}}
% \lverb|1.4e. And I took this opportunity to improve documentation in manual.|
%    \begin{macrocode}
\def\xintDecToStringREZ{\romannumeral0\xintdectostringrez}%
\def\xintdectostringrez#1{\expandafter\XINT_dectostr\romannumeral0\xintrez{#1}}%
%    \end{macrocode}
% \subsection{\csh{xintFloor}, \csh{xintiFloor}}
% \lverb|1.09a, 1.1 for \xintiFloor/\xintFloor. Not efficient if big negative
% decimal exponent. Also sub-efficient if big positive decimal exponent.|
%    \begin{macrocode}
\def\xintFloor {\romannumeral0\xintfloor }%
\def\xintfloor #1% devrais-je faire \xintREZ?
    {\expandafter\XINT_ifloor \romannumeral0\xintrawwithzeros {#1}./1[0]}%
\def\xintiFloor {\romannumeral0\xintifloor }%
\def\xintifloor #1%
    {\expandafter\XINT_ifloor \romannumeral0\xintrawwithzeros {#1}.}%
\def\XINT_ifloor #1/#2.{\xintiiquo {#1}{#2}}%
%    \end{macrocode}
% \subsection{\csh{xintCeil}, \csh{xintiCeil}}
% \lverb|1.09a|
%    \begin{macrocode}
\def\xintCeil {\romannumeral0\xintceil }%
\def\xintceil #1{\xintiiopp {\xintFloor {\xintOpp{#1}}}}%
\def\xintiCeil {\romannumeral0\xinticeil }%
\def\xinticeil #1{\xintiiopp {\xintiFloor {\xintOpp{#1}}}}%
%    \end{macrocode}
% \subsection{\csh{xintNumerator}}
%    \begin{macrocode}
\def\xintNumerator {\romannumeral0\xintnumerator }%
\def\xintnumerator
{%
    \expandafter\XINT_numer\romannumeral0\XINT_infrac
}%
\def\XINT_numer #1%
{%
    \ifcase\XINT_cntSgn #1\xint:
      \expandafter\XINT_numer_B
    \or
      \expandafter\XINT_numer_A
    \else
      \expandafter\XINT_numer_B
    \fi
    {#1}%
}%
\def\XINT_numer_A #1#2#3{\XINT_dsx_addzeros{#1}#2;}%
\def\XINT_numer_B #1#2#3{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintDenominator}}
%    \begin{macrocode}
\def\xintDenominator {\romannumeral0\xintdenominator }%
\def\xintdenominator
{%
    \expandafter\XINT_denom_fork\romannumeral0\XINT_infrac
}%
\def\XINT_denom_fork #1%
{%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \ifnum#1<\xint_c_
      \expandafter\XINT_denom_B
    \else
      \expandafter\XINT_denom_A
    \fi
    #1.%
}%
\def\XINT_denom_A #1.#2#3{ #3}%
\def\XINT_denom_B -#1.#2#3{\XINT_dsx_addzeros{#1}#3;}%
%    \end{macrocode}
% \subsection{\csh{xintTeXFrac}}
% \lverb|1.03 (2013/04/14). Useless typesetting macro.
%
% Renamed  (2021/05/24) from \xintFrac at 1.4g.  Old name deprecated but
% still usable.|
%    \begin{macrocode}
\ifdefined\documentclass
\def\xintfracTeXDeprecation#1#2{%
\PackageWarning{xintfrac}{\string#1 is deprecated. Use \string#2\MessageBreak
                          to suppress this warning}#2%
}%
\else
\edef\xintfracTeXDeprecation#1#2{{\newlinechar10
\immediate\noexpand\write128{&&JPackage xintfrac Warning: \noexpand\string#1 is
  deprecated. Use \noexpand\string#2&&J%
(xintfrac)\xintReplicate{16}{ }to suppress this warning
on input line \noexpand\the\inputlineno.&&J}}#2%
}%
\fi
\def\xintFrac   {\xintfracTeXDeprecation\xintFrac\xintTeXFrac}%
\def\xintTeXFrac{\romannumeral0\xintfrac }%
\def\xintfrac #1%
{%
    \expandafter\XINT_fracfrac_A\romannumeral0\XINT_infrac {#1}%
}%
\def\XINT_fracfrac_A #1{\XINT_fracfrac_B #1\Z }%
\catcode`^=7
\def\XINT_fracfrac_B #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_fracfrac_C 0\XINT_fracfrac_D {10^{#1#2}}%
}%
\def\XINT_fracfrac_C 0\XINT_fracfrac_D #1#2#3%
{%
    \if1\XINT_isOne {#3}%
        \xint_afterfi {\expandafter\xint_stop_atfirstoftwo\xint_gobble_ii }%
    \fi
    \space
    \frac {#2}{#3}%
}%
\def\XINT_fracfrac_D #1#2#3%
{%
    \if1\XINT_isOne {#3}\XINT_fracfrac_E\fi
    \space
    \frac {#2}{#3}#1%
}%
\def\XINT_fracfrac_E \fi\space\frac #1#2{\fi \space #1\cdot }%
%    \end{macrocode}
% \subsection{\csh{xintTeXsignedFrac}}
% \lverb|&
% Renamed  (2021/05/24) from \xintSignedFrac at 1.4g.  Old name deprecated but
% still usable.|
%    \begin{macrocode}
\def\xintSignedFrac   {\xintfracTeXDeprecation\xintSignedFrac\xintTeXsignedFrac}%
\def\xintTeXsignedFrac{\romannumeral0\xintsignedfrac }%
\def\xintsignedfrac #1%
{%
    \expandafter\XINT_sgnfrac_a\romannumeral0\XINT_infrac {#1}%
}%
\def\XINT_sgnfrac_a #1#2%
{%
    \XINT_sgnfrac_b #2\Z {#1}%
}%
\def\XINT_sgnfrac_b #1%
{%
    \xint_UDsignfork
      #1\XINT_sgnfrac_N
       -{\XINT_sgnfrac_P #1}%
    \krof
}%
\def\XINT_sgnfrac_P #1\Z #2%
{%
    \XINT_fracfrac_A {#2}{#1}%
}%
\def\XINT_sgnfrac_N
{%
    \expandafter-\romannumeral0\XINT_sgnfrac_P
}%
%    \end{macrocode}
% \subsection{\csh{xintTeXfromSci}}
% \lverb|1.4g. The main problem is how to name this and related macros.
%
% I use \expanded here, as \xintFracToSci is not f-expandable. But why do I
% bother with the external \expanded?
%
% Some complications as I want this to be usable on output of \xintFracToSci
% hence need to handle the case of a /B.  After some hesitations I ended with
% the following which looks reasonable:
%(- if no scientific part, use \frac (or \over) for A/B
%:- if scientific part, postfix /B as \cdot B^{-1}
%)
% |
%    \begin{macrocode}
\def\xintTeXfromSci#1%
{%
    \expanded{\expandafter\XINT_texfromsci\expanded{#1}/\relax/\xint:}%
}%
\def\XINT_texfromsci #1/#2#3/#4\xint:
{%
    \XINT_texfromsci_a #1e\relax e\xint:
    {\ifx\relax#2\xint_dothis\xint_firstofone\fi
     \xint_orthat{\XINT_texfromsci_frac{#2#3}}}%
    {\unless\ifx\relax#2\cdot{#2#3}^{-1}\fi}%
}%
\def\XINT_texfromsci_a #1e#2#3e#4\xint:#5#6%
{%
    \ifx\relax#2#5{#1}\else#1\cdot10^{#2#3}#6\fi
}%
\ifdefined\frac
  \def\XINT_texfromsci_frac#1#2{\noexpand\frac{#2}{#1}}%
\else
  \def\XINT_texfromsci_frac#1#2{{#2\over#1}}%
\fi
%    \end{macrocode}
% \subsection{\csh{xintTeXOver}}
% \lverb|&
% Renamed  (2021/05/24) from \xintFwOver at 1.4g.  Old name deprecated but
% still usable.|
%    \begin{macrocode}
\def\xintFwOver {\xintfracTeXDeprecation\xintFwOver\xintTeXOver}%
\def\xintTeXOver{\romannumeral0\xintfwover }%
\def\xintfwover #1%
{%
    \expandafter\XINT_fwover_A\romannumeral0\XINT_infrac {#1}%
}%
\def\XINT_fwover_A #1{\XINT_fwover_B #1\Z }%
\def\XINT_fwover_B #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_fwover_C 0\XINT_fwover_D {10^{#1#2}}%
}%
\catcode`^=11
\def\XINT_fwover_C #1#2#3#4#5%
{%
    \if0\XINT_isOne {#5}\xint_afterfi { {#4\over #5}}%
                   \else\xint_afterfi { #4}%
    \fi
}%
\def\XINT_fwover_D #1#2#3%
{%
    \if0\XINT_isOne {#3}\xint_afterfi { {#2\over #3}}%
                   \else\xint_afterfi { #2\cdot }%
    \fi
    #1%
}%
%    \end{macrocode}
% \subsection{\csh{xintTeXsignedOver}}
% \lverb|&
% Renamed  (2021/05/24) from \xintSignedFwOver at 1.4g.  Old name deprecated but
% still usable.|
%    \begin{macrocode}
\def\xintSignedFwOver {\xintfracTeXDeprecation\xintSignedFwOver\xintTeXsignedOver}%
\def\xintTeXsignedOver{\romannumeral0\xintsignedfwover }%
\def\xintsignedfwover #1%
{%
    \expandafter\XINT_sgnfwover_a\romannumeral0\XINT_infrac {#1}%
}%
\def\XINT_sgnfwover_a #1#2%
{%
    \XINT_sgnfwover_b #2\Z {#1}%
}%
\def\XINT_sgnfwover_b #1%
{%
    \xint_UDsignfork
      #1\XINT_sgnfwover_N
       -{\XINT_sgnfwover_P #1}%
    \krof
}%
\def\XINT_sgnfwover_P #1\Z #2%
{%
    \XINT_fwover_A {#2}{#1}%
}%
\def\XINT_sgnfwover_N
{%
    \expandafter-\romannumeral0\XINT_sgnfwover_P
}%
%    \end{macrocode}
% \subsection{\csh{xintREZ}}
% \lverb|Removes trailing zeros from A and B and adjust the N in A/B[N].
%
% The macro really doing the job \XINT_factortens was redone at 1.3a. But
% speed gain really noticeable only beyond about 100 digits.|
%    \begin{macrocode}
\def\xintREZ {\romannumeral0\xintrez }%
\def\xintrez
{%
    \expandafter\XINT_rez_A\romannumeral0\XINT_infrac
}%
\def\XINT_rez_A #1#2%
{%
    \XINT_rez_AB #2\Z {#1}%
}%
\def\XINT_rez_AB #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_rez_zero
      0#1\XINT_rez_neg
       0-{\XINT_rez_B #1}%
    \krof
}%
\def\XINT_rez_zero #1\Z #2#3{ 0/1[0]}%
\def\XINT_rez_neg {\expandafter-\romannumeral0\XINT_rez_B }%
\def\XINT_rez_B #1\Z
{%
    \expandafter\XINT_rez_C\romannumeral0\XINT_factortens {#1}%
}%
\def\XINT_rez_C #1.#2.#3#4%
{%
    \expandafter\XINT_rez_D\romannumeral0\XINT_factortens {#4}#3+#2.#1.%
}%
\def\XINT_rez_D #1.#2.#3.%
{%
    \expandafter\XINT_rez_E\the\numexpr #3-#2.#1.%
}%
\def\XINT_rez_E #1.#2.#3.{ #3/#2[#1]}%
%    \end{macrocode}
% \subsection{\csh{xintE}}
% \lverb|1.07: The fraction is the first argument contrarily to \xintTrunc and
% \xintRound.
%
% 1.1 modifies and moves \xintiiE to xint.sty.|
%    \begin{macrocode}
\def\xintE {\romannumeral0\xinte }%
\def\xinte #1%
{%
    \expandafter\XINT_e \romannumeral0\XINT_infrac {#1}%
}%
\def\XINT_e #1#2#3#4%
{%
    \expandafter\XINT_e_end\the\numexpr #1+#4.{#2}{#3}%
}%
\def\XINT_e_end #1.#2#3{ #2/#3[#1]}%
%    \end{macrocode}
% \subsection{\csh{xintIrr}, \csh{xintPIrr}}
%  \lverb|\xintPIrr (partial Irr, which ignores the decimal part) added at 1.3.|
%    \begin{macrocode}
\def\xintIrr {\romannumeral0\xintirr }%
\def\xintPIrr{\romannumeral0\xintpirr }%
\def\xintirr #1%
{%
    \expandafter\XINT_irr_start\romannumeral0\xintrawwithzeros {#1}\Z
}%
\def\xintpirr #1%
{%
    \expandafter\XINT_pirr_start\romannumeral0\xintraw{#1}%
}%
\def\XINT_irr_start #1#2/#3\Z
{%
    \if0\XINT_isOne {#3}%
      \xint_afterfi
          {\xint_UDsignfork
               #1\XINT_irr_negative
                -{\XINT_irr_nonneg #1}%
           \krof}%
    \else
      \xint_afterfi{\XINT_irr_denomisone #1}%
    \fi
    #2\Z {#3}%
}%
\def\XINT_pirr_start #1#2/#3[%
{%
    \if0\XINT_isOne {#3}%
      \xint_afterfi
          {\xint_UDsignfork
               #1\XINT_irr_negative
                -{\XINT_irr_nonneg #1}%
           \krof}%
    \else
      \xint_afterfi{\XINT_irr_denomisone #1}%
    \fi
    #2\Z {#3}[%
}%
\def\XINT_irr_denomisone #1\Z #2{ #1/1}% changed in 1.08
\def\XINT_irr_negative   #1\Z #2{\XINT_irr_D #1\Z #2\Z -}%
\def\XINT_irr_nonneg     #1\Z #2{\XINT_irr_D #1\Z #2\Z \space}%
\def\XINT_irr_D #1#2\Z #3#4\Z
{%
    \xint_UDzerosfork
       #3#1\XINT_irr_indeterminate
       #30\XINT_irr_divisionbyzero
       #10\XINT_irr_zero
        00\XINT_irr_loop_a
    \krof
    {#3#4}{#1#2}{#3#4}{#1#2}%
}%
\def\XINT_irr_indeterminate #1#2#3#4#5%
{%
    \XINT_signalcondition{DivisionUndefined}{0/0 indeterminate fraction.}{}{ 0/1}%
}%
\def\XINT_irr_divisionbyzero #1#2#3#4#5%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #5#2/0.}{}{ 0/1}%
}%
\def\XINT_irr_zero #1#2#3#4#5{ 0/1}% changed in 1.08
\def\XINT_irr_loop_a #1#2%
{%
    \expandafter\XINT_irr_loop_d
    \romannumeral0\XINT_div_prepare {#1}{#2}{#1}%
}%
\def\XINT_irr_loop_d #1#2%
{%
    \XINT_irr_loop_e #2\Z
}%
\def\XINT_irr_loop_e #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_irr_loop_exit0\XINT_irr_loop_a {#1#2}%
}%
\def\XINT_irr_loop_exit0\XINT_irr_loop_a #1#2#3#4%
{%
    \expandafter\XINT_irr_loop_exitb\expandafter
    {\romannumeral0\xintiiquo {#3}{#2}}%
    {\romannumeral0\xintiiquo {#4}{#2}}%
}%
\def\XINT_irr_loop_exitb #1#2%
{%
   \expandafter\XINT_irr_finish\expandafter {#2}{#1}%
}%
\def\XINT_irr_finish #1#2#3{#3#1/#2}% changed in 1.08
%    \end{macrocode}
% \subsection{\csh{xintifInt}}
%    \begin{macrocode}
\def\xintifInt   {\romannumeral0\xintifint }%
\def\xintifint #1{\expandafter\XINT_ifint\romannumeral0\xintrawwithzeros {#1}.}%
\def\XINT_ifint #1/#2.%
{%
    \if 0\xintiiRem {#1}{#2}%
     \expandafter\xint_stop_atfirstoftwo
    \else
     \expandafter\xint_stop_atsecondoftwo
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintIsInt}}
% \lverb|Added at 1.3d only, for isint() xintexpr function.|
%    \begin{macrocode}
\def\xintIsInt   {\romannumeral0\xintisint }%
\def\xintisint #1%
   {\expandafter\XINT_ifint\romannumeral0\xintrawwithzeros {#1}.10}%
%    \end{macrocode}
% \subsection{\csh{xintJrr}}
%    \begin{macrocode}
\def\xintJrr {\romannumeral0\xintjrr }%
\def\xintjrr #1%
{%
    \expandafter\XINT_jrr_start\romannumeral0\xintrawwithzeros {#1}\Z
}%
\def\XINT_jrr_start #1#2/#3\Z
{%
    \if0\XINT_isOne {#3}\xint_afterfi
          {\xint_UDsignfork
               #1\XINT_jrr_negative
                -{\XINT_jrr_nonneg #1}%
           \krof}%
    \else
      \xint_afterfi{\XINT_jrr_denomisone #1}%
    \fi
    #2\Z {#3}%
}%
\def\XINT_jrr_denomisone #1\Z #2{ #1/1}% changed in 1.08
\def\XINT_jrr_negative   #1\Z #2{\XINT_jrr_D #1\Z #2\Z -}%
\def\XINT_jrr_nonneg     #1\Z #2{\XINT_jrr_D #1\Z #2\Z \space}%
\def\XINT_jrr_D #1#2\Z #3#4\Z
{%
    \xint_UDzerosfork
       #3#1\XINT_jrr_indeterminate
       #30\XINT_jrr_divisionbyzero
       #10\XINT_jrr_zero
        00\XINT_jrr_loop_a
    \krof
    {#3#4}{#1#2}1001%
}%
\def\XINT_jrr_indeterminate #1#2#3#4#5#6#7%
{%
    \XINT_signalcondition{DivisionUndefined}{0/0 indeterminate fraction.}{}{ 0/1}%
}%
\def\XINT_jrr_divisionbyzero #1#2#3#4#5#6#7%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #7#2/0.}{}{ 0/1}%
}%
\def\XINT_jrr_zero #1#2#3#4#5#6#7{ 0/1}% changed in 1.08
\def\XINT_jrr_loop_a #1#2%
{%
    \expandafter\XINT_jrr_loop_b
    \romannumeral0\XINT_div_prepare {#1}{#2}{#1}%
}%
\def\XINT_jrr_loop_b #1#2#3#4#5#6#7%
{%
    \expandafter \XINT_jrr_loop_c \expandafter
        {\romannumeral0\xintiiadd{\XINT_mul_fork #4\xint:#1\xint:}{#6}}%
        {\romannumeral0\xintiiadd{\XINT_mul_fork #5\xint:#1\xint:}{#7}}%
    {#2}{#3}{#4}{#5}%
}%
\def\XINT_jrr_loop_c #1#2%
{%
    \expandafter \XINT_jrr_loop_d \expandafter{#2}{#1}%
}%
\def\XINT_jrr_loop_d #1#2#3#4%
{%
    \XINT_jrr_loop_e #3\Z {#4}{#2}{#1}%
}%
\def\XINT_jrr_loop_e #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_jrr_loop_exit0\XINT_jrr_loop_a {#1#2}%
}%
\def\XINT_jrr_loop_exit0\XINT_jrr_loop_a #1#2#3#4#5#6%
{%
    \XINT_irr_finish {#3}{#4}%
}%
%    \end{macrocode}
% \subsection{\csh{xintTFrac}}
% \lverb|1.09i, for frac in \xintexpr. And \xintFrac is already assigned. T for
% truncation. However, potentially not very efficient with numbers in scientific
% notations, with big exponents. Will have to think it again some day. I
% hesitated how to call the macro. Same convention as in maple, but some people
% reserve fractional part to x - floor(x). Also, not clear if I had to make it
% negative (or zero) if x < 0, or rather always positive. There should be in
% fact such a thing for each rounding function, trunc, round, floor, ceil. |
%    \begin{macrocode}
\def\xintTFrac {\romannumeral0\xinttfrac }%
\def\xinttfrac #1{\expandafter\XINT_tfrac_fork\romannumeral0\xintrawwithzeros {#1}\Z }%
\def\XINT_tfrac_fork #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_tfrac_zero
        0#1{\xintiiopp\XINT_tfrac_P }%
         0-{\XINT_tfrac_P #1}%
    \krof
}%
\def\XINT_tfrac_zero #1\Z { 0/1[0]}%
\def\XINT_tfrac_P #1/#2\Z {\expandafter\XINT_rez_AB
                           \romannumeral0\xintiirem{#1}{#2}\Z {0}{#2}}%
%    \end{macrocode}
% \subsection{\csh{xintTrunc}, \csh{xintiTrunc}}
% \lverb|&
%
% This of course has a long history. Only showing here some comments.
% 
% 1.2i release notes: ever since its inception this macro was stupid for a
% decimal input: it did not handle it separately from the general fraction
% case A/B[N] with B>1, hence ended up doing divisions by powers of ten. But
% this meant that nesting \xintTrunc with itself was very inefficient.
%
% 1.2i version is better. However it still handles B>1, N<0 via adding zeros
% to B and dividing with this extended B. A possibly more efficient approach
% is implemented in \xintXTrunc, but its logic is more complicated, the code
% is quite longer and making it f-expandable would not shorten it... I decided
% for the time being to not complicate things here.
% |
% \lverb|&
%
% 1.4a (2020/02/18) adds handling of a negative first argument.
%
% Zero input still gives single digit 0 output as I did not want to complicate
% the code. But if quantization gives 0, the exponent [D] will be there. Well
% actually eD because of problem that sign of original is preserved in output
% so we can have -0 and I can not use -0[D] notation as it is not legal for
% strict format. So I will use -0eD hence eD generally even though this means
% so slight suboptimality for trunc() function in \xintexpr.
%
% The idea to give a meaning to negative D (in the context of optional
% argument to \xintiexpr) was suggested a long time ago by Kpym (October 20,
% 2015). His suggestion was then to treat it as positive D but trim trailing
% zeroes. But since then, there is \xintDecToString which can be combined with
% \xintREZ, and I feel matters of formatting output require a whole module (or
% rather use existing third-party tools), and I decided to opt rather for an
% operation similar as the quantize() of Python Decimal module. I.e. we
% truncate (or round) to an integer multiple of a given power of 10.
%
% Other reason to decide to do this is that it looks as if I don't even need
% to understand the original code to hack into its ending via \XINT_trunc_G or
% \XINT_itrunc_G. For the latter it looks as if logically I simply have to do
% nothing. For the former I simply have to add some eD postfix.
% |
%    \begin{macrocode}
\def\xintTrunc  {\romannumeral0\xinttrunc }%
\def\xintiTrunc {\romannumeral0\xintitrunc}%
\def\xinttrunc #1{\expandafter\XINT_trunc\the\numexpr#1.\XINT_trunc_G}%
\def\xintitrunc #1{\expandafter\XINT_trunc\the\numexpr#1.\XINT_itrunc_G}%
\def\XINT_trunc #1.#2#3%
{%
    \expandafter\XINT_trunc_a\romannumeral0\XINT_infrac{#3}#1.#2%
}%
\def\XINT_trunc_a #1#2#3#4.#5%
{%
    \if0\XINT_Sgn#2\xint:\xint_dothis\XINT_trunc_zero\fi
    \if1\XINT_is_One#3XY\xint_dothis\XINT_trunc_sp_b\fi
    \xint_orthat\XINT_trunc_b #1+#4.{#2}{#3}#5#4.%
}%
\def\XINT_trunc_zero #1.#2.{ 0}%
\def\XINT_trunc_b     {\expandafter\XINT_trunc_B\the\numexpr}%
\def\XINT_trunc_sp_b  {\expandafter\XINT_trunc_sp_B\the\numexpr}%
\def\XINT_trunc_B #1%
{%
    \xint_UDsignfork
      #1\XINT_trunc_C
       -\XINT_trunc_D
    \krof #1%
}%
\def\XINT_trunc_sp_B #1%
{%
    \xint_UDsignfork
      #1\XINT_trunc_sp_C
       -\XINT_trunc_sp_D
    \krof #1%
}%
\def\XINT_trunc_C -#1.#2#3%
{%
    \expandafter\XINT_trunc_CE
    \romannumeral0\XINT_dsx_addzeros{#1}#3;.{#2}%
}%
\def\XINT_trunc_CE #1.#2{\XINT_trunc_E #2.{#1}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_trunc_sp_C -#1.#2#3{\XINT_trunc_sp_Ca #2.#1.}%
\def\XINT_trunc_sp_Ca #1%
{%
    \xint_UDsignfork
       #1{\XINT_trunc_sp_Cb -}%
        -{\XINT_trunc_sp_Cb \space#1}%
    \krof
}%
\def\XINT_trunc_sp_Cb #1#2.#3.%
{%
    \expandafter\XINT_trunc_sp_Cc
    \romannumeral0\expandafter\XINT_split_fromright_a
    \the\numexpr#3-\numexpr\XINT_length_loop
    #2\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:\xint:
      \xint_c_viii\xint_c_vii\xint_c_vi\xint_c_v
      \xint_c_iv\xint_c_iii\xint_c_ii\xint_c_i\xint_c_\xint_bye
    .#2\xint_bye2345678\xint_bye..#1%
}%
\def\XINT_trunc_sp_Cc #1%
{%
    \if.#1\xint_dothis{\XINT_trunc_sp_Cd 0.}\fi
    \xint_orthat {\XINT_trunc_sp_Cd #1}%
}%
\def\XINT_trunc_sp_Cd #1.#2.#3%
{%
    \XINT_trunc_sp_F #3#1.%
}%
\def\XINT_trunc_D #1.#2%
{%
    \expandafter\XINT_trunc_E
    \romannumeral0\XINT_dsx_addzeros {#1}#2;.%
}%
\def\XINT_trunc_sp_D #1.#2#3%
{%
    \expandafter\XINT_trunc_sp_E
    \romannumeral0\XINT_dsx_addzeros {#1}#2;.%
}%
\def\XINT_trunc_E #1%
{%
    \xint_UDsignfork
       #1{\XINT_trunc_F -}%
        -{\XINT_trunc_F \space#1}%
    \krof
}%
\def\XINT_trunc_sp_E #1%
{%
    \xint_UDsignfork
       #1{\XINT_trunc_sp_F -}%
        -{\XINT_trunc_sp_F\space#1}%
    \krof
}%
\def\XINT_trunc_F #1#2.#3#4%
   {\expandafter#4\romannumeral`&&@\expandafter\xint_firstoftwo
                  \romannumeral0\XINT_div_prepare {#3}{#2}.#1}%
\def\XINT_trunc_sp_F #1#2.#3{#3#2.#1}%
\def\XINT_itrunc_G #1#2.#3#4.%
{%
    \if#10\xint_dothis{ 0}\fi
    \xint_orthat{#3#1}#2%
}%
\def\XINT_trunc_G #1.#2#3#4.%
{%
    \xint_gob_til_minus#3\XINT_trunc_Hc-%
    \expandafter\XINT_trunc_H
    \the\numexpr\romannumeral0\xintlength {#1}-#3#4.#3#4.{#1}#2%
}%
\def\XINT_trunc_Hc-\expandafter\XINT_trunc_H
    \the\numexpr\romannumeral0\xintlength #1.-#2.#3#4{#4#3e#2}%
\def\XINT_trunc_H #1.#2.%
{%
    \ifnum #1 > \xint_c_ \xint_dothis{\XINT_trunc_Ha {#2}}\fi
    \xint_orthat {\XINT_trunc_Hb {-#1}}% -0,--1,--2, ....
}%
\def\XINT_trunc_Ha%
{%
    \expandafter\XINT_trunc_Haa\romannumeral0\xintdecsplit
}%
\def\XINT_trunc_Haa #1#2#3{#3#1.#2}%
\def\XINT_trunc_Hb #1#2#3%
{%
    \expandafter #3\expandafter0\expandafter.%
    \romannumeral\xintreplicate{#1}0#2%
}%
%    \end{macrocode}
% \subsection{\csh{xintTTrunc}}
% \lverb|1.1. Modified in 1.2i, it does simply \xintiTrunc0 with no
% shortcut (the latter having been modified)
%|
%    \begin{macrocode}
\def\xintTTrunc {\romannumeral0\xintttrunc }%
\def\xintttrunc {\xintitrunc\xint_c_}%
%    \end{macrocode}
% \subsection{\cshnolabel{xintNum}}
%    \begin{macrocode}
\let\xintnum \xintttrunc
%    \end{macrocode}
% \subsection{\csh{xintRound}, \csh{xintiRound}}
% \lverb|Modified in 1.2i.
%
% It benefits first of all from the faster \xintTrunc, particularly when the
% input is already a decimal number (denominator B=1).
%
% And the rounding is now done in 1.2 style (with much delay, sorry), like of
% the rewritten \xintInc and \xintDec.
%
% At 1.4a, first argument can be negative. This is handled at \XINT_trunc_G.
%
% |
%    \begin{macrocode}
\def\xintRound  {\romannumeral0\xintround }%
\def\xintiRound {\romannumeral0\xintiround }%
\def\xintround  #1{\expandafter\XINT_round\the\numexpr #1.\XINT_round_A}%
\def\xintiround #1{\expandafter\XINT_round\the\numexpr #1.\XINT_iround_A}%
\def\XINT_round #1.{\expandafter\XINT_round_aa\the\numexpr #1+\xint_c_i.#1.}%
\def\XINT_round_aa #1.#2.#3#4%
{%
    \expandafter\XINT_round_a\romannumeral0\XINT_infrac{#4}#1.#3#2.%
}%
\def\XINT_round_a #1#2#3#4.%
{%
    \if0\XINT_Sgn#2\xint:\xint_dothis\XINT_trunc_zero\fi
    \if1\XINT_is_One#3XY\xint_dothis\XINT_trunc_sp_b\fi
    \xint_orthat\XINT_trunc_b #1+#4.{#2}{#3}%
}%
\def\XINT_round_A{\expandafter\XINT_trunc_G\romannumeral0\XINT_round_B}%
\def\XINT_iround_A{\expandafter\XINT_itrunc_G\romannumeral0\XINT_round_B}%
\def\XINT_round_B #1.%
    {\XINT_dsrr #1\xint_bye\xint_Bye3456789\xint_bye/\xint_c_x\relax.}%
%    \end{macrocode}
% \subsection{\csh{xintXTrunc}}
% \lverb@1.09j [2014/01/06] This is completely expandable but not f-expandable.
% Rewritten for 1.2i (2016/12/04):
%
% - no more use of \xintiloop from xinttools.sty
% (replaced by \xintreplicate... from xintkernel.sty), 
%
% - no more use in 0>N>-D case of a dummy control sequence name via
%   \csname...\endcsname
%
% - handles better the case of an input already a decimal number
%
% Need to transfer code comments into public dtx.
% @
%    \begin{macrocode}
\def\xintXTrunc #1%#2%
{%
    \expandafter\XINT_xtrunc_a
    \the\numexpr #1\expandafter.\romannumeral0\xintraw
}%
\def\XINT_xtrunc_a #1.% ?? faire autre chose
{%
    \expandafter\XINT_xtrunc_b\the\numexpr\ifnum#1<\xint_c_i \xint_c_i-\fi #1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_b #1.#2{\XINT_xtrunc_c #2{#1}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_c #1%
{%
    \xint_UDzerominusfork
        #1-\XINT_xtrunc_zero
        0#1{-\XINT_xtrunc_d {}}%
         0-{\XINT_xtrunc_d #1}%
    \krof
}%[
\def\XINT_xtrunc_zero #1#2]{0.\romannumeral\xintreplicate{#1}0}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_d #1#2#3/#4[#5]%
{%
    \XINT_xtrunc_prepare_a#4\R\R\R\R\R\R\R\R {10}0000001\W
    !{#4};{#5}{#2}{#1#3}%
}%
\def\XINT_xtrunc_prepare_a #1#2#3#4#5#6#7#8#9%
{%
    \xint_gob_til_R #9\XINT_xtrunc_prepare_small\R
    \XINT_xtrunc_prepare_b #9%
}%
\def\XINT_xtrunc_prepare_small\R #1!#2;%
{%
    \ifcase #2
    \or\expandafter\XINT_xtrunc_BisOne
    \or\expandafter\XINT_xtrunc_BisTwo
    \or
    \or\expandafter\XINT_xtrunc_BisFour
    \or\expandafter\XINT_xtrunc_BisFive
    \or
    \or
    \or\expandafter\XINT_xtrunc_BisEight
    \fi\XINT_xtrunc_BisSmall {#2}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_BisOne\XINT_xtrunc_BisSmall #1#2#3#4%
   {\XINT_xtrunc_sp_e {#2}{#4}{#3}}%
\def\XINT_xtrunc_BisTwo\XINT_xtrunc_BisSmall #1#2#3#4%
{%
    \expandafter\XINT_xtrunc_sp_e\expandafter
    {\the\numexpr #2-\xint_c_i\expandafter}\expandafter
    {\romannumeral0\xintiimul 5{#4}}{#3}%
}%
\def\XINT_xtrunc_BisFour\XINT_xtrunc_BisSmall #1#2#3#4%
{%
    \expandafter\XINT_xtrunc_sp_e\expandafter
    {\the\numexpr #2-\xint_c_ii\expandafter}\expandafter
    {\romannumeral0\xintiimul {25}{#4}}{#3}%
}%
\def\XINT_xtrunc_BisFive\XINT_xtrunc_BisSmall #1#2#3#4%
{%
    \expandafter\XINT_xtrunc_sp_e\expandafter
    {\the\numexpr #2-\xint_c_i\expandafter}\expandafter
    {\romannumeral0\xintdouble {#4}}{#3}%
}%
\def\XINT_xtrunc_BisEight\XINT_xtrunc_BisSmall #1#2#3#4%
{%
    \expandafter\XINT_xtrunc_sp_e\expandafter
    {\the\numexpr #2-\xint_c_iii\expandafter}\expandafter
    {\romannumeral0\xintiimul {125}{#4}}{#3}%
}%
\def\XINT_xtrunc_BisSmall #1%
{%
    \expandafter\XINT_xtrunc_e\expandafter
    {\expandafter\XINT_xtrunc_small_a
    \the\numexpr #1/\xint_c_ii\expandafter
    .\the\numexpr \xint_c_x^viii+#1!}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_small_a #1.#2!#3%
{%
    \expandafter\XINT_div_small_b\the\numexpr #1\expandafter
    \xint:\the\numexpr #2\expandafter!%
    \romannumeral0\XINT_div_small_ba #3\R\R\R\R\R\R\R\R{10}0000001\W
       #3\XINT_sepbyviii_Z_end 2345678\relax
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_prepare_b
   {\expandafter\XINT_xtrunc_prepare_c\romannumeral0\XINT_zeroes_forviii }%
\def\XINT_xtrunc_prepare_c #1!%
{%
     \XINT_xtrunc_prepare_d  #1.00000000!{#1}%
}%
\def\XINT_xtrunc_prepare_d #1#2#3#4#5#6#7#8#9%
{%
    \expandafter\XINT_xtrunc_prepare_e
    \xint_gob_til_dot #1#2#3#4#5#6#7#8#9!%
}%
\def\XINT_xtrunc_prepare_e #1!#2!#3#4%
{%
    \XINT_xtrunc_prepare_f #4#3\X {#1}{#3}%
}%
\def\XINT_xtrunc_prepare_f #1#2#3#4#5#6#7#8#9\X
{%
    \expandafter\XINT_xtrunc_prepare_g\expandafter
    \XINT_div_prepare_g
     \the\numexpr  #1#2#3#4#5#6#7#8+\xint_c_i\expandafter
    \xint:\the\numexpr (#1#2#3#4#5#6#7#8+\xint_c_i)/\xint_c_ii\expandafter
    \xint:\the\numexpr #1#2#3#4#5#6#7#8\expandafter
    \xint:\romannumeral0\XINT_sepandrev_andcount
    #1#2#3#4#5#6#7#8#9\XINT_rsepbyviii_end_A 2345678%
                      \XINT_rsepbyviii_end_B 2345678\relax\xint_c_ii\xint_c_i
              \R\xint:\xint_c_xii \R\xint:\xint_c_x  \R\xint:\xint_c_viii \R\xint:\xint_c_vi
              \R\xint:\xint_c_iv  \R\xint:\xint_c_ii \R\xint:\xint_c_\W
    \X
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_prepare_g #1;{\XINT_xtrunc_e {#1}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_e #1#2%
{%
    \ifnum #2<\xint_c_
        \expandafter\XINT_xtrunc_I
    \else
        \expandafter\XINT_xtrunc_II
    \fi  #2\xint:{#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_I -#1\xint:#2#3#4%
{%
    \expandafter\XINT_xtrunc_I_a\romannumeral0#2{#4}{#2}{#1}{#3}%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_I_a #1#2#3#4#5%
{%
    \expandafter\XINT_xtrunc_I_b\the\numexpr #4-#5\xint:#4\xint:{#5}{#2}{#3}{#1}%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_I_b #1%
{%
    \xint_UDsignfork
      #1\XINT_xtrunc_IA_c
       -\XINT_xtrunc_IB_c
    \krof #1%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_IA_c -#1\xint:#2\xint:#3#4#5#6%
{%
   \expandafter\XINT_xtrunc_IA_d
   \the\numexpr#2-\xintLength{#6}\xint:{#6}%
   \expandafter\XINT_xtrunc_IA_xd
   \the\numexpr (#1+\xint_c_ii^v)/\xint_c_ii^vi-\xint_c_i\xint:#1\xint:{#5}{#4}%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_IA_d #1%
{%
    \xint_UDsignfork
      #1\XINT_xtrunc_IAA_e
       -\XINT_xtrunc_IAB_e
    \krof #1%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_IAA_e -#1\xint:#2%
{%
    \romannumeral0\XINT_split_fromleft
    #1.#2\xint_gobble_i\xint_bye2345678\xint_bye..%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_IAB_e #1\xint:#2%
{%
    0.\romannumeral\XINT_rep#1\endcsname0#2%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_IA_xd #1\xint:#2\xint:%
{%
    \expandafter\XINT_xtrunc_IA_xe\the\numexpr #2-\xint_c_ii^vi*#1\xint:#1\xint:%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_IA_xe #1\xint:#2\xint:#3#4%
{%
    \XINT_xtrunc_loop {#2}{#4}{#3}{#1}%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_IB_c #1\xint:#2\xint:#3#4#5#6%
{%
    \expandafter\XINT_xtrunc_IB_d
    \romannumeral0\XINT_split_xfork #1.#6\xint_bye2345678\xint_bye..{#3}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_IB_d #1.#2.#3%
{%
    \expandafter\XINT_xtrunc_IA_d\the\numexpr#3-\xintLength {#1}\xint:{#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_II #1\xint:%
{%
    \expandafter\XINT_xtrunc_II_a\romannumeral\xintreplicate{#1}0\xint:%
}%
\def\XINT_xtrunc_II_a #1\xint:#2#3#4%
{%
    \expandafter\XINT_xtrunc_II_b
    \the\numexpr (#3+\xint_c_ii^v)/\xint_c_ii^vi-\xint_c_i\expandafter\xint:%
    \the\numexpr #3\expandafter\xint:\romannumeral0#2{#4#1}{#2}%
}%
\def\XINT_xtrunc_II_b #1\xint:#2\xint:%
{%
    \expandafter\XINT_xtrunc_II_c\the\numexpr #2-\xint_c_ii^vi*#1\xint:#1\xint:%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_II_c #1\xint:#2\xint:#3#4#5%
{%
    #3.\XINT_xtrunc_loop {#2}{#4}{#5}{#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_loop #1%
{%
    \ifnum #1=\xint_c_ \expandafter\XINT_xtrunc_transition\fi
    \expandafter\XINT_xtrunc_loop_a\the\numexpr #1-\xint_c_i\xint:%
}%
\def\XINT_xtrunc_loop_a #1\xint:#2#3%
{%
    \expandafter\XINT_xtrunc_loop_b\romannumeral0#3%
    {#20000000000000000000000000000000000000000000000000000000000000000}%
    {#1}{#3}%
}%
\def\XINT_xtrunc_loop_b #1#2#3%
{%
    \romannumeral\xintreplicate{\xint_c_ii^vi-\xintLength{#1}}0#1%
    \XINT_xtrunc_loop {#3}{#2}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_transition
    \expandafter\XINT_xtrunc_loop_a\the\numexpr #1\xint:#2#3#4%
{%
    \ifnum #4=\xint_c_ \expandafter\xint_gobble_vi\fi
    \expandafter\XINT_xtrunc_finish\expandafter
    {\romannumeral0\XINT_dsx_addzeros{#4}#2;}{#3}{#4}%
}%
\def\XINT_xtrunc_finish #1#2%
{%
    \expandafter\XINT_xtrunc_finish_a\romannumeral0#2{#1}%
}%
\def\XINT_xtrunc_finish_a #1#2#3%
{%
    \romannumeral\xintreplicate{#3-\xintLength{#1}}0#1%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_sp_e #1%
{%
    \ifnum #1<\xint_c_
        \expandafter\XINT_xtrunc_sp_I
    \else
        \expandafter\XINT_xtrunc_sp_II
    \fi  #1\xint:%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_I -#1\xint:#2#3%
{%
    \expandafter\XINT_xtrunc_sp_I_a\the\numexpr #1-#3\xint:#1\xint:{#3}{#2}%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_I_a #1%
{%
    \xint_UDsignfork
      #1\XINT_xtrunc_sp_IA_b
       -\XINT_xtrunc_sp_IB_b
    \krof #1%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_IA_b -#1\xint:#2\xint:#3#4%
{%
   \expandafter\XINT_xtrunc_sp_IA_c
   \the\numexpr#2-\xintLength{#4}\xint:{#4}\romannumeral\XINT_rep#1\endcsname0%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_IA_c #1%
{%
    \xint_UDsignfork
      #1\XINT_xtrunc_sp_IAA
       -\XINT_xtrunc_sp_IAB
    \krof #1%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_IAA -#1\xint:#2%
{%
    \romannumeral0\XINT_split_fromleft
    #1.#2\xint_gobble_i\xint_bye2345678\xint_bye..%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_IAB #1\xint:#2%
{%
    0.\romannumeral\XINT_rep#1\endcsname0#2%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_IB_b #1\xint:#2\xint:#3#4%
{%
    \expandafter\XINT_xtrunc_sp_IB_c
    \romannumeral0\XINT_split_xfork #1.#4\xint_bye2345678\xint_bye..{#3}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_xtrunc_sp_IB_c #1.#2.#3%
{%
    \expandafter\XINT_xtrunc_sp_IA_c\the\numexpr#3-\xintLength {#1}\xint:{#1}%
}%
%    \end{macrocode}
% \lverb@&
% @
%    \begin{macrocode}
\def\XINT_xtrunc_sp_II #1\xint:#2#3%
{%
    #2\romannumeral\XINT_rep#1\endcsname0.\romannumeral\XINT_rep#3\endcsname0%
}%
%    \end{macrocode}
% \subsection{\csh{xintAdd}}
% \lverb|Big change at 1.3: a/b+c/d uses lcm(b,d) as denominator.|
%    \begin{macrocode}
\def\xintAdd {\romannumeral0\xintadd }%
\def\xintadd #1{\expandafter\XINT_fadd\romannumeral0\xintraw {#1}}%
\def\XINT_fadd #1{\xint_gob_til_zero #1\XINT_fadd_Azero 0\XINT_fadd_a #1}%
\def\XINT_fadd_Azero #1]{\xintraw }%
\def\XINT_fadd_a #1/#2[#3]#4%
   {\expandafter\XINT_fadd_b\romannumeral0\xintraw {#4}{#3}{#1}{#2}}%
\def\XINT_fadd_b #1{\xint_gob_til_zero #1\XINT_fadd_Bzero 0\XINT_fadd_c #1}%
\def\XINT_fadd_Bzero #1]#2#3#4{ #3/#4[#2]}%
\def\XINT_fadd_c #1/#2[#3]#4%
{%
    \expandafter\XINT_fadd_Aa\the\numexpr #4-#3.{#3}{#4}{#1}{#2}%
}%
\def\XINT_fadd_Aa #1%
{%
    \xint_UDzerominusfork
       #1-\XINT_fadd_B
        0#1\XINT_fadd_Bb
        0-\XINT_fadd_Ba
    \krof #1%
}%
\def\XINT_fadd_B   #1.#2#3#4#5#6#7{\XINT_fadd_C {#4}{#5}{#7}{#6}[#3]}%
\def\XINT_fadd_Ba  #1.#2#3#4#5#6#7%
{%
    \expandafter\XINT_fadd_C\expandafter
        {\romannumeral0\XINT_dsx_addzeros {#1}#6;}%
    {#7}{#5}{#4}[#2]%
}%
\def\XINT_fadd_Bb -#1.#2#3#4#5#6#7%
{%
    \expandafter\XINT_fadd_C\expandafter
        {\romannumeral0\XINT_dsx_addzeros {#1}#4;}%
    {#5}{#7}{#6}[#3]%
}%
\def\XINT_fadd_iszero #1[#2]{ 0/1[0]}% ou [#2] originel?
\def\XINT_fadd_C #1#2#3%
{%
    \expandafter\XINT_fadd_D_b
    \romannumeral0\XINT_div_prepare{#2}{#3}{#2}{#2}{#3}{#1}%
}%
%    \end{macrocode}
% \lverb|Basically a clone of the \XINT_irr_loop_a loop. I should modify the
% output of \XINT_div_prepare perhaps to be optimized for checking if
% remainder vanishes.|
%    \begin{macrocode}
\def\XINT_fadd_D_a #1#2%
{%
    \expandafter\XINT_fadd_D_b
    \romannumeral0\XINT_div_prepare {#1}{#2}{#1}%
}%
\def\XINT_fadd_D_b #1#2{\XINT_fadd_D_c #2\Z}%
\def\XINT_fadd_D_c #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_fadd_D_exit0\XINT_fadd_D_a {#1#2}%
}%
\def\XINT_fadd_D_exit0\XINT_fadd_D_a #1#2#3%
{%
    \expandafter\XINT_fadd_E
    \romannumeral0\xintiiquo {#3}{#2}.{#2}%
}%
\def\XINT_fadd_E #1.#2#3%
{%
    \expandafter\XINT_fadd_F
    \romannumeral0\xintiimul{#1}{#3}.{\xintiiQuo{#3}{#2}}{#1}%
}%
\def\XINT_fadd_F #1.#2#3#4#5%
{%
    \expandafter\XINT_fadd_G
    \romannumeral0\xintiiadd{\xintiiMul{#2}{#4}}{\xintiiMul{#3}{#5}}/#1%
}%
\def\XINT_fadd_G #1{%
\def\XINT_fadd_G ##1{\if0##1\expandafter\XINT_fadd_iszero\fi#1##1}%
}\XINT_fadd_G{ }%
%    \end{macrocode}
% \subsection{\csh{xintSub}}
% \lverb|Since 1.3 will use least common multiple of denominators.|
%    \begin{macrocode}
\def\xintSub   {\romannumeral0\xintsub }%
\def\xintsub #1{\expandafter\XINT_fsub\romannumeral0\xintraw {#1}}%
\def\XINT_fsub #1{\xint_gob_til_zero #1\XINT_fsub_Azero 0\XINT_fsub_a #1}%
\def\XINT_fsub_Azero #1]{\xintopp }%
\def\XINT_fsub_a #1/#2[#3]#4%
   {\expandafter\XINT_fsub_b\romannumeral0\xintraw {#4}{#3}{#1}{#2}}%
\def\XINT_fsub_b #1{\xint_UDzerominusfork
                      #1-\XINT_fadd_Bzero
                       0#1\XINT_fadd_c
                       0-{\XINT_fadd_c -#1}%
                     \krof }%
%    \end{macrocode}
% \subsection{\csh{xintSum}}
% \lverb|There was (not documented anymore since 1.09d, 2013/10/22) a macro
% \xintSumExpr, but it has been deleted at 1.2l.
%
% Empty items in the input are not accepted by this macro, but the input
% may be empty.
%
% Refactored slightly at 1.4. \XINT_Sum used in xintexpr code.
%
%
% |
%    \begin{macrocode}
\def\xintSum {\romannumeral0\xintsum }%
\def\xintsum #1{\expandafter\XINT_sum\romannumeral`&&@#1^}%
\def\XINT_Sum{\romannumeral0\XINT_sum}%
\def\XINT_sum#1%
{%
    \xint_gob_til_^ #1\XINT_sum_empty ^%
    \expandafter\XINT_sum_loop\romannumeral0\xintraw{#1}\xint:
}%
\def\XINT_sum_empty ^#1\xint:{ 0/1[0]}%
\def\XINT_sum_loop #1\xint:#2%
{%
    \xint_gob_til_^ #2\XINT_sum_end ^%
    \expandafter\XINT_sum_loop
    \romannumeral0\xintadd{#1}{\romannumeral0\xintraw{#2}}\xint:
}%
\def\XINT_sum_end ^#1\xintadd #2#3\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintMul}}
%    \begin{macrocode}
\def\xintMul {\romannumeral0\xintmul }%
\def\xintmul #1{\expandafter\XINT_fmul\romannumeral0\xintraw {#1}.}%
\def\XINT_fmul #1{\xint_gob_til_zero #1\XINT_fmul_zero 0\XINT_fmul_a #1}%
\def\XINT_fmul_a #1[#2].#3%
   {\expandafter\XINT_fmul_b\romannumeral0\xintraw {#3}#1[#2.]}%
\def\XINT_fmul_b #1{\xint_gob_til_zero #1\XINT_fmul_zero 0\XINT_fmul_c #1}%
\def\XINT_fmul_c #1/#2[#3]#4/#5[#6.]%
{%
    \expandafter\XINT_fmul_d
    \expandafter{\the\numexpr #3+#6\expandafter}%
    \expandafter{\romannumeral0\xintiimul {#5}{#2}}%
    {\romannumeral0\xintiimul {#4}{#1}}%
}%
\def\XINT_fmul_d #1#2#3%
{%
    \expandafter \XINT_fmul_e \expandafter{#3}{#1}{#2}%
}%
\def\XINT_fmul_e #1#2{\XINT_outfrac {#2}{#1}}%
\def\XINT_fmul_zero #1.#2{ 0/1[0]}%
%    \end{macrocode}
% \subsection{\csh{xintSqr}}
% \lverb|1.1 modifs comme xintMul.
%
% |
%    \begin{macrocode}
\def\xintSqr {\romannumeral0\xintsqr }%
\def\xintsqr #1{\expandafter\XINT_fsqr\romannumeral0\xintraw {#1}}%
\def\XINT_fsqr #1{\xint_gob_til_zero #1\XINT_fsqr_zero 0\XINT_fsqr_a #1}%
\def\XINT_fsqr_a #1/#2[#3]%
{%
    \expandafter\XINT_fsqr_b
    \expandafter{\the\numexpr #3+#3\expandafter}%
    \expandafter{\romannumeral0\xintiisqr {#2}}%
    {\romannumeral0\xintiisqr {#1}}%
}%
\def\XINT_fsqr_b #1#2#3{\expandafter \XINT_fmul_e \expandafter{#3}{#1}{#2}}%
\def\XINT_fsqr_zero #1]{ 0/1[0]}%
%    \end{macrocode}
% \subsection{\csh{xintPow}}
% \lverb|&
% 1.2f: to be coherent with the "i" convention \xintiPow should parse also its
% exponent via \xintNum when xintfrac.sty is loaded. This was not the case so
% far. Cependant le problème est que le fait d'appliquer \xintNum rend
% impossible certains inputs qui auraient pu être gérès par \numexpr. Le
% \numexpr externe est ici pour intercepter trop grand input.
% |
%    \begin{macrocode}
\def\xintipow #1#2%
{%
    \expandafter\xint_pow\the\numexpr \xintNum{#2}\expandafter
    .\romannumeral0\xintnum{#1}\xint:
}%
\def\xintPow {\romannumeral0\xintpow }%
\def\xintpow #1%
{%
    \expandafter\XINT_fpow\expandafter {\romannumeral0\XINT_infrac {#1}}%
}%
\def\XINT_fpow #1#2%
{%
    \expandafter\XINT_fpow_fork\the\numexpr \xintNum{#2}\relax\Z #1%
}%
\def\XINT_fpow_fork #1#2\Z
{%
    \xint_UDzerominusfork
      #1-\XINT_fpow_zero
      0#1\XINT_fpow_neg
       0-{\XINT_fpow_pos #1}%
    \krof
    {#2}%
}%
\def\XINT_fpow_zero #1#2#3#4{ 1/1[0]}%
\def\XINT_fpow_pos #1#2#3#4#5%
{%
    \expandafter\XINT_fpow_pos_A\expandafter
    {\the\numexpr #1#2*#3\expandafter}\expandafter
    {\romannumeral0\xintiipow {#5}{#1#2}}%
    {\romannumeral0\xintiipow {#4}{#1#2}}%
}%
\def\XINT_fpow_neg #1#2#3#4%
{%
    \expandafter\XINT_fpow_pos_A\expandafter
    {\the\numexpr -#1*#2\expandafter}\expandafter
    {\romannumeral0\xintiipow {#3}{#1}}%
    {\romannumeral0\xintiipow {#4}{#1}}%
}%
\def\XINT_fpow_pos_A #1#2#3%
{%
    \expandafter\XINT_fpow_pos_B\expandafter {#3}{#1}{#2}%
}%
\def\XINT_fpow_pos_B #1#2{\XINT_outfrac {#2}{#1}}%
%    \end{macrocode}
% \subsection{\csh{xintFac}}
% \lverb|Factorial coefficients: variant which can be chained with other
% xintfrac macros. \xintiFac deprecated at 1.2o and removed at 1.3; \xintFac
% used by xintexpr.sty.|
%    \begin{macrocode}
\def\xintFac  {\romannumeral0\xintfac}%
\def\xintfac  #1{\expandafter\XINT_fac_fork\the\numexpr\xintNum{#1}.[0]}%
%    \end{macrocode}
% \subsection{\csh{xintBinomial}}
% \lverb|1.2f. Binomial coefficients. \xintiBinomial deprecated at 1.2o and
% removed at 1.3;
% \xintBinomial needed by xintexpr.sty.|
%    \begin{macrocode}
\def\xintBinomial {\romannumeral0\xintbinomial}%
\def\xintbinomial #1#2%
{%
    \expandafter\XINT_binom_pre
    \the\numexpr\xintNum{#1}\expandafter.\the\numexpr\xintNum{#2}.[0]%
}%
%    \end{macrocode}
% \subsection{\csh{xintPFactorial}}
% \lverb|1.2f. Partial factorial. For needs of xintexpr.sty.|
%    \begin{macrocode}
\def\xintipfactorial #1#2%
{%
    \expandafter\XINT_pfac_fork
    \the\numexpr\xintNum{#1}\expandafter.\the\numexpr\xintNum{#2}.%
}%
\def\xintPFactorial {\romannumeral0\xintpfactorial}%
\def\xintpfactorial #1#2%
{%
    \expandafter\XINT_pfac_fork
    \the\numexpr\xintNum{#1}\expandafter.\the\numexpr\xintNum{#2}.[0]%
}%
%    \end{macrocode}
% \subsection{\csh{xintPrd}}
% \lverb|
% Refactored at 1.4. After some hesitation the routine still does
% not try to detect on the fly a zero item, to abort the loop. Indeed
% this would add some overhead generally (as we need normalizing the item before
% checking if it vanishes hence we must then grab things once more).
% |
%    \begin{macrocode}
\def\xintPrd {\romannumeral0\xintprd }%
\def\xintprd #1{\expandafter\XINT_prd\romannumeral`&&@#1^}%
\def\XINT_Prd{\romannumeral0\XINT_prd}%
\def\XINT_prd#1%
{%
    \xint_gob_til_^ #1\XINT_prd_empty ^%
    \expandafter\XINT_prd_loop\romannumeral0\xintraw{#1}\xint:
}%
\def\XINT_prd_empty ^#1\xint:{ 1/1[0]}%
\def\XINT_prd_loop #1\xint:#2%
{%
    \xint_gob_til_^ #2\XINT_prd_end ^%
    \expandafter\XINT_prd_loop
    \romannumeral0\xintmul{#1}{\romannumeral0\xintraw{#2}}\xint:
}%
\def\XINT_prd_end ^#1\xintmul #2#3\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintDiv}}
%    \begin{macrocode}
\def\xintDiv {\romannumeral0\xintdiv }%
\def\xintdiv #1%
{%
    \expandafter\XINT_fdiv\expandafter {\romannumeral0\XINT_infrac {#1}}%
}%
\def\XINT_fdiv #1#2%
   {\expandafter\XINT_fdiv_A\romannumeral0\XINT_infrac {#2}#1}%
\def\XINT_fdiv_A #1#2#3#4#5#6%
{%
    \expandafter\XINT_fdiv_B
    \expandafter{\the\numexpr #4-#1\expandafter}%
    \expandafter{\romannumeral0\xintiimul {#2}{#6}}%
    {\romannumeral0\xintiimul {#3}{#5}}%
}%
\def\XINT_fdiv_B #1#2#3%
{%
    \expandafter\XINT_fdiv_C
    \expandafter{#3}{#1}{#2}%
}%
\def\XINT_fdiv_C #1#2{\XINT_outfrac {#2}{#1}}%
%    \end{macrocode}
% \subsection{\csh{xintDivFloor}}
% \lverb|1.1. Changed at 1.2p to not append /1[0] ending but rather output a
% big integer in strict format, like \xintDivTrunc and \xintDivRound.
%
%
% 
% |
%    \begin{macrocode}
\def\xintDivFloor     {\romannumeral0\xintdivfloor }%
\def\xintdivfloor #1#2{\xintifloor{\xintDiv {#1}{#2}}}%
%    \end{macrocode}
% \subsection{\csh{xintDivTrunc}}
% \lverb|1.1. \xintttrunc rather than \xintitrunc0 in 1.1a|
%    \begin{macrocode}
\def\xintDivTrunc     {\romannumeral0\xintdivtrunc }%
\def\xintdivtrunc #1#2{\xintttrunc {\xintDiv {#1}{#2}}}%
%    \end{macrocode}
% \subsection{\csh{xintDivRound}}
% \lverb|1.1|
%    \begin{macrocode}
\def\xintDivRound     {\romannumeral0\xintdivround }%
\def\xintdivround #1#2{\xintiround 0{\xintDiv {#1}{#2}}}%
%    \end{macrocode}
% \subsection{\csh{xintModTrunc}}
% \lverb|1.1. \xintModTrunc {q1}{q2} computes q1 - q2*t(q1/q2) with t(q1/q2)
% equal to the truncated division of two fractions q1 and q2.
%
% Its former name, prior to 1.2p, was \xintMod.
%
% At 1.3, uses least common multiple denominator, like \xintMod (next).|
%    \begin{macrocode}
\def\xintModTrunc {\romannumeral0\xintmodtrunc }%
\def\xintmodtrunc #1{\expandafter\XINT_modtrunc_a\romannumeral0\xintraw{#1}.}%
\def\XINT_modtrunc_a #1#2.#3%
   {\expandafter\XINT_modtrunc_b\expandafter #1\romannumeral0\xintraw{#3}#2.}%
\def\XINT_modtrunc_b #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_modtrunc_divbyzero #1#2}\fi
    \if0#1\xint_dothis\XINT_modtrunc_aiszero\fi
    \if-#2\xint_dothis{\XINT_modtrunc_bneg #1}\fi
          \xint_orthat{\XINT_modtrunc_bpos #1#2}%
}%
\def\XINT_modtrunc_divbyzero #1#2[#3]#4.%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #1#4/(#2[#3]).}{}{ 0/1[0]}%
}%
\def\XINT_modtrunc_aiszero #1.{ 0/1[0]}%
\def\XINT_modtrunc_bneg #1%
{%
    \xint_UDsignfork
         #1{\xintiiopp\XINT_modtrunc_pos {}}%
          -{\XINT_modtrunc_pos #1}%
    \krof
}%
\def\XINT_modtrunc_bpos #1%
{%
    \xint_UDsignfork
            #1{\xintiiopp\XINT_modtrunc_pos {}}%
             -{\XINT_modtrunc_pos #1}%
    \krof
}%
%    \end{macrocode}
% \lverb|Attention. This crucially uses that xint's \xintiiE{x}{e} is defined
% to return x unchanged if e is negative (and x extended by e zeroes if e >=
% 0).|
%    \begin{macrocode}
\def\XINT_modtrunc_pos #1#2/#3[#4]#5/#6[#7].%
{%
    \expandafter\XINT_modtrunc_pos_a
    \the\numexpr\ifnum#7>#4 #4\else #7\fi\expandafter.%
    \romannumeral0\expandafter\XINT_mod_D_b
    \romannumeral0\XINT_div_prepare{#3}{#6}{#3}{#3}{#6}%
    {#1#5}{#7-#4}{#2}{#4-#7}%
}%
\def\XINT_modtrunc_pos_a #1.#2#3#4{\xintiirem {#3}{#4}/#2[#1]}%
%    \end{macrocode}
% \subsection{\csh{xintDivMod}}
% \lverb|1.2p. \xintDivMod{q1}{q2} outputs {floor(q1/q2)}{q1 - q2*floor(q1/q2)}.
% Attention that it relies on \xintiiE{x}{e} returning x if e < 0.
%
% Modified (like \xintAdd and \xintSub) at 1.3 to use a l.c.m for final
% denominator of the "mod" part.|
%    \begin{macrocode}
\def\xintDivMod {\romannumeral0\xintdivmod }%
\def\xintdivmod #1{\expandafter\XINT_divmod_a\romannumeral0\xintraw{#1}.}%
\def\XINT_divmod_a #1#2.#3%
   {\expandafter\XINT_divmod_b\expandafter #1\romannumeral0\xintraw{#3}#2.}%
\def\XINT_divmod_b #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_divmod_divbyzero #1#2}\fi
    \if0#1\xint_dothis\XINT_divmod_aiszero\fi
    \if-#2\xint_dothis{\XINT_divmod_bneg #1}\fi
          \xint_orthat{\XINT_divmod_bpos #1#2}%
}%
\def\XINT_divmod_divbyzero #1#2[#3]#4.%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero: #1#4/(#2[#3]).}{}%
    {{0}{0/1[0]}}% à revoir...
}%
\def\XINT_divmod_aiszero #1.{{0}{0/1[0]}}%
\def\XINT_divmod_bneg #1% f // -g = (-f) // g, f % -g = - ((-f) % g)
{%
    \expandafter\XINT_divmod_bneg_finish
    \romannumeral0\xint_UDsignfork
        #1{\XINT_divmod_bpos {}}%
         -{\XINT_divmod_bpos {-#1}}%
    \krof
}%
\def\XINT_divmod_bneg_finish#1#2%
{%
    \expandafter\xint_exchangetwo_keepbraces\expandafter
    {\romannumeral0\xintiiopp#2}{#1}%
}%
\def\XINT_divmod_bpos #1#2/#3[#4]#5/#6[#7].%
{%
    \expandafter\XINT_divmod_bpos_a
    \the\numexpr\ifnum#7>#4 #4\else #7\fi\expandafter.%
    \romannumeral0\expandafter\XINT_mod_D_b
    \romannumeral0\XINT_div_prepare{#3}{#6}{#3}{#3}{#6}%
    {#1#5}{#7-#4}{#2}{#4-#7}%
}%
\def\XINT_divmod_bpos_a #1.#2#3#4%
{%
    \expandafter\XINT_divmod_bpos_finish
    \romannumeral0\xintiidivision{#3}{#4}{/#2[#1]}%
}%
\def\XINT_divmod_bpos_finish #1#2#3{{#1}{#2#3}}%
%    \end{macrocode}
% \subsection{\csh{xintMod}}
% \lverb|1.2p. \xintMod{q1}{q2} computes q1 - q2*floor(q1/q2). Attention that
% it relies on \xintiiE{x}{e} returning x if e < 0.
%
% Prior to 1.2p, that macro had the meaning now attributed to \xintModTrunc.
%
% Modified (like \xintAdd and \xintSub) at 1.3 to use a l.c.m for final
% denominator.|
%    \begin{macrocode}
\def\xintMod {\romannumeral0\xintmod }%
\def\xintmod #1{\expandafter\XINT_mod_a\romannumeral0\xintraw{#1}.}%
\def\XINT_mod_a #1#2.#3%
   {\expandafter\XINT_mod_b\expandafter #1\romannumeral0\xintraw{#3}#2.}%
\def\XINT_mod_b #1#2% #1 de A, #2 de B.
{%
    \if0#2\xint_dothis{\XINT_mod_divbyzero #1#2}\fi
    \if0#1\xint_dothis\XINT_mod_aiszero\fi
    \if-#2\xint_dothis{\XINT_mod_bneg #1}\fi
          \xint_orthat{\XINT_mod_bpos #1#2}%
}%
%    \end{macrocode}
% \lverb|Attention to not move ModTrunc code beyond that point.|
%    \begin{macrocode}
\let\XINT_mod_divbyzero\XINT_modtrunc_divbyzero
\let\XINT_mod_aiszero  \XINT_modtrunc_aiszero
\def\XINT_mod_bneg #1% f % -g = - ((-f) % g), for g > 0
{%
    \xintiiopp\xint_UDsignfork
        #1{\XINT_mod_bpos {}}%
         -{\XINT_mod_bpos {-#1}}%
    \krof
}%
\def\XINT_mod_bpos #1#2/#3[#4]#5/#6[#7].%
{%
    \expandafter\XINT_mod_bpos_a
    \the\numexpr\ifnum#7>#4 #4\else #7\fi\expandafter.%
    \romannumeral0\expandafter\XINT_mod_D_b
    \romannumeral0\XINT_div_prepare{#3}{#6}{#3}{#3}{#6}%
    {#1#5}{#7-#4}{#2}{#4-#7}%
}%
\def\XINT_mod_D_a #1#2%
{%
    \expandafter\XINT_mod_D_b
    \romannumeral0\XINT_div_prepare {#1}{#2}{#1}%
}%
\def\XINT_mod_D_b #1#2{\XINT_mod_D_c #2\Z}%
\def\XINT_mod_D_c #1#2\Z
{%
    \xint_gob_til_zero #1\XINT_mod_D_exit0\XINT_mod_D_a {#1#2}%
}%
\def\XINT_mod_D_exit0\XINT_mod_D_a #1#2#3%
{%
    \expandafter\XINT_mod_E
    \romannumeral0\xintiiquo {#3}{#2}.{#2}%
}%
\def\XINT_mod_E #1.#2#3%
{%
    \expandafter\XINT_mod_F
    \romannumeral0\xintiimul{#1}{#3}.{\xintiiQuo{#3}{#2}}{#1}%
}%
\def\XINT_mod_F #1.#2#3#4#5#6#7%
{%
    {#1}{\xintiiE{\xintiiMul{#4}{#3}}{#5}}%
        {\xintiiE{\xintiiMul{#6}{#2}}{#7}}%
}%
\def\XINT_mod_bpos_a #1.#2#3#4{\xintiirem {#3}{#4}/#2[#1]}%
%    \end{macrocode}
% \subsection{\csh{xintIsOne}}
% \lverb|New with 1.09a. Could be more efficient. For fractions with big
% powers of tens, it is better to use \xintCmp{f}{1}. Restyled in 1.09i.|
%    \begin{macrocode}
\def\xintIsOne   {\romannumeral0\xintisone }%
\def\xintisone #1{\expandafter\XINT_fracisone
                  \romannumeral0\xintrawwithzeros{#1}\Z }%
\def\XINT_fracisone #1/#2\Z
    {\if0\xintiiCmp {#1}{#2}\xint_afterfi{ 1}\else\xint_afterfi{ 0}\fi}%
%    \end{macrocode}
% \subsection{\csh{xintGeq}}
%    \begin{macrocode}
\def\xintGeq {\romannumeral0\xintgeq }%
\def\xintgeq #1%
{%
    \expandafter\XINT_fgeq\expandafter {\romannumeral0\xintabs {#1}}%
}%
\def\XINT_fgeq #1#2%
{%
    \expandafter\XINT_fgeq_A \romannumeral0\xintabs {#2}#1%
}%
\def\XINT_fgeq_A #1%
{%
    \xint_gob_til_zero #1\XINT_fgeq_Zii 0%
    \XINT_fgeq_B #1%
}%
\def\XINT_fgeq_Zii 0\XINT_fgeq_B #1[#2]#3[#4]{ 1}%
\def\XINT_fgeq_B #1/#2[#3]#4#5/#6[#7]%
{%
    \xint_gob_til_zero #4\XINT_fgeq_Zi 0%
    \expandafter\XINT_fgeq_C\expandafter
    {\the\numexpr #7-#3\expandafter}\expandafter
    {\romannumeral0\xintiimul {#4#5}{#2}}%
    {\romannumeral0\xintiimul {#6}{#1}}%
}%
\def\XINT_fgeq_Zi 0#1#2#3#4#5#6#7{ 0}%
\def\XINT_fgeq_C #1#2#3%
{%
    \expandafter\XINT_fgeq_D\expandafter
    {#3}{#1}{#2}%
}%
\def\XINT_fgeq_D #1#2#3%
{%
    \expandafter\XINT_cntSgnFork\romannumeral`&&@\expandafter\XINT_cntSgn
     \the\numexpr #2+\xintLength{#3}-\xintLength{#1}\relax\xint:
    { 0}{\XINT_fgeq_E #2\Z {#3}{#1}}{ 1}%
}%
\def\XINT_fgeq_E #1%
{%
    \xint_UDsignfork
        #1\XINT_fgeq_Fd
         -{\XINT_fgeq_Fn #1}%
    \krof
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_fgeq_Fd #1\Z #2#3%
{%
    \expandafter\XINT_fgeq_Fe
    \romannumeral0\XINT_dsx_addzeros {#1}#3;\xint:#2\xint:
}%
\def\XINT_fgeq_Fe #1\xint:#2#3\xint:{\XINT_geq_plusplus #2#1\xint:#3\xint:}%
\def\XINT_fgeq_Fn #1\Z #2#3%
{%
    \expandafter\XINT_fgeq_Fo
    \romannumeral0\XINT_dsx_addzeros {#1}#2;\xint:#3\xint:
}%
\def\XINT_fgeq_Fo #1#2\xint:#3\xint:{\XINT_geq_plusplus #1#3\xint:#2\xint:}%
%    \end{macrocode}
% \subsection{\csh{xintMax}}
%    \begin{macrocode}
\def\xintMax {\romannumeral0\xintmax }%
\def\xintmax #1%
{%
    \expandafter\XINT_fmax\expandafter {\romannumeral0\xintraw {#1}}%
}%
\def\XINT_fmax #1#2%
{%
    \expandafter\XINT_fmax_A\romannumeral0\xintraw {#2}#1%
}%
\def\XINT_fmax_A #1#2/#3[#4]#5#6/#7[#8]%
{%
    \xint_UDsignsfork
      #1#5\XINT_fmax_minusminus
       -#5\XINT_fmax_firstneg
       #1-\XINT_fmax_secondneg
        --\XINT_fmax_nonneg_a
    \krof
    #1#5{#2/#3[#4]}{#6/#7[#8]}%
}%
\def\XINT_fmax_minusminus --%
   {\expandafter-\romannumeral0\XINT_fmin_nonneg_b }%
\def\XINT_fmax_firstneg #1-#2#3{ #1#2}%
\def\XINT_fmax_secondneg -#1#2#3{ #1#3}%
\def\XINT_fmax_nonneg_a #1#2#3#4%
{%
    \XINT_fmax_nonneg_b {#1#3}{#2#4}%
}%
\def\XINT_fmax_nonneg_b #1#2%
{%
    \if0\romannumeral0\XINT_fgeq_A #1#2%
          \xint_afterfi{ #1}%
    \else \xint_afterfi{ #2}%
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintMaxof}}
% \lverb|1.2l protects \xintMaxof against items with non terminated
% \the\numexpr expressions.
%
% 1.4 renders the macro compatible with an empty argument and it also defines
% an accessor \XINT_Maxof suitable for xintexpr usage (formerly xintexpr
% had its own macro handling comma separated values, but it changed
% internal representation at 1.4).
%
% |
%    \begin{macrocode}
\def\xintMaxof {\romannumeral0\xintmaxof }%
\def\xintmaxof #1{\expandafter\XINT_maxof\romannumeral`&&@#1^}%
\def\XINT_Maxof{\romannumeral0\XINT_maxof}%
\def\XINT_maxof#1%
{%
    \xint_gob_til_^ #1\XINT_maxof_empty ^%
    \expandafter\XINT_maxof_loop\romannumeral0\xintraw{#1}\xint:
}%
\def\XINT_maxof_empty ^#1\xint:{ 0/1[0]}%
\def\XINT_maxof_loop #1\xint:#2%
{%
    \xint_gob_til_^ #2\XINT_maxof_e ^%
    \expandafter\XINT_maxof_loop
    \romannumeral0\xintmax{#1}{\romannumeral0\xintraw{#2}}\xint:
}%
\def\XINT_maxof_e ^#1\xintmax #2#3\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintMin}}
%    \begin{macrocode}
\def\xintMin {\romannumeral0\xintmin }%
\def\xintmin #1%
{%
    \expandafter\XINT_fmin\expandafter {\romannumeral0\xintraw {#1}}%
}%
\def\XINT_fmin #1#2%
{%
    \expandafter\XINT_fmin_A\romannumeral0\xintraw {#2}#1%
}%
\def\XINT_fmin_A #1#2/#3[#4]#5#6/#7[#8]%
{%
    \xint_UDsignsfork
      #1#5\XINT_fmin_minusminus
       -#5\XINT_fmin_firstneg
       #1-\XINT_fmin_secondneg
        --\XINT_fmin_nonneg_a
    \krof
    #1#5{#2/#3[#4]}{#6/#7[#8]}%
}%
\def\XINT_fmin_minusminus --%
   {\expandafter-\romannumeral0\XINT_fmax_nonneg_b }%
\def\XINT_fmin_firstneg #1-#2#3{ -#3}%
\def\XINT_fmin_secondneg -#1#2#3{ -#2}%
\def\XINT_fmin_nonneg_a #1#2#3#4%
{%
    \XINT_fmin_nonneg_b {#1#3}{#2#4}%
}%
\def\XINT_fmin_nonneg_b #1#2%
{%
    \if0\romannumeral0\XINT_fgeq_A #1#2%
          \xint_afterfi{ #2}%
    \else \xint_afterfi{ #1}%
    \fi
}%
%    \end{macrocode}
% \subsection{\csh{xintMinof}}
% \lverb|1.2l protects \xintMinof against items with non terminated
% \the\numexpr expressions.
%
% 1.4 version is compatible with an empty input (empty items are handled as zero).|
%    \begin{macrocode}
\def\xintMinof {\romannumeral0\xintminof }%
\def\xintminof #1{\expandafter\XINT_minof\romannumeral`&&@#1^}%
\def\XINT_Minof{\romannumeral0\XINT_minof}%
\def\XINT_minof#1%
{%
    \xint_gob_til_^ #1\XINT_minof_empty ^%
    \expandafter\XINT_minof_loop\romannumeral0\xintraw{#1}\xint:
}%
\def\XINT_minof_empty ^#1\xint:{ 0/1[0]}%
\def\XINT_minof_loop #1\xint:#2%
{%
    \xint_gob_til_^ #2\XINT_minof_e ^%
    \expandafter\XINT_minof_loop\romannumeral0\xintmin{#1}{\romannumeral0\xintraw{#2}}\xint:
}%
\def\XINT_minof_e ^#1\xintmin #2#3\xint:{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintCmp}}
%    \begin{macrocode}
\def\xintCmp {\romannumeral0\xintcmp }%
\def\xintcmp #1%
{%
    \expandafter\XINT_fcmp\expandafter {\romannumeral0\xintraw {#1}}%
}%
\def\XINT_fcmp #1#2%
{%
    \expandafter\XINT_fcmp_A\romannumeral0\xintraw {#2}#1%
}%
\def\XINT_fcmp_A #1#2/#3[#4]#5#6/#7[#8]%
{%
    \xint_UDsignsfork
      #1#5\XINT_fcmp_minusminus
       -#5\XINT_fcmp_firstneg
       #1-\XINT_fcmp_secondneg
        --\XINT_fcmp_nonneg_a
    \krof
    #1#5{#2/#3[#4]}{#6/#7[#8]}%
}%
\def\XINT_fcmp_minusminus --#1#2{\XINT_fcmp_B #2#1}%
\def\XINT_fcmp_firstneg #1-#2#3{ -1}%
\def\XINT_fcmp_secondneg -#1#2#3{ 1}%
\def\XINT_fcmp_nonneg_a #1#2%
{%
    \xint_UDzerosfork
      #1#2\XINT_fcmp_zerozero
       0#2\XINT_fcmp_firstzero
       #10\XINT_fcmp_secondzero
        00\XINT_fcmp_pos
    \krof
    #1#2%
}%
\def\XINT_fcmp_zerozero   #1#2#3#4{ 0}%
\def\XINT_fcmp_firstzero  #1#2#3#4{ -1}%
\def\XINT_fcmp_secondzero #1#2#3#4{ 1}%
\def\XINT_fcmp_pos #1#2#3#4%
{%
    \XINT_fcmp_B #1#3#2#4%
}%
\def\XINT_fcmp_B #1/#2[#3]#4/#5[#6]%
{%
    \expandafter\XINT_fcmp_C\expandafter
    {\the\numexpr #6-#3\expandafter}\expandafter
    {\romannumeral0\xintiimul {#4}{#2}}%
    {\romannumeral0\xintiimul {#5}{#1}}%
}%
\def\XINT_fcmp_C #1#2#3%
{%
    \expandafter\XINT_fcmp_D\expandafter
    {#3}{#1}{#2}%
}%
\def\XINT_fcmp_D #1#2#3%
{%
    \expandafter\XINT_cntSgnFork\romannumeral`&&@\expandafter\XINT_cntSgn
    \the\numexpr #2+\xintLength{#3}-\xintLength{#1}\relax\xint:
    { -1}{\XINT_fcmp_E #2\Z {#3}{#1}}{ 1}%
}%
\def\XINT_fcmp_E #1%
{%
    \xint_UDsignfork
        #1\XINT_fcmp_Fd
         -{\XINT_fcmp_Fn #1}%
    \krof
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_fcmp_Fd #1\Z #2#3%
{%
    \expandafter\XINT_fcmp_Fe
    \romannumeral0\XINT_dsx_addzeros {#1}#3;\xint:#2\xint:
}%
\def\XINT_fcmp_Fe #1\xint:#2#3\xint:{\XINT_cmp_plusplus #2#1\xint:#3\xint:}%
\def\XINT_fcmp_Fn #1\Z #2#3%
{%
    \expandafter\XINT_fcmp_Fo
    \romannumeral0\XINT_dsx_addzeros {#1}#2;\xint:#3\xint:
}%
\def\XINT_fcmp_Fo #1#2\xint:#3\xint:{\XINT_cmp_plusplus #1#3\xint:#2\xint:}%
%    \end{macrocode}
% \subsection{\csh{xintAbs}}
%    \begin{macrocode}
\def\xintAbs   {\romannumeral0\xintabs }%
\def\xintabs #1{\expandafter\XINT_abs\romannumeral0\xintraw {#1}}%
%    \end{macrocode}
% \subsection{\csh{xintOpp}}
%    \begin{macrocode}
\def\xintOpp   {\romannumeral0\xintopp }%
\def\xintopp #1{\expandafter\XINT_opp\romannumeral0\xintraw {#1}}%
%    \end{macrocode}
% \subsection{\csh{xintInv}}
% \changed{1.3d}
%    \begin{macrocode}
\def\xintInv   {\romannumeral0\xintinv }%
\def\xintinv #1{\expandafter\XINT_inv\romannumeral0\xintraw {#1}}%
\def\XINT_inv #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_inv_iszero
      0#1\XINT_inv_a
       0-{\XINT_inv_a {}}%
    \krof #1%
}%
\def\XINT_inv_iszero #1]%
   {\XINT_signalcondition{DivisionByZero}{Inverse of zero: inv(#1]).}{}{ 0/1[0]}}%
\def\XINT_inv_a #1#2/#3[#4#5]%
{%
    \xint_UDzerominusfork
      #4-\XINT_inv_expiszero
      0#4\XINT_inv_b
       0-{\XINT_inv_b -#4}%
    \krof #5.{#1#3/#2}%
}%
\def\XINT_inv_expiszero #1.#2{ #2[0]}%
\def\XINT_inv_b #1.#2{ #2[#1]}%
%    \end{macrocode}
% \subsection{\csh{xintSgn}}
%    \begin{macrocode}
\def\xintSgn   {\romannumeral0\xintsgn }%
\def\xintsgn #1{\expandafter\XINT_sgn\romannumeral0\xintraw {#1}\xint:}%
%    \end{macrocode}
% \subsection{\csh{xintGCD}}
% \changed{1.4}
% They replace the former \xintgcdnameimp macros of the same names which
% truncated to integers their arguments. Fraction-producing |gcd()| and
% |lcm()| functions were available since |1.3d| \xintexprnameimp, with
% non-public support macros handling comma separated values.
%
% \changed{1.4d}
% Somewhat strangely \csh{xintGCD} was  formerly \csh{xintGCDof} used with only two
% arguments, as the latter directly implemented a fractionl gcd algorithm
% using \csh{xintMod} repeatedly for two arguments.
%
% Now \csh{xintGCD} contains the pairwise gcd routine and \csh{xintGCDof}
% is only a wrapper. And the pairwise gcd is reduced to integer-only
% computations to hopefully reduce fraction overhead.
%
% Each input is filtered via |\xintPIrr| and |\xintREZ| to reduce size
% of maniuplate integers in algebra.
%
% But hesitation about applying |\xintPIrr| to output, and/or |\xintREZ|.
% (as it is applied on input).
%
% But as the code is now used for frational lcm's we actually need to do
% some reduction of output else lcm's of integers will not be necessarily
% printed by |\xinteval| as integers.
% 
% Well finally I apply |\xintIrr| (but not |\xintREZ| to output).
% Hesitations here (thinking of inputs with large [n] parts, the output
% will have many zeros). So I do this only for the user macro but 
% the core routine as used by |\xintGCDof| will not do it.
%
% Also at |1.4d| the code uses |\expanded|.
%    \begin{macrocode}
\def\xintGCD {\romannumeral0\xintgcd}%
\def\xintgcd #1%
{%
    \expandafter\XINT_fgcd_in
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#1}}}\xint:
}%
\def\XINT_fgcd_in #1#2\xint:#3%
{%
    \expandafter\XINT_fgcd_out
    \romannumeral0\expandafter\XINT_fgcd_chkzeros\expandafter#1%
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#3}}}\xint:#1#2\xint:
}%
\def\XINT_fgcd_out#1[#2]{\xintirr{#1[#2]}[0]}%
\def\XINT_fgcd_chkzeros #1#2%
{%
    \xint_UDzerofork
       #1\XINT_fgcd_aiszero
       #2\XINT_fgcd_biszero
        0\XINT_fgcd_main
    \krof #2%
}%
\def\XINT_fgcd_aiszero #1\xint:#2\xint:{ #1}%
\def\XINT_fgcd_biszero #1\xint:#2\xint:{ #2}%
\def\XINT_fgcd_main #1/#2[#3]\xint:#4/#5[#6]\xint:
{%
    \expandafter\XINT_fgcd_a
      \romannumeral0\XINT_gcd_loop #2\xint:#5\xint:\xint:
      #2\xint:#5\xint:#1\xint:#4\xint:#3.#6.%
}%
\def\XINT_fgcd_a #1\xint:#2\xint:
{%
    \expandafter\XINT_fgcd_b
    \romannumeral0\xintiiquo{#2}{#1}\xint:#1\xint:#2\xint:
}%
\def\XINT_fgcd_b #1\xint:#2\xint:#3\xint:#4\xint:#5\xint:#6\xint:#7.#8.%
{%
   \expanded{%
   \xintiigcd{\xintiiE{\xintiiMul{#5}{\xintiiQuo{#4}{#2}}}{#7-#8}}%
             {\xintiiE{\xintiiMul{#6}{#1}}{#8-#7}}%
    /\xintiiMul{#1}{#4}%
    [\ifnum#7>#8 #8\else #7\fi]%
   }%
}%
%    \end{macrocode}
% \subsection{\csh{xintGCDof}}
% \changed{1.4}
% This inherits from former non public \xintexprnameimp macro called
% |\xintGCDof:csv|, which handled comma separated items.
%
% It handles fractions presented as braced items and is the support macro
% for the |gcd()| function in |\xintexpr| and |\xintfloatexpr|. The support
% macro for the |gcd()| function in |\xintiiexpr| is \csbxint{iiGCDof}, from
% \xintnameimp.
%
% An empty input is allowed but I have some hesitations on the return
% value of 1.
%
% \changed{1.4d}
% Sadly the |1.4| version had multiple problems:
% \begin{itemize}
% \item broken if first argument vanished,
% \item broken if some argument was not in strict format, for example
%       had leading chains of signs or zeros (|\xintGCDof{2}{03}|).
%       This bug originates in the fact the original macro
%       was used only in \xintexprnameimp sanitized context.
% \end{itemize}
%
% Also, output is now always an irreducible fraction (ending with |[0]|).
%    \begin{macrocode}
\def\xintGCDof {\romannumeral0\xintgcdof}%
\def\xintgcdof #1{\expandafter\XINT_fgcdof\romannumeral`&&@#1^}%
\def\XINT_GCDof{\romannumeral0\XINT_fgcdof}%
\def\XINT_fgcdof #1%
{%
    \expandafter\XINT_fgcdof_chkempty\romannumeral`&&@#1\xint:
}%
\def\XINT_fgcdof_chkempty #1%
{%
    \xint_gob_til_^#1\XINT_fgcdof_empty ^\XINT_fgcdof_in #1%
}%
\def\XINT_fgcdof_empty #1\xint:{ 1/1[0]}% hesitation, should it be infinity? O?
\def\XINT_fgcdof_in #1\xint:
{%
    \expandafter\XINT_fgcd_out
    \romannumeral0\expandafter\XINT_fgcdof_loop
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#1}}}\xint:
}%
\def\XINT_fgcdof_loop #1\xint:#2%
{%
    \expandafter\XINT_fgcdof_chkend\romannumeral`&&@#2\xint:#1\xint:\xint:
}%
\def\XINT_fgcdof_chkend #1%
{%
    \xint_gob_til_^#1\XINT_fgcdof_end ^\XINT_fgcdof_loop_pair #1%
}%
\def\XINT_fgcdof_end #1\xint:#2\xint:\xint:{ #2}%
\def\XINT_fgcdof_loop_pair #1\xint:#2%
{%
    \expandafter\XINT_fgcdof_loop
    \romannumeral0\expandafter\XINT_fgcd_chkzeros\expandafter#2%
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#1}}}\xint:#2%
}%
%    \end{macrocode}
% \subsection{\csh{xintLCM}}
% Same comments as for \csh{xintGCD}.
% Entirely redone for |1.4d|.
% Well, actually we can express it in terms of fractional gcd.
%    \begin{macrocode}
\def\xintLCM {\romannumeral0\xintlcm}%
\def\xintlcm #1%
{%
    \expandafter\XINT_flcm_in
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#1}}}\xint:
}%
\def\XINT_flcm_in #1#2\xint:#3%
{%
    \expandafter\XINT_fgcd_out
    \romannumeral0\expandafter\XINT_flcm_chkzeros\expandafter#1%
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#3}}}\xint:#1#2\xint:
}%
\def\XINT_flcm_chkzeros #1#2%
{%
    \xint_UDzerofork
       #1\XINT_flcm_zero
       #2\XINT_flcm_zero
        0\XINT_flcm_main
    \krof #2%
}%
\def\XINT_flcm_zero #1\xint:#2\xint:{ 0/1[0]}%
\def\XINT_flcm_main #1/#2[#3]\xint:#4/#5[#6]\xint:
{%
    \xintinv
    {%
      \romannumeral0\XINT_fgcd_main #2/#1[-#3]\xint:#5/#4[-#6]\xint:
    }%
}%
%    \end{macrocode}
% \subsection{\csh{xintLCMof}}
% See comments for |\xintGCDof|. \xintnameimp provides the integer only
% \csbxint{iiLCMof}.
%
% \changed{1.4d}{}
% Sadly, although a public \xintfracnameimp macro, it did not (since |1.4|)
% sanitize its arguments like other \xintfracnameimp macros.
%
%    \begin{macrocode}
\def\xintLCMof {\romannumeral0\xintlcmof}%
\def\xintlcmof #1{\expandafter\XINT_flcmof\romannumeral`&&@#1^}%
\def\XINT_LCMof{\romannumeral0\XINT_flcmof}%
\def\XINT_flcmof #1%
{%
    \expandafter\XINT_flcmof_chkempty\romannumeral`&&@#1\xint:
}%
\def\XINT_flcmof_chkempty #1%
{%
    \xint_gob_til_^#1\XINT_flcmof_empty ^\XINT_flcmof_in #1%
}%
\def\XINT_flcmof_empty #1\xint:{ 0/1[0]}% hesitation
\def\XINT_flcmof_in #1\xint:
{%
    \expandafter\XINT_fgcd_out
    \romannumeral0\expandafter\XINT_flcmof_loop
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#1}}}\xint:
}%
\def\XINT_flcmof_loop #1\xint:#2%
{%
    \expandafter\XINT_flcmof_chkend\romannumeral`&&@#2\xint:#1\xint:\xint:
}%
\def\XINT_flcmof_chkend #1%
{%
    \xint_gob_til_^#1\XINT_flcmof_end ^\XINT_flcmof_loop_pair #1%
}%
\def\XINT_flcmof_end #1\xint:#2\xint:\xint:{ #2}%
\def\XINT_flcmof_loop_pair #1\xint:#2%
{%
    \expandafter\XINT_flcmof_chkzero
    \romannumeral0\expandafter\XINT_flcm_chkzeros\expandafter#2%
    \romannumeral0\xintrez{\xintPIrr{\xintAbs{#1}}}\xint:#2%
}%
\def\XINT_flcmof_chkzero #1%
{%
    \xint_gob_til_zero#1\XINT_flcmof_zero0\XINT_flcmof_loop#1%
}%
\def\XINT_flcmof_zero#1^{ 0/1[0]}%
%    \end{macrocode}
% \subsection{Floating point macros}
%
% For a long time the float routines dating back to releases |1.07/1.08a|
% (May-June 2013) were not modified.
%
% Since |1.2f| (March 2016) the four operations first round their arguments to
% |\xinttheDigits|-floats (or |P|-floats), not (|\xinttheDigits+2|)-floats or
% (|P+2|)-floats as was the case with earlier releases.
%
% The four operations addition, subtraction, multiplication, division have
% always produced the correct rounding of the theoretical exact value to |P|
% or |\xinttheDigits| digits when the inputs are decimal numbers with at most
% |P| digits, and arbitrary decimal exponent part.
%
% From |1.08a| to |1.2j|, |\xintFloat| (and |\XINTinFloat| which is used to
% parse inputs to other float macros) handled a fractional input |A/B| via an
% initial replacement to |A'/B'| where |A'| and |B'| were |A| and |B|
% truncated to |Q+2| digits (where asked-for precision is |Q|), and then they
% correctly rounded |A'/B'| to |Q| digits. But this meant that this rounding of
% the input could differ (by up to one unit in the last place) from the
% correct rounding of the original |A/B| to the asked-for number of
% digits (which until |1.2f| in uses as auxiliary to the macros for the basic
% operations was 2 more than the prevailing precision).
%
% Since |1.2k| all inputs are correctly rounded to the asked-for number of
% digits (this was, I think, the case in the |1.07| release -- there are no
% code comments -- but was, afaicr, not very efficiently done, and this is why
% the |1.08a| release opeted for truncation of the numerator and denominator.)
%
% Notice that in float expressions, the |/| is treated as operator, hence the
% above discussion makes a difference only for the special input form
% |qfloat(A/B)| or for an |\xintexpr A/B\relax| embedded in the float
% expression, with |A| or |B| having more digits than the prevailing float
% precision.
%
% \begin{framed}
%   Internally there is no inner representation of |P|-floats as such !!!!!
%
%   The input parser will again compute the length of the mantissa on each use
%   !!! This is obviously something that must be improved upon before
%   implementation of higher functions.
%
%   Currently, special tricks are used to quickly recognize inputs having no
%   denominators, or fractions whose numerators and denominators are not too
%   long compared to the target precision |P|, and in particular |P|-floats or
%   quotients of two such.
%
%   Another long-standing issue is that float multiplication will first
%   compute the |2P| or |2P-1| digits of the exact product, and then round it
%   to |P| digits. This is sub-optimal for large |P| particularly as the
%   multiplication algorithm is basically the schoolbook one, hence
%   \emph{worse} than quadratic in the \TeX\ implementation which has extra
%   cost of fetching long sequences of tokens.
% \end{framed}
%
% \lverb|Changes at 1.4e (done 2021/04/15; undone 2021/04/29)
%
% Macros named \XINTinFloat<name> are not public user-level but were designed
% a long time ago for \xintfloatexpr context as a very preliminary step
% towards attempting to preserve some internal format, here A[N] type.
%
% When <name> is lowercased it means it needs a \romannumeral0 trigger
% (\XINTinfloatS keeps an uppercase S).
%
% Most were coded to check for an optional argument [D], and to use
% D=\XINTdigits in its place if absent but it turned out only \XINTinfloatpow,
% \XINTinfloatmul, \XINTinfloatadd were actually used with an optional
% argument and this happened only in macros from the very old xintseries.sty,
% so I changed all of them to not check for optional argument [D] anymore,
% keeping only some private interface for the xintseries.sty use case. Some
% required being used with [D], some still had names ending in "digits"
% indicating they would use \XINTdigits always.
%
% Indeed basically all algebra is done "exactly" and the [D] governs rules of
% float-rounding on input and output.
%
% During development of 1.4e we fleetingly experimented with letting the value
% used in place of D be \XINTdigitsx to 1.4e, i.e. \XINTdigits with guard
% digits, a situation which was motivated by the implementation of
% trigonometrical functions at high level, i.e. using \xintdeffloatfunc which
% had no mechanism to make intermediate calculations with guard digits.
% 
% Simply doing everything "as is" but with 2 guard digits proved very good
% (surprisingly efficient, even) to the trigonometrical functions. However
% using them systematically raises many issues (for example, the correct
% rounding at P digits is destroyed if we obtain it a D=P+2 then round from
% P+2 to P digits so we definitely can not do this as default, so some
% interface is needed to define intermediate functions only using such guard
% digits and keeping them in their output).
%
% Finally, an approach limited to the xinttrig.sty scope was used and I
% removed all \XINTdigitsx related matters from 1.4e. But this left some
% modifications of the interfaces of the "float" macros here which this list
% tries to document, mainly for the author's benefit.
%
%( Macros always using \XINTdigits and now not allowing [P] option
%: \XINTinFloatAdd
%: \XINTinFloatSub
%: \XINTinFloatMul
%: \XINTinFloatSqr
%: \XINTinFloatInv
%: \XINTinFloatDiv
%: \XINTinFloatPow
%: \XINTinFloatPower
%: \XINTinFloatPFactorial
%: \XINTinFloatBinomial
%:
%: Macros which already did not allow [P] option prior to 1.4e refactoring
%: \XINTinFloatFrac (renamed from \XINTinFloatFracdigits)
%: \XINTinFloatE
%: \XINTinFloatMod
%: \XINTinFloatDivFloor
%: \XINTinFloatDivMod
%:
%: Macros requiring a [P]. Some of the "_wopt" named macros are&
%  renamings of macros formerly requiring [P].
%: \XINTinFloat
%: \XINTinFloatS
%: \XINTFloatiLogTen
%: \XINTinRandomFloatS (this one has only the [P] mandatory argument)
%: \XINTinFloatFac
%: \XINTinFloatSqrt
%: \XINTinFloatAdd_wopt, \XINTinfloatadd_wopt
%: \XINTinFloatSub_wopt, \XINTinfloatsub_wopt
%: \XINTinFloatMul_wopt, \XINTinfloatmul_wopt
%: \XINTinFloatSqr_wopt
%: \XINTinfloatpow_wopt (not FloatPow)
%: \XINTinFloatDiv_wopt
%: \XINTinFloatInv_wopt
%:
%: Specially named macros indicating usage of \XINTdigits
%: \XINTinFloatdigits
%: \XINTinFloatSdigits
%: \XINTFloatiLogTendigits
%: \XINTinRandomFloatSdigits
%: \XINTinFloatFacdigits
%: \XINTinFloatSqrtdigits
%)
% |
%
% \subsection{\csh{xintDigits}, \csh{xintSetDigits}}
% \lverb|&
%
% 1.3f allows \xintDigits= in place of \xintDigits:= syntax.
% It defines \xintDigits*[:]= which reloads xinttrig.sty. Perhaps this should
% be default, well.
%
% During 1.4e development I added an interface for guard digits, but I decided
% to drop inclusion from 1.4e release because there were pending issues both
% in documentation and functionalities for which I did not have time left.
%
% 1.4e fixes the issue that \xinttheDigits could not be used in the right hand
% side of \xintDigits[*][:]=...; or inside the argument to \xintSetDigits.
% |
%    \begin{macrocode}
\mathchardef\XINTdigits  16
\chardef\XINTguarddigits 0
\def\xinttheDigits     {\number\XINTdigits}%
%\def\xinttheGuardDigits{\number\XINTguarddigits}%
\def\xinttheGuardDigits{0}% in case used in some of my test files
\def\xintDigits #1={\afterassignment\xintDigits_i\mathchardef\XINT_digits=}%
\def\xintDigits_i#1%
{%
    \let\XINTdigits\XINT_digits
}%
\def\xintSetDigits #1%
{%
    \mathchardef\XINT_digits=\numexpr#1\relax
    \let\XINTdigits=\XINT_digits
}%
%    \end{macrocode}
%
% \subsection{\csh{xintFloat}}
% \lverb|&
% 1.2f and 1.2g brought some refactoring which resulted in faster treatment of
% decimal inputs. 1.2i dropped use of some old routines dating back to pre 1.2
% era in favor of more modern \xintDSRr for rounding. Then 1.2k improves
% again the handling of denominators B with few digits.
%
% But the main change with 1.2k is a complete rewrite of the B>1 case in
% order to achieve again correct rounding in all cases.
%
% The original version from 1.07 (May 2013) computed the exact rounding
% to P digits for all inputs. But from 1.08 on (June 2013), the macro handled
% A/B input by first truncating both A and B to at most P+2 digits. This meant
% that decimal input (arbitrarily long, with scientific part) was correctly
% rounded, but in case of fractional input there could be up to 0.6 unit in
% the last place difference of the produced rounding to the input, hence the
% output could differ from the correct rounding.
%
% Example with 16 digits (the default): \xintFloat {1/17597472569900621233}$newline
% with xintfrac 1.07: 5.682634230727187e-20$newline
% with xintfrac 1.08b--1.2j: 5.682634230727188e-20$newline
% with xintfrac 1.2k: 5.682634230727187e-20$newline
% The exact value is 5.682634230727187499924124...e-20, showing that 1.07 and
% 1.2k
% produce the correct rounding.
%
% Currently the code ends in a more costly branch in about 1 case among 500,
% where it does some extra operations (a multiplication in particular). There
% is a free parameter delta (here set at 4), I have yet to make some numerical
% explorations, to see if it could be favorable to set it to a higher value
% (with delta=5, there is only 1 exceptional case in 5000, etc...).
%
% I have always hesitated about the policy of printing 10.00...0 in case of
% rounding upwards to the next power of ten. Already since 1.2f \XINTinFloat
% always produced a mantissa with exactly P digits (except for the zero
% value). Starting with 1.2k, \xintFloat drops this habit of printing
% 10.00..0 in such cases. Side note: the rounding-up detection worked when the
% input A/B was with numerator A and denominator B having each less than P+2
% digits, or with B=1, else, it could happen that the output was a power of
% ten but not detected to be a rounding up of the original fraction. The value
% was ok, but printed 1.0...0eN with P-1 zeroes, not 10.0...0e(N-1).
%
% I decided it was not worth the effort to enhance the algorithm to detect
% with 100$% fiability all cases of rounding up to next
% power of ten, hence 1.2k dropped this.
%
% To avoid duplication of code, and any extra burden on \XINTinFloat, which is
% the macro used internally by the float macros for parsing their inputs, we
% simply make now \xintFloat a wrapper of \XINTinFloat.|
%    \begin{macrocode}
\def\xintFloat   {\romannumeral0\xintfloat }%
\def\xintfloat #1{\XINT_float_chkopt #1\xint:}%
\def\XINT_float_chkopt #1%
{%
    \ifx [#1\expandafter\XINT_float_opt
       \else\expandafter\XINT_float_noopt
    \fi  #1%
}%
\def\XINT_float_noopt #1\xint:%
{%
    \expandafter\XINT_float_post
    \romannumeral0\XINTinfloat[\XINTdigits]{#1}\XINTdigits.%
}%
\def\XINT_float_opt [\xint:#1]%
{%
    \expandafter\XINT_float_opt_a\the\numexpr #1.%
}%
\def\XINT_float_opt_a #1.#2%
{%
    \expandafter\XINT_float_post
    \romannumeral0\XINTinfloat[#1]{#2}#1.%
}%
\def\XINT_float_post #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_float_zero
       0#1\XINT_float_neg
       0-\XINT_float_pos
    \krof #1%
}%[
\def\XINT_float_zero #1]#2.{ 0.e0}%
\def\XINT_float_neg-{\expandafter-\romannumeral0\XINT_float_pos}%
\def\XINT_float_pos #1#2[#3]#4.%
{%
    \expandafter\XINT_float_pos_done\the\numexpr#3+#4-\xint_c_i.#1.#2;%
}%
\def\XINT_float_pos_done #1.#2;{ #2e#1}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloat}, \csh{XINTinFloatS}, \csh{XINTiLogTen}}
% \lverb|&
% This routine is like \xintFloat but produces an output of the shape A[N]
% which is then parsed faster as input to other float macros. 
% Float operations in \xintfloatexpr...\relax use internally this format.
%
% It must be used in form \XINTinFloat[P]{f}: the optional [P] is
% mandatory.
%
% Since 1.2f, the mantissa always has exactly P digits even in case of
% rounding up to next power of ten. This simplifies other routines.
%
% (but the zero value must always be checked for, as it outputs 0[0])
%
% 1.2g added a variant \XINTinFloatS which, in case of decimal input with less
% than the asked for precision P will not add extra zeros to the mantissa. For
% example it may output 2[0] even if P=500, rather than the canonical
% representation 200...000[-499]. This is how \xintFloatMul and \xintFloatDiv
% parse their inputs, which speeds-up follow-up processing. But \xintFloatAdd
% and \xintFloatSub still use \XINTinFloat for parsing their inputs; anyway
% this will have to be changed again when inner structure will carry upfront
% at least the length of mantissa as data.
%
% Each time \XINTinFloat is called it at least computes a length. Naturally if
% we had some format for floats that would be dispensed of...$newline
%  something like
% <letterP><length of mantissa>.mantissa.exponent, etc... not yet.
%
% Since 1.2k, \XINTinFloat always correctly rounds its argument, even if it
% is a fraction with very big numerator and denominator. See the discussion of
% \xintFloat. 
%
% 1.3e adds \XINTFloatiLogTen.
% |
%    \begin{macrocode}
\def\XINTinFloat {\romannumeral0\XINTinfloat }%
\def\XINTinfloat 
   {\expandafter\XINT_infloat_clean\romannumeral0\XINT_infloat}%
%    \end{macrocode}
% \lverb|Attention que ici le fait que l'on grabbe #1 est important car il
% pourrait y avoir un zéro (en particulier dans le cas où input est nul).|
%    \begin{macrocode}
\def\XINT_infloat_clean #1%
   {\if #1!\xint_dothis\XINT_infloat_clean_a\fi\xint_orthat{ }#1}%
%    \end{macrocode}
% \lverb|Ici on ajoute les zeros pour faire exactement avec P chiffres.
% Car le #1 = P - L avec L la longueur de #2, (ou de abs(#2), ici le #2 peut
% avoir un signe) qui est < P|
%    \begin{macrocode}
\def\XINT_infloat_clean_a !#1.#2[#3]%
{%
    \expandafter\XINT_infloat_done
    \the\numexpr #3-#1\expandafter.%
    \romannumeral0\XINT_dsx_addzeros {#1}#2;;%
}%
\def\XINT_infloat_done #1.#2;{ #2[#1]}%
%    \end{macrocode}
% \lverb|variant which allows output with shorter mantissas.|
%    \begin{macrocode}
\def\XINTinFloatS {\romannumeral0\XINTinfloatS}%
\def\XINTinfloatS 
   {\expandafter\XINT_infloatS_clean\romannumeral0\XINT_infloat}%
\def\XINT_infloatS_clean #1%
   {\if #1!\xint_dothis\XINT_infloatS_clean_a\fi\xint_orthat{ }#1}%
\def\XINT_infloatS_clean_a !#1.{ }%
%    \end{macrocode}
% \lverb|1.3e ajoute \XINTFloatiLogTen. Le comportement pour un input nul est non
% encore finalisé. Il changera lorsque NaN, +Inf, -Inf existeront.
% |
%    \begin{macrocode}
\def\XINTFloatiLogTen {\the\numexpr\XINTfloatilogten}%
\def\XINTfloatilogten [#1]#2%
    {\expandafter\XINT_floatilogten\romannumeral0\XINT_infloat[#1]{#2}#1.}%
\def\XINTFloatiLogTendigits{\the\numexpr\XINTfloatilogten[\XINTdigits]}%
\def\XINT_floatilogten #1{%
    \if #10\xint_dothis\XINT_floatilogten_z\fi
    \if #1!\xint_dothis\XINT_floatilogten_a\fi
    \xint_orthat\XINT_floatilogten_b #1%
}%
\def\XINT_floatilogten_z 0[0]#1.{-"7FFF8000\relax}%
\def\XINT_floatilogten_a !#1.#2[#3]#4.{#3-#1+#4-1\relax}%
\def\XINT_floatilogten_b #1[#2]#3.{#2+#3-1\relax}%
%    \end{macrocode}
% \lverb|début de la routine proprement dite,
% l'argument optionnel est obligatoire.|
%    \begin{macrocode}
\def\XINT_infloat [#1]#2%
{%
    \expandafter\XINT_infloat_a\the\numexpr #1\expandafter.%
    \romannumeral0\XINT_infrac {#2}%
}%
%    \end{macrocode}
% \lverb| #1=P, #2=n, #3=A, #4=B.|
%    \begin{macrocode}
\def\XINT_infloat_a #1.#2#3#4%
{%
%    \end{macrocode}
% \lverb|micro boost au lieu d'utiliser \XINT_isOne{#4}, mais pas bon style.|
%    \begin{macrocode}
    \if1\XINT_is_One#4XY%
      \expandafter\XINT_infloat_sp
    \else\expandafter\XINT_infloat_fork
    \fi #3.{#1}{#2}{#4}%
}%
%    \end{macrocode}
% \lverb|Special quick treatment of B=1 case (1.2f then again 1.2g.)$newline
% maintenant: A.{P}{N}{1}
% Il est possible que A soit nul.
% | 
%    \begin{macrocode}
\def\XINT_infloat_sp #1%
{%
    \xint_UDzerominusfork
     #1-\XINT_infloat_spzero
     0#1\XINT_infloat_spneg
      0-\XINT_infloat_sppos
    \krof #1%
}%
%    \end{macrocode}
% \lverb|Attention surtout pas 0/1[0] ici.|
%    \begin{macrocode}
\def\XINT_infloat_spzero 0.#1#2#3{ 0[0]}%
\def\XINT_infloat_spneg-% 
    {\expandafter\XINT_infloat_spnegend\romannumeral0\XINT_infloat_sppos}%
\def\XINT_infloat_spnegend #1%
    {\if#1!\expandafter\XINT_infloat_spneg_needzeros\fi -#1}%
\def\XINT_infloat_spneg_needzeros -!#1.{!#1.-}%
%    \end{macrocode}
% \lverb|in:  A.{P}{N}{1}$newline
% out: P-L.A.P.N.|
%    \begin{macrocode}
\def\XINT_infloat_sppos #1.#2#3#4%
{%
    \expandafter\XINT_infloat_sp_b\the\numexpr#2-\xintLength{#1}.#1.#2.#3.%
}%
%    \end{macrocode}
% \lverb|#1= P-L. Si c'est positif ou nul il faut retrancher #1 à l'exposant, et
% ajouter autant de zéros. On regarde premier token.
% P-L.A.P.N.|
%    \begin{macrocode}
\def\XINT_infloat_sp_b #1%
{%
    \xint_UDzerominusfork
     #1-\XINT_infloat_sp_quick
     0#1\XINT_infloat_sp_c
      0-\XINT_infloat_sp_needzeros
    \krof #1%
}%
%    \end{macrocode}
% \lverb|Ici P=L. Le cas usuel dans \xintfloatexpr.|
%    \begin{macrocode}
\def\XINT_infloat_sp_quick 0.#1.#2.#3.{ #1[#3]}%
%    \end{macrocode}
% \lverb|Ici #1=P-L est >0. L'exposant sera N-(P-L). #2=A. #3=P. #4=N.$newline
% 18 mars 2016. En fait dans certains contextes il est sous-optimal d'ajouter les
% zéros. Par exemple quand c'est appelé par la multiplication ou la division,
% c'est idiot de convertir 2 en 200000...00000[-499].
% Donc je redéfinis addzeros en needzeroes. Si on appelle sous la forme
% \XINTinFloatS, on ne fait pas l'addition de zeros.|
%    \begin{macrocode}
\def\XINT_infloat_sp_needzeros #1.#2.#3.#4.{!#1.#2[#4]}%
%    \end{macrocode}
% \lverb|L-P=#1.A=#2#3.P=#4.N=#5.$newline
% Ici P<L. Il va falloir arrondir. Attention si on va à la puissance de 10
% suivante. En #1 on a L-P qui est >0. L'exposant final sera N+L-P, 
% sauf dans le cas spécial, il sera alors N+L-P+1. L'ajustement final
% est fait par \XINT_infloat_Y.|
%    \begin{macrocode}
\def\XINT_infloat_sp_c -#1.#2#3.#4.#5.%
{%
    \expandafter\XINT_infloat_Y
    \the\numexpr #5+#1\expandafter.%
    \romannumeral0\expandafter\XINT_infloat_sp_round
    \romannumeral0\XINT_split_fromleft
    (\xint_c_i+#4).#2#3\xint_bye2345678\xint_bye..#2%
}%
\def\XINT_infloat_sp_round #1.#2.%
{%
    \XINT_dsrr#1\xint_bye\xint_Bye3456789\xint_bye/\xint_c_x\relax.%
}%
%    \end{macrocode}
% \lverb|General branch for A/B with B>1 inputs. It achieves correct rounding
% always since 1.2k (done January 2, 2017.) This branch is never taken for A=0
% because \XINT_infrac will have returned B=1 then.|
%    \begin{macrocode}
\def\XINT_infloat_fork #1%
{%
    \xint_UDsignfork
     #1\XINT_infloat_J
     -\XINT_infloat_K
    \krof #1%
}%
\def\XINT_infloat_J-{\expandafter-\romannumeral0\XINT_infloat_K }%
%    \end{macrocode}
% \lverb?A.{P}{n}{B} avec B>1.?
%    \begin{macrocode}
\def\XINT_infloat_K #1.#2%
{%
    \expandafter\XINT_infloat_L
    \the\numexpr\xintLength{#1}\expandafter.\the\numexpr #2+\xint_c_iv.{#1}{#2}%
}%
%    \end{macrocode}
% \lverb?|A|.P+4.{A}{P}{n}{B}. We check if A already has length
% <= P+4.?
%    \begin{macrocode}
\def\XINT_infloat_L #1.#2.%
{%
    \ifnum #1>#2
      \expandafter\XINT_infloat_Ma
    \else
      \expandafter\XINT_infloat_Mb
    \fi #1.#2.%
}%
%    \end{macrocode}
% \lverb?|A|.P+4.{A}{P}{n}{B}. We will keep only the first P+4
% digits of A, denoted A'' in what follows.
% 
% output: u=-0.A''.junk.P+4.|A|.{A}{P}{n}{B}?
%    \begin{macrocode}
\def\XINT_infloat_Ma #1.#2.#3%
{%
    \expandafter\XINT_infloat_MtoN\expandafter-\expandafter0\expandafter.%
    \romannumeral0\XINT_split_fromleft#2.#3\xint_bye2345678\xint_bye..%
    #2.#1.{#3}%
}%
%    \end{macrocode}
% \lverb?|A|.P+4.{A}{P}{n}{B}.$newline
% Here A is short. We set u = P+4-|A|, and A''=A (A' = 10^u A)
%
% output: u.A''..P+4.|A|.{A}{P}{n}{B}?
%    \begin{macrocode}
\def\XINT_infloat_Mb #1.#2.#3%
{%
    \expandafter\XINT_infloat_MtoN\the\numexpr#2-#1.%
    #3..#2.#1.{#3}%
}%
%    \end{macrocode}
% \lverb?input u.A''.junk.P+4.|A|.{A}{P}{n}{B}$newline
% output |B|.P+4.{B}u.A''.P.|A|.n.{A}{B}?
%    \begin{macrocode}
\def\XINT_infloat_MtoN #1.#2.#3.#4.#5.#6#7#8#9%
{%
   \expandafter\XINT_infloat_N
   \the\numexpr\xintLength{#9}.#4.{#9}#1.#2.#7.#5.#8.{#6}{#9}%
}%
\def\XINT_infloat_N #1.#2.%
{%
    \ifnum #1>#2
      \expandafter\XINT_infloat_Oa
    \else
      \expandafter\XINT_infloat_Ob
    \fi #1.#2.%
}%
%    \end{macrocode}
% \lverb?input |B|.P+4.{B}u.A''.P.|A|.n.{A}{B}$newline
% output v=-0.B''.junk.|B|.u.A''.P.|A|.n.{A}{B}?
%    \begin{macrocode}
\def\XINT_infloat_Oa #1.#2.#3%
{%
    \expandafter\XINT_infloat_P\expandafter-\expandafter0\expandafter.%
    \romannumeral0\XINT_split_fromleft#2.#3\xint_bye2345678\xint_bye..%
    #1.%
}%
%    \end{macrocode}
% \lverb?output v=P+4-|B|>=0.B''.junk.|B|.u.A''.P.|A|.n.{A}{B}?
%    \begin{macrocode}
\def\XINT_infloat_Ob #1.#2.#3%
{%
    \expandafter\XINT_infloat_P\the\numexpr#2-#1.#3..#1.%
}%
%    \end{macrocode}
% \lverb?input v.B''.junk.|B|.u.A''.P.|A|.n.{A}{B}$newline
% output Q1.P.|B|.|A|.n.{A}{B}$newline
% Q1 = division euclidienne de  A''.10^{u-v+P+3} par B''.
%
% Special detection of cases with A and B both having length at most P+4: this
% will happen when called from \xintFloatDiv as A and B (produced then via
% \XINTinFloatS) will have at most P digits. We then only need integer division
% with P+1 extra zeros, not P+3.?
%    \begin{macrocode}
\def\XINT_infloat_P #1#2.#3.#4.#5.#6#7.#8.#9.%
{%
   \csname XINT_infloat_Q\if-#1\else\if-#6\else q\fi\fi\expandafter\endcsname
   \romannumeral0\xintiiquo
   {\romannumeral0\XINT_dsx_addzerosnofuss
      {#6#7-#1#2+#9+\xint_c_iii\if-#1\else\if-#6\else-\xint_c_ii\fi\fi}#8;}%
   {#3}.#9.#5.%
}%
%    \end{macrocode}
% \lverb?«quick» branch.?
%    \begin{macrocode}
\def\XINT_infloat_Qq #1.#2.%
{%
    \expandafter\XINT_infloat_Rq
    \romannumeral0\XINT_split_fromleft#2.#1\xint_bye2345678\xint_bye..#2.%
}%
\def\XINT_infloat_Rq #1.#2#3.%
{%
    \ifnum#2<\xint_c_v
         \expandafter\XINT_infloat_SEq
    \else\expandafter\XINT_infloat_SUp
    \fi
    {\if.#3.\xint_c_\else\xint_c_i\fi}#1.%
}%
%    \end{macrocode}
% \lverb?standard branch which will have to handle undecided rounding, if too
% close to a mid-value.?
%    \begin{macrocode}
\def\XINT_infloat_Q #1.#2.%
{%
    \expandafter\XINT_infloat_R
    \romannumeral0\XINT_split_fromleft#2.#1\xint_bye2345678\xint_bye..#2.%
}%
\def\XINT_infloat_R #1.#2#3#4#5.%
{%
    \if.#5.\expandafter\XINT_infloat_Sa\else\expandafter\XINT_infloat_Sb\fi
    #2#3#4#5.#1.%
}%
%    \end{macrocode}
% \lverb?trailing digits.Q.P.|B|.|A|.n.{A}{B}$newline
% #1=trailing digits (they may have leading zeros.)?
%    \begin{macrocode}
\def\XINT_infloat_Sa #1.%
{%
    \ifnum#1>500 \xint_dothis\XINT_infloat_SUp\fi
    \ifnum#1<499 \xint_dothis\XINT_infloat_SEq\fi
    \xint_orthat\XINT_infloat_X\xint_c_
}%
\def\XINT_infloat_Sb #1.%
{%
    \ifnum#1>5009 \xint_dothis\XINT_infloat_SUp\fi
    \ifnum#1<4990 \xint_dothis\XINT_infloat_SEq\fi
    \xint_orthat\XINT_infloat_X\xint_c_i
}%
%    \end{macrocode}
% \lverb?epsilon #2=Q.#3=P.#4=|B|.#5=|A|.#6=n.{A}{B}$newline
% exposant final est n+|A|-|B|-P+epsilon?
%    \begin{macrocode}
\def\XINT_infloat_SEq #1#2.#3.#4.#5.#6.#7#8%
{%
    \expandafter\XINT_infloat_SY
    \the\numexpr #6+#5-#4-#3+#1.#2.%
}%
\def\XINT_infloat_SY #1.#2.{ #2[#1]}%
%    \end{macrocode}
% \lverb?initial digit #2 put aside to check for case of rounding up to
% next power of ten, which will need adjustment of mantissa and exponent.?
%    \begin{macrocode}
\def\XINT_infloat_SUp #1#2#3.#4.#5.#6.#7.#8#9%
{%
    \expandafter\XINT_infloat_Y
    \the\numexpr#7+#6-#5-#4+#1\expandafter.%
    \romannumeral0\xintinc{#2#3}.#2%
}%
%    \end{macrocode}
% \lverb?epsilon Q.P.|B|.|A|.n.{A}{B}$newline
%
% \xintDSH{-x}{U} multiplies U by 10^x. When x is negative, this means
% it truncates (i.e. it drops the last -x digits).
%
% We don't try to optimize too much macro calls here, the odds are 2 per 1000
% for this branch to be taken. Perhaps in future I will use higher free
% parameter d, which currently is set at 4.
%
% #1=epsilon, #2#3=Q, #4=P, #5=|B|, #6=|A|, #7=n, #8=A, #9=B?
%    \begin{macrocode}
\def\XINT_infloat_X #1#2#3.#4.#5.#6.#7.#8#9%
{%
   \expandafter\XINT_infloat_Y
   \the\numexpr #7+#6-#5-#4+#1\expandafter.%
   \romannumeral`&&@\romannumeral0\xintiiiflt
     {\xintDSH{#6-#5-#4+#1}{\xintDouble{#8}}}%
     {\xintiiMul{\xintInc{\xintDouble{#2#3}}}{#9}}%
   \xint_firstofone
   \xintinc{#2#3}.#2%
}%
%    \end{macrocode}
% \lverb?check for rounding up to next power of ten.?
%    \begin{macrocode}
\def\XINT_infloat_Y #1{%
\def\XINT_infloat_Y ##1.##2##3.##4%
{%
   \if##49\if##21\expandafter\expandafter\expandafter\XINT_infloat_Z\fi\fi
   #1##2##3[##1]%
}}\XINT_infloat_Y{ }%
%    \end{macrocode}
% \lverb?#1=1, #2=0.?
%    \begin{macrocode}
\def\XINT_infloat_Z #1#2#3[#4]%
{%
    \expandafter\XINT_infloat_ZZ\the\numexpr#4+\xint_c_i.#3.%
}%
\def\XINT_infloat_ZZ #1.#2.{ 1#2[#1]}% 
%    \end{macrocode}
% \subsection{\csh{xintPFloat}, \csh{xintPFloatE}}
% \lverb|&
%
% xint has not yet incorporated a general formatter as it was
% not a priority during development and external solutions exist
% (I did not check for a while but I think LaTeX3 has implemented
%  a general formatter in the printf or Python ".format" spirit)
%
% But when one starts using really the package, especially in
% an interactive way (xintsession 2021), one needs the default
% output to be as nice as possible.
%
% The \xintPFloat macro was added at 1.1 as a "prettifying printer"
% for floats, basically influenced by Maple.
%
%( The rules were:
%: 0. The input is float-rounded to either Digits or the optional argument
%: 1. zero is printed as "0."
%: 2. x.yz...eK is printed "as is" if K>5 or K<-5.
%: 3. if -5<=K<=5, fixed point decimal notation is used.
%: 4. in cases 2. and 3., no trimming of trailing zeroes.
%)
%
% 1.4b added \xintPFloatE to customize whether to use
% e or E.
%
%
% 1.4e, with some hesitation, decided to make a breaking change and to modify
% the behaviour.
%
%( The new rules:
%: 0. The input is float-rounded to either Digits or the optional argument
%: 1. zero is printed as 0.0
%: 2. x.yz...eK is printed in decimal fixed point if -4<=K<=+5&
%     (notice the change, formerly K=-5 used fixed point notation in output)&
%     else it is printed in scientific notation
%: 3. trailing zeros of the mantissa are trimmed always
%: 4. in case of decimal fixed point for an integer, there&
%     is a trailing ".0"
%: 5. in case of scientific notation with a one-digit trimmed mantissa&
%     there is an added ".0" too
%)
%
% Further, \xintPFloatE can now grab the scientific exponent K which is
% presented to it as explicit tokens (digit tokens, at least one, and an
% optional minus sign) delimited by a dot. It is thus now possible to
% customize at will for example adding a + sign in case of positive scientific
% exponent. The macro must be f-expandable.
%
% |
%    \begin{macrocode}
\def\xintPFloat {\romannumeral0\xintpfloat }%
\def\xintpfloat #1{\XINT_pfloat_chkopt #1\xint:}%
\def\xintPFloat_wopt
{%
    \romannumeral0\expandafter\XINT_pfloat\romannumeral0\XINTinfloatS
}%
\def\XINT_pfloat_chkopt #1%
{%
    \ifx [#1\expandafter\XINT_pfloat_opt
       \else\expandafter\XINT_pfloat_noopt
    \fi  #1%
}%
\def\XINT_pfloat_noopt #1\xint:%
{%
    \expandafter\XINT_pfloat\romannumeral0\XINTinfloatS[\XINTdigits]{#1}%
}%
\def\XINT_pfloat_opt [\xint:#1]%
{%
    \expandafter\XINT_pfloat\romannumeral0\XINTinfloatS[#1]%
}%
\def\XINT_pfloat#1]%
{%
    \expandafter\XINT_pfloat_fork\romannumeral0\xintrez{#1]}%
}%
\def\XINT_pfloat_fork#1%
{%
    \xint_UDzerominusfork
       #1-\XINT_pfloat_zero
       0#1\XINT_pfloat_neg
        0-\XINT_pfloat_pos
    \krof #1%
}%
\def\XINT_pfloat_zero#1]{ 0.0}%
\def\XINT_pfloat_neg-{\expandafter-\romannumeral0\XINT_pfloat_pos}%
\def\XINT_pfloat_pos#1/1[#2]%
{%
    \expandafter\XINT_pfloat_a\the\numexpr\xintLength{#1}.%
    #2.#1.%
}%
\def\XINT_pfloat_a #1.#2#3.%
{%
    \expandafter\XINT_pfloat_b\the\numexpr#1+#2#3-\xint_c_i.%
    #2#1.%
}%
\def\XINT_pfloat_b #1.#2%
{%
    \ifnum #1>\xint_c_v  \xint_dothis\XINT_pfloat_sci\fi
    \ifnum #1<-\xint_c_iv \xint_dothis\XINT_pfloat_sci\fi
    \ifnum #1<\xint_c_   \xint_dothis\XINT_pfloat_N\fi
    \if-#2\xint_dothis\XINT_pfloat_P\fi
    \xint_orthat\XINT_pfloat_Ps
    #1.%
}%
%    \end{macrocode}
% \lverb|&
% #1 is the scientific exponent, #2 is the length of trimmed mantissa.
%
% \xintPFloatE can be replaced by any f-expandable macro with a dot-delimited
% argument.
% |
%    \begin{macrocode}
\def\XINT_pfloat_sci #1.#2.%
{%
    \ifnum#2=\xint_c_i\expandafter\XINT_pfloat_sci_i\expandafter\fi
    \expandafter\XINT_pfloat_sci_a\romannumeral`&&@\xintPFloatE #1.%
}%
\def\XINT_pfloat_sci_a #1.#2#3.{ #2.#3#1}%
%    \end{macrocode}
% \lverb|&
% #1#2=\fi\XINT_pfloat_sci_a
%
% 1-digit mantissa, hesitation between d.0eK or deK|
%    \begin{macrocode}
\edef\XINT_pfloat_sci_i #1#2#3.#4.{#1\space#4.0#3}%
\def\xintPFloatE{e}%
\def\XINT_pfloat_N#1.#2.#3.%
{%
    \csname XINT_pfloat_N_\romannumeral-#1\endcsname #3%
}%
\def\XINT_pfloat_N_i  { 0.}%
\def\XINT_pfloat_N_ii { 0.0}%
\def\XINT_pfloat_N_iii{ 0.00}%
\def\XINT_pfloat_N_iv { 0.000}%
\def\XINT_pfloat_P #1.#2.#3.%
{%
    \csname XINT_pfloat_P_\romannumeral#1\endcsname #3%
}%
\def\XINT_pfloat_P_   #1{ #1.}%
\def\XINT_pfloat_P_i  #1#2{ #1#2.}%
\def\XINT_pfloat_P_ii #1#2#3{ #1#2#3.}%
\def\XINT_pfloat_P_iii#1#2#3#4{ #1#2#3#4.}%
\def\XINT_pfloat_P_iv #1#2#3#4#5{ #1#2#3#4#5.}%
\def\XINT_pfloat_P_v  #1#2#3#4#5#6{ #1#2#3#4#5#6.}%
\def\XINT_pfloat_Ps #1.#2.#3.%
{%
    \csname XINT_pfloat_Ps_\romannumeral#1\endcsname #300000.%
}%
\def\XINT_pfloat_Ps_   #1#2.{ #1.0}%
\def\XINT_pfloat_Ps_i  #1#2#3.{ #1#2.0}%
\def\XINT_pfloat_Ps_ii #1#2#3#4.{ #1#2#3.0}%
\def\XINT_pfloat_Ps_iii#1#2#3#4#5.{ #1#2#3#4.0}%
\def\XINT_pfloat_Ps_iv #1#2#3#4#5#6.{ #1#2#3#4#5.0}%
\def\XINT_pfloat_Ps_v  #1#2#3#4#5#6#7.{ #1#2#3#4#5#6.0}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloatFrac}}
% \lverb|1.09i, for frac function in \xintfloatexpr. This version computes
% exactly from the input the fractional part and then only converts it
% into a float with the asked-for number of digits. I will have to think
% it again some day, certainly.
%
% 1.1 removes optional argument for which there was anyhow no interface, for
% technical reasons having to do with \xintNewExpr.
%
% 1.1a renames the macro as \XINTinFloatFracdigits (from \XINTinFloatFrac) to
% be synchronous with the \XINTinFloatSqrt and \XINTinFloat habits related to
% \xintNewExpr context and issues with macro names.
%
% Note to myself: I still have to rethink the whole thing about what is the best
% to do, the initial way of going through \xinttfrac was just a first
% implementation.
%
% 1.4e renames it back  to \XINTinFloatFrac because of all such similarly named
% macros also using \XINTdigits forcedly.
% |
%    \begin{macrocode}
\def\XINTinFloatFrac {\romannumeral0\XINTinfloatfrac}%
\def\XINTinfloatfrac #1%
{%
    \expandafter\XINT_infloatfrac_a\expandafter {\romannumeral0\xinttfrac{#1}}%
}%
\def\XINT_infloatfrac_a {\XINTinfloat[\XINTdigits]}%
%    \end{macrocode}
% \subsection{\csh{xintFloatAdd}, \csh{XINTinFloatAdd}}
% \lverb|First included in release 1.07.
%
% 1.09ka improved a bit the efficiency. However the add, sub, mul, div
% routines were provisory and supposed to be revised soon.
%
% Which didn't happen until 1.2f. Now, the inputs are first rounded to P
% digits, not P+2 as earlier.
%
% See general introduction for important changes at 1.4e relative to
% the \XINTinFloat<name> macros.
%
%
%|
%    \begin{macrocode}
\def\xintFloatAdd  {\romannumeral0\xintfloatadd}%
\def\xintfloatadd  #1{\XINT_fladd_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatAdd{\romannumeral0\XINTinfloatadd }%
\def\XINTinfloatadd{\XINT_fladd_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINTinFloatAdd_wopt{\romannumeral0\XINTinfloatadd_wopt}%
\def\XINTinfloatadd_wopt[#1]{\expandafter\XINT_fladd_opt_a\the\numexpr#1.\XINTinfloatS}%
\def\XINT_fladd_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_fladd_opt
       \else\expandafter\XINT_fladd_noopt
    \fi  #1#2%
}%
\def\XINT_fladd_noopt #1#2\xint:#3%
{%
    #1[\XINTdigits]%
    {\expandafter\XINT_FL_add_a
     \romannumeral0\XINTinfloat[\XINTdigits]{#2}\XINTdigits.{#3}}%
}%
\def\XINT_fladd_opt #1[\xint:#2]%#3#4%
{%
    \expandafter\XINT_fladd_opt_a\the\numexpr #2.#1%
}%
\def\XINT_fladd_opt_a #1.#2#3#4%
{%
    #2[#1]{\expandafter\XINT_FL_add_a\romannumeral0\XINTinfloat[#1]{#3}#1.{#4}}%
}%
\def\XINT_FL_add_a #1%
{%
    \xint_gob_til_zero #1\XINT_FL_add_zero 0\XINT_FL_add_b #1%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_FL_add_zero #1.#2{#2}%[[
\def\XINT_FL_add_b #1]#2.#3%
{%
    \expandafter\XINT_FL_add_c\romannumeral0\XINTinfloat[#2]{#3}#2.#1]%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_FL_add_c #1%
{%
    \xint_gob_til_zero #1\XINT_FL_add_zero 0\XINT_FL_add_d #1%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_FL_add_d #1[#2]#3.#4[#5]%
{%
    \ifnum\numexpr #2-#3-#5>\xint_c_\xint_dothis\xint_firstoftwo\fi
    \ifnum\numexpr #5-#3-#2>\xint_c_\xint_dothis\xint_secondoftwo\fi
    \xint_orthat\xintAdd {#1[#2]}{#4[#5]}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatSub}, \csh{XINTinFloatSub}}
% \lverb|First done 1.07.
%
% Starting with 1.2f the arguments undergo an intial rounding to the target
% precision P not P+2.|
%
%    \begin{macrocode}
\def\xintFloatSub  {\romannumeral0\xintfloatsub}%
\def\xintfloatsub  #1{\XINT_flsub_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatSub{\romannumeral0\XINTinfloatsub}%
\def\XINTinfloatsub{\XINT_flsub_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINTinFloatSub_wopt{\romannumeral0\XINTinfloatsub_wopt}%
\def\XINTinfloatsub_wopt[#1]{\expandafter\XINT_flsub_opt_a\the\numexpr#1.\XINTinfloatS}%
\def\XINT_flsub_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flsub_opt
       \else\expandafter\XINT_flsub_noopt
    \fi  #1#2%
}%
\def\XINT_flsub_noopt #1#2\xint:#3%
{%
    #1[\XINTdigits]%
    {\expandafter\XINT_FL_add_a
     \romannumeral0\XINTinfloat[\XINTdigits]{#2}\XINTdigits.{\xintOpp{#3}}}%
}%
\def\XINT_flsub_opt #1[\xint:#2]%#3#4%
{%
    \expandafter\XINT_flsub_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flsub_opt_a #1.#2#3#4%
{%
    #2[#1]{\expandafter\XINT_FL_add_a\romannumeral0\XINTinfloat[#1]{#3}#1.{\xintOpp{#4}}}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatMul}, \csh{XINTinFloatMul}}
% \lverb|1.07.
%
% Starting with 1.2f the arguments are rounded to the target precision P not
% P+2.
%
% 1.2g handles the inputs via \XINTinFloatS which will be more efficient when
% the precision is large and the input is for example a small constant like 2.
%
% 1.2k does a micro improvement to the way the macro passes over control
% to its output routine (former version used a higher level \xintE causing
% some extra un-needed processing with two calls to \XINT_infrac where
% one was amply enough).|
%    \begin{macrocode}
\def\xintFloatMul  {\romannumeral0\xintfloatmul}%
\def\xintfloatmul  #1{\XINT_flmul_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatMul{\romannumeral0\XINTinfloatmul}%
\def\XINTinfloatmul{\XINT_flmul_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINTinFloatMul_wopt{\romannumeral0\XINTinfloatmul_wopt}%
\def\XINTinfloatmul_wopt[#1]{\expandafter\XINT_flmul_opt_a\the\numexpr#1.\XINTinfloatS}%
\def\XINT_flmul_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flmul_opt
       \else\expandafter\XINT_flmul_noopt
    \fi  #1#2%
}%
\def\XINT_flmul_noopt #1#2\xint:#3%
{%
    #1[\XINTdigits]%
    {\expandafter\XINT_FL_mul_a
     \romannumeral0\XINTinfloatS[\XINTdigits]{#2}\XINTdigits.{#3}}%
}%
\def\XINT_flmul_opt #1[\xint:#2]%#3#4%
{%
    \expandafter\XINT_flmul_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flmul_opt_a #1.#2#3#4%
{%
    #2[#1]{\expandafter\XINT_FL_mul_a\romannumeral0\XINTinfloatS[#1]{#3}#1.{#4}}%
}%
\def\XINT_FL_mul_a #1[#2]#3.#4%
{%
    \expandafter\XINT_FL_mul_b\romannumeral0\XINTinfloatS[#3]{#4}#1[#2]%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_FL_mul_b #1[#2]#3[#4]{\xintiiMul{#3}{#1}/1[#4+#2]}%
%    \end{macrocode}
% \subsection{\csh{xintFloatSqr}, \csh{XINTinFloatSqr}}
% \lverb|Added only at 1.4e, strangely \xintFloatSqr had never been defined so far.
%
% An \XINTinFloatSqr{#1} was defined in xintexpr.sty directly as
% \XINTinFloatMul[\XINTdigits]{#1}{#1}, to support the sqr() function. The
% {#1}{#1} causes no problem as #1 in this context is always pre-expanded so
% we don't need to worry about this, and the \xintdeffloatfunc mechanism
% should hopefully take care to add the needed argument pre-expansion if need
% be.
%
% Anyway let's do this finally properly here.
% |
%    \begin{macrocode}
\def\xintFloatSqr  {\romannumeral0\xintfloatsqr}%
\def\xintfloatsqr  #1{\XINT_flsqr_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatSqr{\romannumeral0\XINTinfloatsqr}%
\def\XINTinfloatsqr{\XINT_flsqr_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINT_flsqr_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flsqr_opt
       \else\expandafter\XINT_flsqr_noopt
    \fi  #1#2%
}%
\def\XINT_flsqr_noopt #1#2\xint:
{%
    #1[\XINTdigits]%
    {\expandafter\XINT_FL_sqr_a\romannumeral0\XINTinfloatS[\XINTdigits]{#2}}%
}%
\def\XINT_flsqr_opt #1[\xint:#2]%
{%
    \expandafter\XINT_flsqr_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flsqr_opt_a #1.#2#3%
{%
    #2[#1]{\expandafter\XINT_FL_sqr_a\romannumeral0\XINTinfloatS[#1]{#3}}%
}%
\def\XINT_FL_sqr_a #1[#2]{\xintiiSqr{#1}/1[#2+#2]}%
\def\XINTinFloatSqr_wopt[#1]#2{\XINTinFloatS[#1]{\expandafter\XINT_FL_sqr_a\romannumeral0\XINTinfloatS[#1]{#2}}}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloatInv}}
% \lverb|Added belatedly at 1.3e, to support inv() function. We use Short
% output, for rare inv(\xintexpr 1/3\relax) case. I need to think the whole
% thing out at some later date.|
%    \begin{macrocode}
\def\XINTinFloatInv#1{\XINTinFloatS[\XINTdigits]{\xintInv{#1}}}%
\def\XINTinFloatInv_wopt[#1]#2{\XINTinFloatS[#1]{\xintInv{#2}}}%
%    \end{macrocode}
% \subsection{\csh{xintFloatDiv}, \csh{XINTinFloatDiv}}
% \lverb|1.07.
%
% Starting with 1.2f the arguments are rounded to the target precision P not
% P+2.
%
% 1.2g handles the inputs via \XINTinFloatS which will be more efficient when
% the precision is large and the input is for example a small constant like 2.
%
% The actual rounding of the quotient is handled via \xintfloat (or
% \XINTinfloatS).
%
% 1.2k does the same kind of improvement in \XINT_FL_div_b as for
% multiplication: earlier code was unnecessarily high level.
% |
%    \begin{macrocode}
\def\xintFloatDiv  {\romannumeral0\xintfloatdiv}%
\def\xintfloatdiv  #1{\XINT_fldiv_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatDiv{\romannumeral0\XINTinfloatdiv}%
\def\XINTinfloatdiv{\XINT_fldiv_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINTinFloatDiv_wopt[#1]{\romannumeral0\XINT_fldiv_opt_a#1.\XINTinfloatS}%
\def\XINT_fldiv_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_fldiv_opt
       \else\expandafter\XINT_fldiv_noopt
    \fi  #1#2%
}%
%    \end{macrocode}
% \lverb|&
% 1.4g adds here intercept of second argument being zero,
% else a low level error will arise at later stage from the
% the fall-back value returned by core iidivision being 0
% and not having expected number of digits at \XINT_infloat_Qq
% and split from left returning some empty value breaking the
% \ifnum test in \XINT_infloat_Rq.
% |
%    \begin{macrocode}
\def\XINT_fldiv_noopt #1#2\xint:#3%
{%
    #1[\XINTdigits]%
    {\expandafter\XINT_FL_div_aa
     \romannumeral0\XINTinfloatS[\XINTdigits]{#3}\XINTdigits.{#2}}%
}%
\def\XINT_FL_div_aa #1%
{%
    \xint_gob_til_zero#1\XINT_FL_div_Bzero0\XINT_FL_div_a #1%
}%
\def\XINT_FL_div_Bzero0\XINT_FL_div_a#1[#2]#3.#4%
{%
    \XINT_signalcondition{DivisionByZero}{Division by zero (#1[#2]) of #4}{}{ 0[0]}%
}%
\def\XINT_fldiv_opt #1[\xint:#2]%#3#4%
{%
    \expandafter\XINT_fldiv_opt_a\the\numexpr #2.#1%
}%
%    \end{macrocode}
% \lverb|&
% Also here added early check at 1.4g if divisor is zero.
% |
%    \begin{macrocode}
\def\XINT_fldiv_opt_a #1.#2#3#4%
{%
    #2[#1]{\expandafter\XINT_FL_div_aa\romannumeral0\XINTinfloatS[#1]{#4}#1.{#3}}%
}%
\def\XINT_FL_div_a #1[#2]#3.#4%
{%
    \expandafter\XINT_FL_div_b\romannumeral0\XINTinfloatS[#3]{#4}/#1e#2%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_FL_div_b #1[#2]{#1e#2}%
%    \end{macrocode}
% \subsection{\csh{xintFloatPow}, \csh{XINTinFloatPow}}
% \lverb|1.07: initial version. 1.09j has re-organized the core loop.
%
% 2015/12/07. I have hesitated to map ^ in expressions to \xintFloatPow rather
% than \xintFloatPower. But for 1.234567890123456 to the power 2145678912 with
% P=16, using Pow rather than Power seems to bring only about 5$char37 $space
% gain.
%
% This routine requires the exponent x to be compatible with \numexpr parsing.
%
% 1.2f has rewritten the code for better efficiency. Also, now the argument A
% for A^x is first rounded to P digits before switching to the increased
% working precision (which depends upon x).
%
% |
%    \begin{macrocode}
\def\xintFloatPow  {\romannumeral0\xintfloatpow}%
\def\xintfloatpow  #1{\XINT_flpow_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatPow{\romannumeral0\XINTinfloatpow }%
\def\XINTinfloatpow{\XINT_flpow_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINTinfloatpow_wopt[#1]{\expandafter\XINT_flpow_opt_a\the\numexpr#1.\XINTinfloatS}%
\def\XINT_flpow_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flpow_opt
       \else\expandafter\XINT_flpow_noopt
    \fi
    #1#2%
}%
\def\XINT_flpow_noopt  #1#2\xint:#3%
{%
   \expandafter\XINT_flpow_checkB_a
   \the\numexpr #3.\XINTdigits.{#2}{#1[\XINTdigits]}%
}%
\def\XINT_flpow_opt #1[\xint:#2]%
{%
   \expandafter\XINT_flpow_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flpow_opt_a #1.#2#3#4%
{%
   \expandafter\XINT_flpow_checkB_a\the\numexpr #4.#1.{#3}{#2[#1]}%
}%
\def\XINT_flpow_checkB_a #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_flpow_BisZero
      0#1{\XINT_flpow_checkB_b -}%
       0-{\XINT_flpow_checkB_b {}#1}%
    \krof
}%
\def\XINT_flpow_BisZero .#1.#2#3{#3{1[0]}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_checkB_b #1#2.#3.%
{%
    \expandafter\XINT_flpow_checkB_c
    \the\numexpr\xintLength{#2}+\xint_c_iii.#3.#2.{#1}%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_flpow_checkB_c #1.#2.%
{%
    \expandafter\XINT_flpow_checkB_d\the\numexpr#1+#2.#1.#2.%
}%
%    \end{macrocode}
% \lverb|&
%
% 1.2f rounds input to P digits, first.
% |
%    \begin{macrocode}
\def\XINT_flpow_checkB_d #1.#2.#3.#4.#5#6%
{%
    \expandafter \XINT_flpow_aa
    \romannumeral0\XINTinfloat [#3]{#6}{#2}{#1}{#4}{#5}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_aa #1[#2]#3%
{%
    \expandafter\XINT_flpow_ab\the\numexpr #2-#3\expandafter.%
    \romannumeral\XINT_rep #3\endcsname0.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_ab #1.#2.#3.{\XINT_flpow_a #3#2[#1]}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_a #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_flpow_zero
      0#1{\XINT_flpow_b \iftrue}%
       0-{\XINT_flpow_b \iffalse#1}%
    \krof
}%
\def\XINT_flpow_zero #1[#2]#3#4#5#6%
{%
    #6{\if 1#51\xint_dothis {0[0]}\fi
       \xint_orthat
       {\XINT_signalcondition{DivisionByZero}{0 raised to power -#4.}{}{ 0[0]}}%
      }%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_b #1#2[#3]#4#5%
{%
    \XINT_flpow_loopI #5.#3.#2.#4.{#1\ifodd #5 \xint_c_i\fi\fi}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_truncate #1.#2.#3.%
{%
    \expandafter\XINT_flpow_truncate_a
    \romannumeral0\XINT_split_fromleft
    #3.#2\xint_bye2345678\xint_bye..#1.#3.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_truncate_a #1.#2.#3.{#3+\xintLength{#2}.#1.}%
\def\XINT_flpow_loopI #1.%
{%
    \ifnum #1=\xint_c_i\expandafter\XINT_flpow_ItoIII\fi
    \ifodd #1
       \expandafter\XINT_flpow_loopI_odd
    \else
       \expandafter\XINT_flpow_loopI_even
    \fi
    #1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_ItoIII\ifodd #1\fi #2.#3.#4.#5.#6%
{%
    \expandafter\XINT_flpow_III\the\numexpr #6+\xint_c_.#3.#4.#5.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_loopI_even #1.#2.#3.%#4.%
{%
    \expandafter\XINT_flpow_loopI
    \the\numexpr #1/\xint_c_ii\expandafter.%
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#2\expandafter.\romannumeral0\xintiisqr{#3}.%
}%
\def\XINT_flpow_loopI_odd #1.#2.#3.#4.%
{%
    \expandafter\XINT_flpow_loopII
    \the\numexpr #1/\xint_c_ii-\xint_c_i\expandafter.%
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#2\expandafter.\romannumeral0\xintiisqr{#3}.#4.#2.#3.%
}%
\def\XINT_flpow_loopII #1.%
{%
    \ifnum #1 = \xint_c_i\expandafter\XINT_flpow_IItoIII\fi
    \ifodd #1
       \expandafter\XINT_flpow_loopII_odd
    \else
       \expandafter\XINT_flpow_loopII_even
    \fi
    #1.%
}%
\def\XINT_flpow_loopII_even #1.#2.#3.%#4.%
{%
    \expandafter\XINT_flpow_loopII
    \the\numexpr #1/\xint_c_ii\expandafter.%
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#2\expandafter.\romannumeral0\xintiisqr{#3}.%
}%
\def\XINT_flpow_loopII_odd #1.#2.#3.#4.#5.#6.%
{%
    \expandafter\XINT_flpow_loopII_odda
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr#2+#5\expandafter.\romannumeral0\xintiimul{#3}{#6}.#4.%
    #1.#2.#3.%
}%
\def\XINT_flpow_loopII_odda #1.#2.#3.#4.#5.#6.%
{%
    \expandafter\XINT_flpow_loopII
    \the\numexpr #4/\xint_c_ii-\xint_c_i\expandafter.%
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#5\expandafter.\romannumeral0\xintiisqr{#6}.#3.%
    #1.#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_IItoIII\ifodd #1\fi #2.#3.#4.#5.#6.#7.#8%
{%
    \expandafter\XINT_flpow_III\the\numexpr #8+\xint_c_\expandafter.%
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr#3+#6\expandafter.\romannumeral0\xintiimul{#4}{#7}.#5.%
}%
%    \end{macrocode}
% \lverb|This ending is common with \xintFloatPower.
%
% In the case of negative exponent we need to inverse the Q-digits mantissa.
% This requires no special attention now as 1.2k's \xintFloat does correct
% rounding of fractions hence it is easy to bound the total error. It can be
% checked that the algorithm after final rounding to the target precision
% computes a value Z whose distance to the exact theoretical will be less than
% 0.52 ulp(Z) (and worst cases can only be slightly worse than 0.51 ulp(Z)).
%
% In the case of the half-integer exponent (only via the expression
% interface,) the computation (which proceeds via \XINTinFloatPowerH) ends
% with a square root. This square root extraction is done with 3 guard digits
% (the power operations were done with more.) Then the value is rounded to the
% target precision. There is thus this rounding to 3 guard digits (in the case
% of negative exponent the reciprocal is computed before the square-root),
% then the square root is (computed with exact rounding for these 3 guard
% digits), and then there is the final rounding of this to the target
% precision. The total error (for positive as well as negative exponent) has
% been estimated to at worst possibly exceed slightly 0.5125 ulp(Z), and at
% any rate it is less than 0.52 ulp(Z).
%
% |
%    \begin{macrocode}
\def\XINT_flpow_III #1.#2.#3.#4.#5%
{%
    \expandafter\XINT_flpow_IIIend
    \xint_UDsignfork
         #5{{1/#3[-#2]}}%
         -{{#3[#2]}}%
    \krof #1%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpow_IIIend #1#2#3%
    {#3{\if#21\xint_afterfi{\expandafter-\romannumeral`&&@}\fi#1}}%
%    \end{macrocode}
% \subsection{\csh{xintFloatPower}, \csh{XINTinFloatPower}}
% \lverb|1.07. The core loop has been re-organized in 1.09j for some slight
% efficiency gain. The exponent B is given to \xintNum. The ^ in expressions
% is mapped to this routine.
%
% Same modifications as in \xintFloatPow for 1.2f.
%
% 1.2f \XINTinFloatPowerH (now moved to $xintlognameimp, and renamed). It 
% truncated the exponent to an integer of half-integer, and in the latter case
% use Square-root extraction. At 1.2k this was improved as 1.2f stupidly
% rounded to Digits before, not after the square root extraction, 1.2k kept 3
% guard digits for this last step. And the initial step was
% changed to a rounding rather than truncating.
%
% Until 1.4e this \XINTinFloatPowerH was the macro for a^b in expressions,
% but of course it behaved strangely for b not an integer or an half-integer!
% At 1.4e, the non-integer, non-half-integer exponents will be handled via
% log10() and pow10() support macros, see $xintlognameimp. The code
% has now been relocated there.
% |
%    \begin{macrocode}
\def\xintFloatPower  {\romannumeral0\xintfloatpower}%
\def\xintfloatpower  #1{\XINT_flpower_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatPower{\romannumeral0\XINTinfloatpower }%
\def\XINTinfloatpower{\XINT_flpower_opt_a\XINTdigits.\XINTinfloatS}%
%    \end{macrocode}
% \lverb|Start of macro. Check for optional argument.|
%    \begin{macrocode}
\def\XINT_flpower_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flpower_opt
       \else\expandafter\XINT_flpower_noopt
    \fi
    #1#2%
}%
\def\XINT_flpower_noopt  #1#2\xint:#3%
{%
   \expandafter\XINT_flpower_checkB_a
   \romannumeral0\xintnum{#3}.\XINTdigits.{#2}{#1[\XINTdigits]}%
}%
\def\XINT_flpower_opt #1[\xint:#2]%
{%
   \expandafter\XINT_flpower_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flpower_opt_a #1.#2#3#4%
{%
   \expandafter\XINT_flpower_checkB_a
   \romannumeral0\xintnum{#4}.#1.{#3}{#2[#1]}%
}%
\def\XINT_flpower_checkB_a #1%
{%
    \xint_UDzerominusfork
      #1-{\XINT_flpower_BisZero 0}%
      0#1{\XINT_flpower_checkB_b -}%
       0-{\XINT_flpower_checkB_b {}#1}%
    \krof
}%
\def\XINT_flpower_BisZero 0.#1.#2#3{#3{1[0]}}%
\def\XINT_flpower_checkB_b #1#2.#3.%
{%
    \expandafter\XINT_flpower_checkB_c
    \the\numexpr\xintLength{#2}+\xint_c_iii.#3.#2.{#1}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpower_checkB_c #1.#2.%
{%
    \expandafter\XINT_flpower_checkB_d\the\numexpr#1+#2.#1.#2.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpower_checkB_d #1.#2.#3.#4.#5#6%
{%
    \expandafter \XINT_flpower_aa
    \romannumeral0\XINTinfloat [#3]{#6}{#2}{#1}{#4}{#5}%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flpower_aa #1[#2]#3%
{%
    \expandafter\XINT_flpower_ab\the\numexpr #2-#3\expandafter.%
    \romannumeral\XINT_rep #3\endcsname0.#1.%
}%
\def\XINT_flpower_ab #1.#2.#3.{\XINT_flpower_a #3#2[#1]}%
\def\XINT_flpower_a #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_flpow_zero
      0#1{\XINT_flpower_b \iftrue}%
       0-{\XINT_flpower_b \iffalse#1}%
    \krof
}%
\def\XINT_flpower_b #1#2[#3]#4#5%
{%
    \XINT_flpower_loopI #5.#3.#2.#4.{#1\xintiiOdd{#5}\fi}%
}%
\def\XINT_flpower_loopI #1.%
{%
    \if1\XINT_isOne {#1}\xint_dothis\XINT_flpower_ItoIII\fi
    \ifodd\xintLDg{#1} %<- intentional space
       \xint_dothis{\expandafter\XINT_flpower_loopI_odd}\fi
    \xint_orthat{\expandafter\XINT_flpower_loopI_even}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \romannumeral0\XINT_half
    #1\xint_bye\xint_Bye345678\xint_bye
    *\xint_c_v+\xint_c_v)/\xint_c_x-\xint_c_i\relax.%
}%
\def\XINT_flpower_ItoIII #1.#2.#3.#4.#5%
{%
    \expandafter\XINT_flpow_III\the\numexpr #5+\xint_c_.#2.#3.#4.%
}%
\def\XINT_flpower_loopI_even #1.#2.#3.#4.%
{%
    \expandafter\XINT_flpower_toloopI
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#2\expandafter.\romannumeral0\xintiisqr{#3}.#4.#1.%
}%
\def\XINT_flpower_toloopI #1.#2.#3.#4.{\XINT_flpower_loopI #4.#1.#2.#3.}%
\def\XINT_flpower_loopI_odd #1.#2.#3.#4.%
{%
    \expandafter\XINT_flpower_toloopII
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#2\expandafter.\romannumeral0\xintiisqr{#3}.#4.%
    #1.#2.#3.%
}%
\def\XINT_flpower_toloopII #1.#2.#3.#4.{\XINT_flpower_loopII #4.#1.#2.#3.}%
\def\XINT_flpower_loopII #1.%
{%
    \if1\XINT_isOne{#1}\xint_dothis\XINT_flpower_IItoIII\fi
    \ifodd\xintLDg{#1} %<- intentional space
        \xint_dothis{\expandafter\XINT_flpower_loopII_odd}\fi
    \xint_orthat{\expandafter\XINT_flpower_loopII_even}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
    \romannumeral0\XINT_half#1\xint_bye\xint_Bye345678\xint_bye
    *\xint_c_v+\xint_c_v)/\xint_c_x-\xint_c_i\relax.%
}%
\def\XINT_flpower_loopII_even #1.#2.#3.#4.%
{%
    \expandafter\XINT_flpower_toloopII
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#2\expandafter.\romannumeral0\xintiisqr{#3}.#4.#1.%
}%
\def\XINT_flpower_loopII_odd #1.#2.#3.#4.#5.#6.%
{%
    \expandafter\XINT_flpower_loopII_odda
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr#2+#5\expandafter.\romannumeral0\xintiimul{#3}{#6}.#4.%
    #1.#2.#3.%
}%
\def\XINT_flpower_loopII_odda #1.#2.#3.#4.#5.#6.%
{%
    \expandafter\XINT_flpower_toloopII
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr\xint_c_ii*#5\expandafter.\romannumeral0\xintiisqr{#6}.#3.%
    #4.#1.#2.%
}%
\def\XINT_flpower_IItoIII #1.#2.#3.#4.#5.#6.#7%
{%
    \expandafter\XINT_flpow_III\the\numexpr #7+\xint_c_\expandafter.%
    \the\numexpr\expandafter\XINT_flpow_truncate 
    \the\numexpr#2+#5\expandafter.\romannumeral0\xintiimul{#3}{#6}.#4.%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatFac}, \csh{XINTFloatFac}}
% \lverb|&1.2.
%
% Done at 1.2.
%
% At 1.3e \XINTinFloatFac uses \XINTinFloatS for output.
%
% 1.4e adds some overhead for individual evaluations in float context as it
% obeys the guard digits for the default target precision.  It is a waste for
% individual evaluation of one factorial...
% |
%    \begin{macrocode}
\def\xintFloatFac  {\romannumeral0\xintfloatfac}%
\def\xintfloatfac  #1{\XINT_flfac_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatFac{\romannumeral0\XINTinfloatfac}%
\def\XINTinfloatfac[#1]{\expandafter\XINT_flfac_opt_a\the\numexpr#1.\XINTinfloatS}%
\def\XINTinFloatFacdigits{\romannumeral0\XINT_flfac_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINT_flfac_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flfac_opt
       \else\expandafter\XINT_flfac_noopt
    \fi
     #1#2%
}%
\def\XINT_flfac_noopt  #1#2\xint:
{%
   \expandafter\XINT_FL_fac_fork_a
   \the\numexpr \xintNum{#2}.\xint_c_i \XINTdigits\XINT_FL_fac_out{#1[\XINTdigits]}%
}%
\def\XINT_flfac_opt #1[\xint:#2]%
{%
   \expandafter\XINT_flfac_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flfac_opt_a #1.#2#3%
{%
   \expandafter\XINT_FL_fac_fork_a\the\numexpr \xintNum{#3}.\xint_c_i {#1}\XINT_FL_fac_out{#2[#1]}%
}%
\def\XINT_FL_fac_fork_a #1%
{%
     \xint_UDzerominusfork
     #1-\XINT_FL_fac_iszero
     0#1\XINT_FL_fac_isneg
      0-{\XINT_FL_fac_fork_b #1}%
    \krof
}%
\def\XINT_FL_fac_iszero #1.#2#3#4#5{#5{1[0]}}%
%    \end{macrocode}
% \lverb|1.2f XINT_FL_fac_isneg returns 0, earlier versions used 1 here.|
%    \begin{macrocode}
\def\XINT_FL_fac_isneg  #1.#2#3#4#5%
{%
    #5{\XINT_signalcondition{InvalidOperation}
                     {Factorial argument is negative: -#1.}{}{ 0[0]}}%
}%
\def\XINT_FL_fac_fork_b #1.%
{%
    \ifnum #1>\xint_c_x^viii_mone\xint_dothis\XINT_FL_fac_toobig\fi
    \ifnum #1>\xint_c_x^iv\xint_dothis\XINT_FL_fac_vbig \fi
    \ifnum #1>465  \xint_dothis\XINT_FL_fac_big\fi
    \ifnum #1>101  \xint_dothis\XINT_FL_fac_med\fi
                   \xint_orthat\XINT_FL_fac_small
    #1.%
}%
\def\XINT_FL_fac_toobig #1.#2#3#4#5%
{%
    #5{\XINT_signalcondition{InvalidOperation}
                     {Factorial argument is too large: #1>=10^8.}{}{ 0[0]}}%
}%
%    \end{macrocode}
% \lverb?Computations are done with Q blocks of eight digits. When a
% multiplication has a carry, hence creates Q+1 blocks, the least significant
% one is dropped. The goal is to compute an approximate value X' to the exact
% value X, such that the final relative error (X-X')/X will be at most
% 10^{-P-1} with P the desired precision. Then, when we round X' to X'' with P
% significant digits, we can prove that the absolute error |X-X''| is bounded
% (strictly) by 0.6 ulp(X''). (ulp= unit in the last (significant) place). Let
% N be the number of such operations, the formula for Q deduces from the
% previous explanations is that 8Q should be at least P+9+k, with k the number
% of digits of N (in base 10). Note that 1.2 version used P+10+k, for 1.2f I
% reduced to P+9+k. Also, k should be the number of digits of the number N of
% multiplications done, hence for n<=10000 we can take N=n/2, or N/3, or N/4.
% This is rounded above by numexpr and always an overestimate of the actual
% number of approximate multiplications done (the first ones are exact).
% (vérifier ce que je raconte, j'ai la flemme là).
%
% We then want ceil((P+k+n)/8). Using \numexpr rounding division
% (ARRRRRGGGHHHH), if m is a positive integer, ceil(m/8) can be computed as
% (m+3)/8. Thus with m=P+10+k, this gives Q<-(P+13+k)/8. The routine actually
% computes 8(Q-1) for use in \XINT_FL_fac_addzeros.
%
% With 1.2f the formula is m=P+9+k, Q<-(P+12+k)/8, and we use now 4=12-8 rather
% than the earlier 5=13-8. Whatever happens, the value computed in
% \XINT_FL_fac_increaseP is at least 8. There will always be an extra block.
%
% Note: with Digits:=32; Maple gives for 200!:$bgroup$obeylines$obeyspaces$ttbfamily
% > factorial(200.);
% $indent                                                         375
% $indent                    0.78865786736479050355236321393218 10
% My 1.2f routine (and also 1.2) outputs:
% $indent                    7.8865786736479050355236321393219e374
% and this is the correct rounding because for 40 digits it computes
% $indent                    7.886578673647905035523632139321850622951e374
% $egroup
% Maple's result (contrarily to xint) is thus not the correct rounding but
% still it is less than 0.6 ulp wrong.
% ?
%    \begin{macrocode}
\def\XINT_FL_fac_vbig
   {\expandafter\XINT_FL_fac_vbigloop_a
    \the\numexpr \XINT_FL_fac_increaseP \xint_c_i   }%
\def\XINT_FL_fac_big
   {\expandafter\XINT_FL_fac_bigloop_a
    \the\numexpr \XINT_FL_fac_increaseP \xint_c_ii  }%
\def\XINT_FL_fac_med
   {\expandafter\XINT_FL_fac_medloop_a
    \the\numexpr \XINT_FL_fac_increaseP \xint_c_iii }%
\def\XINT_FL_fac_small
   {\expandafter\XINT_FL_fac_smallloop_a
    \the\numexpr \XINT_FL_fac_increaseP \xint_c_iv  }%
\def\XINT_FL_fac_increaseP #1#2.#3#4%
{%
    #2\expandafter.\the\numexpr\xint_c_viii*%
    ((\xint_c_iv+#4+\expandafter\XINT_FL_fac_countdigits
                    \the\numexpr #2/(#1*#3)\relax 87654321\Z)/\xint_c_viii).%
}%
\def\XINT_FL_fac_countdigits #1#2#3#4#5#6#7#8{\XINT_FL_fac_countdone }%
\def\XINT_FL_fac_countdone   #1#2\Z {#1}%
\def\XINT_FL_fac_out #1;![#2]#3%
    {#3{\romannumeral0\XINT_mul_out
         #1;!1\R!1\R!1\R!1\R!%
                   1\R!1\R!1\R!1\R!\W [#2]}}%
\def\XINT_FL_fac_vbigloop_a #1.#2.%
{%
    \XINT_FL_fac_bigloop_a \xint_c_x^iv.#2.%
    {\expandafter\XINT_FL_fac_vbigloop_loop\the\numexpr 100010001\expandafter.%
     \the\numexpr \xint_c_x^viii+#1.}%
}%
\def\XINT_FL_fac_vbigloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_FL_fac_loop_exit\fi
    \expandafter\XINT_FL_fac_vbigloop_loop
    \the\numexpr #1+\xint_c_i\expandafter.%
    \the\numexpr #2\expandafter.\the\numexpr\XINT_FL_fac_mul #1!%
}%
\def\XINT_FL_fac_bigloop_a #1.%
{%
    \expandafter\XINT_FL_fac_bigloop_b \the\numexpr
    #1+\xint_c_i-\xint_c_ii*((#1-464)/\xint_c_ii).#1.%
}%
\def\XINT_FL_fac_bigloop_b #1.#2.#3.%
{%
    \expandafter\XINT_FL_fac_medloop_a
        \the\numexpr #1-\xint_c_i.#3.{\XINT_FL_fac_bigloop_loop #1.#2.}%
}%
\def\XINT_FL_fac_bigloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_FL_fac_loop_exit\fi
    \expandafter\XINT_FL_fac_bigloop_loop
    \the\numexpr #1+\xint_c_ii\expandafter.%
    \the\numexpr #2\expandafter.\the\numexpr\XINT_FL_fac_bigloop_mul #1!%
}%
\def\XINT_FL_fac_bigloop_mul #1!%
{%
    \expandafter\XINT_FL_fac_mul
        \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)!%
}%
\def\XINT_FL_fac_medloop_a #1.%
{%
    \expandafter\XINT_FL_fac_medloop_b
        \the\numexpr #1+\xint_c_i-\xint_c_iii*((#1-100)/\xint_c_iii).#1.%
}%
\def\XINT_FL_fac_medloop_b #1.#2.#3.%
{%
    \expandafter\XINT_FL_fac_smallloop_a
        \the\numexpr #1-\xint_c_i.#3.{\XINT_FL_fac_medloop_loop #1.#2.}%
}%
\def\XINT_FL_fac_medloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_FL_fac_loop_exit\fi
    \expandafter\XINT_FL_fac_medloop_loop
    \the\numexpr #1+\xint_c_iii\expandafter.%
    \the\numexpr #2\expandafter.\the\numexpr\XINT_FL_fac_medloop_mul #1!%
}%
\def\XINT_FL_fac_medloop_mul #1!%
{%
    \expandafter\XINT_FL_fac_mul
    \the\numexpr
        \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_FL_fac_smallloop_a #1.%
{%
    \csname
       XINT_FL_fac_smallloop_\the\numexpr #1-\xint_c_iv*(#1/\xint_c_iv)\relax
    \endcsname #1.%
}%
\expandafter\def\csname XINT_FL_fac_smallloop_1\endcsname #1.#2.%
{%
    \XINT_FL_fac_addzeros #2.100000001!.{2.#1.}{#2}%
}%
\expandafter\def\csname XINT_FL_fac_smallloop_-2\endcsname #1.#2.%
{%
    \XINT_FL_fac_addzeros #2.100000002!.{3.#1.}{#2}%
}%
\expandafter\def\csname XINT_FL_fac_smallloop_-1\endcsname #1.#2.%
{%
    \XINT_FL_fac_addzeros #2.100000006!.{4.#1.}{#2}%
}%
\expandafter\def\csname XINT_FL_fac_smallloop_0\endcsname #1.#2.%
{%
    \XINT_FL_fac_addzeros #2.100000024!.{5.#1.}{#2}%
}%
\def\XINT_FL_fac_addzeros #1.%
{%
    \ifnum #1=\xint_c_viii \expandafter\XINT_FL_fac_addzeros_exit\fi
    \expandafter\XINT_FL_fac_addzeros
    \the\numexpr #1-\xint_c_viii.100000000!%
}%
%    \end{macrocode}
% \lverb|We will manipulate by successive *small* multiplications Q blocks
% 1<8d>!, terminated by 1;!. We need a custom small multiplication which
% tells us when it has create a new block, and the least significant one
% should be dropped.|
%    \begin{macrocode}
\def\XINT_FL_fac_addzeros_exit #1.#2.#3#4{\XINT_FL_fac_smallloop_loop #3#21;![-#4]}%
\def\XINT_FL_fac_smallloop_loop #1.#2.%
{%
    \ifnum #1>#2 \expandafter\XINT_FL_fac_loop_exit\fi
    \expandafter\XINT_FL_fac_smallloop_loop
    \the\numexpr #1+\xint_c_iv\expandafter.%
    \the\numexpr #2\expandafter.\romannumeral0\XINT_FL_fac_smallloop_mul #1!%
}%
\def\XINT_FL_fac_smallloop_mul #1!%
{%
    \expandafter\XINT_FL_fac_mul
    \the\numexpr
        \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%[[
\def\XINT_FL_fac_loop_exit #1!#2]#3{#3#2]}%
\def\XINT_FL_fac_mul 1#1!%
   {\expandafter\XINT_FL_fac_mul_a\the\numexpr\XINT_FL_fac_smallmul 10!{#1}}%
\def\XINT_FL_fac_mul_a #1-#2%
{%
    \if#21\xint_afterfi{\expandafter\space\xint_gob_til_exclam}\else
    \expandafter\space\fi #11;!%
}%
\def\XINT_FL_fac_minimulwc_a #1#2#3#4#5!#6#7#8#9%
{%
    \XINT_FL_fac_minimulwc_b {#1#2#3#4}{#5}{#6#7#8#9}%
}%
\def\XINT_FL_fac_minimulwc_b #1#2#3#4!#5%
{%
    \expandafter\XINT_FL_fac_minimulwc_c
    \the\numexpr \xint_c_x^ix+#5+#2*#4!{{#1}{#2}{#3}{#4}}%
}%
\def\XINT_FL_fac_minimulwc_c 1#1#2#3#4#5#6!#7%
{%
    \expandafter\XINT_FL_fac_minimulwc_d {#1#2#3#4#5}#7{#6}%
}%
\def\XINT_FL_fac_minimulwc_d #1#2#3#4#5%
{%
    \expandafter\XINT_FL_fac_minimulwc_e
    \the\numexpr \xint_c_x^ix+#1+#2*#5+#3*#4!{#2}{#4}%
}%
\def\XINT_FL_fac_minimulwc_e 1#1#2#3#4#5#6!#7#8#9%
{%
    1#6#9\expandafter!%
    \the\numexpr\expandafter\XINT_FL_fac_smallmul
    \the\numexpr \xint_c_x^viii+#1#2#3#4#5+#7*#8!%
}%
\def\XINT_FL_fac_smallmul 1#1!#21#3!%
{%
    \xint_gob_til_sc #3\XINT_FL_fac_smallmul_end;%
    \XINT_FL_fac_minimulwc_a #2!#3!{#1}{#2}%
}%
%    \end{macrocode}
% \lverb|This is the crucial ending. I note that I used here an \ifnum test
% rather than the gob_til_eightzeroes thing. Actually for eight digits there
% is much less difference than for only four.
%
% The "carry" situation is marked by a final !-1 rather than !-2 for no-carry.
% (a \numexpr muste be stopped, and leaving a - as delimiter is good as it
% will not arise earlier.)|
%    \begin{macrocode}
\def\XINT_FL_fac_smallmul_end;\XINT_FL_fac_minimulwc_a #1!;!#2#3[#4]%
{%
   \ifnum #2=\xint_c_
       \expandafter\xint_firstoftwo\else
       \expandafter\xint_secondoftwo
   \fi
   {-2\relax[#4]}%
   {1#2\expandafter!\expandafter-\expandafter1\expandafter
                  [\the\numexpr #4+\xint_c_viii]}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatPFactorial}, \csh{XINTinFloatPFactorial}}
% \lverb|2015/11/29 for 1.2f. Partial factorial pfactorial(a,b)=(a+1)...b,
% only for non-negative integers with a<=b<10^8.
%
% 1.2h (2016/11/20) now avoids raising \xintError:OutOfRangePFac if the
% condition 0<=a<=b<10^8 is violated. Same as for \xintiiPFactorial.
%
% 1.4e extends the precision in floating point context adding some overhead
% but well.|
%    \begin{macrocode}
\def\xintFloatPFactorial  {\romannumeral0\xintfloatpfactorial}%
\def\xintfloatpfactorial  #1{\XINT_flpfac_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatPFactorial{\romannumeral0\XINTinfloatpfactorial }%
\def\XINTinfloatpfactorial{\XINT_flpfac_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINT_flpfac_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flpfac_opt
       \else\expandafter\XINT_flpfac_noopt
    \fi
     #1#2%
}%
\def\XINT_flpfac_noopt  #1#2\xint:#3%
{%
   \expandafter\XINT_FL_pfac_fork
   \the\numexpr \xintNum{#2}\expandafter.%
   \the\numexpr \xintNum{#3}.\xint_c_i{\XINTdigits}{#1[\XINTdigits]}%
}%
\def\XINT_flpfac_opt #1[\xint:#2]%
{%
   \expandafter\XINT_flpfac_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flpfac_opt_a #1.#2#3#4%
{%
   \expandafter\XINT_FL_pfac_fork
   \the\numexpr \xintNum{#3}\expandafter.%
   \the\numexpr \xintNum{#4}.\xint_c_i{#1}{#2[#1]}%
}%
\def\XINT_FL_pfac_fork #1#2.#3#4.%
{%
    \unless\ifnum #1#2<#3#4 \xint_dothis\XINT_FL_pfac_one\fi
    \if-#3\xint_dothis\XINT_FL_pfac_neg \fi
    \if-#1\xint_dothis\XINT_FL_pfac_zero\fi
    \ifnum #3#4>\xint_c_x^viii_mone\xint_dothis\XINT_FL_pfac_outofrange\fi
    \xint_orthat \XINT_FL_pfac_increaseP #1#2.#3#4.%
}%
\def\XINT_FL_pfac_outofrange #1.#2.#3#4#5%
{%
    #5{\XINT_signalcondition{InvalidOperation}
                     {pFactorial with too large argument: #2 >= 10^8.}{}{ 0[0]}}%
}%
\def\XINT_FL_pfac_one  #1.#2.#3#4#5{#5{1[0]}}%
\def\XINT_FL_pfac_zero #1.#2.#3#4#5{#5{0[0]}}%
\def\XINT_FL_pfac_neg -#1.-#2.%
{%
    \ifnum #1>\xint_c_x^viii\xint_dothis\XINT_FL_pfac_outofrange\fi
    \xint_orthat {%
    \ifodd\numexpr#2-#1\relax\xint_afterfi{\expandafter-\romannumeral`&&@}\fi
    \expandafter\XINT_FL_pfac_increaseP}%
    \the\numexpr #2-\xint_c_i\expandafter.\the\numexpr#1-\xint_c_i.%
}%
%    \end{macrocode}
% \lverb|See the comments for \XINT_FL_pfac_increaseP. Case of b=a+1 should be
% filtered out perhaps. We only needed here to copy the \xintPFactorial macros and
% re-use \XINT_FL_fac_mul/\XINT_FL_fac_out. Had to modify a bit
% \XINT_FL_pfac_addzeroes. We can enter here directly with #3 equal to specify
% the precision (the calculated value before final rounding has a relative
% error less than #3.10^{-#4-1}), and #5 would hold the macro doing the final
% rounding (or truncating, if I make a FloatTrunc available) to a given number
% of digits, possibly not #4. By default the #3 is 1, but FloatBinomial calls
% it with #3=4.|
%    \begin{macrocode}
\def\XINT_FL_pfac_increaseP #1.#2.#3#4%
{%
    \expandafter\XINT_FL_pfac_a
    \the\numexpr \xint_c_viii*((\xint_c_iv+#4+\expandafter
                 \XINT_FL_fac_countdigits\the\numexpr (#2-#1-\xint_c_i)%
                    /\ifnum #2>\xint_c_x^iv #3\else(#3*\xint_c_ii)\fi\relax
                 87654321\Z)/\xint_c_viii).#1.#2.%
}%
\def\XINT_FL_pfac_a #1.#2.#3.%
{%
    \expandafter\XINT_FL_pfac_b\the\numexpr \xint_c_i+#2\expandafter.%
    \the\numexpr#3\expandafter.%
    \romannumeral0\XINT_FL_pfac_addzeroes #1.100000001!1;![-#1]%
}%
\def\XINT_FL_pfac_addzeroes #1.%
{%
    \ifnum #1=\xint_c_viii \expandafter\XINT_FL_pfac_addzeroes_exit\fi
    \expandafter\XINT_FL_pfac_addzeroes\the\numexpr #1-\xint_c_viii.100000000!%
}%
\def\XINT_FL_pfac_addzeroes_exit #1.{ }%
\def\XINT_FL_pfac_b #1.%
{%
    \ifnum #1>9999 \xint_dothis\XINT_FL_pfac_vbigloop \fi
    \ifnum #1>463  \xint_dothis\XINT_FL_pfac_bigloop   \fi
    \ifnum #1>98   \xint_dothis\XINT_FL_pfac_medloop   \fi
                   \xint_orthat\XINT_FL_pfac_smallloop #1.%
}%
\def\XINT_FL_pfac_smallloop #1.#2.%
{%
    \ifcase\numexpr #2-#1\relax
        \expandafter\XINT_FL_pfac_end_
    \or \expandafter\XINT_FL_pfac_end_i
    \or \expandafter\XINT_FL_pfac_end_ii
    \or \expandafter\XINT_FL_pfac_end_iii
    \else\expandafter\XINT_FL_pfac_smallloop_a
    \fi #1.#2.%
}%
\def\XINT_FL_pfac_smallloop_a #1.#2.%
{%
    \expandafter\XINT_FL_pfac_smallloop_b
    \the\numexpr #1+\xint_c_iv\expandafter.%
    \the\numexpr #2\expandafter.%
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_FL_pfac_smallloop_b #1.%
{%
    \ifnum #1>98  \expandafter\XINT_FL_pfac_medloop   \else
                  \expandafter\XINT_FL_pfac_smallloop \fi #1.%
}%
\def\XINT_FL_pfac_medloop #1.#2.%
{%
    \ifcase\numexpr #2-#1\relax
        \expandafter\XINT_FL_pfac_end_
    \or \expandafter\XINT_FL_pfac_end_i
    \or \expandafter\XINT_FL_pfac_end_ii
    \else\expandafter\XINT_FL_pfac_medloop_a
    \fi #1.#2.%
}%
\def\XINT_FL_pfac_medloop_a #1.#2.%
{%
    \expandafter\XINT_FL_pfac_medloop_b
    \the\numexpr #1+\xint_c_iii\expandafter.%
    \the\numexpr #2\expandafter.%
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_FL_pfac_medloop_b #1.%
{%
    \ifnum #1>463 \expandafter\XINT_FL_pfac_bigloop   \else
                  \expandafter\XINT_FL_pfac_medloop   \fi #1.%
}%
\def\XINT_FL_pfac_bigloop #1.#2.%
{%
    \ifcase\numexpr #2-#1\relax
        \expandafter\XINT_FL_pfac_end_
    \or \expandafter\XINT_FL_pfac_end_i
    \else\expandafter\XINT_FL_pfac_bigloop_a
    \fi #1.#2.%
}%
\def\XINT_FL_pfac_bigloop_a #1.#2.%
{%
    \expandafter\XINT_FL_pfac_bigloop_b
    \the\numexpr #1+\xint_c_ii\expandafter.%
    \the\numexpr #2\expandafter.%
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)!%
}%
\def\XINT_FL_pfac_bigloop_b #1.%
{%
    \ifnum #1>9999 \expandafter\XINT_FL_pfac_vbigloop  \else
                   \expandafter\XINT_FL_pfac_bigloop   \fi #1.%
}%
\def\XINT_FL_pfac_vbigloop #1.#2.%
{%
    \ifnum #2=#1
         \expandafter\XINT_FL_pfac_end_
    \else\expandafter\XINT_FL_pfac_vbigloop_a
    \fi #1.#2.%
}%
\def\XINT_FL_pfac_vbigloop_a #1.#2.%
{%
    \expandafter\XINT_FL_pfac_vbigloop
    \the\numexpr #1+\xint_c_i\expandafter.%
    \the\numexpr #2\expandafter.%
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr\xint_c_x^viii+#1!%
}%
\def\XINT_FL_pfac_end_iii #1.#2.%
{%
    \expandafter\XINT_FL_fac_out
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)*(#1+\xint_c_iii)!%
}%
\def\XINT_FL_pfac_end_ii #1.#2.%
{%
    \expandafter\XINT_FL_fac_out
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)*(#1+\xint_c_ii)!%
}%
\def\XINT_FL_pfac_end_i #1.#2.%
{%
    \expandafter\XINT_FL_fac_out
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1*(#1+\xint_c_i)!%
}%
\def\XINT_FL_pfac_end_ #1.#2.%
{%
    \expandafter\XINT_FL_fac_out
    \romannumeral0\expandafter\XINT_FL_fac_mul
    \the\numexpr \xint_c_x^viii+#1!%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatBinomial}, \csh{XINTinFloatBinomial}}
% \lverb|1.2f. We compute binomial(x,y) as pfac(x-y,x)/y!, where the numerator
% and denominator are computed with a relative error at most 4.10^{-P-2}, then
% rounded (once I have a float truncation, I will use truncation rather) to
% P+3 digits, and finally the quotient is correctly rounded to P digits. This
% will guarantee that the exact value X differs from the computed one Y by at
% most 0.6 ulp(Y). (2015/12/01).
%
% 2016/11/19 for 1.2h. As for \xintiiBinomial, hard to understand why last
% year I coded this to raise an error if y<0 or y>x ! The question of the
% Gamma function is for another occasion, here x and y must be (small)
% integers.
%
% 1.4e: same remarks as for factorial and partial factorial about added
% overhead due to extra guard digits.
%
%|
%    \begin{macrocode}
\def\xintFloatBinomial  {\romannumeral0\xintfloatbinomial}%
\def\xintfloatbinomial  #1{\XINT_flbinom_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatBinomial{\romannumeral0\XINTinfloatbinomial }%
\def\XINTinfloatbinomial{\XINT_flbinom_opt\XINTinfloatS[\xint:\XINTdigits]}%
\def\XINT_flbinom_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flbinom_opt
       \else\expandafter\XINT_flbinom_noopt
    \fi  #1#2%
}%
\def\XINT_flbinom_noopt #1#2\xint:#3%
{%
    \expandafter\XINT_FL_binom_a
    \the\numexpr\xintNum{#2}\expandafter.\the\numexpr\xintNum{#3}.\XINTdigits.#1%
}%
\def\XINT_flbinom_opt #1[\xint:#2]#3#4%
{%
    \expandafter\XINT_FL_binom_a
    \the\numexpr\xintNum{#3}\expandafter.\the\numexpr\xintNum{#4}\expandafter.%
    \the\numexpr #2.#1%
}%
\def\XINT_FL_binom_a #1.#2.%
{%
    \expandafter\XINT_FL_binom_fork \the\numexpr #1-#2.#2.#1.%
}%
\def\XINT_FL_binom_fork #1#2.#3#4.#5#6.%
{%
    \if-#5\xint_dothis \XINT_FL_binom_neg\fi
    \if-#1\xint_dothis \XINT_FL_binom_zero\fi
    \if-#3\xint_dothis \XINT_FL_binom_zero\fi
    \if0#1\xint_dothis \XINT_FL_binom_one\fi
    \if0#3\xint_dothis \XINT_FL_binom_one\fi
    \ifnum #5#6>\xint_c_x^viii_mone \xint_dothis\XINT_FL_binom_toobig\fi
    \ifnum #1#2>#3#4  \xint_dothis\XINT_FL_binom_ab \fi
                      \xint_orthat\XINT_FL_binom_aa
    #1#2.#3#4.#5#6.%
}%
\def\XINT_FL_binom_neg #1.#2.#3.#4.#5%
{%
    #5[#4]{\XINT_signalcondition{InvalidOperation}
                         {Binomial with negative argument: #3.}{}{ 0[0]}}%
}%
\def\XINT_FL_binom_toobig #1.#2.#3.#4.#5%
{%
    #5[#4]{\XINT_signalcondition{InvalidOperation}
                         {Binomial with too large argument: #3 >= 10^8.}{}{ 0[0]}}%
}%
\def\XINT_FL_binom_one  #1.#2.#3.#4.#5{#5[#4]{1[0]}}%
\def\XINT_FL_binom_zero #1.#2.#3.#4.#5{#5[#4]{0[0]}}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_FL_binom_aa  #1.#2.#3.#4.#5%
{%
    #5[#4]{\xintDiv{\XINT_FL_pfac_increaseP
           #2.#3.\xint_c_iv{#4+\xint_c_i}{\XINTinfloat[#4+\xint_c_iii]}}%
           {\XINT_FL_fac_fork_b
           #1.\xint_c_iv{#4+\xint_c_i}\XINT_FL_fac_out{\XINTinfloat[#4+\xint_c_iii]}}}%
}%
\def\XINT_FL_binom_ab  #1.#2.#3.#4.#5%
{%
    #5[#4]{\xintDiv{\XINT_FL_pfac_increaseP
           #1.#3.\xint_c_iv{#4+\xint_c_i}{\XINTinfloat[#4+\xint_c_iii]}}%
           {\XINT_FL_fac_fork_b
           #2.\xint_c_iv{#4+\xint_c_i}\XINT_FL_fac_out{\XINTinfloat[#4+\xint_c_iii]}}}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatSqrt}, \csh{XINTinFloatSqrt}}
% \lverb|First done for 1.08. 
% 
% The float version was developed at the same time as the integer one and even
% a bit earlier. As a result the integer variant had some sub-optimal parts.
% Anyway, for 1.2f I have rewritten the integer variant, and the float variant
% delegates all preparatory wrok for it until the last step. In particular the
% very low precisions are not penalized anymore from doing computations for at
% least 17 or 18 digits. Both the large and small precisions give quite
% shorter computation times.
%
% Also, after examining more closely the achieved precision I decided to
% extend the float version in order for it to obtain the correct rounding (for
% inputs already of at most P digits with P the precision) of the theoretical
% exact value.
%
% Beyond about 500 digits of precision the efficiency decreases swiftly,
% as is the case generally speaking with xintcore/xint/xintfrac arithmetic
% macros.
%
% Final note: with 1.2f the input is always first rounded to P significant
% places.
%
% 1.4e (2021/04/15) great hesitation about what to do regarding guard digits.
% This will spoil the guaranteed "correct-rounding" property for individual
% calculations... but is interesting for precision as soon as the square root
% is embedded into some larger calculation.  Annoying.  But there is \xintexpr
% which I can left configured to use strictly \xintDigits in contrast to
% \xintfloatexpr.  Ah ok and there will always be sqrt(x,\xinttheDigits) syntax
% if one wants.  And finally I keep sqrt() acting the same in expr and floatexpr.
%
% Attention that at 1.4e \XINTinFloatSqrt is defined to be used ONLY with
% optional argument.
%
%
%
% |
%    \begin{macrocode}
\def\xintFloatSqrt  {\romannumeral0\xintfloatsqrt}%
\def\xintfloatsqrt  #1{\XINT_flsqrt_chkopt \xintfloat #1\xint:}%
\def\XINTinFloatSqrt{\romannumeral0\XINTinfloatsqrt}%
\def\XINTinfloatsqrt[#1]{\expandafter\XINT_flsqrt_opt_a\the\numexpr#1.\XINTinfloatS}%
\def\XINTinFloatSqrtdigits{\romannumeral0\XINT_flsqrt_opt_a\XINTdigits.\XINTinfloatS}%
\def\XINT_flsqrt_chkopt #1#2%
{%
    \ifx [#2\expandafter\XINT_flsqrt_opt
       \else\expandafter\XINT_flsqrt_noopt
    \fi  #1#2%
}%
\def\XINT_flsqrt_noopt #1#2\xint:%
{%
    \expandafter\XINT_FL_sqrt_a 
                \romannumeral0\XINTinfloat[\XINTdigits]{#2}\XINTdigits.#1%
}%
\def\XINT_flsqrt_opt #1[\xint:#2]%#3%
{%
    \expandafter\XINT_flsqrt_opt_a\the\numexpr #2.#1%
}%
\def\XINT_flsqrt_opt_a #1.#2#3%
{%
    \expandafter\XINT_FL_sqrt_a\romannumeral0\XINTinfloat[#1]{#3}#1.#2%
}%
\def\XINT_FL_sqrt_a #1%
{%
    \xint_UDzerominusfork
     #1-\XINT_FL_sqrt_iszero
     0#1\XINT_FL_sqrt_isneg
      0-{\XINT_FL_sqrt_pos #1}%
    \krof
}%[
\def\XINT_FL_sqrt_iszero #1]#2.#3{#3[#2]{0[0]}}%
\def\XINT_FL_sqrt_isneg #1]#2.#3%
{%
   #3[#2]{\XINT_signalcondition{InvalidOperation}
                        {Square root of negative: -#1].}{}{ 0[0]}}%
}%
%    \end{macrocode}
%\lverb|&
% |
%    \begin{macrocode}
\def\XINT_FL_sqrt_pos #1[#2]#3.%
{%
    \expandafter\XINT_flsqrt
    \the\numexpr #3\ifodd #2 \xint_dothis {+\xint_c_iii.(#2+\xint_c_i).0}\fi
    \xint_orthat {+\xint_c_ii.#2.{}}#100.#3.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt #1.#2.%
{%
    \expandafter\XINT_flsqrt_a
    \the\numexpr #2/\xint_c_ii-(#1-\xint_c_i)/\xint_c_ii.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_a #1.#2.#3#4.#5.%
{%
    \expandafter\XINT_flsqrt_b
    \the\numexpr (#2-\xint_c_i)/\xint_c_ii\expandafter.%
    \romannumeral0\XINT_sqrt_start #2.#4#3.#5.#2.#4#3.#5.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_b #1.#2#3%
{%
   \expandafter\XINT_flsqrt_c
   \romannumeral0\xintiisub
    {\XINT_dsx_addzeros {#1}#2;}%
    {\xintiiDivRound{\XINT_dsx_addzeros {#1}#3;}%
                    {\XINT_dbl#2\xint_bye2345678\xint_bye*\xint_c_ii\relax}}.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_c #1.#2.%
{%
    \expandafter\XINT_flsqrt_d
    \romannumeral0\XINT_split_fromleft#2.#1\xint_bye2345678\xint_bye..%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_d #1.#2#3.%
{%
    \ifnum #2=\xint_c_v
    \expandafter\XINT_flsqrt_f\else\expandafter\XINT_flsqrt_finish\fi 
    #2#3.#1.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_finish #1#2.#3.#4.#5.#6.#7.#8{#8[#6]{#3#1[#7]}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_f 5#1.%
   {\expandafter\XINT_flsqrt_g\romannumeral0\xintinum{#1}\relax.}%
\def\XINT_flsqrt_g #1#2#3.{\if\relax#2\xint_dothis{\XINT_flsqrt_h #1}\fi
                           \xint_orthat{\XINT_flsqrt_finish 5.}}%
\def\XINT_flsqrt_h #1{\ifnum #1<\xint_c_iii\xint_dothis{\XINT_flsqrt_again}\fi
                      \xint_orthat{\XINT_flsqrt_finish 5.}}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_again #1.#2.%
{%
    \expandafter\XINT_flsqrt_again_a\the\numexpr #2+\xint_c_viii.%
}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_flsqrt_again_a #1.#2.#3.%
{%
    \expandafter\XINT_flsqrt_b
    \the\numexpr (#1-\xint_c_i)/\xint_c_ii\expandafter.%
    \romannumeral0\XINT_sqrt_start #1.#200000000.#3.%
                                   #1.#200000000.#3.%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatE}, \csh{XINTinFloatE}}
% \lverb|1.07: The fraction is the first argument contrarily to \xintTrunc and
% \xintRound.
%
% 1.2k had to rewrite this since there is no more a \XINT_float_a macro.
%
% Attention about \XINTinFloatE: it is for use by xintexpr.sty.
% With input 0 it produces on output an 0[N], not 0[0].
% |
%    \begin{macrocode}
\def\xintFloatE   {\romannumeral0\xintfloate }%
\def\xintfloate #1{\XINT_floate_chkopt #1\xint:}%
\def\XINT_floate_chkopt #1%
{%
    \ifx [#1\expandafter\XINT_floate_opt
       \else\expandafter\XINT_floate_noopt
    \fi  #1%
}%
\def\XINT_floate_noopt #1\xint:%
{%
    \expandafter\XINT_floate_post
    \romannumeral0\XINTinfloat[\XINTdigits]{#1}\XINTdigits.%
}%
\def\XINT_floate_opt [\xint:#1]%
{%
    \expandafter\XINT_floate_opt_a\the\numexpr #1.%
}%
\def\XINT_floate_opt_a #1.#2%
{%
    \expandafter\XINT_floate_post
    \romannumeral0\XINTinfloat[#1]{#2}#1.%
}%
\def\XINT_floate_post #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_floate_zero
       0#1\XINT_floate_neg
       0-\XINT_floate_pos
    \krof #1%
}%[
\def\XINT_floate_zero #1]#2.#3{ 0.e0}%
\def\XINT_floate_neg-{\expandafter-\romannumeral0\XINT_floate_pos}%
%    \end{macrocode}
% \lverb|&
% |
%    \begin{macrocode}
\def\XINT_floate_pos #1#2[#3]#4.#5%
{%
    \expandafter\XINT_float_pos_done\the\numexpr#3+#4+#5-\xint_c_i.#1.#2;%
}%
\def\XINTinFloatE {\romannumeral0\XINTinfloate }%
\def\XINTinfloate
   {\expandafter\XINT_infloate\romannumeral0\XINTinfloat[\XINTdigits]}%
\def\XINT_infloate #1[#2]#3%
   {\expandafter\XINT_infloate_end\the\numexpr #3+#2.{#1}}%
\def\XINT_infloate_end #1.#2{ #2[#1]}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloatMod}}
% \lverb|1.1. Pour emploi dans xintexpr. Code shortened at 1.2p.|
%    \begin{macrocode}
\def\XINTinFloatMod {\romannumeral0\XINTinfloatmod [\XINTdigits]}%
\def\XINTinfloatmod [#1]#2#3%
{%
    \XINTinfloat[#1]{\xintMod
        {\romannumeral0\XINTinfloat[#1]{#2}}%
        {\romannumeral0\XINTinfloat[#1]{#3}}}%
}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloatDivFloor}}
% \lverb|1.2p. Formerly // and /: in \xintfloatexpr used \xintDivFloor and
% \xintMod, hence did not round their operands to float precision beforehand.|
%    \begin{macrocode}
\def\XINTinFloatDivFloor {\romannumeral0\XINTinfloatdivfloor [\XINTdigits]}%
\def\XINTinfloatdivfloor [#1]#2#3%
{%
    \xintdivfloor
          {\romannumeral0\XINTinfloat[#1]{#2}}%
          {\romannumeral0\XINTinfloat[#1]{#3}}%
}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloatDivMod}}
% \lverb|1.2p. Pour emploi dans xintexpr, donc je ne prends pas la peine de
% faire l'expansion du modulo, qui se produira dans le \csname.
%
% Hésitation sur le quotient, faut-il l'arrondir immédiatement ?
% Finalement non, le produire comme un integer.
%
% Breaking change at 1.4 as output format is not comma separated anymore.
% Attention also that it uses \expanded.
%
% No time now at the time of completion of the big 1.4 rewrite of xintexpr
% to test whether code efficiency here can be improved to expand the second
% item of output.|
%    \begin{macrocode}
\def\XINTinFloatDivMod {\romannumeral0\XINTinfloatdivmod [\XINTdigits]}%
\def\XINTinfloatdivmod [#1]#2#3%
{%
    \expandafter\XINT_infloatdivmod
    \romannumeral0\xintdivmod
          {\romannumeral0\XINTinfloat[#1]{#2}}%
          {\romannumeral0\XINTinfloat[#1]{#3}}%
    {#1}%
}%
\def\XINT_infloatdivmod #1#2#3{\expanded{{#1}{\XINTinFloat[#3]{#2}}}}%
%    \end{macrocode}
% \subsection{\csh{xintifFloatInt}}
% \lverb|1.3a for ifint() function in \xintfloatexpr.|
%    \begin{macrocode}
\def\xintifFloatInt {\romannumeral0\xintiffloatint}%
\def\xintiffloatint #1{\expandafter\XINT_iffloatint
                       \romannumeral0\xintrez{\XINTinFloatS[\XINTdigits]{#1}}}%
\def\XINT_iffloatint #1#2/1[#3]%
{%
  \if 0#1\xint_dothis\xint_stop_atfirstoftwo\fi
  \ifnum#3<\xint_c_\xint_dothis\xint_stop_atsecondoftwo\fi
  \xint_orthat\xint_stop_atfirstoftwo
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatIsInt}}
% \lverb|1.3d for isint() function in \xintfloatexpr.|
%    \begin{macrocode}
\def\xintFloatIsInt {\romannumeral0\xintfloatisint}%
\def\xintfloatisint #1{\expandafter\XINT_iffloatint
     \romannumeral0\xintrez{\XINTinFloatS[\XINTdigits]{#1}}10}%
%    \end{macrocode}
% \subsection{\csh{xintFloatIntType}}
% \lverb|1.4e for fractional powers. Expands to \xint_c_mone if argument is not an
% integer, to \xint_c_ if it is an even integer and to \xint_c_i if it is an
% odd integer.|
%    \begin{macrocode}
\def\xintFloatIntType {\romannumeral`&&@\xintfloatinttype}%
\def\xintfloatinttype #1%
{%
    \expandafter\XINT_floatinttype
    \romannumeral0\xintrez{\XINTinFloatS[\XINTdigits]{#1}}%
}%
\def\XINT_floatinttype #1#2/1[#3]%
{%
    \if 0#1\xint_dothis\xint_c_\fi
    \ifnum#3<\xint_c_\xint_dothis\xint_c_mone\fi
    \ifnum#3>\xint_c_\xint_dothis\xint_c_\fi
    \ifodd\xintLDg{#1#2} \xint_dothis\xint_c_i\fi
    \xint_orthat\xint_c_
}%
%    \end{macrocode}
% \subsection{\csh{XINTinFloatdigits}, \csh{XINTinFloatSdigits}}
% \lverb|For \xintNewExpr/\xintdeffloatfunc matters, mainly.
%
% |
%    \begin{macrocode}
\def\XINTinFloatdigits {\XINTinFloat [\XINTdigits]}%
\def\XINTinFloatSdigits{\XINTinFloatS[\XINTdigits]}%
%    \end{macrocode}
% \subsection{(WIP) \csh{XINTinRandomFloatS}, \csh{XINTinRandomFloatSdigits}}
% \lverb|1.3b. Support for random() function.
%
% Thus as it is a priori only for xintexpr usage, it expands inside \csname
% context, but as we need to get rid of initial zeros we use \xintRandomDigits
% not \xintXRandomDigits (\expanded would have a use case here).
%
% And anyway as we want to be able to use random() in
% \xintdeffunc/\xintNewExpr, it is good to have f-expandable macros, so we add
% the small overhead to make it f-expandable.
%
% We don't have to be very efficient in removing leading zeroes, as there is
% only 10$%
% chance for each successive one. Besides we use (current) internal storage
% format of the type A[N], where A is not required to be with \xintDigits
% digits, so N will simply be -\xintDigits and needs no adjustment.
%
% In case we use in future with #1 something else than \xintDigits we do
% the 0-(#1) construct.
%
% I had some qualms about doing a random float like this which means that
% when there are leading zeros in the random digits the (virtual) mantissa
% ends up with trailing zeros. That did not feel right but I checked random()
% in Python (which of course uses radix 2), and indeed this is what happens
% there.
%
% 
% |
%    \begin{macrocode}
\def\XINTinRandomFloatS{\romannumeral0\XINTinrandomfloatS}%
\def\XINTinRandomFloatSdigits{\XINTinRandomFloatS[\XINTdigits]}%
\def\XINTinrandomfloatS[#1]%
{%
    \expandafter\XINT_inrandomfloatS\the\numexpr\xint_c_-(#1)\xint:
}%
\def\XINT_inrandomfloatS-#1\xint:
{%
    \expandafter\XINT_inrandomfloatS_a
    \romannumeral0\xintrandomdigits{#1}[-#1]%
}%
%    \end{macrocode}
% \lverb|We add one macro to handle a tiny bit faster 90$%
% of cases, after all
% we also use one extra macro for the completely improbable all 0 case.|
%    \begin{macrocode}
\def\XINT_inrandomfloatS_a#1%
{%
    \if#10\xint_dothis{\XINT_inrandomfloatS_b}\fi
    \xint_orthat{ #1}%
}%[
\def\XINT_inrandomfloatS_b#1%
{%
    \if#1[\xint_dothis{\XINT_inrandomfloatS_zero}\fi% ]
    \if#10\xint_dothis{\XINT_inrandomfloatS_b}\fi
    \xint_orthat{ #1}%
}%[
\def\XINT_inrandomfloatS_zero#1]{ 0[0]}%
%    \end{macrocode}
% \subsection{(WIP) \csh{XINTinRandomFloatSixteen}}
% \lverb|1.3b. Support for qrand() function.|
%    \begin{macrocode}
\def\XINTinRandomFloatSixteen%
{%
    \romannumeral0\expandafter\XINT_inrandomfloatS_a
    \romannumeral`&&@\expandafter\XINT_eightrandomdigits
                \romannumeral`&&@\XINT_eightrandomdigits[-16]%
}%
\let\XINTinFloatMaxof\XINT_Maxof
\let\XINTinFloatMinof\XINT_Minof
\let\XINTinFloatSum\XINT_Sum
\let\XINTinFloatPrd\XINT_Prd
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintfrac}
% \cleardoublepage\let\xintfracnameUp\undefined
%\gardesactifs
%\let</xintfrac>\relax
%\let<*xintseries>\gardesinactifs
%</xintfrac>^^A---------------------------------------------------
%<*xintseries>^^A-------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintseriesnameUp\endcsname
% \section{Package \xintseriesnameimp implementation}
% \RaisedLabel{sec:seriesimp}
%
% \localtableofcontents
%
% The commenting is currently (\xintdocdate) very sparse.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xintseries.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintfrac.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintseries}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintseries.sty
      \ifx\w\relax % but xintfrac.sty not yet loaded.
         \def\z{\endgroup\input xintfrac.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintfrac.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xintfrac}}%
          \fi
      \else
        \aftergroup\endinput % xintseries already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintseries}%
  [2021/07/13 v1.4j Expandable partial sums with xint package (JFB)]%
%    \end{macrocode}
% \subsection{\csh{xintSeries}}
%    \begin{macrocode}
\def\xintSeries {\romannumeral0\xintseries }%
\def\xintseries #1#2%
{%
    \expandafter\XINT_series\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_series #1#2#3%
{%
   \ifnum #2<#1
      \xint_afterfi { 0/1[0]}%
   \else
      \xint_afterfi {\XINT_series_loop {#1}{0}{#2}{#3}}%
   \fi
}%
\def\XINT_series_loop #1#2#3#4%
{%
    \ifnum #3>#1 \else \XINT_series_exit \fi
    \expandafter\XINT_series_loop\expandafter
    {\the\numexpr #1+1\expandafter }\expandafter
    {\romannumeral0\xintadd {#2}{#4{#1}}}%
    {#3}{#4}%
}%
\def\XINT_series_exit \fi #1#2#3#4#5#6#7#8%
{%
    \fi\xint_gobble_ii #6%
}%
%    \end{macrocode}
% \subsection{\csh{xintiSeries}}
%    \begin{macrocode}
\def\xintiSeries {\romannumeral0\xintiseries }%
\def\xintiseries #1#2%
{%
    \expandafter\XINT_iseries\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_iseries #1#2#3%
{%
   \ifnum #2<#1
      \xint_afterfi { 0}%
   \else
      \xint_afterfi {\XINT_iseries_loop {#1}{0}{#2}{#3}}%
   \fi
}%
\def\XINT_iseries_loop #1#2#3#4%
{%
    \ifnum #3>#1 \else \XINT_iseries_exit \fi
    \expandafter\XINT_iseries_loop\expandafter
    {\the\numexpr #1+1\expandafter }\expandafter
    {\romannumeral0\xintiiadd {#2}{#4{#1}}}%
    {#3}{#4}%
}%
\def\XINT_iseries_exit \fi #1#2#3#4#5#6#7#8%
{%
    \fi\xint_gobble_ii #6%
}%
%    \end{macrocode}
% \subsection{\csh{xintPowerSeries}}
% \lverb|&
% The 1.03 version was very lame and created a build-up of denominators.
% (this was at a time \xintAdd always multiplied denominators, by the way)
% The Horner scheme for polynomial evaluation is used in 1.04, this
% cures the denominator problem and drastically improves the efficiency
% of the macro.
% Modified in 1.06 to give the indices first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
% 1.08a adds the forgotten optimization following that previous change.|
%    \begin{macrocode}
\def\xintPowerSeries {\romannumeral0\xintpowerseries }%
\def\xintpowerseries #1#2%
{%
    \expandafter\XINT_powseries\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_powseries #1#2#3#4%
{%
   \ifnum #2<#1
      \xint_afterfi { 0/1[0]}%
   \else
      \xint_afterfi
      {\XINT_powseries_loop_i {#3{#2}}{#1}{#2}{#3}{#4}}%
   \fi
}%
\def\XINT_powseries_loop_i #1#2#3#4#5%
{%
    \ifnum #3>#2 \else\XINT_powseries_exit_i\fi
    \expandafter\XINT_powseries_loop_ii\expandafter
    {\the\numexpr #3-1\expandafter}\expandafter
    {\romannumeral0\xintmul {#1}{#5}}{#2}{#4}{#5}%
}%
\def\XINT_powseries_loop_ii #1#2#3#4%
{%
   \expandafter\XINT_powseries_loop_i\expandafter
   {\romannumeral0\xintadd {#4{#1}}{#2}}{#3}{#1}{#4}%
}%
\def\XINT_powseries_exit_i\fi #1#2#3#4#5#6#7#8#9%
{%
    \fi \XINT_powseries_exit_ii  #6{#7}%
}%
\def\XINT_powseries_exit_ii #1#2#3#4#5#6%
{%
    \xintmul{\xintPow {#5}{#6}}{#4}%
}%
%    \end{macrocode}
% \subsection{\csh{xintPowerSeriesX}}
% \lverb|&
% Same as \xintPowerSeries except for the initial expansion of the x parameter.
% Modified in 1.06 to give the indices first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
% 1.08a adds the forgotten optimization following that previous change.|
%    \begin{macrocode}
\def\xintPowerSeriesX {\romannumeral0\xintpowerseriesx }%
\def\xintpowerseriesx #1#2%
{%
    \expandafter\XINT_powseriesx\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_powseriesx #1#2#3#4%
{%
   \ifnum #2<#1
      \xint_afterfi { 0/1[0]}%
   \else
      \xint_afterfi
      {\expandafter\XINT_powseriesx_pre\expandafter
                  {\romannumeral`&&@#4}{#1}{#2}{#3}%
      }%
   \fi
}%
\def\XINT_powseriesx_pre #1#2#3#4%
{%
    \XINT_powseries_loop_i {#4{#3}}{#2}{#3}{#4}{#1}%
}%
%    \end{macrocode}
% \subsection{\csh{xintRationalSeries}}
% \lverb|&
% This computes F(a)+...+F(b) on the basis of the value of F(a) and the
% ratios F(n)/F(n-1). As in \xintPowerSeries we use an iterative scheme which
% has the great advantage to avoid denominator build-up. This makes exact
% computations possible with exponential type series, which would be completely
% inaccessible to \xintSeries.
% #1=a, #2=b, #3=F(a), #4=ratio function
% Modified in 1.06 to give the indices first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
% 1.08a adds the forgotten optimization following that previous change.|
%    \begin{macrocode}
\def\xintRationalSeries {\romannumeral0\xintratseries }%
\def\xintratseries #1#2%
{%
    \expandafter\XINT_ratseries\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_ratseries #1#2#3#4%
{%
   \ifnum #2<#1
      \xint_afterfi { 0/1[0]}%
   \else
      \xint_afterfi
      {\XINT_ratseries_loop {#2}{1}{#1}{#4}{#3}}%
   \fi
}%
\def\XINT_ratseries_loop #1#2#3#4%
{%
    \ifnum #1>#3 \else\XINT_ratseries_exit_i\fi
    \expandafter\XINT_ratseries_loop\expandafter
    {\the\numexpr #1-1\expandafter}\expandafter
    {\romannumeral0\xintadd {1}{\xintMul {#2}{#4{#1}}}}{#3}{#4}%
}%
\def\XINT_ratseries_exit_i\fi #1#2#3#4#5#6#7#8%
{%
    \fi \XINT_ratseries_exit_ii  #6%
}%
\def\XINT_ratseries_exit_ii #1#2#3#4#5%
{%
    \XINT_ratseries_exit_iii #5%
}%
\def\XINT_ratseries_exit_iii #1#2#3#4%
{%
    \xintmul{#2}{#4}%
}%
%    \end{macrocode}
% \subsection{\csh{xintRationalSeriesX}}
% \lverb|&
% a,b,initial,ratiofunction,x$\
% This computes F(a,x)+...+F(b,x) on the basis of the value of F(a,x) and the
% ratios F(n,x)/F(n-1,x). The argument x is first expanded and it is the value
% resulting from this which is used then throughout. The initial term F(a,x)
% must be defined as one-parameter macro which will be given x.
% Modified in 1.06 to give the indices first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
% 1.08a adds the forgotten optimization following that previous change.|
%    \begin{macrocode}
\def\xintRationalSeriesX {\romannumeral0\xintratseriesx }%
\def\xintratseriesx #1#2%
{%
    \expandafter\XINT_ratseriesx\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_ratseriesx #1#2#3#4#5%
{%
   \ifnum #2<#1
      \xint_afterfi { 0/1[0]}%
   \else
      \xint_afterfi
      {\expandafter\XINT_ratseriesx_pre\expandafter
                   {\romannumeral`&&@#5}{#2}{#1}{#4}{#3}%
      }%
   \fi
}%
\def\XINT_ratseriesx_pre #1#2#3#4#5%
{%
    \XINT_ratseries_loop {#2}{1}{#3}{#4{#1}}{#5{#1}}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFxPtPowerSeries}}
% \lverb|&
% I am not two happy with this piece of code. Will make it more economical
% another day.
% Modified in 1.06 to give the indices first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
% 1.08a: forgot last time some optimization from the change to \numexpr.|
%    \begin{macrocode}
\def\xintFxPtPowerSeries {\romannumeral0\xintfxptpowerseries }%
\def\xintfxptpowerseries #1#2%
{%
    \expandafter\XINT_fppowseries\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_fppowseries #1#2#3#4#5%
{%
   \ifnum #2<#1
      \xint_afterfi { 0}%
   \else
      \xint_afterfi
        {\expandafter\XINT_fppowseries_loop_pre\expandafter
           {\romannumeral0\xinttrunc {#5}{\xintPow {#4}{#1}}}%
          {#1}{#4}{#2}{#3}{#5}%
        }%
   \fi
}%
\def\XINT_fppowseries_loop_pre #1#2#3#4#5#6%
{%
    \ifnum #4>#2 \else\XINT_fppowseries_dont_i \fi
    \expandafter\XINT_fppowseries_loop_i\expandafter
    {\the\numexpr #2+\xint_c_i\expandafter}\expandafter
    {\romannumeral0\xintitrunc {#6}{\xintMul {#5{#2}}{#1}}}%
    {#1}{#3}{#4}{#5}{#6}%
}%
\def\XINT_fppowseries_dont_i \fi\expandafter\XINT_fppowseries_loop_i
    {\fi \expandafter\XINT_fppowseries_dont_ii }%
\def\XINT_fppowseries_dont_ii #1#2#3#4#5#6#7{\xinttrunc {#7}{#2[-#7]}}%
\def\XINT_fppowseries_loop_i #1#2#3#4#5#6#7%
{%
    \ifnum #5>#1 \else \XINT_fppowseries_exit_i \fi
    \expandafter\XINT_fppowseries_loop_ii\expandafter
    {\romannumeral0\xinttrunc {#7}{\xintMul {#3}{#4}}}%
    {#1}{#4}{#2}{#5}{#6}{#7}%
}%
\def\XINT_fppowseries_loop_ii #1#2#3#4#5#6#7%
{%
    \expandafter\XINT_fppowseries_loop_i\expandafter
    {\the\numexpr #2+\xint_c_i\expandafter}\expandafter
    {\romannumeral0\xintiiadd {#4}{\xintiTrunc {#7}{\xintMul {#6{#2}}{#1}}}}%
    {#1}{#3}{#5}{#6}{#7}%
}%
\def\XINT_fppowseries_exit_i\fi\expandafter\XINT_fppowseries_loop_ii
    {\fi \expandafter\XINT_fppowseries_exit_ii }%
\def\XINT_fppowseries_exit_ii #1#2#3#4#5#6#7%
{%
    \xinttrunc {#7}
    {\xintiiadd {#4}{\xintiTrunc {#7}{\xintMul {#6{#2}}{#1}}}[-#7]}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFxPtPowerSeriesX}}
% \lverb|&
% a,b,coeff,x,D$\
% Modified in 1.06 to give the indices first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
% 1.08a adds the forgotten optimization following that previous change.|
%    \begin{macrocode}
\def\xintFxPtPowerSeriesX {\romannumeral0\xintfxptpowerseriesx }%
\def\xintfxptpowerseriesx #1#2%
{%
    \expandafter\XINT_fppowseriesx\expandafter
    {\the\numexpr #1\expandafter}\expandafter{\the\numexpr #2}%
}%
\def\XINT_fppowseriesx #1#2#3#4#5%
{%
   \ifnum #2<#1
      \xint_afterfi { 0}%
   \else
      \xint_afterfi
        {\expandafter \XINT_fppowseriesx_pre \expandafter
         {\romannumeral`&&@#4}{#1}{#2}{#3}{#5}%
        }%
   \fi
}%
\def\XINT_fppowseriesx_pre #1#2#3#4#5%
{%
    \expandafter\XINT_fppowseries_loop_pre\expandafter
       {\romannumeral0\xinttrunc {#5}{\xintPow {#1}{#2}}}%
       {#2}{#1}{#3}{#4}{#5}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatPowerSeries}}
% \lverb|1.08a. I still have to re-visit \xintFxPtPowerSeries; temporarily I
% just adapted the code to the case of floats.
%
% Usage of new names
% \XINTinfloatpow_wopt \XINTinfloatmul_wopt, \XINTinfloatadd_wopt to track
% xintfrac.sty changes at 1.4e.
% |
%    \begin{macrocode}
\def\xintFloatPowerSeries {\romannumeral0\xintfloatpowerseries }%
\def\xintfloatpowerseries #1{\XINT_flpowseries_chkopt #1\xint:}%
\def\XINT_flpowseries_chkopt #1%
{%
    \ifx [#1\expandafter\XINT_flpowseries_opt
       \else\expandafter\XINT_flpowseries_noopt
    \fi
    #1%
}%
\def\XINT_flpowseries_noopt  #1\xint:#2%
{%
    \expandafter\XINT_flpowseries\expandafter
    {\the\numexpr #1\expandafter}\expandafter
    {\the\numexpr #2}\XINTdigits
}%
\def\XINT_flpowseries_opt [\xint:#1]#2#3%
{%
    \expandafter\XINT_flpowseries\expandafter
    {\the\numexpr #2\expandafter}\expandafter
    {\the\numexpr #3\expandafter}{\the\numexpr #1}%
}%
\def\XINT_flpowseries #1#2#3#4#5%
{%
   \ifnum #2<#1
      \xint_afterfi { 0.e0}%
   \else
      \xint_afterfi
        {\expandafter\XINT_flpowseries_loop_pre\expandafter
           {\romannumeral0\XINTinfloatpow_wopt[#3]{#5}{#1}}%
          {#1}{#5}{#2}{#4}{#3}%
        }%
   \fi
}%
\def\XINT_flpowseries_loop_pre #1#2#3#4#5#6%
{%
    \ifnum #4>#2 \else\XINT_flpowseries_dont_i \fi
    \expandafter\XINT_flpowseries_loop_i\expandafter
    {\the\numexpr #2+\xint_c_i\expandafter}\expandafter
    {\romannumeral0\XINTinfloatmul_wopt[#6]{#5{#2}}{#1}}%
    {#1}{#3}{#4}{#5}{#6}%
}%
\def\XINT_flpowseries_dont_i \fi\expandafter\XINT_flpowseries_loop_i
    {\fi \expandafter\XINT_flpowseries_dont_ii }%
\def\XINT_flpowseries_dont_ii #1#2#3#4#5#6#7{\xintfloat [#7]{#2}}%
\def\XINT_flpowseries_loop_i #1#2#3#4#5#6#7%
{%
    \ifnum #5>#1 \else \XINT_flpowseries_exit_i \fi
    \expandafter\XINT_flpowseries_loop_ii\expandafter
    {\romannumeral0\XINTinfloatmul_wopt[#7]{#3}{#4}}%
    {#1}{#4}{#2}{#5}{#6}{#7}%
}%
\def\XINT_flpowseries_loop_ii #1#2#3#4#5#6#7%
{%
    \expandafter\XINT_flpowseries_loop_i\expandafter
    {\the\numexpr #2+\xint_c_i\expandafter}\expandafter
    {\romannumeral0\XINTinfloatadd_wopt[#7]{#4}%
                        {\XINTinfloatmul_wopt[#7]{#6{#2}}{#1}}}%
    {#1}{#3}{#5}{#6}{#7}%
}%
\def\XINT_flpowseries_exit_i\fi\expandafter\XINT_flpowseries_loop_ii
    {\fi \expandafter\XINT_flpowseries_exit_ii }%
\def\XINT_flpowseries_exit_ii #1#2#3#4#5#6#7%
{%
    \xintfloatadd[#7]{#4}{\XINTinfloatmul_wopt[#7]{#6{#2}}{#1}}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFloatPowerSeriesX}}
% \lverb|1.08a
%
% See \xintFloatPowerSeries for 1.4e comments.|
%    \begin{macrocode}
\def\xintFloatPowerSeriesX {\romannumeral0\xintfloatpowerseriesx }%
\def\xintfloatpowerseriesx #1{\XINT_flpowseriesx_chkopt #1\xint:}%
\def\XINT_flpowseriesx_chkopt #1%
{%
    \ifx [#1\expandafter\XINT_flpowseriesx_opt
       \else\expandafter\XINT_flpowseriesx_noopt
    \fi
    #1%
}%
\def\XINT_flpowseriesx_noopt  #1\xint:#2%
{%
    \expandafter\XINT_flpowseriesx\expandafter
    {\the\numexpr #1\expandafter}\expandafter
    {\the\numexpr #2}\XINTdigits
}%
\def\XINT_flpowseriesx_opt [\xint:#1]#2#3%
{%
    \expandafter\XINT_flpowseriesx\expandafter
    {\the\numexpr #2\expandafter}\expandafter
    {\the\numexpr #3\expandafter}{\the\numexpr #1}%
}%
\def\XINT_flpowseriesx #1#2#3#4#5%
{%
   \ifnum #2<#1
      \xint_afterfi { 0.e0}%
   \else
      \xint_afterfi
        {\expandafter \XINT_flpowseriesx_pre \expandafter
         {\romannumeral`&&@#5}{#1}{#2}{#4}{#3}%
        }%
   \fi
}%
\def\XINT_flpowseriesx_pre #1#2#3#4#5%
{%
    \expandafter\XINT_flpowseries_loop_pre\expandafter
       {\romannumeral0\XINTinfloatpow_wopt[#5]{#1}{#2}}%
       {#2}{#1}{#3}{#4}{#5}%
}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintseries}
% \cleardoublepage\let\xintseriesnameUp\undefined
%\gardesactifs
%\let</xintseries>\relax
%\let<*xintcfrac>\gardesinactifs
%</xintseries>^^A-------------------------------------------------
%<*xintcfrac>^^A--------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintcfracnameUp\endcsname
% \section{Package \xintcfracnameimp implementation}
% \RaisedLabel{sec:cfracimp}
%
% \localtableofcontents
%
% The commenting is currently (\xintdocdate) very sparse. Release |1.09m|
% (|2014/02/26|) has modified a few things: |\xintFtoCs| and
% |\xintCntoCs| insert spaces after the commas, |\xintCstoF| and
% |\xintCstoCv| authorize spaces in the input also before the commas,
% |\xintCntoCs| does not brace the produced coefficients, new macros
% |\xintFtoC|, |\xintCtoF|, |\xintCtoCv|, |\xintFGtoC|, and
% |\xintGGCFrac|.
%
% There is partial dependency on \xinttoolsnameimp due to |\xintCstoF| and
% |\xintCsToCv|.
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \let\z\endgroup
  \expandafter\let\expandafter\x\csname ver@xintcfrac.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintfrac.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintcfrac}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintcfrac.sty
      \ifx\w\relax % but xintfrac.sty not yet loaded.
         \def\z{\endgroup\input xintfrac.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintfrac.sty not yet loaded.
            \def\z{\endgroup\RequirePackage{xintfrac}}%
          \fi
      \else
        \aftergroup\endinput % xintcfrac already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes% defined in xintkernel.sty
%    \end{macrocode}
% \subsection{Package identification}
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintcfrac}%
  [2021/07/13 v1.4j Expandable continued fractions with xint package (JFB)]%
%    \end{macrocode}
% \subsection{\csh{xintCFrac}}
%    \begin{macrocode}
\def\xintCFrac {\romannumeral0\xintcfrac }%
\def\xintcfrac #1%
{%
    \XINT_cfrac_opt_a #1\xint:
}%
\def\XINT_cfrac_opt_a #1%
{%
    \ifx[#1\XINT_cfrac_opt_b\fi \XINT_cfrac_noopt #1%
}%
\def\XINT_cfrac_noopt #1\xint:
{%
    \expandafter\XINT_cfrac_A\romannumeral0\xintrawwithzeros {#1}\Z
    \relax\relax
}%
\def\XINT_cfrac_opt_b\fi\XINT_cfrac_noopt [\xint:#1]%
{%
    \fi\csname XINT_cfrac_opt#1\endcsname
}%
\def\XINT_cfrac_optl #1%
{%
    \expandafter\XINT_cfrac_A\romannumeral0\xintrawwithzeros {#1}\Z
    \relax\hfill
}%
\def\XINT_cfrac_optc #1%
{%
    \expandafter\XINT_cfrac_A\romannumeral0\xintrawwithzeros {#1}\Z
    \relax\relax
}%
\def\XINT_cfrac_optr #1%
{%
    \expandafter\XINT_cfrac_A\romannumeral0\xintrawwithzeros {#1}\Z
    \hfill\relax
}%
\def\XINT_cfrac_A #1/#2\Z
{%
    \expandafter\XINT_cfrac_B\romannumeral0\xintiidivision {#1}{#2}{#2}%
}%
\def\XINT_cfrac_B #1#2%
{%
    \XINT_cfrac_C #2\Z {#1}%
}%
\def\XINT_cfrac_C #1%
{%
    \xint_gob_til_zero #1\XINT_cfrac_integer 0\XINT_cfrac_D #1%
}%
\def\XINT_cfrac_integer 0\XINT_cfrac_D 0#1\Z #2#3#4#5{ #2}%
\def\XINT_cfrac_D #1\Z #2#3{\XINT_cfrac_loop_a {#1}{#3}{#1}{{#2}}}%
\def\XINT_cfrac_loop_a
{%
    \expandafter\XINT_cfrac_loop_d\romannumeral0\XINT_div_prepare
}%
\def\XINT_cfrac_loop_d #1#2%
{%
    \XINT_cfrac_loop_e #2.{#1}%
}%
\def\XINT_cfrac_loop_e #1%
{%
    \xint_gob_til_zero #1\xint_cfrac_loop_exit0\XINT_cfrac_loop_f #1%
}%
\def\XINT_cfrac_loop_f #1.#2#3#4%
{%
    \XINT_cfrac_loop_a {#1}{#3}{#1}{{#2}#4}%
}%
\def\xint_cfrac_loop_exit0\XINT_cfrac_loop_f #1.#2#3#4#5#6%
   {\XINT_cfrac_T #5#6{#2}#4\Z }%
\def\XINT_cfrac_T #1#2#3#4%
{%
  \xint_gob_til_Z #4\XINT_cfrac_end\Z\XINT_cfrac_T #1#2{#4+\cfrac{#11#2}{#3}}%
}%
\def\XINT_cfrac_end\Z\XINT_cfrac_T #1#2#3%
{%
    \XINT_cfrac_end_b #3%
}%
\def\XINT_cfrac_end_b \Z+\cfrac#1#2{ #2}%
%    \end{macrocode}
% \subsection{\csh{xintGCFrac}}
% Updated at |1.4g| to follow-up on renaming of |\xintFrac| into |\xintTeXFrac|.
%    \begin{macrocode}
\def\xintGCFrac {\romannumeral0\xintgcfrac }%
\def\xintgcfrac #1{\XINT_gcfrac_opt_a #1\xint:}%
\def\XINT_gcfrac_opt_a #1%
{%
    \ifx[#1\XINT_gcfrac_opt_b\fi \XINT_gcfrac_noopt #1%
}%
\def\XINT_gcfrac_noopt #1\xint:%
{%
    \XINT_gcfrac #1+!/\relax\relax
}%
\def\XINT_gcfrac_opt_b\fi\XINT_gcfrac_noopt [\xint:#1]%
{%
    \fi\csname XINT_gcfrac_opt#1\endcsname
}%
\def\XINT_gcfrac_optl #1%
{%
    \XINT_gcfrac #1+!/\relax\hfill
}%
\def\XINT_gcfrac_optc #1%
{%
    \XINT_gcfrac #1+!/\relax\relax
}%
\def\XINT_gcfrac_optr #1%
{%
    \XINT_gcfrac #1+!/\hfill\relax
}%
\def\XINT_gcfrac
{%
    \expandafter\XINT_gcfrac_enter\romannumeral`&&@%
}%
\def\XINT_gcfrac_enter {\XINT_gcfrac_loop {}}%
\def\XINT_gcfrac_loop #1#2+#3/%
{%
    \xint_gob_til_exclam #3\XINT_gcfrac_endloop!%
    \XINT_gcfrac_loop {{#3}{#2}#1}%
}%
\def\XINT_gcfrac_endloop!\XINT_gcfrac_loop #1#2#3%
{%
    \XINT_gcfrac_T #2#3#1!!%
}%
\def\XINT_gcfrac_T #1#2#3#4{\XINT_gcfrac_U #1#2{\xintTeXFrac{#4}}}%
\def\XINT_gcfrac_U #1#2#3#4#5%
{%
    \xint_gob_til_exclam #5\XINT_gcfrac_end!\XINT_gcfrac_U
              #1#2{\xintTeXFrac{#5}%
               \ifcase\xintSgn{#4}
               +\or+\else-\fi
               \cfrac{#1\xintTeXFrac{\xintAbs{#4}}#2}{#3}}%
}%
\def\XINT_gcfrac_end!\XINT_gcfrac_U #1#2#3%
{%
    \XINT_gcfrac_end_b #3%
}%
\def\XINT_gcfrac_end_b #1\cfrac#2#3{ #3}%
%    \end{macrocode}
% \subsection{\csh{xintGGCFrac}}
% \lverb|New with 1.09m|
%    \begin{macrocode}
\def\xintGGCFrac {\romannumeral0\xintggcfrac }%
\def\xintggcfrac #1{\XINT_ggcfrac_opt_a #1\xint:}%
\def\XINT_ggcfrac_opt_a #1%
{%
    \ifx[#1\XINT_ggcfrac_opt_b\fi \XINT_ggcfrac_noopt #1%
}%
\def\XINT_ggcfrac_noopt #1\xint:
{%
    \XINT_ggcfrac #1+!/\relax\relax
}%
\def\XINT_ggcfrac_opt_b\fi\XINT_ggcfrac_noopt [\xint:#1]%
{%
    \fi\csname XINT_ggcfrac_opt#1\endcsname
}%
\def\XINT_ggcfrac_optl #1%
{%
    \XINT_ggcfrac #1+!/\relax\hfill
}%
\def\XINT_ggcfrac_optc #1%
{%
    \XINT_ggcfrac #1+!/\relax\relax
}%
\def\XINT_ggcfrac_optr #1%
{%
    \XINT_ggcfrac #1+!/\hfill\relax
}%
\def\XINT_ggcfrac
{%
    \expandafter\XINT_ggcfrac_enter\romannumeral`&&@%
}%
\def\XINT_ggcfrac_enter {\XINT_ggcfrac_loop {}}%
\def\XINT_ggcfrac_loop #1#2+#3/%
{%
    \xint_gob_til_exclam #3\XINT_ggcfrac_endloop!%
    \XINT_ggcfrac_loop {{#3}{#2}#1}%
}%
\def\XINT_ggcfrac_endloop!\XINT_ggcfrac_loop #1#2#3%
{%
    \XINT_ggcfrac_T #2#3#1!!%
}%
\def\XINT_ggcfrac_T #1#2#3#4{\XINT_ggcfrac_U #1#2{#4}}%
\def\XINT_ggcfrac_U #1#2#3#4#5%
{%
    \xint_gob_til_exclam #5\XINT_ggcfrac_end!\XINT_ggcfrac_U
              #1#2{#5+\cfrac{#1#4#2}{#3}}%
}%
\def\XINT_ggcfrac_end!\XINT_ggcfrac_U #1#2#3%
{%
    \XINT_ggcfrac_end_b #3%
}%
\def\XINT_ggcfrac_end_b #1\cfrac#2#3{ #3}%
%    \end{macrocode}
% \subsection{\csh{xintGCtoGCx}}
%    \begin{macrocode}
\def\xintGCtoGCx {\romannumeral0\xintgctogcx }%
\def\xintgctogcx #1#2#3%
{%
    \expandafter\XINT_gctgcx_start\expandafter {\romannumeral`&&@#3}{#1}{#2}%
}%
\def\XINT_gctgcx_start #1#2#3{\XINT_gctgcx_loop_a {}{#2}{#3}#1+!/}%
\def\XINT_gctgcx_loop_a #1#2#3#4+#5/%
{%
    \xint_gob_til_exclam #5\XINT_gctgcx_end!%
    \XINT_gctgcx_loop_b {#1{#4}}{#2{#5}#3}{#2}{#3}%
}%
\def\XINT_gctgcx_loop_b #1#2%
{%
    \XINT_gctgcx_loop_a {#1#2}%
}%
\def\XINT_gctgcx_end!\XINT_gctgcx_loop_b #1#2#3#4{ #1}%
%    \end{macrocode}
% \subsection{\csh{xintFtoCs}}
% \lverb|Modified in 1.09m: a space is added after the inserted commas.|
%    \begin{macrocode}
\def\xintFtoCs {\romannumeral0\xintftocs }%
\def\xintftocs #1%
{%
    \expandafter\XINT_ftc_A\romannumeral0\xintrawwithzeros {#1}\Z
}%
\def\XINT_ftc_A #1/#2\Z
{%
    \expandafter\XINT_ftc_B\romannumeral0\xintiidivision {#1}{#2}{#2}%
}%
\def\XINT_ftc_B #1#2%
{%
    \XINT_ftc_C #2.{#1}%
}%
\def\XINT_ftc_C #1%
{%
    \xint_gob_til_zero #1\XINT_ftc_integer 0\XINT_ftc_D #1%
}%
\def\XINT_ftc_integer 0\XINT_ftc_D 0#1.#2#3{ #2}%
\def\XINT_ftc_D #1.#2#3{\XINT_ftc_loop_a {#1}{#3}{#1}{#2, }}% 1.09m adds a space
\def\XINT_ftc_loop_a
{%
    \expandafter\XINT_ftc_loop_d\romannumeral0\XINT_div_prepare
}%
\def\XINT_ftc_loop_d #1#2%
{%
    \XINT_ftc_loop_e #2.{#1}%
}%
\def\XINT_ftc_loop_e #1%
{%
    \xint_gob_til_zero #1\xint_ftc_loop_exit0\XINT_ftc_loop_f #1%
}%
\def\XINT_ftc_loop_f #1.#2#3#4%
{%
    \XINT_ftc_loop_a {#1}{#3}{#1}{#4#2, }% 1.09m has an added space here
}%
\def\xint_ftc_loop_exit0\XINT_ftc_loop_f #1.#2#3#4{ #4#2}%
%    \end{macrocode}
% \subsection{\csh{xintFtoCx}}
%    \begin{macrocode}
\def\xintFtoCx {\romannumeral0\xintftocx }%
\def\xintftocx #1#2%
{%
    \expandafter\XINT_ftcx_A\romannumeral0\xintrawwithzeros {#2}\Z {#1}%
}%
\def\XINT_ftcx_A #1/#2\Z
{%
    \expandafter\XINT_ftcx_B\romannumeral0\xintiidivision {#1}{#2}{#2}%
}%
\def\XINT_ftcx_B #1#2%
{%
    \XINT_ftcx_C #2.{#1}%
}%
\def\XINT_ftcx_C #1%
{%
    \xint_gob_til_zero #1\XINT_ftcx_integer 0\XINT_ftcx_D #1%
}%
\def\XINT_ftcx_integer 0\XINT_ftcx_D 0#1.#2#3#4{ #2}%
\def\XINT_ftcx_D #1.#2#3#4{\XINT_ftcx_loop_a {#1}{#3}{#1}{{#2}#4}{#4}}%
\def\XINT_ftcx_loop_a
{%
    \expandafter\XINT_ftcx_loop_d\romannumeral0\XINT_div_prepare
}%
\def\XINT_ftcx_loop_d #1#2%
{%
    \XINT_ftcx_loop_e #2.{#1}%
}%
\def\XINT_ftcx_loop_e #1%
{%
    \xint_gob_til_zero #1\xint_ftcx_loop_exit0\XINT_ftcx_loop_f #1%
}%
\def\XINT_ftcx_loop_f #1.#2#3#4#5%
{%
    \XINT_ftcx_loop_a {#1}{#3}{#1}{#4{#2}#5}{#5}%
}%
\def\xint_ftcx_loop_exit0\XINT_ftcx_loop_f #1.#2#3#4#5{ #4{#2}}%
%    \end{macrocode}
% \subsection{\csh{xintFtoC}}
% \lverb|New in 1.09m: this is the same as \xintFtoCx with empty separator. I
% had temporarily during preparation of 1.09m removed braces from \xintFtoCx,
% but I recalled later why that was useful (see doc), thus let's just here do
% \xintFtoCx {}|
%    \begin{macrocode}
\def\xintFtoC {\romannumeral0\xintftoc }%
\def\xintftoc {\xintftocx {}}%
%    \end{macrocode}
% \subsection{\csh{xintFtoGC}}
%    \begin{macrocode}
\def\xintFtoGC {\romannumeral0\xintftogc }%
\def\xintftogc {\xintftocx {+1/}}%
%    \end{macrocode}
% \subsection{\csh{xintFGtoC}}
% \lverb|New with 1.09m of 2014/02/26. Computes the common initial coefficients
% for the two fractions f and g, and outputs them as a sequence of braced
% items.|
%    \begin{macrocode}
\def\xintFGtoC {\romannumeral0\xintfgtoc}%
\def\xintfgtoc#1%
{%
    \expandafter\XINT_fgtc_a\romannumeral0\xintrawwithzeros {#1}\Z
}%
\def\XINT_fgtc_a #1/#2\Z #3%
{%
    \expandafter\XINT_fgtc_b\romannumeral0\xintrawwithzeros {#3}\Z #1/#2\Z { }%
}%
\def\XINT_fgtc_b #1/#2\Z
{%
    \expandafter\XINT_fgtc_c\romannumeral0\xintiidivision {#1}{#2}{#2}%
}%
\def\XINT_fgtc_c #1#2#3#4/#5\Z
{%
    \expandafter\XINT_fgtc_d\romannumeral0\xintiidivision
                                         {#4}{#5}{#5}{#1}{#2}{#3}%
}%
\def\XINT_fgtc_d #1#2#3#4%#5#6#7%
{%
    \xintifEq {#1}{#4}{\XINT_fgtc_da {#1}{#2}{#3}{#4}}%
                      {\xint_thirdofthree}%
}%
\def\XINT_fgtc_da #1#2#3#4#5#6#7%
{%
     \XINT_fgtc_e {#2}{#5}{#3}{#6}{#7{#1}}%
}%
\def\XINT_fgtc_e #1%
{%
    \xintiiifZero {#1}{\expandafter\xint_firstofone\xint_gobble_iii}%
                    {\XINT_fgtc_f {#1}}%
}%
\def\XINT_fgtc_f #1#2%
{%
   \xintiiifZero {#2}{\xint_thirdofthree}{\XINT_fgtc_g {#1}{#2}}%
}%
\def\XINT_fgtc_g #1#2#3%
{%
    \expandafter\XINT_fgtc_h\romannumeral0\XINT_div_prepare {#1}{#3}{#1}{#2}%
}%
\def\XINT_fgtc_h #1#2#3#4#5%
{%
    \expandafter\XINT_fgtc_d\romannumeral0\XINT_div_prepare
                       {#4}{#5}{#4}{#1}{#2}{#3}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFtoCC}}
%    \begin{macrocode}
\def\xintFtoCC {\romannumeral0\xintftocc }%
\def\xintftocc #1%
{%
    \expandafter\XINT_ftcc_A\expandafter {\romannumeral0\xintrawwithzeros {#1}}%
}%
\def\XINT_ftcc_A #1%
{%
    \expandafter\XINT_ftcc_B
    \romannumeral0\xintrawwithzeros {\xintAdd {1/2[0]}{#1[0]}}\Z {#1[0]}%
}%
\def\XINT_ftcc_B #1/#2\Z
{%
    \expandafter\XINT_ftcc_C\expandafter {\romannumeral0\xintiiquo {#1}{#2}}%
}%
\def\XINT_ftcc_C #1#2%
{%
    \expandafter\XINT_ftcc_D\romannumeral0\xintsub {#2}{#1}\Z {#1}%
}%
\def\XINT_ftcc_D #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_ftcc_integer
      0#1\XINT_ftcc_En
       0-{\XINT_ftcc_Ep #1}%
    \krof
}%
\def\XINT_ftcc_Ep #1\Z #2%
{%
    \expandafter\XINT_ftcc_loop_a\expandafter
    {\romannumeral0\xintdiv {1[0]}{#1}}{#2+1/}%
}%
\def\XINT_ftcc_En #1\Z #2%
{%
    \expandafter\XINT_ftcc_loop_a\expandafter
    {\romannumeral0\xintdiv {1[0]}{#1}}{#2+-1/}%
}%
\def\XINT_ftcc_integer #1\Z #2{ #2}%
\def\XINT_ftcc_loop_a #1%
{%
    \expandafter\XINT_ftcc_loop_b
    \romannumeral0\xintrawwithzeros {\xintAdd {1/2[0]}{#1}}\Z {#1}%
}%
\def\XINT_ftcc_loop_b #1/#2\Z
{%
    \expandafter\XINT_ftcc_loop_c\expandafter
    {\romannumeral0\xintiiquo {#1}{#2}}%
}%
\def\XINT_ftcc_loop_c #1#2%
{%
    \expandafter\XINT_ftcc_loop_d
    \romannumeral0\xintsub {#2}{#1[0]}\Z {#1}%
}%
\def\XINT_ftcc_loop_d #1%
{%
    \xint_UDzerominusfork
      #1-\XINT_ftcc_end
      0#1\XINT_ftcc_loop_N
       0-{\XINT_ftcc_loop_P #1}%
    \krof
}%
\def\XINT_ftcc_end #1\Z #2#3{ #3#2}%
\def\XINT_ftcc_loop_P #1\Z #2#3%
{%
    \expandafter\XINT_ftcc_loop_a\expandafter
    {\romannumeral0\xintdiv {1[0]}{#1}}{#3#2+1/}%
}%
\def\XINT_ftcc_loop_N #1\Z #2#3%
{%
    \expandafter\XINT_ftcc_loop_a\expandafter
    {\romannumeral0\xintdiv {1[0]}{#1}}{#3#2+-1/}%
}%
%    \end{macrocode}
% \subsection{\csh{xintCtoF}, \csh{xintCstoF}}
% \lverb|1.09m uses \xintCSVtoList on the argument of \xintCstoF to allow
% spaces also before the commas. And the original \xintCstoF code became the
% one of the new \xintCtoF dealing with a braced rather than comma separated
% list.|
%    \begin{macrocode}
\def\xintCstoF {\romannumeral0\xintcstof }%
\def\xintcstof #1%
{%
    \expandafter\XINT_ctf_prep \romannumeral0\xintcsvtolist{#1}!%
}%
\def\xintCtoF {\romannumeral0\xintctof }%
\def\xintctof #1%
{%
    \expandafter\XINT_ctf_prep \romannumeral`&&@#1!%
}%
\def\XINT_ctf_prep
{%
    \XINT_ctf_loop_a 1001%
}%
\def\XINT_ctf_loop_a #1#2#3#4#5%
{%
    \xint_gob_til_exclam #5\XINT_ctf_end!%
    \expandafter\XINT_ctf_loop_b
    \romannumeral0\xintrawwithzeros {#5}.{#1}{#2}{#3}{#4}%
}%
\def\XINT_ctf_loop_b #1/#2.#3#4#5#6%
{%
    \expandafter\XINT_ctf_loop_c\expandafter
    {\romannumeral0\XINT_mul_fork #2\xint:#4\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#3\xint:}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#6\xint:}%
                              {\XINT_mul_fork #1\xint:#4\xint:}}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#5\xint:}%
                              {\XINT_mul_fork #1\xint:#3\xint:}}%
}%
\def\XINT_ctf_loop_c #1#2%
{%
    \expandafter\XINT_ctf_loop_d\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_ctf_loop_d #1#2%
{%
    \expandafter\XINT_ctf_loop_e\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_ctf_loop_e #1#2%
{%
    \expandafter\XINT_ctf_loop_a\expandafter{#2}#1%
}%
\def\XINT_ctf_end #1.#2#3#4#5{\xintrawwithzeros {#2/#3}}% 1.09b removes [0]
%    \end{macrocode}
% \subsection{\csh{xintiCstoF}}
%    \begin{macrocode}
\def\xintiCstoF {\romannumeral0\xinticstof }%
\def\xinticstof #1%
{%
    \expandafter\XINT_icstf_prep \romannumeral`&&@#1,!,%
}%
\def\XINT_icstf_prep
{%
    \XINT_icstf_loop_a 1001%
}%
\def\XINT_icstf_loop_a #1#2#3#4#5,%
{%
    \xint_gob_til_exclam #5\XINT_icstf_end!%
    \expandafter
    \XINT_icstf_loop_b \romannumeral`&&@#5.{#1}{#2}{#3}{#4}%
}%
\def\XINT_icstf_loop_b #1.#2#3#4#5%
{%
    \expandafter\XINT_icstf_loop_c\expandafter
    {\romannumeral0\xintiiadd {#5}{\XINT_mul_fork #1\xint:#3\xint:}}%
    {\romannumeral0\xintiiadd {#4}{\XINT_mul_fork #1\xint:#2\xint:}}%
    {#2}{#3}%
}%
\def\XINT_icstf_loop_c #1#2%
{%
    \expandafter\XINT_icstf_loop_a\expandafter {#2}{#1}%
}%
\def\XINT_icstf_end#1.#2#3#4#5{\xintrawwithzeros {#2/#3}}% 1.09b removes [0]
%    \end{macrocode}
% \subsection{\csh{xintGCtoF}}
%    \begin{macrocode}
\def\xintGCtoF {\romannumeral0\xintgctof }%
\def\xintgctof #1%
{%
    \expandafter\XINT_gctf_prep \romannumeral`&&@#1+!/%
}%
\def\XINT_gctf_prep
{%
    \XINT_gctf_loop_a 1001%
}%
\def\XINT_gctf_loop_a #1#2#3#4#5+%
{%
    \expandafter\XINT_gctf_loop_b
    \romannumeral0\xintrawwithzeros {#5}.{#1}{#2}{#3}{#4}%
}%
\def\XINT_gctf_loop_b #1/#2.#3#4#5#6%
{%
    \expandafter\XINT_gctf_loop_c\expandafter
    {\romannumeral0\XINT_mul_fork #2\xint:#4\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#3\xint:}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#6\xint:}%
                              {\XINT_mul_fork #1\xint:#4\xint:}}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#5\xint:}%
                              {\XINT_mul_fork #1\xint:#3\xint:}}%
}%
\def\XINT_gctf_loop_c #1#2%
{%
    \expandafter\XINT_gctf_loop_d\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_gctf_loop_d #1#2%
{%
    \expandafter\XINT_gctf_loop_e\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_gctf_loop_e #1#2%
{%
    \expandafter\XINT_gctf_loop_f\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_gctf_loop_f #1#2/%
{%
    \xint_gob_til_exclam #2\XINT_gctf_end!%
    \expandafter\XINT_gctf_loop_g
    \romannumeral0\xintrawwithzeros {#2}.#1%
}%
\def\XINT_gctf_loop_g #1/#2.#3#4#5#6%
{%
    \expandafter\XINT_gctf_loop_h\expandafter
    {\romannumeral0\XINT_mul_fork #1\xint:#6\xint:}%
    {\romannumeral0\XINT_mul_fork #1\xint:#5\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#4\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#3\xint:}%
}%
\def\XINT_gctf_loop_h #1#2%
{%
    \expandafter\XINT_gctf_loop_i\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_gctf_loop_i #1#2%
{%
    \expandafter\XINT_gctf_loop_j\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_gctf_loop_j #1#2%
{%
    \expandafter\XINT_gctf_loop_a\expandafter {#2}#1%
}%
\def\XINT_gctf_end #1.#2#3#4#5{\xintrawwithzeros {#2/#3}}% 1.09b removes [0]
%    \end{macrocode}
% \subsection{\csh{xintiGCtoF}}
%    \begin{macrocode}
\def\xintiGCtoF {\romannumeral0\xintigctof }%
\def\xintigctof #1%
{%
    \expandafter\XINT_igctf_prep \romannumeral`&&@#1+!/%
}%
\def\XINT_igctf_prep
{%
    \XINT_igctf_loop_a 1001%
}%
\def\XINT_igctf_loop_a #1#2#3#4#5+%
{%
    \expandafter\XINT_igctf_loop_b
    \romannumeral`&&@#5.{#1}{#2}{#3}{#4}%
}%
\def\XINT_igctf_loop_b #1.#2#3#4#5%
{%
    \expandafter\XINT_igctf_loop_c\expandafter
    {\romannumeral0\xintiiadd {#5}{\XINT_mul_fork #1\xint:#3\xint:}}%
    {\romannumeral0\xintiiadd {#4}{\XINT_mul_fork #1\xint:#2\xint:}}%
    {#2}{#3}%
}%
\def\XINT_igctf_loop_c #1#2%
{%
    \expandafter\XINT_igctf_loop_f\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_igctf_loop_f #1#2#3#4/%
{%
    \xint_gob_til_exclam #4\XINT_igctf_end!%
    \expandafter\XINT_igctf_loop_g
    \romannumeral`&&@#4.{#2}{#3}#1%
}%
\def\XINT_igctf_loop_g #1.#2#3%
{%
    \expandafter\XINT_igctf_loop_h\expandafter
    {\romannumeral0\XINT_mul_fork #1\xint:#3\xint:}%
    {\romannumeral0\XINT_mul_fork #1\xint:#2\xint:}%
}%
\def\XINT_igctf_loop_h #1#2%
{%
    \expandafter\XINT_igctf_loop_i\expandafter {#2}{#1}%
}%
\def\XINT_igctf_loop_i #1#2#3#4%
{%
    \XINT_igctf_loop_a {#3}{#4}{#1}{#2}%
}%
\def\XINT_igctf_end #1.#2#3#4#5{\xintrawwithzeros {#4/#5}}% 1.09b removes [0]
%    \end{macrocode}
% \subsection{\csh{xintCtoCv}, \csh{xintCstoCv}}
% \lverb|1.09m uses \xintCSVtoList on the argument of \xintCstoCv to allow
% spaces also before the commas. The original \xintCstoCv code became the
% one of the new \xintCtoF dealing with a braced rather than comma separated
% list.|
%    \begin{macrocode}
\def\xintCstoCv {\romannumeral0\xintcstocv }%
\def\xintcstocv #1%
{%
    \expandafter\XINT_ctcv_prep\romannumeral0\xintcsvtolist{#1}!%
}%
\def\xintCtoCv {\romannumeral0\xintctocv }%
\def\xintctocv #1%
{%
    \expandafter\XINT_ctcv_prep\romannumeral`&&@#1!%
}%
\def\XINT_ctcv_prep
{%
    \XINT_ctcv_loop_a {}1001%
}%
\def\XINT_ctcv_loop_a #1#2#3#4#5#6%
{%
    \xint_gob_til_exclam #6\XINT_ctcv_end!%
    \expandafter\XINT_ctcv_loop_b
    \romannumeral0\xintrawwithzeros {#6}.{#2}{#3}{#4}{#5}{#1}%
}%
\def\XINT_ctcv_loop_b #1/#2.#3#4#5#6%
{%
    \expandafter\XINT_ctcv_loop_c\expandafter
    {\romannumeral0\XINT_mul_fork #2\xint:#4\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#3\xint:}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#6\xint:}%
                              {\XINT_mul_fork #1\xint:#4\xint:}}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#5\xint:}%
                              {\XINT_mul_fork #1\xint:#3\xint:}}%
}%
\def\XINT_ctcv_loop_c #1#2%
{%
    \expandafter\XINT_ctcv_loop_d\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_ctcv_loop_d #1#2%
{%
    \expandafter\XINT_ctcv_loop_e\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_ctcv_loop_e #1#2%
{%
    \expandafter\XINT_ctcv_loop_f\expandafter{#2}#1%
}%
\def\XINT_ctcv_loop_f #1#2#3#4#5%
{%
    \expandafter\XINT_ctcv_loop_g\expandafter
    {\romannumeral0\xintrawwithzeros {#1/#2}}{#5}{#1}{#2}{#3}{#4}%
}%
\def\XINT_ctcv_loop_g #1#2{\XINT_ctcv_loop_a {#2{#1}}}% 1.09b removes [0]
\def\XINT_ctcv_end #1.#2#3#4#5#6{ #6}%
%    \end{macrocode}
% \subsection{\csh{xintiCstoCv}}
%    \begin{macrocode}
\def\xintiCstoCv {\romannumeral0\xinticstocv }%
\def\xinticstocv #1%
{%
    \expandafter\XINT_icstcv_prep \romannumeral`&&@#1,!,%
}%
\def\XINT_icstcv_prep
{%
    \XINT_icstcv_loop_a {}1001%
}%
\def\XINT_icstcv_loop_a #1#2#3#4#5#6,%
{%
    \xint_gob_til_exclam #6\XINT_icstcv_end!%
    \expandafter
    \XINT_icstcv_loop_b \romannumeral`&&@#6.{#2}{#3}{#4}{#5}{#1}%
}%
\def\XINT_icstcv_loop_b #1.#2#3#4#5%
{%
    \expandafter\XINT_icstcv_loop_c\expandafter
    {\romannumeral0\xintiiadd {#5}{\XINT_mul_fork #1\xint:#3\xint:}}%
    {\romannumeral0\xintiiadd {#4}{\XINT_mul_fork #1\xint:#2\xint:}}%
    {{#2}{#3}}%
}%
\def\XINT_icstcv_loop_c #1#2%
{%
    \expandafter\XINT_icstcv_loop_d\expandafter {#2}{#1}%
}%
\def\XINT_icstcv_loop_d #1#2%
{%
    \expandafter\XINT_icstcv_loop_e\expandafter
    {\romannumeral0\xintrawwithzeros {#1/#2}}{{#1}{#2}}%
}%
\def\XINT_icstcv_loop_e #1#2#3#4{\XINT_icstcv_loop_a {#4{#1}}#2#3}%
\def\XINT_icstcv_end #1.#2#3#4#5#6{ #6}%  1.09b removes [0]
%    \end{macrocode}
% \subsection{\csh{xintGCtoCv}}
%    \begin{macrocode}
\def\xintGCtoCv {\romannumeral0\xintgctocv }%
\def\xintgctocv #1%
{%
    \expandafter\XINT_gctcv_prep \romannumeral`&&@#1+!/%
}%
\def\XINT_gctcv_prep
{%
    \XINT_gctcv_loop_a {}1001%
}%
\def\XINT_gctcv_loop_a #1#2#3#4#5#6+%
{%
    \expandafter\XINT_gctcv_loop_b
    \romannumeral0\xintrawwithzeros {#6}.{#2}{#3}{#4}{#5}{#1}%
}%
\def\XINT_gctcv_loop_b #1/#2.#3#4#5#6%
{%
    \expandafter\XINT_gctcv_loop_c\expandafter
    {\romannumeral0\XINT_mul_fork #2\xint:#4\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#3\xint:}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#6\xint:}%
                              {\XINT_mul_fork #1\xint:#4\xint:}}%
    {\romannumeral0\xintiiadd {\XINT_mul_fork #2\xint:#5\xint:}%
                              {\XINT_mul_fork #1\xint:#3\xint:}}%
}%
\def\XINT_gctcv_loop_c #1#2%
{%
    \expandafter\XINT_gctcv_loop_d\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_gctcv_loop_d #1#2%
{%
    \expandafter\XINT_gctcv_loop_e\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_gctcv_loop_e #1#2%
{%
    \expandafter\XINT_gctcv_loop_f\expandafter {#2}#1%
}%
\def\XINT_gctcv_loop_f #1#2%
{%
    \expandafter\XINT_gctcv_loop_g\expandafter
    {\romannumeral0\xintrawwithzeros {#1/#2}}{{#1}{#2}}%
}%
\def\XINT_gctcv_loop_g #1#2#3#4%
{%
    \XINT_gctcv_loop_h {#4{#1}}{#2#3}% 1.09b removes [0]
}%
\def\XINT_gctcv_loop_h #1#2#3/%
{%
    \xint_gob_til_exclam #3\XINT_gctcv_end!%
    \expandafter\XINT_gctcv_loop_i
    \romannumeral0\xintrawwithzeros {#3}.#2{#1}%
}%
\def\XINT_gctcv_loop_i #1/#2.#3#4#5#6%
{%
    \expandafter\XINT_gctcv_loop_j\expandafter
    {\romannumeral0\XINT_mul_fork #1\xint:#6\xint:}%
    {\romannumeral0\XINT_mul_fork #1\xint:#5\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#4\xint:}%
    {\romannumeral0\XINT_mul_fork #2\xint:#3\xint:}%
}%
\def\XINT_gctcv_loop_j #1#2%
{%
    \expandafter\XINT_gctcv_loop_k\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_gctcv_loop_k #1#2%
{%
    \expandafter\XINT_gctcv_loop_l\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_gctcv_loop_l #1#2%
{%
    \expandafter\XINT_gctcv_loop_m\expandafter {\expandafter{#2}#1}%
}%
\def\XINT_gctcv_loop_m #1#2{\XINT_gctcv_loop_a {#2}#1}%
\def\XINT_gctcv_end #1.#2#3#4#5#6{ #6}%
%    \end{macrocode}
% \subsection{\csh{xintiGCtoCv}}
%    \begin{macrocode}
\def\xintiGCtoCv {\romannumeral0\xintigctocv }%
\def\xintigctocv #1%
{%
    \expandafter\XINT_igctcv_prep \romannumeral`&&@#1+!/%
}%
\def\XINT_igctcv_prep
{%
    \XINT_igctcv_loop_a {}1001%
}%
\def\XINT_igctcv_loop_a #1#2#3#4#5#6+%
{%
    \expandafter\XINT_igctcv_loop_b
    \romannumeral`&&@#6.{#2}{#3}{#4}{#5}{#1}%
}%
\def\XINT_igctcv_loop_b #1.#2#3#4#5%
{%
    \expandafter\XINT_igctcv_loop_c\expandafter
    {\romannumeral0\xintiiadd {#5}{\XINT_mul_fork #1\xint:#3\xint:}}%
    {\romannumeral0\xintiiadd {#4}{\XINT_mul_fork #1\xint:#2\xint:}}%
    {{#2}{#3}}%
}%
\def\XINT_igctcv_loop_c #1#2%
{%
    \expandafter\XINT_igctcv_loop_f\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_igctcv_loop_f #1#2#3#4/%
{%
    \xint_gob_til_exclam #4\XINT_igctcv_end_a!%
    \expandafter\XINT_igctcv_loop_g
    \romannumeral`&&@#4.#1#2{#3}%
}%
\def\XINT_igctcv_loop_g #1.#2#3#4#5%
{%
    \expandafter\XINT_igctcv_loop_h\expandafter
    {\romannumeral0\XINT_mul_fork #1\xint:#5\xint:}%
    {\romannumeral0\XINT_mul_fork #1\xint:#4\xint:}%
    {{#2}{#3}}%
}%
\def\XINT_igctcv_loop_h #1#2%
{%
    \expandafter\XINT_igctcv_loop_i\expandafter {\expandafter{#2}{#1}}%
}%
\def\XINT_igctcv_loop_i #1#2{\XINT_igctcv_loop_k #2{#2#1}}%
\def\XINT_igctcv_loop_k #1#2%
{%
    \expandafter\XINT_igctcv_loop_l\expandafter
    {\romannumeral0\xintrawwithzeros {#1/#2}}%
}%
\def\XINT_igctcv_loop_l #1#2#3{\XINT_igctcv_loop_a {#3{#1}}#2}%1.09i removes [0]
\def\XINT_igctcv_end_a #1.#2#3#4#5%
{%
    \expandafter\XINT_igctcv_end_b\expandafter
    {\romannumeral0\xintrawwithzeros {#2/#3}}%
}%
\def\XINT_igctcv_end_b #1#2{ #2{#1}}% 1.09b removes [0]
%    \end{macrocode}
% \subsection{\csh{xintFtoCv}}
% \lverb|Still uses \xinticstocv \xintFtoCs rather than \xintctocv \xintFtoC.|
%    \begin{macrocode}
\def\xintFtoCv {\romannumeral0\xintftocv }%
\def\xintftocv #1%
{%
    \xinticstocv {\xintFtoCs {#1}}%
}%
%    \end{macrocode}
% \subsection{\csh{xintFtoCCv}}
%    \begin{macrocode}
\def\xintFtoCCv {\romannumeral0\xintftoccv }%
\def\xintftoccv #1%
{%
    \xintigctocv {\xintFtoCC {#1}}%
}%
%    \end{macrocode}
% \subsection{\csh{xintCntoF}}
% \lverb|&
% Modified in 1.06 to give the N first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.|
%    \begin{macrocode}
\def\xintCntoF {\romannumeral0\xintcntof }%
\def\xintcntof #1%
{%
    \expandafter\XINT_cntf\expandafter {\the\numexpr #1}%
}%
\def\XINT_cntf #1#2%
{%
   \ifnum #1>\xint_c_
      \xint_afterfi {\expandafter\XINT_cntf_loop\expandafter
                     {\the\numexpr #1-1\expandafter}\expandafter
                     {\romannumeral`&&@#2{#1}}{#2}}%
   \else
      \xint_afterfi
         {\ifnum #1=\xint_c_
              \xint_afterfi {\expandafter\space \romannumeral`&&@#2{0}}%
          \else \xint_afterfi { }% 1.09m now returns nothing.
          \fi}%
   \fi
}%
\def\XINT_cntf_loop #1#2#3%
{%
    \ifnum #1>\xint_c_ \else \XINT_cntf_exit \fi
    \expandafter\XINT_cntf_loop\expandafter
    {\the\numexpr #1-1\expandafter }\expandafter
    {\romannumeral0\xintadd {\xintDiv {1[0]}{#2}}{#3{#1}}}%
    {#3}%
}%
\def\XINT_cntf_exit \fi
    \expandafter\XINT_cntf_loop\expandafter
    #1\expandafter #2#3%
{%
    \fi\xint_gobble_ii #2%
}%
%    \end{macrocode}
% \subsection{\csh{xintGCntoF}}
% \lverb|Modified in 1.06 to give the N argument first to a \numexpr rather
% than expanding twice. I just use \the\numexpr and maintain the previous code
% after that.|
%    \begin{macrocode}
\def\xintGCntoF {\romannumeral0\xintgcntof }%
\def\xintgcntof #1%
{%
    \expandafter\XINT_gcntf\expandafter {\the\numexpr #1}%
}%
\def\XINT_gcntf #1#2#3%
{%
   \ifnum #1>\xint_c_
      \xint_afterfi {\expandafter\XINT_gcntf_loop\expandafter
                     {\the\numexpr #1-1\expandafter}\expandafter
                     {\romannumeral`&&@#2{#1}}{#2}{#3}}%
   \else
      \xint_afterfi
         {\ifnum #1=\xint_c_
              \xint_afterfi {\expandafter\space\romannumeral`&&@#2{0}}%
          \else \xint_afterfi { }% 1.09m now returns nothing rather than 0/1[0]
          \fi}%
   \fi
}%
\def\XINT_gcntf_loop #1#2#3#4%
{%
    \ifnum #1>\xint_c_ \else \XINT_gcntf_exit \fi
    \expandafter\XINT_gcntf_loop\expandafter
    {\the\numexpr #1-1\expandafter }\expandafter
    {\romannumeral0\xintadd {\xintDiv {#4{#1}}{#2}}{#3{#1}}}%
    {#3}{#4}%
}%
\def\XINT_gcntf_exit \fi
    \expandafter\XINT_gcntf_loop\expandafter
    #1\expandafter #2#3#4%
{%
    \fi\xint_gobble_ii #2%
}%
%    \end{macrocode}
% \subsection{\csh{xintCntoCs}}
% \lverb|Modified in 1.09m: added spaces after the commas in the produced list.
% Moreover the coefficients are not braced anymore. A slight induced limitation
% is that the macro argument should not contain some explicit comma (cf.
% \XINT_cntcs_exit_b), hence \xintCntoCs {\macro,} with \def\macro,#1{<stuff>}
% would crash. Not a very serious limitation, I believe. |
%    \begin{macrocode}
\def\xintCntoCs {\romannumeral0\xintcntocs }%
\def\xintcntocs #1%
{%
    \expandafter\XINT_cntcs\expandafter {\the\numexpr #1}%
}%
\def\XINT_cntcs #1#2%
{%
   \ifnum #1<0
      \xint_afterfi { }% 1.09i: a 0/1[0] was here, now the macro returns nothing
   \else
      \xint_afterfi {\expandafter\XINT_cntcs_loop\expandafter
                     {\the\numexpr #1-\xint_c_i\expandafter}\expandafter
                     {\romannumeral`&&@#2{#1}}{#2}}% produced coeff not braced
   \fi
}%
\def\XINT_cntcs_loop #1#2#3%
{%
    \ifnum #1>-\xint_c_i \else \XINT_cntcs_exit \fi
    \expandafter\XINT_cntcs_loop\expandafter
    {\the\numexpr #1-\xint_c_i\expandafter}\expandafter
    {\romannumeral`&&@#3{#1}, #2}{#3}% space added, 1.09m
}%
\def\XINT_cntcs_exit \fi
    \expandafter\XINT_cntcs_loop\expandafter
    #1\expandafter #2#3%
{%
    \fi\XINT_cntcs_exit_b #2%
}%
\def\XINT_cntcs_exit_b #1,{}% romannumeral stopping space already there
%    \end{macrocode}
% \subsection{\csh{xintCntoGC}}
% \lverb|&
% Modified in 1.06 to give the N first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.
%
% 1.09m maintains the braces, as the coeff are allowed to be fraction and the
% slash can not be naked in the GC format, contrarily to what happens in
% \xintCntoCs. Also the separators given to \xintGCtoGCx may then fetch the
% coefficients as argument, as they are braced.|
%    \begin{macrocode}
\def\xintCntoGC {\romannumeral0\xintcntogc }%
\def\xintcntogc #1%
{%
    \expandafter\XINT_cntgc\expandafter {\the\numexpr #1}%
}%
\def\XINT_cntgc #1#2%
{%
   \ifnum #1<0
      \xint_afterfi { }% 1.09i there was as strange 0/1[0] here, removed
   \else
      \xint_afterfi {\expandafter\XINT_cntgc_loop\expandafter
                     {\the\numexpr #1-\xint_c_i\expandafter}\expandafter
                     {\expandafter{\romannumeral`&&@#2{#1}}}{#2}}%
   \fi
}%
\def\XINT_cntgc_loop #1#2#3%
{%
    \ifnum #1>-\xint_c_i \else \XINT_cntgc_exit \fi
    \expandafter\XINT_cntgc_loop\expandafter
    {\the\numexpr #1-\xint_c_i\expandafter }\expandafter
    {\expandafter{\romannumeral`&&@#3{#1}}+1/#2}{#3}%
}%
\def\XINT_cntgc_exit \fi
    \expandafter\XINT_cntgc_loop\expandafter
    #1\expandafter #2#3%
{%
    \fi\XINT_cntgc_exit_b #2%
}%
\def\XINT_cntgc_exit_b #1+1/{ }%
%    \end{macrocode}
% \subsection{\csh{xintGCntoGC}}
% \lverb|&
% Modified in 1.06 to give the N first to a \numexpr rather than expanding
% twice. I just use \the\numexpr and maintain the previous code after that.|
%    \begin{macrocode}
\def\xintGCntoGC {\romannumeral0\xintgcntogc }%
\def\xintgcntogc #1%
{%
    \expandafter\XINT_gcntgc\expandafter {\the\numexpr #1}%
}%
\def\XINT_gcntgc #1#2#3%
{%
   \ifnum #1<0
      \xint_afterfi { }% 1.09i now returns nothing
   \else
      \xint_afterfi {\expandafter\XINT_gcntgc_loop\expandafter
                     {\the\numexpr #1-\xint_c_i\expandafter}\expandafter
                     {\expandafter{\romannumeral`&&@#2{#1}}}{#2}{#3}}%
   \fi
}%
\def\XINT_gcntgc_loop #1#2#3#4%
{%
    \ifnum #1>-\xint_c_i \else \XINT_gcntgc_exit \fi
    \expandafter\XINT_gcntgc_loop_b\expandafter
    {\expandafter{\romannumeral`&&@#4{#1}}/#2}{#3{#1}}{#1}{#3}{#4}%
}%
\def\XINT_gcntgc_loop_b #1#2#3%
{%
    \expandafter\XINT_gcntgc_loop\expandafter
    {\the\numexpr #3-\xint_c_i \expandafter}\expandafter
    {\expandafter{\romannumeral`&&@#2}+#1}%
}%
\def\XINT_gcntgc_exit \fi
    \expandafter\XINT_gcntgc_loop_b\expandafter #1#2#3#4#5%
{%
    \fi\XINT_gcntgc_exit_b #1%
}%
\def\XINT_gcntgc_exit_b #1/{ }%
%    \end{macrocode}
% \subsection{\csh{xintCstoGC}}
%    \begin{macrocode}
\def\xintCstoGC {\romannumeral0\xintcstogc }%
\def\xintcstogc #1%
{%
    \expandafter\XINT_cstc_prep \romannumeral`&&@#1,!,%
}%
\def\XINT_cstc_prep #1,{\XINT_cstc_loop_a {{#1}}}%
\def\XINT_cstc_loop_a #1#2,%
{%
    \xint_gob_til_exclam #2\XINT_cstc_end!%
    \XINT_cstc_loop_b {#1}{#2}%
}%
\def\XINT_cstc_loop_b #1#2{\XINT_cstc_loop_a {#1+1/{#2}}}%
\def\XINT_cstc_end!\XINT_cstc_loop_b #1#2{ #1}%
%    \end{macrocode}
% \subsection{\csh{xintGCtoGC}}
%    \begin{macrocode}
\def\xintGCtoGC {\romannumeral0\xintgctogc }%
\def\xintgctogc #1%
{%
    \expandafter\XINT_gctgc_start \romannumeral`&&@#1+!/%
}%
\def\XINT_gctgc_start {\XINT_gctgc_loop_a {}}%
\def\XINT_gctgc_loop_a #1#2+#3/%
{%
    \xint_gob_til_exclam #3\XINT_gctgc_end!%
    \expandafter\XINT_gctgc_loop_b\expandafter
    {\romannumeral`&&@#2}{#3}{#1}%
}%
\def\XINT_gctgc_loop_b #1#2%
{%
    \expandafter\XINT_gctgc_loop_c\expandafter
    {\romannumeral`&&@#2}{#1}%
}%
\def\XINT_gctgc_loop_c #1#2#3%
{%
    \XINT_gctgc_loop_a {#3{#2}+{#1}/}%
}%
\def\XINT_gctgc_end!\expandafter\XINT_gctgc_loop_b
{%
    \expandafter\XINT_gctgc_end_b
}%
\def\XINT_gctgc_end_b #1#2#3{ #3{#1}}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintcfrac}
% \cleardoublepage\let\xintcfracnameUp\undefined
%\gardesactifs
%\let</xintcfrac>\relax
%\let<*xintexpr>\gardesinactifs
%</xintcfrac>^^A--------------------------------------------------
%<*xintexpr>^^A---------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintexprnameUp\endcsname
% \section{Package \xintexprnameimp implementation}
% \RaisedLabel{sec:exprimp}
% \etocarticlestylenomarks
% \etocstandardlines
% \etocsetnexttocdepth {subsubsection}
%
%
% This is release \expandafter|\xintbndlversion| of
% \expandafter|\xintbndldate|.
%
% \localtableofcontents
%
% \etocsettocstyle{}{}
%
% \subsection{READ ME! Important warnings and explanations relative to the
% status of the code source at the time of the 1.4 release}
% \def\mybeginitemize{\begin{itemize}\catcode`\% 9 }%
% \def\myenditemize{\end{itemize}}
% \lverb|&
% At release 1.4 the csname encapsulation of intermediate evaluations during
% parsing of expressions is dropped, and $xintexprnameimp requires the
% \expanded primitive. This means that there is no more impact on the string
% pool. And as internal storage now uses simply core \TeX{} syntax with braces
% rather than comma separated items inside a csname dummy control sequence,
% it became much easier to let the [...] syntax be associated to a true internal type of «tuple»
% or «list».
%
% The output of \xintexpr (after \romannumeral0 or \romannumeral-`0 triggered
% expansion or double expansion) is thus modified at 1.4. It now looks like
% this:
% 
% \XINTfstop \XINTexprprint .{{<number>}} in simplest case
%
% \XINTfstop \XINTexprprint .{{...}...{...}} in general case
%
% where ... stands for nested braces ultimately ending in {<num. rep.>} leaves.
% The <num. rep.> stands for some internal representation of numeric data. It
% may be empty, and currently as well as probably in future uses only catcode
% 12 tokens (no spaces currently).
%
% {{}} corresponds (in input as in output) to []. The external TeX braces also
% serve as set-theoretical braces. The comma is concatenation, so for example
% [], [] will become {{}{}}, or rather {}{} if sub-unit of something else.
%
% The associated vocabulary is explained in the user manual and we avoid
% too much duplication here. $xintfracnameimp numerical macros receiving an
% empty argument usually handle it as being 0, but this is not the case of the
% $xintcorenameimp macros supporting \xintiiexpr, they usually break if
% exercised on some empty argument.
%
% The above expansion result \XINTfstop \XINTexprprint .{{<num1>}{<num2}...}
% uses only normal catcodes:
% the backslash, regular braces, and catcode 12 characters. Scientific
% notation is internally converted to raw $xintfracnameimp representation
% [N].
%
% Additional data may be located before the dot; this is the case only for
% \xintfloatexpr currently. As
% xintexpr actually defines three parsers \xintexpr, \xintiiexpr and
% \xintfloatexpr but tries to share as much code as possible, some overhead
% is induced to fit all into the same mold.
%
% \XINTfstop stops \romannumeral-`0 (or 0) type spanned expansion, and is
% invariant under \edef, but simply disappears in typesetting context. It is
% thus now legal to use \xintexpr directly in typesetting flow.
% 
% \XINTexprprint is \protected.
%
% The f-expansion of an \xintexpr <expression>\relax is a complete expansion,
% i.e. one whose result remains invariant under \edef. But if exposed to
% finitely many expansion steps (at least two) there is a «blinking» \noexpand
% upfront depending on parity of number of steps.
%
% \xintthe\xintexpr <expression>\relax or \xinteval{<expression>} serve as
% formerly to deliver the explicit digits, or more exactly some prettifying
% view of the actual <internal number representation>. For example
% \xintthe\xintboolexpr will (this is tentative) use True and False in output.
%
% Nested contents like this
%
% {{1}{{2}{3}{{4}{5}{6}}}{9}}
%
% $noindent will get delivered using nested square brackets like that
%
% 1, [2, 3, [4, 5, 6]], 9
%
% $noindent and as conversely \xintexpr 1, [2, 3, [4, 5, 6]], 9\relax expands to
%
% \XINTfstop \XINTexprprint .{{1}{{2}{3}{{4}{5}{6}}}{9}}
%
% $noindent we obtain the gratifying result that
%
% \xinteval{1, [2, 3, [4, 5, 6]], 9}
%
% $noindent expands to 
%
% 1, [2, 3, [4, 5, 6]], 9
% 
% See user manual for explanations on the plasticity of \xintexpr syntax
% regarding functions with multiple arguments, and the 1.4 «unpacking»
% Python-like * prefix operator.
%
% |
%
% \begin{framed}
%   I have suppressed (from the public dtx) many big chunks of
%   comments. Some became obsolete and need to be updated, others are
%   currently of value only to the author as a historical record.
%
%   ATTENTION! As the removal process itself took too much time, I ended up
%   leaving as is many comments which are obsoleted and wrong to various
%   degrees after the |1.4| release. Precedence levels of operators have all
%   been doubled to make room for new constructs
%
%   Even comments added during 1.4 developement may now be obsolete because
%   the preparation of 1.4 took a few weeks and that's enough of duration to
%   provide the author many chances to contradict in the code what has been
%   already commented upon.
%
%   Thus don't believe (fully) anything which is said here!\IMPORTANTf{}
% \end{framed}
%
%
% \lverb|Warning: in text below and also in left-over old comments I may refer
% to «until» and «op» macros; due to the change of data storage at 1.4, I
% needed to refactor a bit the way expansion is controlled, and the situation
% now is mainly governed by «op», «exec», «check-» and «checkp» macros the
% latter three replacing the two «until_a» and «until_b» of former code. This
% allows to diminish the number of times an accumulated result will be grabbed
% in order to propagate expansion to its right. Formerly this was not an issue
% because such things were only a single token! I do not describe here how
% this is all articulated but it is not hard to see it from the code (the
% hardest thing in all such matter was in 2013 to actually write how the
% expansion would be initially launched because to do that one basically has to
% understand the mechanism in its whole and such things are not easy to
% develop piecemeal). Another thing to keep in mind is that operators in truth
% have a left precedence (i.e. the precedence they show to operators arising
% earlier) and a right precedence (which determines how they react to
% operators coming after them from the right). Only the first one is usually
% encapsulated in a chardef, the second one is most of the times identical to
% the first one and if not it is only virtual but implemented via \ifcase of
% \ifnum branching. A final remark is that some things are achieved by special
% «op» macros, which are a favorite tool to hack into the normal regular flow
% of things, via injection of special syntax elements. I did not rename these
% macros for avoiding too large git diffs, and besides the nice thing is that
% the 1.4 refactoring minimally had to modify them, and all hacky things using
% them kept on working with not a single modification. And a post-scriptum is
% that advanced features crucially exploit injecting sub-\xintexpr-essions, as
% all is expandable there is no real «context» (only a minimal one) which one
% would have to perhaps store and restore and doing this sub-expression
% injection is rather cheap and efficient operation.|
%
% \subsection{Old comments}
%
% These general comments were last updated at the end of the |1.09x| series in
% 2014. The principles remain in place to this day but refer to
% \href{http://www.ctan.org/pkg/xint/CHANGES.html}{CHANGES.html} for some
% significant evolutions since.
%
% The first version was released in June 2013. I was greatly helped in this task
% of writing an expandable parser of infix operations by the comments provided
% in |l3fp-parse.dtx| (in its version as available in April-May 2013). One will
% recognize in particular the idea of the `until' macros; I have not looked into
% the actual |l3fp| code beyond the very useful comments provided in its
% documentation.
%
% A main worry was that my data has no a priori bound on its size; to keep the
% code reasonably efficient, I experimented with a technique of storing and
% retrieving data expandably as \emph{names} of control sequences. Intermediate
% computation results are stored as control sequences |\.=a/b[n]|.
%
%
% Roughly speaking, the parser mechanism is as follows: at any given time the
% last found ``operator'' has its associated |until| macro awaiting some news
% from the token flow; first |getnext| expands forward in the hope to construct
% some number, which may come from a parenthesized sub-expression, from some
% braced material, or from a digit by digit scan. After this number has been
% formed the next operator is looked for by the |getop| macro. Once |getop| has
% finished its job, |until| is presented with three tokens: the first one is the
% precedence level of the new found operator (which may be an end of expression
% marker), the second is the operator character token (earlier versions had here
% already some macro name, but in order to keep as much common code to expr and
% floatexpr common as possible, this was modified) of the new found operator, and
% the third one is the newly found number (which was encountered just before the
% new operator).
%
% The |until| macro of the earlier operator examines the precedence level of the
% new found one, and either executes the earlier operator (in the case of a
% binary operation, with the found number and a previously stored one) or it
% delays execution, giving the hand to the |until| macro of the operator having
% been found of higher precedence.
%
% A minus sign acting as prefix gets converted into a (unary) operator
% inheriting the precedence level of the previous operator.
%
% Once the end of the expression is found (it has to be marked by a |\relax|)
% the final result is output as four tokens (five tokens since |1.09j|) the
% first one a catcode 11 exclamation mark, the second one an error generating
% macro, the third one is a protection mechanism, the fourth one a printing
% macro and the fifth is |\.=a/b[n]|. The prefix |\xintthe| makes the output
% printable by killing the first three tokens.
%
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%
% The code for reload detection was initially copied from \textsc{Heiko
% Oberdiek}'s packages, then modified.
%
% The method for catcodes was also initially directly inspired by these
% packages.
%
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \def\z {\endgroup}%
  \expandafter\let\expandafter\x\csname ver@xintexpr.sty\endcsname
  \expandafter\let\expandafter\w\csname ver@xintfrac.sty\endcsname
  \expandafter\let\expandafter\t\csname ver@xinttools.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info: #2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  % I don't think engine exists providing \expanded but not \numexpr
  \ifx\csname expanded\endcsname\relax
     \y{xintexpr}{\expanded not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\x\relax   % plain-TeX, first loading of xintexpr.sty
      \ifx\w\relax % but xintfrac.sty not yet loaded.
         \expandafter\def\expandafter\z\expandafter
                    {\z\input xintfrac.sty\relax}%
      \fi
      \ifx\t\relax % but xinttools.sty not yet loaded.
         \expandafter\def\expandafter\z\expandafter
                    {\z\input xinttools.sty\relax}%
      \fi
    \else
      \def\empty {}%
      \ifx\x\empty % LaTeX, first loading,
      % variable is initialized, but \ProvidesPackage not yet seen
          \ifx\w\relax % xintfrac.sty not yet loaded.
            \expandafter\def\expandafter\z\expandafter
                           {\z\RequirePackage{xintfrac}}%
          \fi
          \ifx\t\relax % xinttools.sty not yet loaded.
            \expandafter\def\expandafter\z\expandafter
                           {\z\RequirePackage{xinttools}}%
          \fi
      \else
        \aftergroup\endinput % xintexpr already loaded.
      \fi
    \fi
  \fi
\z%
\XINTsetupcatcodes%
%    \end{macrocode}
% \subsection{Package identification}
% \lverb|&
% \XINT_Cmp alias for \xintiiCmp needed for some forgotten reason related to
% \xintNewExpr (FIX THIS!)
%
% |
%    \begin{macrocode}
\XINT_providespackage
\ProvidesPackage{xintexpr}%
  [2021/07/13 v1.4j Expandable expression parser (JFB)]%
\catcode`! 11
\let\XINT_Cmp \xintiiCmp
\def\XINTfstop{\noexpand\XINTfstop}%
%    \end{macrocode}
% \subsection{\csh{xintDigits*}, \csh{xintSetDigits*}, \csh{xintreloadscilibs}}
% \lverb|1.3f. 1.4e added some \xintGuardDigits and \XINTdigitsx mechanism but
% it was finally removed, due to pending issues of user interface,
% functionality, and documentation (the worst part) for whose resolution
% no time was left.
%|
%    \begin{macrocode}
\def\xintreloadscilibs{\xintreloadxintlog\xintreloadxinttrig}%
\def\xintDigits {\futurelet\XINT_token\xintDigits_i}%
\def\xintDigits_i#1={\afterassignment\xintDigits_j\mathchardef\XINT_digits=}%
\def\xintDigits_j#1%
{%
    \let\XINTdigits=\XINT_digits
    \ifx*\XINT_token\expandafter\xintreloadscilibs\fi
}%
\let\xintfracSetDigits\xintSetDigits
\def\xintSetDigits#1#{\if\relax\detokenize{#1}\relax\expandafter\xintfracSetDigits
                      \else\expandafter\xintSetDigits_a\fi}%
\def\xintSetDigits_a#1%
{%
    \mathchardef\XINT_digits=\numexpr#1\relax
    \let\XINTdigits\XINT_digits
    \xintreloadscilibs
}%
%    \end{macrocode}
% \subsection{\csh{XINTdigitsormax}}
% \lverb|1.4f. To not let xintlog and xinttrig work with, and produce,
% long mantissas exceeeding the supported range for accuracy of the math
% functions. The official maximal value is 62, let's set the cut-off at 64.
%
% A priori, no need for \expandafter, always ends up expanded in \numexpr (I
% saw also in an \edef in xinttrig as argument to \xintReplicate prior
% to its \numexpr).
%|
%    \begin{macrocode}
\def\XINTdigitsormax{\ifnum\XINTdigits>\xint_c_ii^vi\xint_c_ii^vi\else\XINTdigits\fi}%
%    \end{macrocode}
% \subsection{Support for output and transform of nested braced contents as
% core data type}
% New at 1.4, of course. The former |\csname.=...\endcsname| encapsulation
% technique made very difficult implementation of nested structures.
% \subsubsection{Bracketed list rendering with prettifying of leaves from nested
% braced contents}
% \lverb|1.4 The braces in \XINT:expr:toblistwith are there because there is
% an \expanded trigger.
%
% 1.4d: support for polexpr 0.8 polynomial type.
% |
%    \begin{macrocode}
\def\XINT:expr:toblistwith#1#2%
{%
    {\expandafter\XINT:expr:toblist_checkempty
     \expanded{\noexpand#1!\expandafter}\detokenize{#2}^}%
}%
\def\XINT:expr:toblist_checkempty #1!#2%
{%
    \if ^#2\expandafter\xint_gob_til_^\else\expandafter\XINT:expr:toblist_a\fi
    #1!#2%
}%
\catcode`< 1 \catcode`> 2 \catcode`{ 12 \catcode`} 12
\def\XINT:expr:toblist_a #1{#2%
<%
    \if{#2\xint_dothis<[\XINT:expr:toblist_a>\fi
    \if P#2\xint_dothis<\XINT:expr:toblist_pol>\fi
    \xint_orthat\XINT:expr:toblist_b #1#2%
>%
\def\XINT:expr:toblist_pol #1!#2.{#3}}%
<%
    pol([\XINT:expr:toblist_b #1!#3}^])\XINT:expr:toblist_c #1!}%
>%
\def\XINT:expr:toblist_b #1!#2}%
<%
    \if\relax#2\relax\xintexprEmptyItem\else#1<#2>\fi\XINT:expr:toblist_c #1!}%
>%
\def\XINT:expr:toblist_c #1}#2%
<%
    \if ^#2\xint_dothis<\xint_gob_til_^>\fi
    \if{#2\xint_dothis<, \XINT:expr:toblist_a>\fi
    \xint_orthat<]\XINT:expr:toblist_c>#1#2%
>%
\catcode`{ 1 \catcode`} 2 \catcode`< 12 \catcode`> 12
%    \end{macrocode}
% \subsubsection{Flattening nested
% braced contents}
% \lverb|1.4b I hesitated whether using this technique or some variation of
% the method of the ListSel macros. I chose this one which I downscaled from
% toblistwith, I will revisit later. I only have a few minutes right now.
%
% Call form is \expanded\XINT:expr:flatten
%
% See \XINT_expr_func_flat. I hesitated with «flattened», but short names
% are faster parsed.
% |
%    \begin{macrocode}
\def\XINT:expr:flatten#1%
{%
   {{\expandafter\XINT:expr:flatten_checkempty\detokenize{#1}^}}%
}%
\def\XINT:expr:flatten_checkempty #1%
{%
    \if ^#1\expandafter\xint_gobble_i\else\expandafter\XINT:expr:flatten_a\fi
    #1%
}%
\begingroup % should I check lccode s generally if corrupted context at load?
\catcode`[ 1 \catcode`] 2 \lccode`[`{ \lccode`]`}
\catcode`< 1 \catcode`> 2 \catcode`{ 12 \catcode`} 12
\lowercase<\endgroup
\def\XINT:expr:flatten_a {#1%
<%
    \if{#1\xint_dothis<\XINT:expr:flatten_a>\fi
    \xint_orthat\XINT:expr:flatten_b #1%
>%
\def\XINT:expr:flatten_b #1}%
<%
    [#1]\XINT:expr:flatten_c }%
>%
\def\XINT:expr:flatten_c }#1%
<%
    \if ^#1\xint_dothis<\xint_gobble_i>\fi
    \if{#1\xint_dothis<\XINT:expr:flatten_a>\fi
    \xint_orthat<\XINT:expr:flatten_c>#1%
>%
>% back to normal catcodes
%    \end{macrocode}
% \subsubsection{Braced contents rendering via a \TeX{} alignment with prettifying of leaves}
% \lverb|1.4.
%
% Breaking change at 1.4a as helper macros were renamed and their meanings
% refactored: no more \xintexpraligntab nor \xintexpraligninnercomma or
% \xintexpralignoutercomma but \xintexpraligninnersep, etc...
%
% At 1.4c I remove the \protected from \xintexpralignend.  I had made note a
% year ago that it served nothing.  Let's trust myself on this one (risky one
% year later!)
% .|
%    \begin{macrocode}
\catcode`& 4
\protected\def\xintexpralignbegin       {\halign\bgroup\tabskip2ex\hfil##&&##\hfil\cr}%
\def\xintexpralignend                   {\crcr\egroup}%
\protected\def\xintexpralignlinesep     {,\cr}%
\protected\def\xintexpralignleftbracket {[}%
\protected\def\xintexpralignrightbracket{]}%
\protected\def\xintexpralignleftsep     {&}%
\protected\def\xintexpralignrightsep    {&}%
\protected\def\xintexpraligninnersep    {,&}%
\catcode`& 7
\def\XINT:expr:toalignwith#1#2%
{%
    {\expandafter\XINT:expr:toalign_checkempty
     \expanded{\noexpand#1!\expandafter}\detokenize{#2}^\expandafter}%
    \xintexpralignend
}%
\def\XINT:expr:toalign_checkempty #1!#2%
{%
    \if ^#2\expandafter\xint_gob_til_^\else\expandafter\XINT:expr:toalign_a\fi
    #1!#2%
}%
\catcode`< 1 \catcode`> 2 \catcode`{ 12 \catcode`} 12
\def\XINT:expr:toalign_a #1{#2%
<%
    \if{#2\xint_dothis<\xintexpralignleftbracket\XINT:expr:toalign_a>\fi
    \xint_orthat<\xintexpralignleftsep\XINT:expr:toalign_b>#1#2%
>%
\def\XINT:expr:toalign_b #1!#2}%
<%
    \if\relax#2\relax\xintexprEmptyItem\else#1<#2>\fi\XINT:expr:toalign_c #1!}%
>%
\def\XINT:expr:toalign_c #1}#2%
<%
    \if ^#2\xint_dothis<\xint_gob_til_^>\fi
    \if {#2\xint_dothis<\xintexpraligninnersep\XINT:expr:toalign_A>\fi
    \xint_orthat<\xintexpralignrightsep\xintexpralignrightbracket\XINT:expr:toalign_C>#1#2%
>%
\def\XINT:expr:toalign_A #1{#2%
<%
    \if{#2\xint_dothis<\xintexpralignleftbracket\XINT:expr:toalign_A>\fi
    \xint_orthat\XINT:expr:toalign_b #1#2%
>%
\def\XINT:expr:toalign_C #1}#2%
<%
    \if ^#2\xint_dothis<\xint_gob_til_^>\fi
    \if {#2\xint_dothis<\xintexpralignlinesep\XINT:expr:toalign_a>\fi
    \xint_orthat<\xintexpralignrightbracket\XINT:expr:toalign_C>#1#2%
>%
\catcode`{ 1 \catcode`} 2 \catcode`< 12 \catcode`> 12
%    \end{macrocode}
% \subsubsection{Transforming all leaves within nested
% braced contents}
%
% \lverb|1.4. Leaves must be of catcode 12... This is currently not a
% constraint (or rather not a new constraint) for xintexpr because formerly
% anyhow all data went through csname encapsulation and extraction via string.
%
% In order to share code with the functioning of universal functions, which
% will be allowed to transform a number into an ople, the applied macro
% is supposed to apply one level of bracing to its ouput. Thus to apply this
% with an xintfrac macro such as \xintiRound{0} one needs first to define
% a wrapper which will expand it inside an added brace pair:
%
% \def\foo#1{{\xintiRound{0}{#1}}}%
%
% As the things will expand inside expanded, propagating expansion is not an
% issue.
%
% This code is used by \xintiexpr and \xintfloatexpr in case of optional
% argument and by the «Universal functions».
% |
%
%    \begin{macrocode}
\def\XINT:expr:mapwithin#1#2%
{%
    {{\expandafter\XINT:expr:mapwithin_checkempty
      \expanded{\noexpand#1!\expandafter}\detokenize{#2}^}}%
}%
\def\XINT:expr:mapwithin_checkempty #1!#2%
{%
    \if ^#2\expandafter\xint_gob_til_^\else\expandafter\XINT:expr:mapwithin_a\fi
    #1!#2%
}%
\begingroup % should I check lccode s generally if corrupted context at load?
\catcode`[ 1 \catcode`] 2 \lccode`[`{ \lccode`]`}
\catcode`< 1 \catcode`> 2 \catcode`{ 12 \catcode`} 12
\lowercase<\endgroup
\def\XINT:expr:mapwithin_a #1{#2%
<%
    \if{#2\xint_dothis<[\iffalse]\fi\XINT:expr:mapwithin_a>\fi%
    \xint_orthat\XINT:expr:mapwithin_b #1#2%
>%
\def\XINT:expr:mapwithin_b #1!#2}%
<%
    #1<#2>\XINT:expr:mapwithin_c #1!}%
>%
\def\XINT:expr:mapwithin_c #1}#2%
<%
    \if ^#2\xint_dothis<\xint_gob_til_^>\fi
    \if{#2\xint_dothis<\XINT:expr:mapwithin_a>\fi%
    \xint_orthat<\iffalse[\fi]\XINT:expr:mapwithin_c>#1#2%
>%
>% back to normal catcodes
%    \end{macrocode}
% \subsection{Top level user \TeX{} interface: \cshnolabel{xinteval},
% \cshnolabel{xintfloateval}, \cshnolabel{xintiieval}}
% \localtableofcontents
% \subsubsection{\csh{xintexpr}, \csh{xintiexpr}, \csh{xintfloatexpr},
% \csh{xintiiexpr}}
% \lverb|&
%
% \xintiexpr and \xintfloatexpr have an optional argument since 1.1.
%
% ATTENTION! 1.3d renamed \xinteval to \xintexpro etc...
%
%
% Usage of \xintiRound{0} for \xintiexpr without optional [D] means that
% \xintiexpr ... \relax wrapper can be used to insert rounded-to-integers
% values in \xintiiexpr context: no post-fix [0] which would break it.
%
% 1.4a add support for the optional argument [D] for \xintiexpr being negative
% D, with same meaning as the 1.4a modified \xintRound from xintfrac.sty.
%
% \xintiexpr mechanism was refactored at 1.4e so that rounding due to [D]
% optional argument uses raw format, not fixed point format on output,
% delegating fixed point conversion to an \XINTiexprprint now separated from
% \XINTexprprint.
%
% In case of negative [D], \xintiexpr [D]...\relax internally has the [0]
% post-fix so it can not be inserted as sub-expression in \xintiiexpr without
% a num() or \xintiexpr ...\relax (extra) wrapper.
%
% |
%    \begin{macrocode}
\def\xintexpr       {\romannumeral0\xintexpro      }%
\def\xintiexpr      {\romannumeral0\xintiexpro     }%
\def\xintfloatexpr  {\romannumeral0\xintfloatexpro }%
\def\xintiiexpr     {\romannumeral0\xintiiexpro    }%
\def\xintexpro   {\expandafter\XINT_expr_wrap\romannumeral0\xintbareeval }%
\def\xintiiexpro {\expandafter\XINT_iiexpr_wrap\romannumeral0\xintbareiieval }%
\def\xintiexpro #1%
{%
    \ifx [#1\expandafter\XINT_iexpr_withopt\else\expandafter\XINT_iexpr_noopt
    \fi #1%
}%
\def\XINT_iexpr_noopt
{%
    \expandafter\XINT_iexpr_iiround\romannumeral0\xintbareeval
}%
\def\XINT_iexpr_iiround
{%
    \expandafter\XINT_expr_wrap
    \expanded
    \XINT:NEhook:x:mapwithin\XINT:expr:mapwithin{\XINTiRoundzero_braced}%
}%
\def\XINTiRoundzero_braced#1{{\xintiRound{0}{#1}}}%
\def\XINT_iexpr_withopt [#1]%
{%
    \expandafter\XINT_iexpr_round
    \the\numexpr \xint_zapspaces #1 \xint_gobble_i\expandafter.%
    \romannumeral0\xintbareeval
}%
\def\XINT_iexpr_round #1.%
{%
    \ifnum#1=\xint_c_\xint_dothis{\XINT_iexpr_iiround}\fi
    \xint_orthat{\XINT_iexpr_round_a #1.}%
}%
\def\XINT_iexpr_round_a #1.%
{%
    \expandafter\XINT_iexpr_wrap
    \expanded
    \XINT:NEhook:x:mapwithin\XINT:expr:mapwithin{\XINTiRound_braced{#1}}%
}%
\def\XINTiRound_braced#1#2{{\xintiRound{#1}{#2}[\the\numexpr\ifnum#1<\xint_c_i0\else-#1\fi]}}%
\def\xintfloatexpro #1%
{%
    \ifx [#1\expandafter\XINT_flexpr_withopt\else\expandafter\XINT_flexpr_noopt
    \fi #1%
}%
\def\XINT_flexpr_noopt
{%
    \expandafter\XINT_flexpr_wrap\the\numexpr\XINTdigits\expandafter.%
    \romannumeral0\xintbarefloateval
}%
\def\XINT_flexpr_withopt [#1]%
{%
    \expandafter\XINT_flexpr_withopt_a
    \the\numexpr\xint_zapspaces #1 \xint_gobble_i\expandafter.%
    \romannumeral0\xintbarefloateval
}%
\def\XINT_flexpr_withopt_a #1#2.%
{%
    \expandafter\XINT_flexpr_withopt_b\the\numexpr\if#1-\XINTdigits\fi#1#2.%
}%
\def\XINT_flexpr_withopt_b #1.%
{%
    \expandafter\XINT_flexpr_wrap
    \the\numexpr#1\expandafter.%
    \expanded
    \XINT:NEhook:x:mapwithin\XINT:expr:mapwithin{\XINTinFloat_braced[#1]}%
}%
\def\XINTinFloat_braced[#1]#2{{\XINTinFloat[#1]{#2}}}%
%    \end{macrocode}
% \subsubsection{\csh{XINT_expr_wrap}, \csh{XINT_iiexpr_wrap}, \csh{XINT_flexpr_wrap}}
% \lverb|1.3e removes some leading space tokens which served nothing. There is
% no \XINT_iexpr_wrap, because \XINT_expr_wrap is used directly.
%
% 1.4e has \XINT_iexpr_wrap separated from \XINT_expr_wrap, thus simplifying
% internal matters as output printer for \xintexpr will not have to handle
% fixed point input but only extended-raw type input (i.e. A, A/B, A[N] or A/B[N]).
% |
%    \begin{macrocode}
\def\XINT_expr_wrap   {\XINTfstop\XINTexprprint.}%
\def\XINT_iexpr_wrap  {\XINTfstop\XINTiexprprint.}%
\def\XINT_iiexpr_wrap {\XINTfstop\XINTiiexprprint.}%
\def\XINT_flexpr_wrap {\XINTfstop\XINTflexprprint}%
%    \end{macrocode}
% \subsubsection{\csh{XINTexprprint}, \csh{XINTiexprprint}, \csh{XINTiiexprprint}, \csh{XINTflexprprint}}
% \lverb|
% 
% Comments (still) currently under reconstruction.
%
% 1.4: this now requires \expanded context.
%
% 1.4e has a separate \XINTiexprprint and \xintiexprPrintOne.
%
% 1.4e has a breaking change of \XINTflexprprint and \xintfloatexprPrintOne
% which now requires \xintfloatexprPrintOne[D]{x} usage, with first argument
% in brackets.
%
%  | 
%    \begin{macrocode}
\protected\def\XINTexprprint.%
    {\XINT:NEhook:x:toblist\XINT:expr:toblistwith\xintexprPrintOne}%
\let\xintexprPrintOne\xintFracToSci
\protected\def\XINTiexprprint.%
    {\XINT:NEhook:x:toblist\XINT:expr:toblistwith\xintiexprPrintOne}%
\let\xintiexprPrintOne\xintDecToString
\def\xintexprEmptyItem{[]}%
\protected\def\XINTiiexprprint.%
    {\XINT:NEhook:x:toblist\XINT:expr:toblistwith\xintiiexprPrintOne}%
\let\xintiiexprPrintOne\xint_firstofone
\protected\def\XINTflexprprint #1.%
    {\XINT:NEhook:x:toblist\XINT:expr:toblistwith{\xintfloatexprPrintOne[#1]}}%
\let\xintfloatexprPrintOne\xintPFloat_wopt
\protected\def\XINTboolexprprint.%
    {\XINT:NEhook:x:toblist\XINT:expr:toblistwith\xintboolexprPrintOne}%
\def\xintboolexprPrintOne#1{\xintiiifNotZero{#1}{True}{False}}%
%    \end{macrocode}
% \subsubsection{\csh{xintthe}, \csh{xintthealign}, \csh{xinttheexpr}, \csh{xinttheiexpr}, \csh{xintthefloatexpr},
% \csh{xinttheiiexpr}}
% \lverb|The reason why \xinttheiexpr et \xintthefloatexpr are handled
% differently is that they admit an optional argument which acts via a custom
% «printing» stage.
%
% We exploit here that \expanded expands forward until finding an implicit or
% explicit brace, and that this expansion overrules \protected macros, forcing
% them to expand, similarly as \romannumeral expands \protected macros, and
% contrarily to what happens *within* the actual \expanded scope. I discovered
% this fact by testing (with pdftex) and I don't know where this is documented
% apart from the source code of the relevant engines. This is useful to us
% because there are contexts where we will want to apply a complete expansion
% before printing, but in purely numerical context this is not needed (if I
% converted correctly after dropping at 1.4 the \csname governed expansions;
% however I rely at various places on the fact that the xint macros are
% f-expandable, so I have tried to not use zillions of expanded all over the
% place), hence it is not needed to add the expansion overhead by default. But
% the \expanded here will allow \xintNewExpr to create macro with suitable
% modification or the printing step, via some hook rather than having to
% duplicate all macros here with some new «NE» meaning (aliasing does not work
% or causes big issues due to desire to support \xinteval also in «NE» context
% as sub-constituent. The \XINT:NEhook:x:toblist is something else which
% serves to achieve this support of *sub* \xinteval, it serves nothing for
% the actual produced macros. For \xintdeffunc, things are simpler, but still
% we support the [N] optional argument of \xintiexpr and \xintfloatexpr, which
% required some work...
%
% The \expanded upfront ensures \xintthe mechanism does expand
% completely in two steps.
% |
%    \begin{macrocode}
\def\xintthe      #1{\expanded\expandafter\xint_gobble_i\romannumeral`&&@#1}%
\def\xintthealign #1{\expandafter\xintexpralignbegin
                     \expanded\expandafter\XINT:expr:toalignwith
                     \romannumeral0\expandafter\expandafter\expandafter\expandafter
                                   \expandafter\expandafter\expandafter\xint_gob_andstop_ii
                     \expandafter\xint_gobble_i\romannumeral`&&@#1}%
\def\xinttheexpr
   {\expanded\expandafter\XINTexprprint\expandafter.\romannumeral0\xintbareeval}%
\def\xinttheiexpr    
   {\expanded\expandafter\xint_gobble_i\romannumeral`&&@\xintiexpr}%
\def\xintthefloatexpr
   {\expanded\expandafter\xint_gobble_i\romannumeral`&&@\xintfloatexpr}%
\def\xinttheiiexpr
   {\expanded\expandafter\XINTiiexprprint\expandafter.\romannumeral0\xintbareiieval}%
%    \end{macrocode}
% \subsubsection{\csh{thexintexpr}, \csh{thexintiexpr}, \csh{thexintfloatexpr},
% \csh{thexintiiexpr}}
% \lverb|New with 1.2h. I have been for the last three years very strict
% regarding macros with \xint or \XINT, but well.
%
% 1.4. Definitely I don't like those. I will remove them at 1.5.|
%    \begin{macrocode}
\let\thexintexpr     \xinttheexpr
\let\thexintiexpr    \xinttheiexpr
\let\thexintfloatexpr\xintthefloatexpr
\let\thexintiiexpr   \xinttheiiexpr
%    \end{macrocode}
% \subsubsection{\csh{xintbareeval}, \csh{xintbarefloateval}, \csh{xintbareiieval}}
% \lverb|At 1.4 added one expansion step via _start macros. Triggering is expected to be
% via either \romannumeral`^^@ or \romannumeral0 is also ok|
%    \begin{macrocode}
\def\xintbareeval     {\XINT_expr_start  }%
\def\xintbarefloateval{\XINT_flexpr_start}%
\def\xintbareiieval   {\XINT_iiexpr_start}%
%    \end{macrocode}
% \subsubsection{\csh{xintthebareeval}, \csh{xintthebarefloateval},
% \csh{xintthebareiieval}}
% \lverb|For matters of \XINT_NewFunc
% |
%    \begin{macrocode}
\def\XINT_expr_unlock     {\expandafter\xint_firstofone\romannumeral`&&@}%
\def\xintthebareeval      {\romannumeral0\expandafter\xint_stop_atfirstofone\romannumeral0\xintbareeval}%
\def\xintthebareiieval    {\romannumeral0\expandafter\xint_stop_atfirstofone\romannumeral0\xintbareiieval}%
\def\xintthebarefloateval {\romannumeral0\expandafter\xint_stop_atfirstofone\romannumeral0\xintbarefloateval}%
\def\xintthebareroundedfloateval
{%
    \romannumeral0\expandafter\xintthebareroundedfloateval_a\romannumeral0\xintbarefloateval
}%
\def\xintthebareroundedfloateval_a
{%
    \expandafter\xint_stop_atfirstofone
    \expanded\XINT:NEhook:x:mapwithin\XINT:expr:mapwithin{\XINTinFloatSdigits_braced}%
}%
\def\XINTinFloatSdigits_braced#1{{\XINTinFloatS[\XINTdigits]{#1}}}%
%    \end{macrocode}
% \subsubsection{\csh{xinteval}, \csh{xintieval}, \csh{xintfloateval},
% \csh{xintiieval}}
% \lverb|Refactored at 1.4.
%
% The \expanded upfront ensures \xinteval still expands
% completely in two steps. No \romannumeral trigger here, in relation to the
% fact that \XINTexprprint is no f-expandable, only e-expandable.
%
% (and attention that \xintexpr\relax is now legal, and an empty ople can be
% produced in output also from \xintexpr [17][1]\relax for example)
% |
%    \begin{macrocode}
\def\xinteval #1%
   {\expanded\expandafter\XINTexprprint\expandafter.\romannumeral0\xintbareeval#1\relax}%
\def\xintieval #1% 
   {\expanded\expandafter\xint_gobble_i\romannumeral`&&@\xintiexpr#1\relax}%
\def\xintfloateval #1%
   {\expanded\expandafter\xint_gobble_i\romannumeral`&&@\xintfloatexpr#1\relax}%
\def\xintiieval #1%
   {\expanded\expandafter\XINTiiexprprint\expandafter.\romannumeral0\xintbareiieval#1\relax}%
%    \end{macrocode}
% \subsubsection{\csh{xintboolexpr}, \csh{XINT_boolexpr_print}, \csh{xinttheboolexpr},
% \csh{thexintboolexpr}}
% \lverb|ATTENTION! 1.3d renamed \xinteval to \xintexpro etc...
%
% Attention, the conversion to 1 or 0 is done only by the print macro.
% Perhaps I should force it also inside raw result.|
%    \begin{macrocode}
\def\xintboolexpr
{%
    \romannumeral0\expandafter\XINT_boolexpr_done\romannumeral0\xintexpro
}%
\def\XINT_boolexpr_done #1.{\XINTfstop\XINTboolexprprint.}%
\def\xinttheboolexpr
{%
    \expanded\expandafter\XINTboolexprprint\expandafter.\romannumeral0\xintbareeval
}%
\let\thexintboolexpr\xinttheboolexpr
%    \end{macrocode}
% \subsubsection{\csh{xintifboolexpr}, \csh{xintifboolfloatexpr}, \csh{xintifbooliiexpr}}
% \lverb|They do not accept comma separated expressions input.|
%    \begin{macrocode}
\def\xintifboolexpr      #1{\romannumeral0\xintiiifnotzero {\xinttheexpr #1\relax}}%
\def\xintifboolfloatexpr #1{\romannumeral0\xintiiifnotzero {\xintthefloatexpr #1\relax}}%
\def\xintifbooliiexpr    #1{\romannumeral0\xintiiifnotzero {\xinttheiiexpr #1\relax}}%
%    \end{macrocode}
% \subsubsection{\csh{xintifsgnexpr}, \csh{xintifsgnfloatexpr}, \csh{xintifsgniiexpr}}
% \changed{1.3d}
% \lverb|They do not accept comma separated expressions.|
%    \begin{macrocode}
\def\xintifsgnexpr      #1{\romannumeral0\xintiiifsgn {\xinttheexpr #1\relax}}%
\def\xintifsgnfloatexpr #1{\romannumeral0\xintiiifsgn {\xintthefloatexpr #1\relax}}%
\def\xintifsgniiexpr    #1{\romannumeral0\xintiiifsgn {\xinttheiiexpr #1\relax}}%
%    \end{macrocode}
% \subsubsection{Small bits we have to put somewhere}
% \lverb|&
% Some renaming and modifications here with release 1.2 to switch from
% using chains of \romannumeral-`0 in order to gather numbers, possibly
% hexadecimals, to using a \csname governed expansion. In this way no more
% limit at 5000 digits, and besides this is a logical move because the
% \xintexpr parser is already based on \csname...\endcsname storage of numbers
% as one token.
%
% The limitation at 5000 digits didn't worry me too much because it was not
% very realistic to launch computations with thousands of digits... such
% computations are still slow with 1.2 but less so now. Chains or
% \romannumeral are still used for the gathering of function names and other
% stuff which I have half-forgotten because the parser does many things.
%
% In the earlier versions we used the lockscan macro after a chain of
% \romannumeral-`0 had ended gathering digits; this uses has been replaced by
% direct processing inside a \csname...\endcsname and the macro is kept only
% for matters of dummy variables.
%
% Currently, the parsing of hexadecimal numbers needs two nested
% \csname...\endcsname, first to gather the letters (possibly with a hexadecimal
% fractional part), and in a second stage to apply \xintHexToDec to do the
% actual conversion. This should be faster than updating on the fly the number
% (which would be hard for the fraction part...).
% |
%    \begin{macrocode}
\def\XINT_embrace#1{{#1}}%
\def\xint_gob_til_! #1!{}% ! with catcode 11
\def\xintError:noopening
{%
    \XINT_expandableerror{Extra ). This is serious and prospects are bleak.}%
}%
%    \end{macrocode}
% \paragraph{\csh{xintthecoords}}
% \lverb|1.1 Wraps up an even number of comma separated items into pairs of
% TikZ coordinates; for use in the following way:
%
% coordinates {\xintthecoords\xintfloatexpr ... \relax}
%
% The crazyness with the \csname and unlock is due to TikZ somewhat STRANGE
% control of the TOTAL number of expansions which should not exceed the very low
% value of 100 !! As we implemented \XINT_thecoords_b in an "inline" style for
% efficiency, we need to hide its expansions.
%
% Not to be used as \xintthecoords\xintthefloatexpr, only as
% \xintthecoords\xintfloatexpr (or \xintiexpr etc...). Perhaps \xintthecoords
% could make an extra check, but one should not accustom users to too loose
% requirements!|
%    \begin{macrocode}
\def\xintthecoords#1%
   {\romannumeral`&&@\expandafter\XINT_thecoords_a\romannumeral0#1}%
\def\XINT_thecoords_a #1#2.#3% #2.=\XINTfloatprint<digits>. etc...
   {\expanded{\expandafter\XINT_thecoords_b\expanded#2.{#3},!,!,^}}%
\def\XINT_thecoords_b #1#2,#3#4,%
   {\xint_gob_til_! #3\XINT_thecoords_c ! (#1#2, #3#4)\XINT_thecoords_b }%
\def\XINT_thecoords_c #1^{}%
%    \end{macrocode}
% \paragraph{\csh{xintthespaceseparated}}
% \lverb|1.4a This is a utility macro which was distributed previously
% separately for usage with PSTricks \listplot|
%    \begin{macrocode}
\def\xintthespaceseparated#1%
   {\expanded\expandafter\xintthespaceseparated_a\romannumeral0#1}%
\def\xintthespaceseparated_a #1#2.#3%
   {{\expandafter\xintthespaceseparated_b\expanded#2.{#3},!,!,!,!,!,!,!,!,!,^}}%
\def\xintthespaceseparated_b #1,#2,#3,#4,#5,#6,#7,#8,#9,%
   {\xint_gob_til_! #9\xintthespaceseparated_c !%
    #1#2#3#4#5#6#7#8#9%
    \xintthespaceseparated_b}%
%    \end{macrocode}
% \lverb|1.4c I add a space here to stop the \romannumeral`$&$&@ in case of
% empty input.
% But this space induces an extra un-needed space token after 9, 18, 27,...
% items before the last group of less than 9 items.
%
% Fix (at 1.4h) is simple because I already use \expanded anyhow: I don't need
% at all the \romannumeral`$&$&@ which was first in \xintthespaceseparated,
% let's move the first \expanded which was in \xintthespaceseparated_a to
% \xintthespaceseparated, and remove the extra space here in _c.
%
% (alternative would have been to put the space after #1 and accept a
% systematic trailing space, at least it is more aesthetic).
%
% Again, I did have a test file, but it was not incorporated in my test suite,
% so I discovered the problem accidentally by compiling all files in an archive.|
%    \begin{macrocode}
\def\xintthespaceseparated_c !#1!#2^{#1}%
%    \end{macrocode}
% \subsection{Hooks into the numeric parser for usage by the
% \cshnolabel{xintdeffunc} symbolic parser}
% \lverb|This is new with 1.3 and considerably refactored at 1.4. See
% «Mysterious stuff».
% |
%    \begin{macrocode}
\let\XINT:NEhook:f:one:from:one\expandafter
\let\XINT:NEhook:f:one:from:one:direct\empty
\let\XINT:NEhook:f:one:from:two\expandafter
\let\XINT:NEhook:f:one:from:two:direct\empty
\let\XINT:NEhook:x:one:from:two\empty
\let\XINT:NEhook:f:one:and:opt:direct      \empty
\let\XINT:NEhook:f:tacitzeroifone:direct   \empty
\let\XINT:NEhook:f:iitacitzeroifone:direct \empty
\let\XINT:NEhook:x:select:obey\empty
\let\XINT:NEhook:x:listsel\empty
\let\XINT:NEhook:f:reverse\empty
%    \end{macrocode}
% \lverb|At 1.4 it was \def\XINT:NEhook:f:from:delim:u #1#2^{#1#2^}
% which was trick to allow automatic unpacking of a nutple argument
% to multi-arguments functions such as gcd() or max(). But this
% sacrificied the usage with a single numeric argument.|
% \changed{1.4i}
% \lverb|More sophisticated code to check if the argument ople
% was actually a single number. Notice that this forces numeric
% types to actually use catcode 12 tokens, and polexpr diverges
% a bit using P, but actually always testing with \if not \ifx.
%
% This is used by gcd(), lcm(), max(), min(), `+`(), `*`(), all(), any(),
% xor().
%
% The nil and None will give the same result due to the initial brace
% stripping done by \XINT:NEhook:f:from:delim:u (there was even a prior brace
% stripping to provide the #2 which is empty here for the nil and {} for the
% None).  |
%    \begin{macrocode}
\def\XINT:NEhook:f:from:delim:u #1#2^%
{%
    \expandafter\XINT_fooof_checkifnumber\expandafter#1\string#2^%
}%
\def\XINT_fooof_checkifnumber#1#2%
{%
    \expandafter#1%
    \romannumeral0\expanded{\if ^#2^\else
              \if\bgroup#2\noexpand\XINT_fooof_no\else
              \noexpand\XINT_fooof_yes#2\fi\fi}%
}%
\def\XINT_fooof_yes#1^{{#1}^}%
\def\XINT_fooof_no{\expandafter{\iffalse}\fi}%
%    \end{macrocode}
% \changed{1.4i}
% \lverb|Same changes as for the other multiple arguments functions,
% making them again usable with a single numeric input.
%
% Was at 1.4 \def\XINT:NEhook:f:noeval:from:braced:u#1#2^{#1{#2}} which
% is not compatible with a single numeric input.
%
% Used by len(), first(), last() but it is a potential implementation bug that
% the three share this as the location where expansion takes places is one
% level deeper for the support macro of len(). 
%
% The None is here handled as nil, i.e. it is unpacked, which is fine
% as the documentations says nutples are unpacked.
%
%|
%    \begin{macrocode}
\def\XINT:NEhook:f:LFL #1{\expandafter#1\expandafter}%
\def\XINT:NEhook:r:check #1^%
{%
    \expandafter\XINT:NEhook:r:check_a\string#1^%
}%
\let\XINT:NEsaved:r:check \XINT:NEhook:r:check
\def\XINT:NEhook:r:check_a #1%
{%
    \if ^#1\xint_dothis\xint_c_\fi
    \if\bgroup#1\xint_dothis\XINT:NEhook:r:check_no\fi
    \xint_orthat{\XINT:NEhook:r:check_yes#1}%
}%
\def\XINT:NEhook:r:check_no
{%
    \expandafter\XINT:NEhook:r:check_no_b
    \expandafter\xint_c_\expandafter{\iffalse}\fi
}%
\def\XINT:NEhook:r:check_no_b#1^{#1}%
\def\XINT:NEhook:r:check_yes#1^{\xint_c_{#1}}%
\let\XINT:NEhook:branch\expandafter
\let\XINT:NEhook:seqx\empty
\let\XINT:NEhook:iter\expandafter
\let\XINT:NEhook:opx\empty
\let\XINT:NEhook:rseq\expandafter
\let\XINT:NEhook:iterr\expandafter
\let\XINT:NEhook:rrseq\expandafter
\let\XINT:NEhook:x:toblist\empty
\let\XINT:NEhook:x:mapwithin\empty
\let\XINT:NEhook:x:ndmapx\empty
%    \end{macrocode}
% \subsection{\csh{XINT_expr_getnext}: fetch some value then an operator and
% present them to last waiter with the found operator precedence, then
% the operator, then the value}
% \lverb|Big change in 1.1, no attempt to detect braced stuff anymore as the
% [N] notation is implemented otherwise. Now, braces should not be used at
% all; one level removed, then \romannumeral-`0 expansion.
%
% Refactored at 1.4 to put expansion of \XINT_expr_getop after the fetched
% number, thus avoiding it to have to fetch it (which could happen then
% multiple times, it was not really important when it was only one token in
% pre-1.4 xintexpr).
%
% Allow \xintexpr\relax at 1.4.
%
% Refactored at 1.4 the articulation
% \XINT_expr_getnext/XINT_expr_func/XINT_expr_getop. For some legacy reason
% the first token picked by getnext was soon turned to catcode 12 The next
% ones after the first were not a priori stringified but the first token was,
% and this made allowing things such as \xintexpr\relax, \xintexpr,,\relax,
% [], 1+(), [:] etc... complicated and requiring each time specific measures.
%
% The \expandafter chain in \XINT_expr_put_op_first is an overhead related
% to an 1.4 attempt, the "varvalue" mechanism. I.e.: expansion of
% \XINT_expr_var_foo is {\XINT_expr_varvalue_foo } and then for example
% \XINT_expr_varvalue_foo expands to {4/1[0]}.  The mechanism was originally
% conceived to have only one token with idea its makes things faster. But the
% xintfrac macros break with syntax such as \xintMul\foo\bar and \foo
% expansion giving braces. So at 1.4c I added here these \expandafter, but
% this is REALLY not satisfactory because the \expandafter are needed it seems
% only for this variable "varvalue" mechanism.
%
% See also the discussion of \XINT_expr_op__ which distinguishes variables from
% functions.
%
% After a 1.4g refactoring it would be possible to drop here the
% \expandafter if the \XINT_expr_var_foo macro was defined to f-expand to
% {actual expanded value (as ople)} for example explicit {{3}}. I have to
% balance the relative weights of doing always the \expandafter but they are
% needed only for the case the value was encapsulated in a variable, and of
% never doing the \expandafter and ensure f-expansion of the _var_foo gives
% explicit value (now that the refactoring let it be f-expanded, and the case
% of fake variables omit and abort in particular was safely separated instead
% of being treated like other and imposing restrictions on general variable
% handling), and then there is the overhead of possibly moving around many
% digits in the #1 of \XINT_expr_put_op_first.
%
%  |
%    \begin{macrocode}
\def\XINT_expr_getnext #1%
{%
    \expandafter\XINT_expr_put_op_first\romannumeral`&&@%
    \expandafter\XINT_expr_getnext_a\romannumeral`&&@#1%
}%
\def\XINT_expr_put_op_first #1#2#3{\expandafter#2\expandafter#3\expandafter{#1}}%
\def\XINT_expr_getnext_a #1%
{%
    \ifx\relax #1\xint_dothis\XINT_expr_foundprematureend\fi
    \ifx\XINTfstop#1\xint_dothis\XINT_expr_subexpr\fi
    \ifcat\relax#1\xint_dothis\XINT_expr_countetc\fi
    \xint_orthat{}\XINT_expr_getnextfork #1%
}%
\def\XINT_expr_foundprematureend\XINT_expr_getnextfork #1{{}\xint_c_\relax}%
\def\XINT_expr_subexpr #1.#2%
{%
    \expanded{\unexpanded{{#2}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
%    \end{macrocode}
% \lverb|1.2 adds \ht, \dp, \wd and the eTeX font things. 1.4 avoids big
% nested \if's, simply for code readability.
%
% This "fetch as number" is dangerous as long as list is not complete... at
% 1.4g I belatedly add \catcode|
%    \begin{macrocode}
\def\XINT_expr_countetc\XINT_expr_getnextfork#1%
{%
    \if0\ifx\count#11\fi
        \ifx\numexpr#11\fi
        \ifx\catcode#11\fi
        \ifx\dimen#11\fi
        \ifx\dimexpr#11\fi
        \ifx\skip#11\fi
        \ifx\glueexpr#11\fi
        \ifx\fontdimen#11\fi
        \ifx\ht#11\fi
        \ifx\dp#11\fi
        \ifx\wd#11\fi
        \ifx\fontcharht#11\fi
        \ifx\fontcharwd#11\fi
        \ifx\fontchardp#11\fi
        \ifx\fontcharic#11\fi
       0\expandafter\XINT_expr_fetch_as_number\fi
   \expandafter\XINT_expr_getnext_a\number #1%
}%
\def\XINT_expr_fetch_as_number
    \expandafter\XINT_expr_getnext_a\number #1%
{%
    \expanded{{{\number#1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
%    \end{macrocode}
% \lverb|This is the key initial dispatch component. It has been refactored at
% 1.4g to give priority to identifying letter and digit tokens first.  It
% thus combines former \XINT_expr_getnextfork, \XINT_expr_scan_nbr_or_func and
% \XINT_expr_scanfunc. A branch of the latter having become
% \XINT_expr_startfunc. The handling of non-catcode 11 underscore _ has
% changed: it is now skipped completely like the +. Formerly it would cause an
% infinite loop because it triggered first insertion of a nil variable, (being
% confused with a possible operator at a location where one looks for a
% value), then tacit multiplication (being now interpreted as starting some
% name), and then it came back to getnextfork creating loop. The @ of catcode
% 12 could have caused the same issue if it was not handled especially because
% it is used in the syntax as special variable for recursion hence was
% recognized even if of catcode 12. Anyway I could have handled the _ like the
% @, to avoid this problem of infinite loop with a non-letter underscore used
% as first character but decided finally to have it be ignored (it is already
% ignored if among digits, but it can be a constituent of a function of
% variable name). It is not ignored of course if of catcode 11. It may then
% start a variable or function name, but only for use by the package (by
% polexpr for example), not by users.
%
% Then the matter is handed over to specialized routines: gathering digits
% of a number (inclusive of a decimal mark, an exponential part) or letters
% of a function or variable.  And we have to intercept some tokens to implement
% various functionalities.
%
% In each dothis/orthat structure, the first encountered branches are usually
% handled slower than the next, because \if..\fi test cost less than grabbing
% tokens. The exception is in the first one where letters pass through
% slightly faster than digits, presumably because the \ifnum test is more
% costly.  Prior to this 1.4g refactoring the case of a starting letter of
% a variable or function name was handled last, it is now handled first.  Now,
% this is only first letter...
%
% Here are the various possibilities in order that they are appear below
% (the indicative order of speed of treatment is given as a number).
%
%( -1 tokens of catcode letter start a variable or function name
%: -2 digits (I apply \string for the test, but I will have to review,&
%    it seems natural anyhow to require digits to be of catcode 12 and&
%    this is in fact basically done by the package, \numexpr does not&
%    work if not the case.),
%: -7 support for Python-like * "unpacking" unary operator (added at 1.4),
%: -6 support for [ as opener for the [..] nutple constructor (1.4),
%: -5 support for the minus as unary operator of variable precedence,
%: -4 support for @ as first character of special variables even if not letter,
%: -3 support for opening parentheses (possibly triggering tacit multiplication),
%: -13 support for skipping over ignored + character,
%: -12 support for numbers starting with a decimal point,
%: -11 support for the `+`() and `*`() functions,
%: -10 support for the !() function,
%: -9 support for the ?() function,
%: -8 support for " for input of hexadecimal numbers. But $xintbinhexnameimp&
%    must be loaded explicitly by user.
%: -17 support for \xintdeffunc via special handling of # token,
%: -16 support for ignoring _ if not of catcode 11 and at start of&
%    numbers or names (this 1.4g change fixes \xinteval{_4} creating&
%    infinite loop)
%: -15 support for inserting "nil" in front of operators, as needed in particular&
%    for the Python slicing syntax. This covers the comma, the :, the ] and&
%    the ) and also the ; although I don't think using ; to delimit nil is licit.
%: -14 support for inserting 0 as missing value if / or ^ are encountered&
%    directly.  This 1.4g changes avoids \xinteval{/3} causing&
%    unrecoverable low level errors from \xintDiv receiving only one argument.
%    I did not see here other bad syntax to protect.
%)
%
% The handling of "nil" insertion penalizes Python slicing but anyway time
% differences in the 14-15-16-17 group are less than
% 5$%. The alternative will be
% to do some positive test for the targets (:, ], the comma and closing
% parenthesis) and do this in the prior group but this then penalizes
% others. Anyway. This is all negligible compared to actual computations...
%
% Note: the above may not be in sync with code as it is extremely
% time-consuming to maintain correspondence in case of re-factoring.  |
%    \begin{macrocode}
\def\XINT_expr_getnextfork #1%
{%
    \ifcat a#1\xint_dothis\XINT_expr_startfunc\fi
    \ifnum \xint_c_ix<1\string#1 \xint_dothis\XINT_expr_startint\fi
    \xint_orthat\XINT_expr_getnextfork_a #1%
}%
\def\XINT_expr_getnextfork_a #1%
{%
    \if#1*\xint_dothis {{}\xint_c_ii^v 0}\fi
    \if#1[\xint_dothis {{}\xint_c_ii^v \XINT_expr_itself_obracket}\fi
    \if#1-\xint_dothis {{}{}-}\fi
    \if#1@\xint_dothis{\XINT_expr_startfunc @}\fi
    \if#1(\xint_dothis {{}\xint_c_ii^v (}\fi
    \xint_orthat{\XINT_expr_getnextfork_b#1}%
}%
\catcode96 11 % `
\def\XINT_expr_getnextfork_b #1%
{%
    \if#1+\xint_dothis \XINT_expr_getnext_a\fi
    \if#1.\xint_dothis \XINT_expr_startdec\fi
    \if#1`\xint_dothis {\XINT_expr_onliteral_`}\fi
    \if#1!\xint_dothis {\XINT_expr_startfunc !}\fi
    \if#1?\xint_dothis {\XINT_expr_startfunc ?}\fi
    \if#1"\xint_dothis \XINT_expr_starthex\fi
    \xint_orthat{\XINT_expr_getnextfork_c#1}%
}%
\def\XINT_tmpa #1{%
\def\XINT_expr_getnextfork_c ##1%
{%
    \if##1#1\xint_dothis \XINT_expr_getmacropar\fi
    \if##1_\xint_dothis \XINT_expr_getnext_a\fi
    \if0\if##1/1\fi\if##1^1\fi0\xint_dothis{\XINT_expr_insertnil##1}\fi
    \xint_orthat{\XINT_expr_missing_arg##1}%
}%
}\expandafter\XINT_tmpa\string#%
%    \end{macrocode}
% \lverb@
% The ` syntax is here used for special constructs like `+`(..), `*`(..) where
% + or * will be treated as functions. Current implementation picks only one
% token (could have been braced stuff), here it will be + or *, and via
% \XINT_expr_op_` this then becomes a suitable
% \XINT_{expr|iiexpr|flexpr}_func_+ (or *). Documentation says to use
% `+`(...), but `+(...) is also valid. The opening parenthesis must be there,
% it is not allowed to require some expansion.
% @
%    \begin{macrocode}
\def\XINT_expr_onliteral_` #1#2({{#1}\xint_c_ii^v `}%
\catcode96 12 % `
%    \end{macrocode}
% \lverb@ Prior to 1.4g, I was using a \lowercase technique to insert the
% catcode 12 #, but this is a bit risky when one does not ensure a priori control of all
% lccodes.  @
%    \begin{macrocode}
\def\XINT_tmpa #1{%
\def\XINT_expr_getmacropar ##1%
{%
    \expandafter{\expandafter{\expandafter#1\expandafter
    ##1\expandafter}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
}\expandafter\XINT_tmpa\string#%
\def\XINT_expr_insertnil #1%
{%
    \expandafter{\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1%
}%
\def\XINT_expr_missing_arg#1%
{%
    \expanded{\XINT_expandableerror{Expected a value, got nothing before `#1'. Inserting 0.}{{0}}\expandafter}%
    \romannumeral`&&@\XINT_expr_getop_a#1%
}%
%    \end{macrocode}
% \subsection{\csh{XINT_expr_startint}}
% \localtableofcontents
% \lverb@1.2 release has replaced chains of \romannumeral-`0 by \csname
% governed expansion. Thus there is no more the limit at about 5000 digits for
% parsed numbers.
%
% In order to avoid having to lock and unlock in succession to handle the
% scientific part and adjust the exponent according to the number of digits of
% the decimal part, the parsing of this decimal part counts on the fly the
% number of digits it encounters.
%
% There is some slight annoyance with \xintiiexpr which should never be given
% a [n] inside its \csname.=<digits>\endcsname storage of numbers (because its
% arithmetic uses the ii macros which know nothing about the [N] notation).
% Hence if the parser has only seen digits when hitting something else than
% the dot or e (or E), it will not insert a [0]. Thus we very slightly
% compromise the efficiency of \xintexpr and \xintfloatexpr in order to be
% able to share the same code with \xintiiexpr.
%
% Indeed, the parser at this location is completely common to all, it does not
% know if it is working inside \xintexpr or \xintiiexpr. On the other hand if
% a dot or a e (or E) is met, then the (common) parser has no scrupules ending
% this number with a [n], this will provoke an error later if that was within
% an \xintiiexpr, as soon as an arithmetic macro is used.
%
% As the gathered numbers have no spaces, no pluses, no minuses, the only
% remaining issue is with leading zeroes, which are discarded on the fly. The
% hexadecimal numbers leading zeroes are stripped in a second stage by the
% \xintHexToDec macro.
%
% With 1.2, \xinttheexpr . \relax does not work anymore (it did in earlier
% releases). There must be digits either before or after the decimal mark. Thus
% both \xinttheexpr 1.\relax and \xinttheexpr .1\relax are legal.
%
% Attention at this location #1 was of catcode 12 in all versions prior to
% 1.4.
%
% We assume anyhow that catcodes of digits are 12... 
% @
%    \begin{macrocode}
\def\XINT_expr_startint #1%
{%
    \if #10\expandafter\XINT_expr_gobz_a\else\expandafter\XINT_expr_scanint_a\fi #1%
}%
\def\XINT_expr_scanint_a #1#2%
    {\expanded\bgroup{{\iffalse}}\fi #1% spare a \string
     \expandafter\XINT_expr_scanint_main\romannumeral`&&@#2}%
\def\XINT_expr_gobz_a #1#2%
    {\expanded\bgroup{{\iffalse}}\fi
     \expandafter\XINT_expr_gobz_scanint_main\romannumeral`&&@#2}%
\def\XINT_expr_startdec #1%
    {\expanded\bgroup{{\iffalse}}\fi
     \expandafter\XINT_expr_scandec_a\romannumeral`&&@#1}%
%    \end{macrocode}
% \subsubsection{Integral part (skipping zeroes)}
% \lverb|1.2 has modified the code to give highest priority to digits, the
% accelerating impact is non-negligeable. I don't think the doubled \string is
% a serious penalty.
%
% (reference to \string is obsolete: it is only used in the test but the
% tokens are not submitted to \string anymore)|
%    \begin{macrocode}
\def\XINT_expr_scanint_main #1%
{%
    \ifcat \relax #1\expandafter\XINT_expr_scanint_hit_cs \fi
    \ifnum\xint_c_ix<1\string#1 \else\expandafter\XINT_expr_scanint_next\fi
    #1\XINT_expr_scanint_again
}%
\def\XINT_expr_scanint_again #1%
{%
    \expandafter\XINT_expr_scanint_main\romannumeral`&&@#1%
}%
%    \end{macrocode}
% |1.4f| had |_getop| here, but let's jump directly to |_getop_a|.
%    \begin{macrocode}
\def\XINT_expr_scanint_hit_cs \ifnum#1\fi#2\XINT_expr_scanint_again
{%
    \iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#2%
}%
%    \end{macrocode}
% \lverb|With 1.2d the tacit multiplication in front of a variable name or
% function name is now done with a higher precedence, intermediate between the
% common one of * and / and the one of ^. Thus x/2y is like x/(2y), but x^2y
% is like x^2*y and 2y! is not (2y)! but 2*y!.
%
% Finally, 1.2d has moved away from the _scan macros all the business of the
% tacit multiplication in one unique place via \XINT_expr_getop. For this, the
% ending token is not first given to \string as was done earlier before
% handing over back control to \XINT_expr_getop. Earlier we had to identify
% the catcode 11 ! signaling a sub-expression here. With no \string applied
% we can do it in \XINT_expr_getop. As a corollary of this displacement,
% parsing of big numbers should be a tiny bit faster now.
%
% Extended for 1.2l to ignore underscore character _ if encountered within
% digits; so it can serve as separator for better readability.|
%
% \lverb|It is not obvious at 1.4 to support [] for three things: packing,
% slicing, ... and raw xintfrac syntax A/B[N]. The only good way would be to
% actually really separate completely \xintexpr, \xintfloatexpr and
% \xintiiexpr code which would allow to handle both / and [] from A/B[N] as we
% handle e and E. But triplicating the code is something I need to think
% about. It is not possible as in pre 1.4 to consider [ only as an operator of
% same precedence as multiplication and division which was the way we did
% this, but we can use the technique of fake operators. Thus we intercept
% hitting a [ here, which is not too much of a problem as anyhow we dropped
% temporarily 3*[1,2,3]+5 syntax so we don't have to worry that 3[1,2,3]
% should do tacit multiplication. I think only way in future will be to really
% separate the code of the three parsers (or drop entirely support for A/B[N];
% as 1.4 has modified output of \xinteval to not use this notation this is not
% too dramatic).
%
% Anyway we find a way to inject here the former handling of [N], which will
% use a delimited macro to directly fetch until the closing]. We do still need
% some fake operator because A/B[N] is (A/B) times 10^N and the /B is allowed
% to be missing. We hack this using the $ which is not used currently as
% operator elsewhere in the syntax and need to hook into \XINT_expr_getop_b.
% No finally I use the null char. It must be of catcode 12.
%
% 1.4f had _getop here, but let's jump directly to _getop_a.|
%    \begin{macrocode}
\def\XINT_expr_scanint_next #1\XINT_expr_scanint_again
{%
    \if    [#1\xint_dothis\XINT_expr_rawxintfrac\fi
    \if    _#1\xint_dothis\XINT_expr_scanint_again\fi
    \if    e#1\xint_dothis{[\the\numexpr0\XINT_expr_scanexp_a +}\fi
    \if    E#1\xint_dothis{[\the\numexpr0\XINT_expr_scanexp_a +}\fi
    \if    .#1\xint_dothis{\XINT_expr_startdec_a .}\fi
    \xint_orthat
    {\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1}%
}%
\def\XINT_expr_rawxintfrac
{%
    \iffalse{{{\fi}}\expandafter}\csname XINT_expr_precedence_&&@\endcsname&&@%
}%
\def\XINT_expr_gobz_scanint_main #1%
{%
    \ifcat \relax #1\expandafter\XINT_expr_gobz_scanint_hit_cs\fi
    \ifnum\xint_c_x<1\string#1 \else\expandafter\XINT_expr_gobz_scanint_next\fi
    #1\XINT_expr_scanint_again
}%
\def\XINT_expr_gobz_scanint_again #1%
{%
    \expandafter\XINT_expr_gobz_scanint_main\romannumeral`&&@#1%
}%
%    \end{macrocode}
% |1.4f| had |_getop| here, but let's jump directly to |_getop_a|.
%    \begin{macrocode}
\def\XINT_expr_gobz_scanint_hit_cs\ifnum#1\fi#2\XINT_expr_scanint_again
{%
    0\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#2%
}%
\def\XINT_expr_gobz_scanint_next #1\XINT_expr_scanint_again
{%
    \if    [#1\xint_dothis{\expandafter0\XINT_expr_rawxintfrac}\fi
    \if    _#1\xint_dothis\XINT_expr_gobz_scanint_again\fi
    \if    e#1\xint_dothis{0[\the\numexpr0\XINT_expr_scanexp_a +}\fi
    \if    E#1\xint_dothis{0[\the\numexpr0\XINT_expr_scanexp_a +}\fi
    \if    .#1\xint_dothis{\XINT_expr_gobz_startdec_a .}\fi
    \if    0#1\xint_dothis\XINT_expr_gobz_scanint_again\fi
    \xint_orthat
    {0\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1}%
}%
%    \end{macrocode}
% \subsubsection{Fractional part}
% \lverb|Annoying duplication of code to allow 0. as input.
%
% 1.2a corrects a very bad bug in 1.2 \XINT_expr_gobz_scandec_b which should
% have stripped leading zeroes in the fractional part but didn't; as a result
% \xinttheexpr 0.01\relax returned 0 =:-((( Thanks to Kroum Tzanev who
% reported the issue. Does it improve things if I say the bug was introduced
% in 1.2, it wasn't present before ?
%
% 1.4f had _getop here, but let's jump directly to _getop_a.|
%    \begin{macrocode}
\def\XINT_expr_startdec_a .#1%
{%
    \expandafter\XINT_expr_scandec_a\romannumeral`&&@#1%
}%
\def\XINT_expr_scandec_a #1%
{%
    \if .#1\xint_dothis{\iffalse{{{\fi}}\expandafter}%
                        \romannumeral`&&@\XINT_expr_getop_a..}\fi
    \xint_orthat {\XINT_expr_scandec_main 0.#1}%
}%
\def\XINT_expr_gobz_startdec_a .#1%
{%
    \expandafter\XINT_expr_gobz_scandec_a\romannumeral`&&@#1%
}%
\def\XINT_expr_gobz_scandec_a #1%
{%
    \if .#1\xint_dothis
    {0\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a..}\fi
    \xint_orthat {\XINT_expr_gobz_scandec_main 0.#1}%
}%
\def\XINT_expr_scandec_main #1.#2%
{%
    \ifcat \relax #2\expandafter\XINT_expr_scandec_hit_cs\fi
    \ifnum\xint_c_ix<1\string#2 \else\expandafter\XINT_expr_scandec_next\fi
    #2\expandafter\XINT_expr_scandec_again\the\numexpr #1-\xint_c_i.%
}%
\def\XINT_expr_scandec_again #1.#2%
{%
    \expandafter\XINT_expr_scandec_main
    \the\numexpr #1\expandafter.\romannumeral`&&@#2%
}%
%    \end{macrocode}
%|1.4f| had |_getop| here, but let's jump directly to |_getop_a|.
%    \begin{macrocode}
\def\XINT_expr_scandec_hit_cs\ifnum#1\fi
    #2\expandafter\XINT_expr_scandec_again\the\numexpr#3-\xint_c_i.%
{%
    [#3]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#2%
}%
\def\XINT_expr_scandec_next #1#2\the\numexpr#3-\xint_c_i.%
{%
    \if    _#1\xint_dothis{\XINT_expr_scandec_again#3.}\fi
    \if    e#1\xint_dothis{[\the\numexpr#3\XINT_expr_scanexp_a +}\fi
    \if    E#1\xint_dothis{[\the\numexpr#3\XINT_expr_scanexp_a +}\fi
    \xint_orthat 
    {[#3]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1}%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_expr_gobz_scandec_main #1.#2%
{%
    \ifcat \relax #2\expandafter\XINT_expr_gobz_scandec_hit_cs\fi
    \ifnum\xint_c_ix<1\string#2 \else\expandafter\XINT_expr_gobz_scandec_next\fi
    \if0#2\expandafter\xint_firstoftwo\else\expandafter\xint_secondoftwo\fi
    {\expandafter\XINT_expr_gobz_scandec_main}%
    {#2\expandafter\XINT_expr_scandec_again}\the\numexpr#1-\xint_c_i.%
}%
%    \end{macrocode}
% |1.4f| had |_getop| here, but let's jump directly to |_getop_a|.
%    \begin{macrocode}
\def\XINT_expr_gobz_scandec_hit_cs \ifnum#1\fi\if0#2#3\xint_c_i.%
{%
    0[0]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#2%
}%
\def\XINT_expr_gobz_scandec_next\if0#1#2\fi #3\numexpr#4-\xint_c_i.%
{%
    \if    _#1\xint_dothis{\XINT_expr_gobz_scandec_main #4.}\fi
    \if    e#1\xint_dothis{0[\the\numexpr0\XINT_expr_scanexp_a +}\fi
    \if    E#1\xint_dothis{0[\the\numexpr0\XINT_expr_scanexp_a +}\fi
    \xint_orthat
    {0[0]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1}%
}%
%    \end{macrocode}
% \subsubsection{Scientific notation}
% \lverb|Some pluses and minuses are allowed at the start of the scientific
% part, however not later, and no parenthesis.
%
% ATTENTION! 1e\numexpr2+3\relax or 1e\xintiexpr i\relax, i=1..5
% are not allowed and 1e1\numexpr2\relax does 1e1 * \numexpr2\relax.
% Use \the\numexpr, \xinttheiexpr, etc...
% |
%    \begin{macrocode}
\def\XINT_expr_scanexp_a #1#2%
{%
    #1\expandafter\XINT_expr_scanexp_main\romannumeral`&&@#2%
}%
\def\XINT_expr_scanexp_main #1%
{%
    \ifcat \relax #1\expandafter\XINT_expr_scanexp_hit_cs\fi
    \ifnum\xint_c_ix<1\string#1 \else\expandafter\XINT_expr_scanexp_next\fi
    #1\XINT_expr_scanexp_again
}%
\def\XINT_expr_scanexp_again #1%
{%
    \expandafter\XINT_expr_scanexp_main_b\romannumeral`&&@#1%
}%
%    \end{macrocode}
% |1.4f| had |_getop| here, but let's jump directly to |_getop_a|.
%    \begin{macrocode}
\def\XINT_expr_scanexp_hit_cs\ifnum#1\fi#2\XINT_expr_scanexp_again
{%
    ]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#2%
}%
\def\XINT_expr_scanexp_next #1\XINT_expr_scanexp_again
{%
    \if    _#1\xint_dothis  \XINT_expr_scanexp_again   \fi
    \if    +#1\xint_dothis {\XINT_expr_scanexp_a +}\fi
    \if    -#1\xint_dothis {\XINT_expr_scanexp_a -}\fi
    \xint_orthat
    {]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1}%
}%
\def\XINT_expr_scanexp_main_b #1%
{%
    \ifcat \relax #1\expandafter\XINT_expr_scanexp_hit_cs_b\fi
    \ifnum\xint_c_ix<1\string#1 \else\expandafter\XINT_expr_scanexp_next_b\fi
    #1\XINT_expr_scanexp_again_b
}%
%    \end{macrocode}
% |1.4f| had |_getop| here, but let's jump directly to |_getop_a|.
%    \begin{macrocode}
\def\XINT_expr_scanexp_hit_cs_b\ifnum#1\fi#2\XINT_expr_scanexp_again_b
{%
    ]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#2%
}%
\def\XINT_expr_scanexp_again_b #1%
{%
    \expandafter\XINT_expr_scanexp_main_b\romannumeral`&&@#1%
}%
\def\XINT_expr_scanexp_next_b #1\XINT_expr_scanexp_again_b
{%
    \if _#1\xint_dothis\XINT_expr_scanexp_again\fi
    \xint_orthat
    {]\iffalse{{{\fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop_a#1}%
}%
%    \end{macrocode}
% \subsubsection{Hexadecimal numbers}
% \lverb|1.2d has moved most of the handling of tacit multiplication to
% \XINT_expr_getop, but we have to do some of it here, because we apply
% \string before calling \XINT_expr_scanhexI_aa. I do not insert the *
% in \XINT_expr_scanhexI_a, because it is its higher precedence variant which
% will is expected, to do the same as when a non-hexadecimal number prefixes a
% sub-expression. Tacit multiplication in front of variable or function names
% will not work (because of this \string).
%
% Extended for 1.2l to ignore underscore character _ if encountered within
% digits.
%
% (some above remarks have been obsoleted for some long time, no more applied
% \string since 1.4)
%
% Notice that internal representation adds a [N] part only in case input
% used "DDD.dddd form, for compatibility with \xintiiexpr which is not
% compatible with such internal representation.
%
% At 1.4g a very long-standing bug was fixed: input such as "\foo broke the
% parser because (incredibly) the \foo token was picked up unexpanded and
% ended up as is in an \ifcat !
%
% Another long-standing bug was fixed at 1.4g: contrarily to the decimal case,
% here in the hexadecimal input leading zeros were not trimmed. This was ok,
% because formerly \xintHexToDec trimmed leading zeros, but at 1.2m 2017/07/31
% xintbinhex.sty was modified and this ceased being the case. But I forgot to
% upgrade the parser here at that time. Leading zeros would in many
% circumstances (presence of a fractional part, or \xintiiexpr context) lead
% to wrong results.  Leading zeros are now trimmed during input.
%
% |
%    \begin{macrocode}
\def\XINT_expr_hex_in #1.#2#3;%
{%
    \expanded{{{\if#2>%
      \xintHexToDec{#1}%
    \else
      \xintiiMul{\xintiiPow{625}{\xintLength{#3}}}{\xintHexToDec{#1#3}}%
      [\the\numexpr-4*\xintLength{#3}]%
    \fi}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
%    \end{macrocode}
% \lverb|Let's not forget to grab-expand next token first as is normal rule
% of operation. Formerly called \XINT_expr_scanhex_I and had " upfront.|
%    \begin{macrocode}
\def\XINT_expr_starthex #1%
{%
    \expandafter\XINT_expr_hex_in\expanded\bgroup
    \expandafter\XINT_expr_scanhexIgobz_a\romannumeral`&&@#1%
}%
\def\XINT_expr_scanhexIgobz_a #1%
{%
    \ifcat #1\relax
      0.>;\iffalse{\fi\expandafter}\expandafter\xint_gobble_i\fi
    \XINT_expr_scanhexIgobz_aa #1%
}%
\def\XINT_expr_scanhexIgobz_aa #1%
{%
    \if\ifnum`#1>`0
       \ifnum`#1>`9
       \ifnum`#1>`@
       \ifnum`#1>`F
       0\else1\fi\else0\fi\else1\fi\else0\fi 1%
       \xint_dothis\XINT_expr_scanhexI_b
    \fi
    \if 0#1\xint_dothis\XINT_expr_scanhexIgobz_bgob\fi
    \if _#1\xint_dothis\XINT_expr_scanhexIgobz_bgob\fi
    \if .#1\xint_dothis\XINT_expr_scanhexIgobz_toII\fi
    \xint_orthat
     {\XINT_expandableerror
        {Expected an hexadecimal digit but got `#1'. Using `0'.}%
      0.>;\iffalse{\fi}}%
    #1%
}%
\def\XINT_expr_scanhexIgobz_bgob #1#2%
{%
    \expandafter\XINT_expr_scanhexIgobz_a\romannumeral`&&@#2%
}%
\def\XINT_expr_scanhexIgobz_toII .#1%
{%
    0..\expandafter\XINT_expr_scanhexII_a\romannumeral`&&@#1%
}%
\def\XINT_expr_scanhexI_a #1%
{%
    \ifcat #1\relax
      .>;\iffalse{\fi\expandafter}\expandafter\xint_gobble_i\fi
    \XINT_expr_scanhexI_aa #1%
}%
\def\XINT_expr_scanhexI_aa #1%
{%
    \if\ifnum`#1>`/
       \ifnum`#1>`9
       \ifnum`#1>`@
       \ifnum`#1>`F
       0\else1\fi\else0\fi\else1\fi\else0\fi 1%
       \expandafter\XINT_expr_scanhexI_b
    \else
       \if _#1\xint_dothis{\expandafter\XINT_expr_scanhexI_bgob}\fi
       \if .#1\xint_dothis{\expandafter\XINT_expr_scanhexI_toII}\fi
       \xint_orthat {.>;\iffalse{\fi\expandafter}}%
    \fi
    #1%
}%
\def\XINT_expr_scanhexI_b #1#2%
{%
    #1\expandafter\XINT_expr_scanhexI_a\romannumeral`&&@#2%
}%
\def\XINT_expr_scanhexI_bgob #1#2%
{%
    \expandafter\XINT_expr_scanhexI_a\romannumeral`&&@#2%
}%
\def\XINT_expr_scanhexI_toII .#1%
{%
    ..\expandafter\XINT_expr_scanhexII_a\romannumeral`&&@#1%
}%
\def\XINT_expr_scanhexII_a #1%
{%
    \ifcat #1\relax\xint_dothis{;\iffalse{\fi}#1}\fi
    \xint_orthat {\XINT_expr_scanhexII_aa #1}%
}%
\def\XINT_expr_scanhexII_aa #1%
{%
    \if\ifnum`#1>`/
       \ifnum`#1>`9
       \ifnum`#1>`@
       \ifnum`#1>`F
       0\else1\fi\else0\fi\else1\fi\else0\fi 1%
       \expandafter\XINT_expr_scanhexII_b
    \else
       \if _#1\xint_dothis{\expandafter\XINT_expr_scanhexII_bgob}\fi
       \xint_orthat{;\iffalse{\fi\expandafter}}%
    \fi
    #1%
}%
\def\XINT_expr_scanhexII_b #1#2%
{%
    #1\expandafter\XINT_expr_scanhexII_a\romannumeral`&&@#2%
}%
\def\XINT_expr_scanhexII_bgob #1#2%
{%
    \expandafter\XINT_expr_scanhexII_a\romannumeral`&&@#2%
}%
%    \end{macrocode}
% \subsubsection{\csh{XINT_expr_startfunc}: collecting names of functions and
% variables}
%
% \lverb|At 1.4 the first token left over has not been submitted to
% \string. We also know it is not a control sequence. So we can test catcode
% to identify if operator is found. And it is allowed to hit some operator
% such as a closing parenthesis we will then insert the «nil» value (edited: which
% however will cause certain breakage of the infix binary operators: I notice
% I did not insert None {{}} but nil {}, perhaps by oversight).
%
% There was prior to 1.4 solely the dispatch in \XINT_expr_scanfunc_b
% but now we do it immediately and issue \XINT_expr_func only in certain
% cases.
%
% Comments here have been removed because 1.4g did a refactoring
% and renamed \XINT_expr_scanfunc to \XINT_expr_startfunc, moving half of
% it earlier inside the getnextfork macros.
% |
%    \begin{macrocode}
\def\XINT_expr_startfunc #1{\expandafter\XINT_expr_func\expanded\bgroup#1\XINT_expr_scanfunc_a}%
\def\XINT_expr_scanfunc_a #1%
{%
    \expandafter\XINT_expr_scanfunc_b\romannumeral`&&@#1%
}%
%    \end{macrocode}
% \lverb|This handles: 1) (indirectly) tacit multiplication by a variable in
% front a of sub-expression, 2) (indirectly) tacit multiplication in front of
% a \count etc..., 3) functions which are recognized via an encountered opening
% parenthesis (but later this must be disambiguated from variables with tacit
% multiplication) 4) 5) 6) 7) acceptable components of a variable or function
% names: @, underscore, digits, letters (or chars of category code letter.)
%
% The short lived 1.2d which followed the even shorter lived 1.2c managed to
% introduce a bug here as it removed the check for catcode 11 !, which must be
% recognized if ! is not to be taken as part of a variable name. Don't know
% what I was thinking, it was the time when I was moving the handling of tacit
% mutliplication entirely to the \XINT_expr_getop side. Fixed in 1.2e.
%
% I almost decided to remove the \ifcat\relax test whose rôle is to avoid the
% \string#1 to do something bad is the escape char is a digit! Perhaps I will
% remove it at some point ! I truly almost did it, but also the case of no
% escape char is a problem (\string\0, if \0 is a count ...)
%
% The (indirectly) above means that via \XINT_expr_func then \XINT_expr_op__
% one goes back to \XINT_expr_getop then \XINT_expr_getop_b which is the
% location where tacit multiplication is now centralized. This makes the
% treatment of tacit multiplication for situations such as <variable>\count or
% <variable>\xintexpr..\relax, perhaps a bit sub-optimal, but first the
% variable name must be gathered, second the variable must expand to its
% value.|
%    \begin{macrocode}
\def\XINT_expr_scanfunc_b #1%
{%
  \ifcat \relax#1\xint_dothis{\iffalse{\fi}(_#1}\fi
  \if (#1\xint_dothis{\iffalse{\fi}(`}\fi
  \if 1\ifcat a#10\fi
       \ifnum\xint_c_ix<1\string#1 0\fi
       \if @#10\fi
       \if _#10\fi
      1%
      \xint_dothis{\iffalse{\fi}(_#1}\fi
  \xint_orthat {#1\XINT_expr_scanfunc_a}%
}%
%    \end{macrocode}
% \subsubsection{\csh{XINT_expr_func}: dispatch to variable replacement or to
% function execution}
% \lverb@Comments written 2015/11/12: earlier there was an \ifcsname test for
% checking if we had a variable in front of a (, for tacit multiplication for
% example in x(y+z(x+w)) to work. But after I had implemented functions (that
% was yesterday...), I had the problem if was impossible to re-declare a
% variable name such as "f" as a function name. The problem is that here we
% can not test if the function is available because we don't know if we are in
% expr, iiexpr or floatexpr. The \xint_c_ii^v causes all fetching operations
% to stop and control is handed over to the routines which will be expr,
% iiexpr ou floatexpr specific, i.e. the \XINT_{expr|iiexpr|flexpr}_op_{`|_}
% which are invoked by the until_<op>_b macros earlier in the stream.
% Functions may exist for one but not the two other parsers. Variables are
% declared via one parser and usable in the others, but naturally \xintiiexpr
% has its restrictions.
%
% Thinking about this again I decided to treat a priori cases such as x(...)
% as functions, after having assigned to each variable a low-weight macro
% which will convert this into _getop\.=<value of x>*(...). To activate that
% macro at the right time I could for this exploit the "onliteral" intercept,
% which is parser independent (1.2c).
%
% This led to me necessarily to rewrite partially the seq, add, mul, subs,
% iter ... routines as now the variables fetch only one token. I think the
% thing is more efficient.
%
% 1.2c had \def\XINT_expr_func #1(#2{\xint_c_ii^v #2{#1}}
%
% In \XINT_expr_func the #2 is _ if #1 must be a variable name, or #2=` if #1
% must be either a function name or possibly a variable name which will then
% have to be followed by tacit multiplication before the opening parenthesis.
%
% The \xint_c_ii^v is there because _op_` must know in which parser
% it works. Dispendious for _. Hence I modify for 1.2d. @
%    \begin{macrocode}
\def\XINT_expr_func #1(#2{\if _#2\xint_dothis{\XINT_expr_op__{#1}}\fi
                          \xint_orthat{{#1}\xint_c_ii^v #2}}%
%    \end{macrocode}
% \subsection{\csh{XINT_expr_op_`}: launch function or
% pseudo-function, or evaluate variable and insert operator of multiplication
% in front of parenthesized contents}
%
% \lverb|The "onliteral" intercepts is for bool, togl, protect, ... but also
% for add, mul, seq, etc... Genuine functions have expr, iiexpr and flexpr
% versions (or only one or two of the three) and trigger here the use of the
% suitable parser-dependant form.  The former (pseudo functions and functions
% handling dummy variables) first trigger a parser independent mechanism.
%
% With 1.2c "onliteral" is also used to disambiguate a variable followed
% by an opening parenthesis from a function and then apply tacit multiplication.
% However as I use only a \ifcsname test, in order to be able to
% re-define a variable as function, I move the check for being a function
% first. Each variable name now has its onliteral_<name> associated macro.
% This used to be decided much earlier at the time of
% \XINT_expr_func.
%
% The advantage of 1.2c code is that the same name can be used for
% a variable or a function.
% |
%
% \changed[2021/06/11]{1.4i}
%
% \lverb|The 1.2c abuse of «onliteral» for both tacit multiplication in front
% of an opening parenthesis and «generic» functions or pseudo-functions meant
% that the latter were vulnerable against user redefinition of a function name
% as a variable name.  This applied to subs, subsm, subsn, seq, add, mul,
% ndseq, ndmap, ndfillraw, bool, togl, protect, qint, qfrac, qfloat, qraw,
% random, qrand, rbit and the most susceptible in real life was probably "seq".
%
% Now variables have an associated «var*» named macro, not «onliteral».
%
% In passing I refactor here in a \romannumeral inspired way how \csname and
% TeX booleans are intertwined, minimizing \expandafter usage.|
%
%    \begin{macrocode}
\def\XINT_tmpa #1#2#3{%
  \def #1##1%
  {%
    \csname
      XINT_\ifcsname XINT_#3_func_##1\endcsname
        #3_func_##1\expandafter\endcsname\romannumeral`&&@\expandafter#2%
      \romannumeral\else
      \ifcsname XINT_expr_onliteral_##1\endcsname
        expr_onliteral_##1\expandafter\endcsname\romannumeral
      \else
      \ifcsname XINT_expr_var*_##1\endcsname
        expr_var*_##1\expandafter\endcsname\romannumeral
      \else
        #3_func_\XINT_expr_unknown_function {##1}%
                \expandafter\endcsname\romannumeral`&&@\expandafter#2%
      \romannumeral
      \fi\fi\fi\xint_c_
  }%
}%
\xintFor #1 in {expr,flexpr,iiexpr} \do {%
     \expandafter\XINT_tmpa
                 \csname XINT_#1_op_`\expandafter\endcsname
                 \csname XINT_#1_oparen\endcsname
                 {#1}%
}%
\def\XINT_expr_unknown_function #1%
   {\XINT_expandableerror{`#1' is unknown, say `Isome_func' or I use 0.}}%
\def\XINT_expr_func_ #1#2#3{#1#2{{0}}}%
\let\XINT_flexpr_func_\XINT_expr_func_
\let\XINT_iiexpr_func_\XINT_expr_func_
%    \end{macrocode}
% 
% \subsection{\csh{XINT_expr_op__}: replace a variable by its value and
% then fetch next operator}
% \lverb|The 1.1 mechanism for \XINT_expr_var_<varname> has been
% modified in 1.2c. The <varname> associated macro is now only expanded
% once, not twice. We arrive here via \XINT_expr_func.
%
% At 1.4 \XINT_expr_getop is launched with accumulated result on its left.
% But the omit and abort keywords are implemented via fake variables
% which rely on possibility to modify incoming upfront tokens. If we did here
% something such as
% $centeredline$bgroup_var_#1\expandafter\endcsname\romannumeral`^^@\XINT_expr_getop$egroup
% the premature expansion of getop would break the var_omit and
% var_abort mechanism. Thus we revert
% to former code which locates an \XINT_expr_getop (call it _legacy)
% before the tokens from the variable expansion (in xintexpr < 1.4 the
% normal variables expanded
% to a single token so the overhead was not serious) so we can expand fake
% variables first.
%
% Abusing variables to manipulate the incoming token stream is a bit bad,
% usually I prefer functions for this (such as the break() function) but then
% I have to define 3 macros for the 3 parsers.
%
% This trick of fake variables puts thus a general overhead at various
% locations, and the situation here is REALLY not satisfactory. But 1.4 has
% (had) to be released now. 
%
% Even if I could put the \csname XINT_expr_var_foo\endcsname upfront, which
% would then be f-expanded, this would still need \XINT_expr_put_op_first to
% use its \expandafter's as long as \XINT_expr_var_foo expands to
% {\XINT_expr_varvalue_foo} with a not-yet expanded \XINT_expr_var_value.
%
% I could let \XINT_expr_var_foo expand to
% \expandafter{\XINT_expr_varvalue_foo} allowing then (if it gets f-expanded)
% probably to drop the \expandafter in \XINT_expr_put_op_first. But I can not
% consider this option in the form 
% $centeredline$bgroup_var_foo\expandafter\endcsname\romannumeral`^^@\XINT_expr_getop$egroup
% until the issue with fake variables such as omit and
% abort which must act before \XINT_expr_getop has some workaround.
% This could be implemented here with some extra branch, i.e. there would not
% be some \XINT_expr_var_omit but something else filtered out in the \else
% branch here.
%
% The above comments mention only omit and abort, but the case of real dummy
% variables also needs consideration.
%
% At 1.4g, I test first for existence of \XINT_expr_onliteral_foo. 
%
% Updated for 1.4i: now rather existence of \XINT_expr_var*_foo is tested.
%
% This is
% a trick which allosw to distinguish actual or dummy variables from really
% fake variables omit and abort (must check if there are others). For the real
% or dummy variables we can trigger the expansion of the \XINT_expr_getop
% before the one of the variable.  I could test vor varvalue_foo but this
% applies only to real variables not dummy variables.  Actual and dummy
% variables are thus handled slightly faster at 1.4g as there is less
% induced moving around (the \expandafter chain in \XINT_expr_put_op_first
% still applies at this stage, as I have not yet re-examined the var/varlue
% mechanism). And the test for var_foo is moved directly inside the \csname
% construct in the \else branch which now handles together
% fake variables and non-existing variables.
%
% I only have to make sure dummy variables are really safe being handled this
% way with the getop action having being done before they expand, but it looks
% ok. Attention it is crucial that if \XINT_expr_getop  finds a \relax it inserts
% \xint_c_\relax so the \relax token is still there!
%
% With this refactoring the \XINT_expr_getop_legacy is applied only in case
% of non-existent variables or fake variables omit/abort or things such as 
% nil, None, false, true, False, True.
%
% If user in interactive mode fixes the variable name, the \XINT_expr_var_foo
% expanded once with deliver {\XINT_expr_varvalue_foo} (if not dummy), and the
% braces are maintained by \XINT_expr_getop_legacy.
%
% |
%    \begin{macrocode}
\def\XINT_expr_op__  #1% op__ with two _'s
{%
    \ifcsname XINT_expr_var*_#1\endcsname
       \csname XINT_expr_var_#1\expandafter\endcsname
       \romannumeral`&&@\expandafter\XINT_expr_getop
    \else
      \expandafter\expandafter\expandafter\XINT_expr_getop_legacy
      \csname XINT_expr_var_%
              \ifcsname XINT_expr_var_#1\endcsname#1\else\XINT_expr_unknown_variable{#1}\fi
      \expandafter\endcsname
    \fi
}%
\def\XINT_expr_unknown_variable #1%
   {\XINT_expandableerror {`#1' unknown, say `Isome_var' or I use 0.}}%
\def\XINT_expr_var_{{{0}}}%
\let\XINT_flexpr_op__ \XINT_expr_op__
\let\XINT_iiexpr_op__ \XINT_expr_op__
\def\XINT_expr_getop_legacy #1%
{%
    \expanded{\unexpanded{{#1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
%    \end{macrocode}
% \subsection{\csh{XINT_expr_getop}: fetch the next operator or closing
% parenthesis or end of expression}
% \lverb|Release 1.1 implements multi-character operators.
%
% 1.2d adds tacit mutiplication also in front of variable or functions names
% starting with a letter, not only a @ or a _ as was already the case. This is
% for (x+y)z situations. It also applies higher precedence in cases like x/2y
% or x/2@, or x/2max(3,5), or x/2\xintexpr 3\relax.
%
% In fact, finally I decide that all sorts of tacit multiplication will always
% use the higher precedence.
%
% Indeed I hesitated somewhat: with the current code one does not know if
% \XINT_expr_getop as invoked after a closing parenthesis or because a number
% parsing ended, and I felt distinguishing the two was unneeded extra stuff.
% This means cases like (a+b)/(c+d)(e+f) will first multiply the last two
% parenthesized terms.
%
%
% 1.2q adds tacit multiplication in cases such as (1+1)3 or 5!7!
%
% 1.4 has simplified coding here as \XINT_expr_getop expansion happens
% at a time when a fetched value has already being stored.
%
% Prior to 1.4g there was an \if _#1\xint_dothis\xint_secondofthree\fi
% because the _ can be used to start names, for private use by package (for
% example by polexpr). But this test was silly because these usages are only
% with a _ of catcode 11. And allowing non-catcode 11 _ also to trigger tacit
% multiplication caused an infinite loop in collaboration with
% \XINT_expr_scanfunc, see explanations there (now removed after refactoring,
% see \XINT_expr_startfunc).
%
% The situation with the @ is different because we must allow it even as
% catcode 12 as a name, as it used in the syntax and must work the same if of
% catcode 11 or 12. No infinite loop because it is filtered out by one of the
% \XINT_expr_getnextfork macros.
%
% The check for : to send it to thirdofthree "getop" branch is needed, last
% time I checked, because during some part of at least \xintdeffunc, some
% scantokens are done which need to work with the : of catcode 11, and it
% would be misconstrued to start a name if not filtered out.
%
% |
%    \begin{macrocode}
\def\XINT_expr_getop #1%
{%
    \expandafter\XINT_expr_getop_a\romannumeral`&&@#1%
}%
\catcode`* 11
\def\XINT_expr_getop_a #1%
{%
    \ifx   \relax #1\xint_dothis\xint_firstofthree\fi
    \ifcat \relax #1\xint_dothis\xint_secondofthree\fi
    \ifnum\xint_c_ix<1\string#1 \xint_dothis\xint_secondofthree\fi
    \if    :#1\xint_dothis      \xint_thirdofthree\fi
    \if    @#1\xint_dothis      \xint_secondofthree\fi
    \if    (#1\xint_dothis      \xint_secondofthree\fi %)
    \ifcat a#1\xint_dothis      \xint_secondofthree\fi
    \xint_orthat \xint_thirdofthree
%    \end{macrocode}
% \lverb|Formerly \XINT_expr_foundend as firstofthree but at 1.4g let's
% simply insert \xint_c_ as the #1 is \relax (and anyhow a place-holder
% according to remark in definition of \XINT_expr_foundend|
%    \begin{macrocode}
    \xint_c_
%    \end{macrocode}
% \lverb|Tacit multiplication with higher precedence. Formerly
% \XINT_expr_precedence_*** was used, renamed to \XINT_expr_prec_tacit at
% 1.4g in case a backport is done of the \bnumdefinfix from bnumexpr.|
%    \begin{macrocode}
    {\XINT_expr_prec_tacit *}%
%    \end{macrocode}
% \lverb|This is only location which jumps to \XINT_expr_getop_b. At 1.4f and
% perhaps for old legacy reasons this was \expandafter\XINT_expr_getop_b
% \string#1 but I see no reason now for applying \string to #1. Removed at
% 1.4g. And the #1 now moved out of the secondofthree and thirdofthree
% branches.|
%    \begin{macrocode}
    \XINT_expr_getop_b
    #1%
}%
\catcode`* 12
%    \end{macrocode}
% \lverb|\relax is a place holder here. At 1.4g, we don't use
% \XINT_expr_foundend anymore in \XINT_expr_getop_a which was slightly
% refactored, but it is used elsewhere.
%
% Attention that keeping a \relax around if \XINT_expr_getop hits it is
% crucial to good functioning of dummy variables after 1.4g refactoring of
% \XINT_expr_op__, the \relax being used as delimiter by dummy variables, and
% \XINT_expr_getop is now expanded before the variable itself does its thing.
% |
%    \begin{macrocode}
\def\XINT_expr_foundend {\xint_c_ \relax}%
%    \end{macrocode}
% \lverb|? is a very special operator with top precedence which will check if
% the next token is another ?, while avoiding removing a brace pair from token
% stream due to its syntax. Pre 1.1 releases used : rather than ??, but we
% need : for Python like slices of lists.|
%
% \lverb|null char is used as hack to implement A/B[N] raw input at 1.4. See
% also \XINT_expr_scanint_c.|
%
% \lverb|Memo: 1.4g, the token fetched by \XINT_expr_getop_b has not anymore
% been previously submitted in \XINT_expr_getop_a to \string.|
%    \begin{macrocode}
\def\XINT_expr_getop_b#1{\def\XINT_expr_getop_b ##1%
{%
     \if &&@##1\xint_dothis{#1&&@}\fi
     \if   '##1\xint_dothis{\XINT_expr_binopwrd }\fi
     \if   ?##1\xint_dothis{\XINT_expr_precedence_? ?}\fi
     \xint_orthat          {\XINT_expr_scanop_a ##1}%
}}\expandafter\XINT_expr_getop_b\csname XINT_expr_precedence_&&@\endcsname
\def\XINT_expr_binopwrd #1'%
{%
    \expandafter\XINT_expr_foundop_a
    \csname XINT_expr_itself_\xint_zapspaces #1 \xint_gobble_i\endcsname
}%
\def\XINT_expr_scanop_a #1#2%
{%
    \expandafter\XINT_expr_scanop_b\expandafter#1\romannumeral`&&@#2%
}%
%    \end{macrocode}
% \lverb|Multi-character operators have an associated itself macro at each
% stage of decomposition starting at two characters. Here, nothing imposes to
% the operator characters not to be of catcode letter, this constraint applies
% only on the first character and is done via \XINT_expr_getop_a, to handle in
% particular tacit multiplication in front of variable or function names.
%
% But it would be dangerous to allow letters in operator characters, again due
% to existence of variables and functions, and anyhow there is no user
% interface to add such custom operators.  However in bnumexpr, such a
% constraint does not exist.
%
% I don't worry too much about efficiency here... and at 1.4g I have
% re-written for code readability only.  Once we see that #1#2 is not a
% candidate to be or start an operator, we need to check if single-character
% operator #1 is really an operator and this is done via the existence of the
% precedence token.
%
% Unfortunately the 1.4g refactoring of the scanop macros had a bad bug:
% \XINT_expr_scanop_c inserted \romannumeral`^^@ in stream but did not grab a
% token first so a space would stop the \romannumeral and then the #2 in
% \XINT_expr_scanop_d was not pre-expanded and ended up alone in \ifcat.  It
% is too distant in the past the time when I wrote the core of xintexpr in
% 2013... older and dumber now.|
%    \begin{macrocode}
\def\XINT_expr_scanop_b #1#2%
{%
    \unless\ifcat#2\relax
           \ifcsname XINT_expr_itself_#1#2\endcsname
           \XINT_expr_scanop_c
    \fi\fi
    \XINT_expr_foundop_a #1#2%
}%
\def\XINT_expr_scanop_c #1#2#3#4#5#6% #1#2=\fi\fi
{%
    #1#2%
    \expandafter\XINT_expr_scanop_d\csname XINT_expr_itself_#4#5\expandafter\endcsname
    \romannumeral`&&@#6%
}%
\def\XINT_expr_scanop_d #1#2%
{%
    \unless\ifcat#2\relax
           \ifcsname XINT_expr_itself_#1#2\endcsname
           \XINT_expr_scanop_c
    \fi\fi
    \XINT_expr_foundop #1#2%
}%
\def\XINT_expr_foundop_a #1%
{%
    \ifcsname XINT_expr_precedence_#1\endcsname
        \csname XINT_expr_precedence_#1\expandafter\endcsname
        \expandafter #1%
    \else
        \expandafter\XINT_expr_getop\romannumeral`&&@%
        \xint_afterfi{\XINT_expandableerror
        {Expected an operator but got `#1'. Ignoring.}}%
     \fi
}%
\def\XINT_expr_foundop #1{\csname XINT_expr_precedence_#1\endcsname #1}%
%    \end{macrocode}
% \subsection{Expansion spanning; opening and closing parentheses}
% \lverb|&
%
% These comments apply to all definitions coming next relative
% to execution of operations from parsing of syntax.
%
% Refactored (and unified) at 1.4. In particular
% the 1.4 scheme uses op, exec, check-, and checkp. Formerly
% it was until_a (check-) and until_b (now split into checkp and exec).
%
% This way neither check- nor checkp have to
% grab the accumulated number so far (top of stack if you like) and besides
% one never has to go back to check- from checkp (and neither from check-).
%
%
% 
% Prior to 1.4, accumulated
% intermediate results were stored as one token, but now we have to use
% \expanded to propagate expansion beyond possibly arbitrary long braced
% nested data. With the 1.4 refactoring we do this only once and only grab
% a second time the data if we actually have to act upon it.
%
% Version 1.1 had a hack inside the until macros for handling the omit
% and abort in iterations over dummy variables. This has been removed by
% 1.2c, see the subsection where omit and abort are discussed.
%
% Exceptionally, the check- is here abbreviated to check.
% |
%
%    \begin{macrocode}
\catcode`) 11
\def\XINT_tmpa #1#2#3#4#5#6%
{%
    \def#1% start
    {%
        \expandafter#2\romannumeral`&&@\XINT_expr_getnext
    }%
    \def#2##1% check
    {%
        \xint_UDsignfork
          ##1{\expandafter#3\romannumeral`&&@#4}%
            -{#3##1}%
        \krof
    }%
    \def#3##1##2% checkp
    {%
        \ifcase ##1%
           \expandafter\XINT_expr_done
        \or\expandafter#5%
        \else
           \expandafter#3\romannumeral`&&@\csname XINT_#6_op_##2\expandafter\endcsname
        \fi
    }%
    \def#5%
    {%
        \XINT_expandableerror
        {Extra ) removed. Hit <return>, fingers crossed.}%
        \expandafter#2\romannumeral`&&@\expandafter\XINT_expr_put_op_first
        \romannumeral`&&@\XINT_expr_getop_legacy
    }%
}%
\let\XINT_expr_done\space
\xintFor #1 in {expr,flexpr,iiexpr} \do {%
    \expandafter\XINT_tmpa
    \csname XINT_#1_start\expandafter\endcsname
    \csname XINT_#1_check\expandafter\endcsname
    \csname XINT_#1_checkp\expandafter\endcsname
    \csname XINT_#1_op_-xii\expandafter\endcsname
    \csname XINT_#1_extra_)\endcsname
    {#1}%
}%
%    \end{macrocode}
% \lverb|&
% Here also we take some shortcuts relative to general philosophy and have no explicit
% exec macro.|
%    \begin{macrocode}
\def\XINT_tmpa #1#2#3#4#5#6#7%
{%
    \def #1##1% op_(
    {%
        \expandafter #4\romannumeral`&&@\XINT_expr_getnext
    }%
    \def #2##1% op_)
    {%
        \expanded{\unexpanded{\XINT_expr_put_op_first{##1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
    }%
    \def #3% oparen
    {%
        \expandafter #4\romannumeral`&&@\XINT_expr_getnext
    }%
    \def #4##1% check-
    {%
        \xint_UDsignfork
            ##1{\expandafter#5\romannumeral`&&@#6}%
              -{#5##1}%
        \krof
    }%
    \def #5##1##2% checkp
    {%
        \ifcase ##1\expandafter\XINT_expr_missing_)
        \or \csname XINT_#7_op_##2\expandafter\endcsname
        \else
          \expandafter #5\romannumeral`&&@\csname XINT_#7_op_##2\expandafter\endcsname
        \fi
    }%
}%
\def\XINT_expr_missing_) 
   {\XINT_expandableerror{End of expression found, but some ) was missing there.}%
    \xint_c_ \XINT_expr_done }%
\xintFor #1 in {expr,flexpr,iiexpr} \do {%
    \expandafter\XINT_tmpa
    \csname XINT_#1_op_(\expandafter\endcsname
    \csname XINT_#1_op_)\expandafter\endcsname
    \csname XINT_#1_oparen\expandafter\endcsname
    \csname XINT_#1_check-_)\expandafter\endcsname
    \csname XINT_#1_checkp_)\expandafter\endcsname
    \csname XINT_#1_op_-xii\endcsname
    {#1}%
}%
\let\XINT_expr_precedence_)\xint_c_i
\catcode`) 12
%    \end{macrocode}
% \subsection{The comma as binary operator}
% \lverb|New with 1.09a. Refactored at 1.4.|
%    \begin{macrocode}
\def\XINT_tmpa #1#2#3#4#5#6%
{%
    \def #1##1% \XINT_expr_op_,
    {%
      \expanded{\unexpanded{#2{##1}}\expandafter}%
      \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
    }%
    \def #2##1##2##3##4{##2##3{##1##4}}% \XINT_expr_exec_,
    \def #3##1% \XINT_expr_check-_,
    {%
      \xint_UDsignfork
        ##1{\expandafter#4\romannumeral`&&@#5}%
          -{#4##1}%
      \krof
    }%
    \def #4##1##2% \XINT_expr_checkp_,
    {%
      \ifnum ##1>\xint_c_iii
        \expandafter#4%
           \romannumeral`&&@\csname XINT_#6_op_##2\expandafter\endcsname
      \else
        \expandafter##1\expandafter##2%
      \fi
    }%
}%
\xintFor #1 in {expr,flexpr,iiexpr} \do {%
\expandafter\XINT_tmpa
    \csname XINT_#1_op_,\expandafter\endcsname
    \csname XINT_#1_exec_,\expandafter\endcsname
    \csname XINT_#1_check-_,\expandafter\endcsname
    \csname XINT_#1_checkp_,\expandafter\endcsname
    \csname XINT_#1_op_-xii\endcsname {#1}%
}%
\expandafter\let\csname XINT_expr_precedence_,\endcsname\xint_c_iii
%    \end{macrocode}
% \subsection{The minus as prefix operator of variable precedence level}
% \lverb|Inherits the precedence level of the previous infix operator, if the
% latter has at least the precedence level of binary + and -, i.e. currently
% 12.
%
% Refactored at 1.4.
%
% At 1.4g I belatedly observe that I have been defining architecture for
% op_-xvi but such operator can never be created, because there are no infix
% operators of precedence level 16.  Perhaps in the past this was really
% needed?  But now such 16 is precedence level of tacit multiplication which
% is implemented simply by the \XINT_expr_prec_tacit token, there is no macro
% check-_*** which would need an op_-xvi.
%
% For the record: at least one scenario exists
% which creates tacit multiplication in front of a unary -, it is 2\count0
% which first generates tacit multiplication then applies \number to \count0,
% but the operator
% is still *, so this triggers only \XINT_expr_op_-xiv, not -xvi.
%
% At 1.4g we need 17 and not 18 anymore as the precedence of unary minus
% following power operators ^ and **. The needed \xint_c_xvii creation was
% added to xintkernel.sty.|
%    \begin{macrocode}
\def\XINT_tmpb #1#2#3#4#5#6#7%
{%
    \def #1% \XINT_expr_op_-<level>
    {%
      \expandafter #2\romannumeral`&&@\expandafter#3%
      \romannumeral`&&@\XINT_expr_getnext
    }%
    \def #2##1##2##3% \XINT_expr_exec_-<level>
    {%
      \expandafter ##1\expandafter ##2\expandafter
       {%
        \romannumeral`&&@\XINT:NEhook:f:one:from:one
        {\romannumeral`&&@#7##3}%
       }%
    }%
    \def #3##1% \XINT_expr_check-_-<level>
    {%
      \xint_UDsignfork
        ##1{\expandafter #4\romannumeral`&&@#1}%
          -{#4##1}%
      \krof
    }%
    \def #4##1##2% \XINT_expr_checkp_-<level>
    {%
      \ifnum ##1>#5%
        \expandafter #4%
        \romannumeral`&&@\csname XINT_#6_op_##2\expandafter\endcsname
      \else
        \expandafter ##1\expandafter ##2%
      \fi
    }%
}%
\def\XINT_tmpa #1#2#3%
{%
    \expandafter\XINT_tmpb
    \csname XINT_#1_op_-#3\expandafter\endcsname
    \csname XINT_#1_exec_-#3\expandafter\endcsname
    \csname XINT_#1_check-_-#3\expandafter\endcsname
    \csname XINT_#1_checkp_-#3\expandafter\endcsname
    \csname xint_c_#3\endcsname {#1}#2%
}%
\xintApplyInline{\XINT_tmpa {expr}\xintOpp}{{xii}{xiv}{xvii}}%
\xintApplyInline{\XINT_tmpa {flexpr}\xintOpp}{{xii}{xiv}{xvii}}%
\xintApplyInline{\XINT_tmpa {iiexpr}\xintiiOpp}{{xii}{xiv}{xvii}}%
%    \end{macrocode}
% \subsection{The \texorpdfstring{\protect\lowast}{*}{} as Python-like «unpacking» prefix operator}
%
% \lverb|New with 1.4. Prior to 1.4 the internal data structure was
% the one of \csname encapsulated comma separated numbers. No hierarchical
% structure was (easily) possible. At 1.4, we can use TeX braces because there
% is no detokenization to catcode 12.|
%
% 
%    \begin{macrocode}
\def\XINT_tmpa#1#2#3%
{%
    \def#1##1{\expandafter#2\romannumeral`&&@\XINT_expr_getnext}%
    \def#2##1##2%
    {%
      \ifnum ##1>\xint_c_xx
        \expandafter #2%
        \romannumeral`&&@\csname XINT_#3_op_##2\expandafter\endcsname
      \else
        \expandafter##1\expandafter##2\romannumeral0\expandafter\XINT:NEhook:unpack
      \fi
    }%
}%
\def\XINT:NEhook:unpack{\xint_stop_atfirstofone}%
\xintFor* #1 in {{expr}{flexpr}{iiexpr}}:
    {\expandafter\XINT_tmpa\csname XINT_#1_op_0\expandafter\endcsname
                           \csname XINT_#1_until_unpack\endcsname {#1}}%
%    \end{macrocode}
% \subsection{Infix operators}
%
% \localtableofcontents
%
% \lverb|1.2d adds the *** for tying via tacit multiplication, for example
% x/2y. Actually I don't need the _itself mechanism for ***, only a
% precedence.
%
% At 1.4b we must make sure that the ! in expansion of \XINT_expr_itself_!= is
% of catcode 12 and not of catcode 11.  This is because implementation of
% chaining of comparison operators proceeds via inserting the itself macro
% directly into upcoming token stream, whereas formerly such itself macros
% would be expanded only in a \csname...\endcsname context.|
%    \begin{macrocode}
\catcode`& 12 \catcode`! 12
\xintFor* #1 in {{==}{!=}{<=}{>=}{&&}{||}{//}{/:}{..}{..[}{].}{]..}}%
    \do {\expandafter\def\csname XINT_expr_itself_#1\endcsname {#1}}%
\catcode`& 7  \catcode`! 11
%    \end{macrocode}
% \subsubsection{\&\&, \textbar\textbar, //, /:, +,
% \textendash, \texorpdfstring{\protect\lowast}{*}, /, \textasciicircum,
% \texorpdfstring{\protect\lowast\protect\lowast}{**}{}, \textquotesingle and\textquotesingle, \textquotesingle
% or\textquotesingle, \textquotesingle xor\textquotesingle, and
% \textquotesingle mod\textquotesingle}
%
% \lverb?
% Single character boolean operators $& and | had been deprecated since
% 1.1 and finally got removed (together with = comparison test)
% from syntax at 1.4g.
%
% Also, at 1.4g I finally decide to enact the switch to right
% associativity for the power operators ^ and **.
%
% This goes via inserting into the checkp macros not anymore the precedence
% chardef token (which now only serves as left precedence, inserted in the
% token stream) but in its place an \xint_c_<roman> token holding the right
% precedence. Which is also transmitted to spanned unary minus operators.
%
% Here only levels 12, 14, and 17 are created as right precedences.
%
% #6 and #7 got permuted and the new #7 is directly a control sequence.
% Also #3 and #4 are now integers which need \romannumeral. The change in
% \XINT_expr_defbin_c does not propagate as it is re-defined shortly thereafter.
% ?
%    \begin{macrocode}
\def\XINT_expr_defbin_c #1#2#3#4#5#6#7#8%
{%
  \def #1##1% \XINT_expr_op_<op>
  {%
    \expanded{\unexpanded{#2{##1}}\expandafter}%
    \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
  }%
  \def #2##1##2##3##4% \XINT_expr_exec_<op>
  {%
    \expandafter##2\expandafter##3\expandafter
      {\romannumeral`&&@\XINT:NEhook:f:one:from:two{\romannumeral`&&@#7##1##4}}%
  }%
  \def #3##1% \XINT_expr_check-_<op>
  {%
    \xint_UDsignfork
      ##1{\expandafter#4\romannumeral`&&@#5}%
        -{#4##1}%
    \krof
  }%
  \def #4##1##2% \XINT_expr_checkp_<op>
  {%
    \ifnum ##1>#6%
      \expandafter#4%
      \romannumeral`&&@\csname XINT_#8_op_##2\expandafter\endcsname
    \else 
      \expandafter ##1\expandafter ##2%
    \fi
  }%
}%
\def\XINT_expr_defbin_b #1#2#3#4#5%
{%
  \expandafter\XINT_expr_defbin_c
  \csname XINT_#1_op_#2\expandafter\endcsname
  \csname XINT_#1_exec_#2\expandafter\endcsname
  \csname XINT_#1_check-_#2\expandafter\endcsname
  \csname XINT_#1_checkp_#2\expandafter\endcsname
  \csname XINT_#1_op_-\romannumeral\ifnum#4>12 #4\else12\fi\expandafter\endcsname
  \csname xint_c_\romannumeral#4\endcsname
  #5%
  {#1}%
  \expandafter % done 3 times but well
  \let\csname XINT_expr_precedence_#2\expandafter\endcsname
      \csname xint_c_\romannumeral#3\endcsname
}%
\XINT_expr_defbin_b {expr}  {||} {6} {6} \xintOR
\XINT_expr_defbin_b {flexpr}{||} {6} {6} \xintOR
\XINT_expr_defbin_b {iiexpr}{||} {6} {6} \xintOR
\catcode`& 12
\XINT_expr_defbin_b {expr}  {&&} {8} {8} \xintAND
\XINT_expr_defbin_b {flexpr}{&&} {8} {8} \xintAND
\XINT_expr_defbin_b {iiexpr}{&&} {8} {8} \xintAND
\catcode`& 7
\XINT_expr_defbin_b {expr}  {xor}{6} {6} \xintXOR
\XINT_expr_defbin_b {flexpr}{xor}{6} {6} \xintXOR
\XINT_expr_defbin_b {iiexpr}{xor}{6} {6} \xintXOR
\XINT_expr_defbin_b {expr}  {//} {14}{14}\xintDivFloor
\XINT_expr_defbin_b {flexpr}{//} {14}{14}\XINTinFloatDivFloor
\XINT_expr_defbin_b {iiexpr}{//} {14}{14}\xintiiDivFloor
\XINT_expr_defbin_b {expr}  {/:} {14}{14}\xintMod
\XINT_expr_defbin_b {flexpr}{/:} {14}{14}\XINTinFloatMod
\XINT_expr_defbin_b {iiexpr}{/:} {14}{14}\xintiiMod
\XINT_expr_defbin_b {expr}   +   {12}{12}\xintAdd
\XINT_expr_defbin_b {flexpr} +   {12}{12}\XINTinFloatAdd
\XINT_expr_defbin_b {iiexpr} +   {12}{12}\xintiiAdd
\XINT_expr_defbin_b {expr}   -   {12}{12}\xintSub
\XINT_expr_defbin_b {flexpr} -   {12}{12}\XINTinFloatSub
\XINT_expr_defbin_b {iiexpr} -   {12}{12}\xintiiSub
\XINT_expr_defbin_b {expr}   *   {14}{14}\xintMul
\XINT_expr_defbin_b {flexpr} *   {14}{14}\XINTinFloatMul
\XINT_expr_defbin_b {iiexpr} *   {14}{14}\xintiiMul
\let\XINT_expr_prec_tacit \xint_c_xvi
\XINT_expr_defbin_b {expr}   /   {14}{14}\xintDiv
\XINT_expr_defbin_b {flexpr} /   {14}{14}\XINTinFloatDiv
\XINT_expr_defbin_b {iiexpr} /   {14}{14}\xintiiDivRound
%    \end{macrocode}
% \lverb|At 1.4g, right associativity is implemented via
% a lowered right precedence here.|
%    \begin{macrocode}
\XINT_expr_defbin_b {expr}   ^   {18}{17}\xintPow
\XINT_expr_defbin_b {flexpr} ^   {18}{17}\XINTinFloatSciPow
\XINT_expr_defbin_b {iiexpr} ^   {18}{17}\xintiiPow
%    \end{macrocode}
% \lverb|1.4g This is a trick (which was in old version of bnumexpr, I
% wonder why I did not have it here) but it will make error messages in case
% of **<token> confusing.  The ^ here is of catcode 11 but it does not matter.|
%    \begin{macrocode}
\expandafter\def\csname XINT_expr_itself_**\endcsname{^}%
\catcode`& 12
%    \end{macrocode}
% \lverb|For this which contributes to implementing 'and', 'or', etc... see
% \XINT_expr_binopwrd.|
%    \begin{macrocode}
\xintFor #1 in {and,or,xor,mod} \do
{%
   \expandafter\def\csname XINT_expr_itself_#1\endcsname {#1}%
}%
\expandafter\let\csname XINT_expr_precedence_and\expandafter\endcsname
                \csname XINT_expr_precedence_&&\endcsname
\expandafter\let\csname XINT_expr_precedence_or\expandafter\endcsname
                \csname XINT_expr_precedence_||\endcsname
\expandafter\let\csname XINT_expr_precedence_mod\expandafter\endcsname
                \csname XINT_expr_precedence_/:\endcsname
\xintFor #1 in {expr, flexpr, iiexpr} \do
{%
   \expandafter\let\csname XINT_#1_op_and\expandafter\endcsname
                   \csname XINT_#1_op_&&\endcsname
   \expandafter\let\csname XINT_#1_op_or\expandafter\endcsname
                   \csname XINT_#1_op_||\endcsname
   \expandafter\let\csname XINT_#1_op_mod\expandafter\endcsname
                   \csname XINT_#1_op_/:\endcsname
}%
\catcode`& 7
%    \end{macrocode}
% \subsubsection{.., ..[, and ].. for a..b and a..[b]..c syntax}
% \lverb|The 1.4 exec_..[ macros (which do no further expansion!) had silly
% \expandafter doing nothing for the sole reason of sharing a common
% \XINT_expr_defbin_c as used previously for the +, - etc... operators. At
% 1.4b we take the time to set things straight and do other similar
% simplifications.|
%    \begin{macrocode}
\def\XINT_expr_defbin_c #1#2#3#4#5#6#7%
{%
  \def #1##1% \XINT_expr_op_..[
  {%
    \expanded{\unexpanded{#2{##1}}\expandafter}%
    \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
  }%
  \def #2##1##2##3##4% \XINT_expr_exec_..[
  {%
    ##2##3{{##1##4}}%
  }%
  \def #3##1% \XINT_expr_check-_..[
  {%
    \xint_UDsignfork
      ##1{\expandafter#4\romannumeral`&&@#5}%
        -{#4##1}%
    \krof
  }%
  \def #4##1##2% \XINT_expr_checkp_..[
  {%
    \ifnum ##1>#6%
      \expandafter#4%
      \romannumeral`&&@\csname XINT_#7_op_##2\expandafter\endcsname
    \else 
      \expandafter ##1\expandafter ##2%
    \fi
  }%
}%
\def\XINT_expr_defbin_b #1%
{%
  \expandafter\XINT_expr_defbin_c
  \csname XINT_#1_op_..[\expandafter\endcsname
  \csname XINT_#1_exec_..[\expandafter\endcsname
  \csname XINT_#1_check-_..[\expandafter\endcsname
  \csname XINT_#1_checkp_..[\expandafter\endcsname
  \csname XINT_#1_op_-xii\expandafter\endcsname
  \csname XINT_expr_precedence_..[\endcsname
  {#1}%
}%
\XINT_expr_defbin_b {expr}%
\XINT_expr_defbin_b {flexpr}%
\XINT_expr_defbin_b {iiexpr}%
\expandafter\let\csname XINT_expr_precedence_..[\endcsname\xint_c_vi
\def\XINT_expr_defbin_c #1#2#3#4#5#6#7#8%
{%
  \def #1##1% \XINT_expr_op_<op>
  {%
    \expanded{\unexpanded{#2{##1}}\expandafter}%
    \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
  }%
  \def #2##1##2##3##4% \XINT_expr_exec_<op>
  {%
    \expandafter##2\expandafter##3\expanded
    {{\XINT:NEhook:x:one:from:two#8##1##4}}%
  }%
  \def #3##1% \XINT_expr_check-_<op>
  {%
    \xint_UDsignfork
      ##1{\expandafter#4\romannumeral`&&@#5}%
        -{#4##1}%
    \krof
  }%
  \def #4##1##2% \XINT_expr_checkp_<op>
  {%
    \ifnum ##1>#6%
      \expandafter#4%
      \romannumeral`&&@\csname XINT_#7_op_##2\expandafter\endcsname
    \else 
      \expandafter ##1\expandafter ##2%
    \fi
  }%
}%
\def\XINT_expr_defbin_b #1#2#3%
{%
  \expandafter\XINT_expr_defbin_c
  \csname XINT_#1_op_#2\expandafter\endcsname
  \csname XINT_#1_exec_#2\expandafter\endcsname
  \csname XINT_#1_check-_#2\expandafter\endcsname
  \csname XINT_#1_checkp_#2\expandafter\endcsname
  \csname XINT_#1_op_-xii\expandafter\endcsname
  \csname XINT_expr_precedence_#2\endcsname
  {#1}#3%
  \expandafter\let
  \csname XINT_expr_precedence_#2\expandafter\endcsname\xint_c_vi
}%
\XINT_expr_defbin_b {expr}   {..}\xintSeq:tl:x
\XINT_expr_defbin_b {flexpr} {..}\xintSeq:tl:x
\XINT_expr_defbin_b {iiexpr} {..}\xintiiSeq:tl:x
\XINT_expr_defbin_b {expr}  {]..}\xintSeqB:tl:x
\XINT_expr_defbin_b {flexpr}{]..}\xintSeqB:tl:x
\XINT_expr_defbin_b {iiexpr}{]..}\xintiiSeqB:tl:x
%    \end{macrocode}
% \subsubsection{<, >, ==, <=, >=, != with Python-like chaining}
% \lverb|1.4b
% This is preliminary implementation of chaining of comparison
% operators like Python and (I think) l3fp do. I am not too happy
% with how many times the (second) operand (already evaluated) is fetched.
% |
%    \begin{macrocode}
\def\XINT_expr_defbin_d #1#2%
{%
  \def #1##1##2##3##4% \XINT_expr_exec_<op>
  {%
    \expandafter##2\expandafter##3\expandafter
      {\romannumeral`&&@\XINT:NEhook:f:one:from:two{\romannumeral`&&@#2##1##4}}%
  }%
}%
\def\XINT_expr_defbin_c #1#2#3#4#5#6#7#8#9%
{%
  \def #1##1% \XINT_expr_op_<op>
  {%
    \expanded{\unexpanded{#2{##1}}\expandafter}%
    \romannumeral`&&@\expandafter#7%
    \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
  }%
  \def #3##1% \XINT_expr_check-_<op>
  {%
    \xint_UDsignfork
      ##1{\expandafter#4\romannumeral`&&@#5}%
        -{#4##1}%
    \krof
  }%
  \def #4##1##2% \XINT_expr_checkp_<op>
  {%
    \ifnum ##1>#6%
      \expandafter#4%
      \romannumeral`&&@\csname XINT_#9_op_##2\expandafter\endcsname
    \else
      \expandafter ##1\expandafter ##2%
    \fi
  }%
  \let #6\xint_c_x
  \def #7##1% \XINT_expr_checkc_<op>
  {%
    \ifnum ##1=\xint_c_x\expandafter#8\fi ##1%
  }%
  \edef #8##1##2##3% \XINT_expr_execc_<op>
  {%
    \csname XINT_#9_precedence_\string&\string&\endcsname
    \expandafter\noexpand\csname XINT_#9_itself_\string&\string&\endcsname
    {##3}%
    \XINTfstop.{##3}##2%
  }%
  \XINT_expr_defbin_d #2% \XINT_expr_exec_<op>
}%
\def\XINT_expr_defbin_b #1#2%#3%
{%
  \expandafter\XINT_expr_defbin_c
  \csname XINT_#1_op_#2\expandafter\endcsname
  \csname XINT_#1_exec_#2\expandafter\endcsname
  \csname XINT_#1_check-_#2\expandafter\endcsname
  \csname XINT_#1_checkp_#2\expandafter\endcsname
  \csname XINT_#1_op_-xii\expandafter\endcsname
  \csname XINT_expr_precedence_#2\expandafter\endcsname
  \csname XINT_#1_checkc_#2\expandafter\endcsname
  \csname XINT_#1_execc_#2\endcsname
  {#1}%#3%
}%
%    \end{macrocode}
% \lverb?
% Single character comparison operator = had been deprecated for many
% years (but I can't find since when precisely; $& and | were deprecated at
% 1.1 as since in CHANGES.md) and it finally got removed from syntax at
% 1.4g, as well as $& and |.
% ?
% \lverb|Attention that third token here is left in stream by defbin_b, then
% also by defbin_c and is picked up as #2 of defbin_d. Had to work around TeX
% accepting only 9 arguments. Why did it not start counting at #0 like all
% decent mathematicians do?|
%    \begin{macrocode}
\XINT_expr_defbin_b {expr}  <\xintLt
\XINT_expr_defbin_b {flexpr}<\xintLt
\XINT_expr_defbin_b {iiexpr}<\xintiiLt
\XINT_expr_defbin_b {expr}  >\xintGt
\XINT_expr_defbin_b {flexpr}>\xintGt
\XINT_expr_defbin_b {iiexpr}>\xintiiGt
\XINT_expr_defbin_b {expr}  {==}\xintEq
\XINT_expr_defbin_b {flexpr}{==}\xintEq
\XINT_expr_defbin_b {iiexpr}{==}\xintiiEq
\XINT_expr_defbin_b {expr}  {<=}\xintLtorEq
\XINT_expr_defbin_b {flexpr}{<=}\xintLtorEq
\XINT_expr_defbin_b {iiexpr}{<=}\xintiiLtorEq
\XINT_expr_defbin_b {expr}  {>=}\xintGtorEq
\XINT_expr_defbin_b {flexpr}{>=}\xintGtorEq
\XINT_expr_defbin_b {iiexpr}{>=}\xintiiGtorEq
\XINT_expr_defbin_b {expr}  {!=}\xintNotEq
\XINT_expr_defbin_b {flexpr}{!=}\xintNotEq
\XINT_expr_defbin_b {iiexpr}{!=}\xintiiNotEq
%    \end{macrocode}
% \subsubsection{Support macros for .., ..[ and ]..}
%
%
%\paragraph{\csh{xintSeq:tl:x}}
%\lverb|Commence par remplacer a par ceil(a) et b par floor(b) et renvoie
% ensuite les entiers entre les deux, possiblement en décroissant, et
% extrémités comprises. Si a=b est non entier en obtient donc ceil(a) et
% floor(a). Ne renvoie jamais une liste vide.
%
% Note: le a..b dans \xintfloatexpr utilise cette routine.|
%    \begin{macrocode}
\def\xintSeq:tl:x #1#2%
{%
    \expandafter\XINT_Seq:tl:x
    \the\numexpr \xintiCeil{#1}\expandafter.\the\numexpr \xintiFloor{#2}.%
}%
\def\XINT_Seq:tl:x #1.#2.%
{%
   \ifnum #2=#1 \xint_dothis\XINT_Seq:tl:x_z\fi
   \ifnum #2<#1 \xint_dothis\XINT_Seq:tl:x_n\fi
   \xint_orthat\XINT_Seq:tl:x_p
   #1.#2.%
}%
\def\XINT_Seq:tl:x_z #1.#2.{{#1/1[0]}}%
\def\XINT_Seq:tl:x_p #1.#2.%
{%
    {#1/1[0]}\ifnum #1=#2 \XINT_Seq:tl:x_e\fi
    \expandafter\XINT_Seq:tl:x_p \the\numexpr #1+\xint_c_i.#2.%
}%
\def\XINT_Seq:tl:x_n #1.#2.%
{%
    {#1/1[0]}\ifnum #1=#2 \XINT_Seq:tl:x_e\fi
    \expandafter\XINT_Seq:tl:x_n \the\numexpr #1-\xint_c_i.#2.%
}%
\def\XINT_Seq:tl:x_e#1#2.#3.{#1}%
%    \end{macrocode}
%\paragraph{\csh{xintiiSeq:tl:x}}
%    \begin{macrocode}
\def\xintiiSeq:tl:x #1#2%
{%
    \expandafter\XINT_iiSeq:tl:x
    \the\numexpr \xintiCeil{#1}\expandafter.\the\numexpr \xintiFloor{#2}.%
}%
\def\XINT_iiSeq:tl:x #1.#2.%
{%
   \ifnum #2=#1 \xint_dothis\XINT_iiSeq:tl:x_z\fi
   \ifnum #2<#1 \xint_dothis\XINT_iiSeq:tl:x_n\fi
   \xint_orthat\XINT_iiSeq:tl:x_p
   #1.#2.%
}%
\def\XINT_iiSeq:tl:x_z #1.#2.{{#1}}%
\def\XINT_iiSeq:tl:x_p #1.#2.%
{%
    {#1}\ifnum #1=#2 \XINT_Seq:tl:x_e\fi
    \expandafter\XINT_iiSeq:tl:x_p \the\numexpr #1+\xint_c_i.#2.%
}%
\def\XINT_iiSeq:tl:x_n #1.#2.%
{%
    {#1}\ifnum #1=#2 \XINT_Seq:tl:x_e\fi
    \expandafter\XINT_iiSeq:tl:x_n \the\numexpr #1-\xint_c_i.#2.%
}%
%    \end{macrocode}
% \lverb|Contrarily to a..b which is limited to small integers, this works
% with a, b, and d (big) fractions. It will produce a «nil» list, if a>b and
% d<0 or a<b and d>0.|
%
% \paragraph{\csh{xintSeqA}, \csh{xintiiSeqA}}
%
%    \begin{macrocode}
\def\xintSeqA        {\expandafter\XINT_SeqA\romannumeral0\xintraw}%
\def\xintiiSeqA    #1{\expandafter\XINT_iiSeqA\romannumeral`&&@#1;}%
\def\XINT_SeqA  #1]#2{\expandafter\XINT_SeqA_a\romannumeral0\xintraw {#2}#1]}%
\def\XINT_iiSeqA#1;#2{\expandafter\XINT_SeqA_a\romannumeral`&&@#2;#1;}%
\def\XINT_SeqA_a #1{\xint_UDzerominusfork
                                   #1-{z}%
                                   0#1{n}%
                                   0-{p}%
                    \krof #1}%
%    \end{macrocode}
%\paragraph{\csh{xintSeqB:tl:x}}
%\lverb|At 1.4, delayed expansion of start and step done here and not before,
% for matters of \xintdeffunc and «NEhooks».
%
% The float variant at 1.4 is made identical to the exact variant. I.e.
% stepping is exact and comparison to the range limit too. But recall that a/b
% input will be converted to a float. To handle 1/3 step for example still
% better to use \xintexpr 1..1/3..10\relax for example inside the \xintfloateval.|
%    \begin{macrocode}
\def\xintSeqB:tl:x  #1{\expandafter\XINT_SeqB:tl:x\romannumeral`&&@\xintSeqA#1}%
\def\XINT_SeqB:tl:x #1{\csname XINT_SeqB#1:tl:x\endcsname}%
\def\XINT_SeqBz:tl:x #1]#2]#3{{#2]}}%
\def\XINT_SeqBp:tl:x #1]#2]#3{\expandafter\XINT_SeqBp:tl:x_a\romannumeral0\xintraw{#3}#2]#1]}%
\def\XINT_SeqBp:tl:x_a #1]#2]#3]%
{%
    \xintifCmp{#1]}{#2]}%
    {}{{#2]}}{{#2]}\expandafter\XINT_SeqBp:tl:x_b\romannumeral0\xintadd{#3]}{#2]}#1]#3]}%
}%
\def\XINT_SeqBp:tl:x_b #1]#2]#3]%
{%
    \xintifCmp{#1]}{#2]}%
    {{#1]}\expandafter\XINT_SeqBp:tl:x_b\romannumeral0\xintadd{#3]}{#1]}#2]#3]}{{#1]}}{}%
}%
\def\XINT_SeqBn:tl:x #1]#2]#3{\expandafter\XINT_SeqBn:tl:x_a\romannumeral0\xintraw{#3}#2]#1]}%
\def\XINT_SeqBn:tl:x_a #1]#2]#3]%
{%
    \xintifCmp{#1]}{#2]}%
    {{#2]}\expandafter\XINT_SeqBn:tl:x_b\romannumeral0\xintadd{#3]}{#2]}#1]#3]}{{#2]}}{}%
}%
\def\XINT_SeqBn:tl:x_b #1]#2]#3]%
{%
    \xintifCmp{#1]}{#2]}%
    {}{{#1]}}{{#1]}\expandafter\XINT_SeqBn:tl:x_b\romannumeral0\xintadd{#3]}{#1]}#2]#3]}%
}%
%    \end{macrocode}
% \paragraph{\csh{xintiiSeqB:tl:x}}
%    \begin{macrocode}
\def\xintiiSeqB:tl:x  #1{\expandafter\XINT_iiSeqB:tl:x\romannumeral`&&@\xintiiSeqA#1}%
\def\XINT_iiSeqB:tl:x #1{\csname XINT_iiSeqB#1:tl:x\endcsname}%
\def\XINT_iiSeqBz:tl:x #1;#2;#3{{#2}}%
\def\XINT_iiSeqBp:tl:x #1;#2;#3{\expandafter\XINT_iiSeqBp:tl:x_a\romannumeral`&&@#3;#2;#1;}%
\def\XINT_iiSeqBp:tl:x_a #1;#2;#3;%
{%
    \xintiiifCmp{#1}{#2}%
    {}{{#2}}{{#2}\expandafter\XINT_iiSeqBp:tl:x_b\romannumeral0\xintiiadd{#3}{#2};#1;#3;}%
}%
\def\XINT_iiSeqBp:tl:x_b #1;#2;#3;%
{%
    \xintiiifCmp{#1}{#2}%
    {{#1}\expandafter\XINT_iiSeqBp:tl:x_b\romannumeral0\xintiiadd{#3}{#1};#2;#3;}{{#1}}{}%
}%
\def\XINT_iiSeqBn:tl:x #1;#2;#3{\expandafter\XINT_iiSeqBn:tl:x_a\romannumeral`&&@#3;#2;#1;}%
\def\XINT_iiSeqBn:tl:x_a #1;#2;#3;%
{%
    \xintiiifCmp{#1}{#2}%
    {{#2}\expandafter\XINT_iiSeqBn:tl:x_b\romannumeral0\xintiiadd{#3}{#2};#1;#3;}{{#2}}{}%
}%
\def\XINT_iiSeqBn:tl:x_b #1;#2;#3;%
{%
    \xintiiifCmp{#1}{#2}%
    {}{{#1}}{{#1}\expandafter\XINT_iiSeqBn:tl:x_b\romannumeral0\xintiiadd{#3}{#1};#2;#3;}%
}%
%    \end{macrocode}
% \subsection{Square brackets [\,] both as a container and a Python slicer}
% Refactored at |1.4|
%
% \lverb|The architecture allows to implement separately a «left» and a «right»
% precedence and this is crucial.|
% \localtableofcontents
%
%
% \subsubsection{[...] as «oneple» constructor}
%
% \lverb|In the definition of \XINT_expr_op_obracket the parameter
% is trash {}. The [ is intercepted by the getnextfork and handled
% via the \xint_c_ii^v highest precedence trick to get op_obracket
% executed.
% |
%    \begin{macrocode}
\def\XINT_expr_itself_obracket{obracket}%
\catcode`] 11 \catcode`[ 11
\def\XINT_expr_defbin_c #1#2#3#4#5#6%
{%
    \def #1##1%
    {%
      \expandafter#3\romannumeral`&&@\XINT_expr_getnext
    }%
    \def #2##1% op_]
    {%
      \expanded{\unexpanded{\XINT_expr_put_op_first{{##1}}}\expandafter}%
         \romannumeral`&&@\XINT_expr_getop
    }%
    \def #3##1%  until_cbracket_a
    {%
      \xint_UDsignfork
         ##1{\expandafter#4\romannumeral`&&@#5}%  #5 = op_-xii
           -{#4##1}%
      \krof
    }%
    \def #4##1##2%  until_cbracket_b
    {%
      \ifcase ##1\expandafter\XINT_expr_missing_]
      \or \expandafter\XINT_expr_missing_]
      \or \expandafter#2%
      \else
      \expandafter #4%
        \romannumeral`&&@\csname XINT_#6_op_##2\expandafter\endcsname
      \fi
    }%
}%
\def\XINT_expr_defbin_b #1%
{%
    \expandafter\XINT_expr_defbin_c
    \csname XINT_#1_op_obracket\expandafter\endcsname
    \csname XINT_#1_op_]\expandafter\endcsname
    \csname XINT_#1_until_cbracket_a\expandafter\endcsname
    \csname XINT_#1_until_cbracket_b\expandafter\endcsname
    \csname XINT_#1_op_-xii\endcsname
    {#1}%
}%
\XINT_expr_defbin_b {expr}%
\XINT_expr_defbin_b {flexpr}%
\XINT_expr_defbin_b {iiexpr}%
\def\XINT_expr_missing_]
   {\XINT_expandableerror{Ooops, looks like we are missing a ]. Aborting!}%
    \xint_c_ \XINT_expr_done}%
\let\XINT_expr_precedence_]\xint_c_ii
%    \end{macrocode}
% \subsubsection{[...] brackets and : operator for NumPy-like slicing and item
% indexing syntax}
% \lverb|&
% The opening bracket [ for the nutple constructor is filtered out by
% \XINT_expr_getnextfork and becomes «obracket» which behaves with
% precedence level 2. For the [..] Python slicer on the other hand, a real
% operator [ is defined with precedence level 4 (it must be higher than
% precedence level of commas) on its right and maximal precedence on its left.
%
% Important: although slicing and indexing shares many rules with Python/NumPy
% there are some significant differences: in particular there can not be any
% out-of-range error generated, slicing applies also to «oples» and not only
% to «nutple», and nested lists do not have to have their leaves at a constant
% depth. See the user manual.
%
% Currently, NumPy-like nested (basic) slicing is implemented, i.e [a:b, c:d,
% N, e:f, M] type syntax with Python rules regarding negative integers. This
% is parsed as an expression and can arise from expansion or contain
% calculations.
%
% Currently stepping, Ellipsis, and simultaneous multi-index extracting are
% not yet implemented.
%
% There are some subtle things here with possibility of variables been passed
% by reference.|
%    \begin{macrocode}
\def\XINT_expr_defbin_c #1#2#3#4#5#6%
{%
    \def #1##1% \XINT_expr_op_[
    {%
        \expanded{\unexpanded{#2{##1}}\expandafter}%
        \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
    }%
    \def #2##1##2##3##4% \XINT_expr_exec_]
    {%
         \expandafter\XINT_expr_put_op_first
         \expanded
          {%
           {\XINT:NEhook:x:listsel\XINT_ListSel_top ##1__##4&({##1}\expandafter}%
           \expandafter
          }%
         \romannumeral`&&@\XINT_expr_getop
    }%
    \def #3##1% \XINT_expr_check-_]
    {%
      \xint_UDsignfork
        ##1{\expandafter#4\romannumeral`&&@#5}%
          -{#4##1}%
      \krof
    }%
    \def #4##1##2% \XINT_expr_checkp_]
    {%
      \ifcase ##1\XINT_expr_missing_]
        \or \XINT_expr_missing_]
        \or \expandafter##1\expandafter##2%
        \else \expandafter#4%
              \romannumeral`&&@\csname XINT_#6_op_##2\expandafter\endcsname
      \fi
    }%
}%
\let\XINT_expr_precedence_[ \xint_c_xx
\def\XINT_expr_defbin_b #1%
{%
    \expandafter\XINT_expr_defbin_c
    \csname XINT_#1_op_[\expandafter\endcsname
    \csname XINT_#1_exec_]\expandafter\endcsname
    \csname XINT_#1_check-_]\expandafter\endcsname
    \csname XINT_#1_checkp_]\expandafter\endcsname
    \csname XINT_#1_op_-xii\endcsname
    {#1}%
}%
\XINT_expr_defbin_b {expr}%
\XINT_expr_defbin_b {flexpr}%
\XINT_expr_defbin_b {iiexpr}%
\catcode`] 12 \catcode`[ 12
%    \end{macrocode}
% \lverb|At 1.4 the getnext, scanint, scanfunc, getop chain got revisited to
% trigger automatic insertion of the nil variable if needed, without having in
% situations like here to define operators to support «[:» or «:]». And as we
% want to implement nested slicing à la NumPy, we would have had to handle
% also «:,» for example. Thus here we simply have to define the sole operator
% «:» and it will be some sort of inert joiner preparing a slicing spec.|
%    \begin{macrocode}
\def\XINT_expr_defbin_c #1#2#3#4#5#6%
{%
    \def #1##1% \XINT_expr_op_:
    {%
        \expanded{\unexpanded{#2{##1}}\expandafter}%
        \romannumeral`&&@\expandafter#3\romannumeral`&&@\XINT_expr_getnext
    }%
    \def #2##1##2##3##4% \XINT_expr_exec_:
    {%
       ##2##3{:##1{0};##4:_}%
    }%
    \def #3##1% \XINT_expr_check-_:
    {\xint_UDsignfork
        ##1{\expandafter#4\romannumeral`&&@#5}%
          -{#4##1}%
     \krof
    }%
    \def #4##1##2% \XINT_expr_checkp_:
    {%
      \ifnum ##1>\XINT_expr_precedence_:
        \expandafter #4\romannumeral`&&@%
                       \csname XINT_#6_op_##2\expandafter\endcsname
      \else
        \expandafter##1\expandafter##2%
      \fi
    }%
}%
\let\XINT_expr_precedence_: \xint_c_vi
\def\XINT_expr_defbin_b #1%
{%
    \expandafter\XINT_expr_defbin_c
    \csname XINT_#1_op_:\expandafter\endcsname
    \csname XINT_#1_exec_:\expandafter\endcsname
    \csname XINT_#1_check-_:\expandafter\endcsname
    \csname XINT_#1_checkp_:\expandafter\endcsname
    \csname XINT_#1_op_-xii\endcsname {#1}%
}%
\XINT_expr_defbin_b {expr}%
\XINT_expr_defbin_b {flexpr}%
\XINT_expr_defbin_b {iiexpr}%
%    \end{macrocode}
% \subsubsection{Macro layer implementing indexing and slicing}
% \lverb|xintexpr applies slicing not only to «objects» (which can be passed
% as arguments to functions) but also to «oples».
%
% Our «nlists» are not necessarily regular N-dimensional arrays à la NumPy.
% Leaves can be at arbitrary depths. If we were handling regular «ndarrays»,
% we could proceed a bit differently.
%
% For the related
% explanations, refer to the user manual.
%
% Notice that currently the code uses f-expandable (and not using \expanded)
% macros \xintApply, \xintApplyUnbraced, \xintKeep, \xintTrim, \xintNthOne from
% $xinttoolsnameimp.
%
% But the whole expansion happens inside an \expanded context, so possibly
% some gain could be achieved with x-expandable variants (xintexpr < 1.4
% had an \xintKeep:x:csv).
%
% I coded \xintApply:x and \xintApplyUnbraced:x in $xinttoolsnameimp, Brief
% testing indicated they were perhaps a bit better for 5x5x5x5 and 15x15x15x15
% arrays of 8 digits numbers and for 30x30x15 with 16 digits numbers: say
% 1$% gain... this seems to raise to between 4$% and
% 5$% for 400x400 array of 1 digit...
%
% Currently sticking with old macros.
% |
%    \begin{macrocode}
\def\XINT_ListSel_deeper #1%
{%
    \if :#1\xint_dothis\XINT_ListSel_slice_next\fi
    \xint_orthat {\XINT_ListSel_extract_next {#1}}%
}%
\def\XINT_ListSel_slice_next #1(%
{%
    \xintApply{\XINT_ListSel_recurse{:#1}}%
}%
\def\XINT_ListSel_extract_next #1(%
{%
    \xintApplyUnbraced{\XINT_ListSel_recurse{#1}}%
}%
\def\XINT_ListSel_recurse #1#2%
{%
    \XINT_ListSel_check #2__#1({#2}\expandafter\empty\empty
}%
\def\XINT_ListSel_check{\expandafter\XINT_ListSel_check_a \string}%
\def\XINT_ListSel_check_a #1%
{%
    \if #1\bgroup\xint_dothis\XINT_ListSel_check_is_ok\fi
    \xint_orthat\XINT_ListSel_check_leaf
}%
\def\XINT_ListSel_check_leaf #1\expandafter{\expandafter}%
\def\XINT_ListSel_check_is_ok
{%
    \expandafter\XINT_ListSel_check_is_ok_a\expandafter{\string}%
}%
\def\XINT_ListSel_check_is_ok_a #1__#2%
{%
    \if :#2\xint_dothis{\XINT_ListSel_slice}\fi
    \xint_orthat {\XINT_ListSel_nthone {#2}}%
}%
\def\XINT_ListSel_top #1#2%
{%
    \if _\noexpand#2%
           \expandafter\XINT_ListSel_top_one_or_none\string#1.\else
           \expandafter\XINT_ListSel_top_at_least_two\fi
}%
\def\XINT_ListSel_top_at_least_two #1__{\XINT_ListSel_top_ople}%
\def\XINT_ListSel_top_one_or_none #1%
{%
    \if #1_\xint_dothis\XINT_ListSel_top_nil\fi
    \if #1.\xint_dothis\XINT_ListSel_top_nutple_a\fi
    \if #1\bgroup\xint_dothis\XINT_ListSel_top_nutple\fi
    \xint_orthat\XINT_ListSel_top_number
}%
\def\XINT_ListSel_top_nil #1\expandafter#2\expandafter{\fi\expandafter}%
\def\XINT_ListSel_top_nutple
{%
    \expandafter\XINT_ListSel_top_nutple_a\expandafter{\string}%
}%
\def\XINT_ListSel_top_nutple_a #1_#2#3(#4%
{%
    \fi\if :#2\xint_dothis{{\XINT_ListSel_slice #3(#4}}\fi
    \xint_orthat {\XINT_ListSel_nthone {#2}#3(#4}%
}%
\def\XINT_ListSel_top_number #1_{\fi\XINT_ListSel_top_ople}%
\def\XINT_ListSel_top_ople #1%
{%
    \if :#1\xint_dothis\XINT_ListSel_slice\fi
    \xint_orthat {\XINT_ListSel_nthone {#1}}%
}%
\def\XINT_ListSel_slice #1%
{%
    \expandafter\XINT_ListSel_slice_a \expandafter{\romannumeral0\xintnum{#1}}%
}%
\def\XINT_ListSel_slice_a #1#2;#3#4%
{%
   \if _#4\expandafter\XINT_ListSel_s_b
          \else\expandafter\XINT_ListSel_slice_b\fi 
   #1;#3%
}%
\def\XINT_ListSel_s_b #1#2;#3#4%
{%
    \if &#4\expandafter\XINT_ListSel_s_last\fi
    \XINT_ListSel_s_c #1{#1#2}{#4}%
}%
\def\XINT_ListSel_s_last\XINT_ListSel_s_c #1#2#3(#4%
{%
   \if-#1\expandafter\xintKeep\else\expandafter\xintTrim\fi {#2}{#4}%
}%
\def\XINT_ListSel_s_c #1#2#3(#4%
{%
   \expandafter\XINT_ListSel_deeper
   \expanded{\unexpanded{#3}(\expandafter}\expandafter{%
   \romannumeral0%
   \if-#1\expandafter\xintkeep\else\expandafter\xinttrim\fi {#2}{#4}}%
}%
%    \end{macrocode}
% \lverb|&
%
% \xintNthElt from xinttools (knowingly) strips one level of
% braces when fetching kth «item» from {v1}...{vN}. If we expand
% {\xintNthElt{k}{{v1}...{vN}}} (notice external braces):
%
%( if k is out of range we end up with {}
%: if k is in range and the kth braced item was {} we end up with {}
%: if k is in range and the kth braced item was {17} we end up with {17}
%)
%
% Problem is that individual numbers such as 17 are stored {{17}}. So
% we must have one more brace pair and in the first two cases we end up
% with {{}}. But in the first case we should end up with the empty ople
% {}, not the empty bracketed ople {{}}.
%
% I have thus added \xintNthOne to $xinttoolsnameimp which does not
% strip brace pair from an extracted item.
%
% Attention: \XINT_nthonepy_a does no expansion on second argument.
% But here arguments are either numerical or already expanded.
% Normally.
%
% |
%    \begin{macrocode}
\def\XINT_ListSel_nthone #1#2%
{%
    \if &#2\expandafter\XINT_ListSel_nthone_last\fi
    \XINT_ListSel_nthone_a {#1}{#2}%
}%
\def\XINT_ListSel_nthone_a #1#2(#3%
{%
    \expandafter\XINT_ListSel_deeper
    \expanded{\unexpanded{#2}(\expandafter}\expandafter{%
    \romannumeral0\expandafter\XINT_nthonepy_a\the\numexpr\xintNum{#1}.{#3}}%
}%
\def\XINT_ListSel_nthone_last\XINT_ListSel_nthone_a #1#2(%#3%
{%
    \romannumeral0\expandafter\XINT_nthonepy_a\the\numexpr\xintNum{#1}.%{#3}
}%
%    \end{macrocode}
% \lverb|The macros here are basically f-expandable and use the
% f-expandable \xintKeep and \xintTrim. Prior to xint 1.4, there was
% here an x-expandable \xintKeep:x:csv dealing with comma separated
% items, for time being we make do with our f-expandable toolkit.
% |
%    \begin{macrocode}
\def\XINT_ListSel_slice_b #1;#2_#3%
{%
    \if &#3\expandafter\XINT_ListSel_slice_last\fi
    \expandafter\XINT_ListSel_slice_c \expandafter{\romannumeral0\xintnum{#2}};#1;{#3}%
}%
\def\XINT_ListSel_slice_last\expandafter\XINT_ListSel_slice_c #1;#2;#3(%#4
{%
   \expandafter\XINT_ListSel_slice_last_c #1;#2;%{#4}
}%
\def\XINT_ListSel_slice_last_c #1;#2;#3%
{%
    \romannumeral0\XINT_ListSel_slice_d #2;#1;{#3}%
}%
\def\XINT_ListSel_slice_c #1;#2;#3(#4%
{%
   \expandafter\XINT_ListSel_deeper
   \expanded{\unexpanded{#3}(\expandafter}\expandafter{%
   \romannumeral0\XINT_ListSel_slice_d #2;#1;{#4}}%
}%
\def\XINT_ListSel_slice_d #1#2;#3#4;%
{%
    \xint_UDsignsfork
       #1#3\XINT_ListSel_N:N
        #1-\XINT_ListSel_N:P
        -#3\XINT_ListSel_P:N
         --\XINT_ListSel_P:P
    \krof #1#2;#3#4;%
}%
\def\XINT_ListSel_P:P #1;#2;#3%
{%
    \unless\ifnum #1<#2 \expandafter\xint_gob_andstop_iii\fi
    \xintkeep{#2-#1}{\xintTrim{#1}{#3}}%
}%
\def\XINT_ListSel_N:N #1;#2;#3%
{%
    \expandafter\XINT_ListSel_N:N_a
    \the\numexpr #2-#1\expandafter;\the\numexpr#1+\xintLength{#3};{#3}%
}%
\def\XINT_ListSel_N:N_a #1;#2;#3%
{%
    \unless\ifnum #1>\xint_c_ \expandafter\xint_gob_andstop_iii\fi
    \xintkeep{#1}{\xintTrim{\ifnum#2<\xint_c_\xint_c_\else#2\fi}{#3}}%
}%
\def\XINT_ListSel_N:P #1;#2;#3%
{%
    \expandafter\XINT_ListSel_N:P_a
    \the\numexpr #1+\xintLength{#3};#2;{#3}%
}%
\def\XINT_ListSel_N:P_a #1#2;%
   {\if -#1\expandafter\XINT_ListSel_O:P\fi\XINT_ListSel_P:P #1#2;}%
\def\XINT_ListSel_O:P\XINT_ListSel_P:P #1;{\XINT_ListSel_P:P 0;}%
\def\XINT_ListSel_P:N #1;#2;#3%
{%
    \expandafter\XINT_ListSel_P:N_a
    \the\numexpr #2+\xintLength{#3};#1;{#3}%
}%
\def\XINT_ListSel_P:N_a #1#2;#3;%
   {\if -#1\expandafter\XINT_ListSel_P:O\fi\XINT_ListSel_P:P #3;#1#2;}%
\def\XINT_ListSel_P:O\XINT_ListSel_P:P #1;#2;{\XINT_ListSel_P:P #1;0;}%
%    \end{macrocode}
% \subsection{Support for raw A/B[N]}
% \lverb|
% Releases earlier than 1.1 required the use of braces around A/B[N]
% input. The [N] is now implemented directly. *BUT* this uses a delimited macro!
% thus N is not allowed to be itself an expression (I could add it...).
% \xintE, \xintiiE, and \XINTinFloatE all put #2 in a \numexpr. But attention
% to the fact that \numexpr stops at spaces separating digits:
% \the\numexpr 3 + 7 9\relax gives 109\relax !! Hence we have to be
% careful.
%
% \numexpr will not handle catcode 11 digits, but adding a \detokenize will
% suddenly make illicit for N to rely on macro expansion.
%
% At 1.4, [ is already overloaded and it is not easy to support this. We do
% this by a kludge maintaining more or less former (very not efficient) way
% but using $$ sign which is free for time being. No, finally I use the null
% character, should be safe enough! (I hesitated about using R with catcode
% 12).
%
% As for ? operator we needed
% to hack into \XINT_expr_getop_b for intercepting that pseudo operator. See
% also \XINT_expr_scanint_c (\XINT_expr_rawxintfrac).|
%    \begin{macrocode}
\catcode0 11
\let\XINT_expr_precedence_&&@ \xint_c_xiv
\def\XINT_expr_op_&&@ #1#2]%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded{{{\xintE#1{\xint_zapspaces #2 \xint_gobble_i}}}%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_iiexpr_op_&&@ #1#2]%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded{{{\xintiiE#1{\xint_zapspaces #2 \xint_gobble_i}}}%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_flexpr_op_&&@ #1#2]%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded{{{\XINTinFloatE#1{\xint_zapspaces #2 \xint_gobble_i}}}%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\catcode0 12
%    \end{macrocode}
% \subsection{? as two-way and ?? as three-way «short-circuit» conditionals}
% \lverb|
% Comments undergoing reconstruction.
% |
%
%    \begin{macrocode}
\let\XINT_expr_precedence_? \xint_c_xx
\catcode`- 11
\def\XINT_expr_op_?  {\XINT_expr_op__? \XINT_expr_op_-xii}%
\def\XINT_flexpr_op_?{\XINT_expr_op__? \XINT_flexpr_op_-xii}%
\def\XINT_iiexpr_op_?{\XINT_expr_op__? \XINT_iiexpr_op_-xii}%
\catcode`- 12
\def\XINT_expr_op__? #1#2#3%
    {\XINT_expr_op__?_a #3!\xint_bye\XINT_expr_exec_? {#1}{#2}{#3}}%
\def\XINT_expr_op__?_a #1{\expandafter\XINT_expr_op__?_b\detokenize{#1}}%
\def\XINT_expr_op__?_b #1%
    {\if ?#1\expandafter\XINT_expr_op__?_c\else\expandafter\xint_bye\fi }%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_expr_op__?_c #1{\xint_gob_til_! #1\XINT_expr_op_?? !\xint_bye}%
\def\XINT_expr_op_?? !\xint_bye\xint_bye\XINT_expr_exec_?{\XINT_expr_exec_??}%
\catcode`- 11
\def\XINT_expr_exec_? #1#2%
{%
    \expandafter\XINT_expr_check-_after?\expandafter#1%
    \romannumeral`&&@\expandafter\XINT_expr_getnext\romannumeral0\xintiiifnotzero#2%
}%
\def\XINT_expr_exec_?? #1#2#3%
{%
    \expandafter\XINT_expr_check-_after?\expandafter#1%
    \romannumeral`&&@\expandafter\XINT_expr_getnext\romannumeral0\xintiiifsgn#2%
}%
\def\XINT_expr_check-_after? #1{%
\def\XINT_expr_check-_after? ##1##2%
{%
    \xint_UDsignfork
        ##2{##1}%
        #1{##2}%
    \krof
}}\expandafter\XINT_expr_check-_after?\string -%
\catcode`- 12
%    \end{macrocode}
% \subsection{! as postfix factorial operator}
% \lverb|&
% |
%    \begin{macrocode}
\let\XINT_expr_precedence_! \xint_c_xx
\def\XINT_expr_op_! #1%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded{{\romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintFac#1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_flexpr_op_! #1%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded{{\romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatFacdigits#1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_iiexpr_op_! #1%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded{{\romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiFac#1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
%    \end{macrocode}
% \lverb|At 1.4g, fix for input "x! == y" via a fake operator !==. The !
% is of catcode 11 but this does not matter here. The definition of
% \XINT_expr_itself_!== is required by the functioning of the scanop macros.
%
% We don't have to worry about "x! = y" as the single-character Boolean
% comparison = operator has been removed from syntax.  Fixing it would have
% required obeying space tokens when parsing operators.  For "x! == y" case,
% obeying space tokens would not solve "x!==y" input case anyhow.|
%    \begin{macrocode}
\expandafter
\def\csname XINT_expr_precedence_!==\expandafter\endcsname
    \csname XINT_expr_itself_!==\endcsname {\XINT_expr_precedence_! !==}%
\expandafter\def\csname XINT_expr_itself_!==\endcsname{!==}%
%    \end{macrocode}
% \subsection{User defined variables}
% \localtableofcontents
%
% \subsubsection{\csh{xintdefvar}, \csh{xintdefiivar},
% \csh{xintdeffloatvar}}
% \changed{1.1}
%
% \changed[2017/12/01]{1.2p} Extends |\xintdefvar| et al.\@ to accept
% simultaneous assignments to multiple variables.
%
% \changed[2018/06/17]{1.3c}
% Use \csbxint{exprSafeCatcodes} (to palliate issue with
% active semi-colon from Babel+French if in body of a \LaTeX{} document).
%
% And allow usage with both syntaxes |name:=expr;| or |name=expr;|. Also the
% colon may have catcode 11, 12, or 13 with  no issue.  
% Variable names may contain letters, digits, underscores, and must not start
% with a digit. Names starting with |@| or an underscore are reserved.
%
% \begin{itemize}[nosep]
%   \item currently |@|, |@1|, |@2|, |@3|, and |@4| are reserved because they
%     have special meanings for use in iterations,
%   \item |@@|, |@@@|, |@@@@| are also reserved but
% are technically functions, not variables: a user may possibly define |@@| as
% a variable name, but if it is followed by parentheses, the function
% interpretation will be applied (rather than the variable interpretation
% followed by a tacit multiplication),
%   \item since |1.2l|, the underscore |_| may be used as separator of digits in
%     long numbers.
%     Hence a variable whose name starts with |_| will not play well with the
%     mechanism of tacit multiplication of variables by numbers: the underscore
%     will be removed from input stream by the number scanner, thus creating 
%     an undefined or wrong variable name, or none at all if the variable
%     name was an initial |_| followed by digits.
% \end{itemize}
%
% \lverb|Note that the optional argument [P] as usable with
%  \xintfloatexpr is **not**
% supported by \xintdeffloatvar. One must do \xintdeffloatvar
% foo = \xintfloatexpr[16] blabla \relax; to achieve the effect.|
%
% \changed[2020/01/27]{1.4}
% The expression will be fetched up to final semi-colon in a manner
% allowing inner semi-colons as used in the iter(), rseq(), subsm(),
% subsn() etc... syntax. They don't need to be hidden within a 
% braced pair anymore.
%
% \changed{1.4}
% Automatic unpacking in case of simultaneous assignments if the expression
% evaluates to a nutple.
%
% Notes (added much later on 2021/06/10 during preparation of |1.4i|):
% \begin{enumerate}
% \item the code did not try to intercept illicit syntax such as
%   |\xintdefvar a,b,c:=<number>;|. It blindly «unpacked» the number handling it as if
%   it was a nutple.  The extended functionality added at |1.4i| requires to
%   check for such a situation, as the syntax is not illicit anymore.
% \item the code was broken in case the expression to evaluate was an ople of
%   length 10 or more, due to a silly mistake at some point during |1.4|
%   development which replaced some |\ifnum| by an |\if|, perhaps due to
%   mental confusion with the fact that functions can have at most 9
%   arguments, but here the code is about defining variables.  Anyway this got
%   fixed as corollary to the |1.4i| extension.
% \end{enumerate}
%
%
% \changed[2021/02/20]{1.4c} One year later I realized I had broken tacit
% multiplication for situations such as |variable(1+2)|.  As hinted at in
% comments above before |1.4| release I had been doing some deep refactoring
% here, which I cancelled almost completely in the end... but not quite, and
% as a result there was a problem that some macro holding braced contents was
% expanded to late, once it was in old core routines of xintfrac not expecting
% other things than digits. I do an emergency bugfix here with some
% |\expandafter|'s but I don't have the code in my brain at this time, and
% don't have the luxury now to invest into it. Let's hope this does not induce
% breakage elsewhere, and that the February 2020 |1.4| did not break something
% else.
% 
% \changed[2021/04/17]{1.4e}
% \lverb|&
% Modifies \xintdeffloatvar to round to the prevailing
% precision (formerly, any operation would induce rounding, but in case of
% things such as \xintdeffloatvar foo:=\xintexpr 1/100!\relax; there was no
% automatic rounding.  One could use 0+ syntax to trigger it, and for oples,
% some trick like \xintfloatexpr[\XINTdigits]...\relax extra wrapper.
% |
% \changed[2021/05/22]{1.4g}
% \lverb|&
% The \expandafter\expandafter\expandafter et al. chain
% which was kept by \XINT_expr_defvar_one_b for expanding only at time of use
% the \XINT_expr_var_foo in \XINT_expr_onliteral_foo were senseless overhead
% added at 1.4c. This is used only for real variables, not dummy variables or
% fake variables and it is simpler to have the \XINT_expr_var_foo
% pre-expanded. So let's use some \edef here.
%
% The \XINT_expr_onliteral_foo is expanded as result of action of \XINT_expr_op_`
% (or \XINT_flexpr_op_`, \XINT_iiexpr_op_`) which itself was triggered consuming
% already an \XINT_expr_put_op_first, so its expansion has to produce tokens
% as expected after \XINT_expr_put_op_first: <precedence token><op token>{expanded value}.
% 
% |
%
% \changed[2021/06/10]{1.4i}
%
% Implement extended notion of simultaneous assignments: if there are more
% variables than values, define the extra variables to be nil. If there are
% less variables than values let the last variable
% be defined as the ople concatenating all non reclaimed values.
%
% If there are at least two variables, the right hand side, if it turns out to
% be a nutple, is (as since |1.4|) automatically unpacked, then the above
% rules apply.
%
% \changed[2021/06/11]{1.4i}
%
% \lverb|Fix the long-standing «seq renaming bug» via a change here of the
% name of auxiliary macro.  Previously «onliteral_<varname>» now
% «var*_<varname>». I hesitated with using «var_varname*» rather.
%
% Hesitated adding \XINT_expr_letvar_one (motivation: case of simultaneous
% assignments leading to defining «nil» variables).  Finally, no.|
%
%    \begin{macrocode}
\catcode`* 11
\def\XINT_expr_defvar_one #1#2%
{%
    \XINT_global
    \expandafter\edef\csname XINT_expr_varvalue_#1\endcsname {#2}%
    \XINT_expr_defvar_one_b {#1}%
}%
\def\XINT_expr_defvar_one_b #1%
{%
    \XINT_global
    \expandafter\edef\csname XINT_expr_var_#1\endcsname
        {{\expandafter\noexpand\csname XINT_expr_varvalue_#1\endcsname}}%
    \XINT_global
    \expandafter\edef\csname XINT_expr_var*_#1\endcsname
        {\XINT_expr_prec_tacit *\csname XINT_expr_var_#1\endcsname(}%
    \ifxintverbose\xintMessage{xintexpr}{Info}
       {Variable #1 \ifxintglobaldefs globally \fi
        defined with value \csname XINT_expr_varvalue_#1\endcsname.}%
    \fi
}%
\catcode`* 12
\catcode`~ 13
\catcode`: 12
\def\XINT_expr_defvar_getname #1:#2~%
{%
    \endgroup
    \def\XINT_defvar_tmpa{#1}\edef\XINT_defvar_tmpc{\xintCSVLength{#1}}%
}%
\def\XINT_expr_defvar #1#2%
{%
    \def\XINT_defvar_tmpa{#2}%
    \expandafter\XINT_expr_defvar_a\expanded{\unexpanded{{#1}}\expandafter}%
    \romannumeral\XINT_expr_fetch_to_semicolon
}%
\def\XINT_expr_defvar_a #1#2%
{%
    \xintexprRestoreCatcodes
%    \end{macrocode}
% \lverb|Maybe SafeCatcodes was without effect because the colon and the rest
% are from some earlier macro definition. Give a safe definition to active
% colon (even if in math mode with a math active colon..).
%
% The \XINT_expr_defvar_getname closes the group opened here.|
%    \begin{macrocode}
    \begingroup\lccode`~`: \lowercase{\let~}\empty
    \edef\XINT_defvar_tmpa{\XINT_defvar_tmpa}%
    \edef\XINT_defvar_tmpa{\xint_zapspaces_o\XINT_defvar_tmpa}%
    \expandafter\XINT_expr_defvar_getname
                \detokenize\expandafter{\XINT_defvar_tmpa}:~%
    \ifcase\XINT_defvar_tmpc\space
      \xintMessage {xintexpr}{Error}
      {Aborting: not allowed to declare variable with empty name.}%
    \or
     \XINT_global
     \expandafter
     \edef\csname XINT_expr_varvalue_\XINT_defvar_tmpa\endcsname{#1#2\relax}%
     \XINT_expr_defvar_one_b\XINT_defvar_tmpa
    \else
     \edef\XINT_defvar_tmpb{#1#2\relax}%
     \edef\XINT_defvar_tmpd{\expandafter\xintLength\expandafter{\XINT_defvar_tmpb}}%
     \ifnum\XINT_defvar_tmpd=\xint_c_i
         \oodef\XINT_defvar_tmpb{\expandafter\xint_firstofone\XINT_defvar_tmpb}%
         \if0\expandafter\expandafter\expandafter\XINT_defvar_checkifnutple
                     \expandafter\string\XINT_defvar_tmpb _\xint_bye
           \odef\XINT_defvar_tmpb{\expandafter{\XINT_defvar_tmpb}}%
         \else
           \edef\XINT_defvar_tmpd{\expandafter\xintLength\expandafter{\XINT_defvar_tmpb}}%
         \fi
     \fi
     \xintAssignArray\xintCSVtoList\XINT_defvar_tmpa\to\XINT_defvar_tmpvar
     \def\XINT_defvar_tmpe{1}%
     \expandafter\XINT_expr_defvar_multiple\XINT_defvar_tmpb\relax
   \fi
}%
\def\XINT_defvar_checkifnutple#1%
{%
    \if#1_1\fi
    \if#1\bgroup1\fi
    0\xint_bye
}%
\def\XINT_expr_defvar_multiple
{%
    \ifnum\XINT_defvar_tmpe<\XINT_defvar_tmpc\space
        \expandafter\XINT_expr_defvar_multiple_one
    \else
        \expandafter\XINT_expr_defvar_multiple_last\expandafter\empty
    \fi
}%
\def\XINT_expr_defvar_multiple_one
{%
    \ifnum\XINT_defvar_tmpe>\XINT_defvar_tmpd\space
       \expandafter\XINT_expr_defvar_one
       \csname XINT_defvar_tmpvar\XINT_defvar_tmpe\endcsname{}%
       \edef\XINT_defvar_tmpe{\the\numexpr\XINT_defvar_tmpe+1}%
       \expandafter\XINT_expr_defvar_multiple
    \else
       \expandafter\XINT_expr_defvar_multiple_one_a
    \fi
}%
\def\XINT_expr_defvar_multiple_one_a #1%
{%
    \expandafter\XINT_expr_defvar_one
       \csname XINT_defvar_tmpvar\XINT_defvar_tmpe\endcsname{{#1}}%
    \edef\XINT_defvar_tmpe{\the\numexpr\XINT_defvar_tmpe+1}%
    \XINT_expr_defvar_multiple
}%
\def\XINT_expr_defvar_multiple_last #1\relax
{%
    \expandafter\XINT_expr_defvar_one
       \csname XINT_defvar_tmpvar\XINT_defvar_tmpe\endcsname{#1}%
    \xintRelaxArray\XINT_defvar_tmpvar
    \let\XINT_defvar_tmpa\empty
    \let\XINT_defvar_tmpb\empty
    \let\XINT_defvar_tmpc\empty
    \let\XINT_defvar_tmpd\empty
    \let\XINT_defvar_tmpe\empty
}%
\catcode`~ 3
\catcode`: 11
%    \end{macrocode}
% \lverb|This SafeCatcodes is mainly in the hope that semi-colon ending the
% expression can still be sanitized.
% 
% Pre 1.4e definition:
%(\def\xintdeffloatvar      {\xintexprSafeCatcodes\xintdeffloatvar_a}%
%:\def\xintdeffloatvar_a #1={\XINT_expr_defvar\xintthebarefloateval{#1}}%
%)
% This would keep the value (or values) with extra digits, now.
% If this is actually wanted one can use
% \xintdefvar foo:=\xintfloatexpr...\relax;
% syntax, but recalling that only operations trigger the rounding inside
% \xintfloatexpr. Some tricks are needed for no operations case if multiple or
% nested values. But for a single one, one can use simply the float()
% function.
% 
% |
%    \begin{macrocode}
\def\xintdefvar     {\xintexprSafeCatcodes\xintdefvar_a}%
\def\xintdefvar_a#1={\XINT_expr_defvar\xintthebareeval{#1}}%
\def\xintdefiivar     {\xintexprSafeCatcodes\xintdefiivar_a}%
\def\xintdefiivar_a#1={\XINT_expr_defvar\xintthebareiieval{#1}}%
\def\xintdeffloatvar  {\xintexprSafeCatcodes\xintdeffloatvar_a}%
\def\xintdeffloatvar_a #1={\XINT_expr_defvar\xintthebareroundedfloateval{#1}}%
%    \end{macrocode}
% \subsubsection{\csh{xintunassignvar}}
% \changed{1.2e}
%
% \changed{1.3d}
% Embarrassingly I had for a long time a misunderstanding of |\ifcsname|
% (let's blame its documentation) and I was not aware that it chooses FALSE
% branch if tested control sequence has been |\let| to |\undefined|... So
% earlier version didn't do the right thing (and had another bug: failure to
% protect |\.=0| from expansion).
%
% The |\ifcsname| tests are done in \csbXINT{_expr_op__} and
% \csbXINT{_expr_op_`}.
%
% \changed{1.4i}
% Track |s/onliteral/var*/| change in macro names.
%    \begin{macrocode}
\def\xintunassignvar #1{%
   \edef\XINT_unvar_tmpa{#1}%
   \edef\XINT_unvar_tmpa {\xint_zapspaces_o\XINT_unvar_tmpa}%
   \ifcsname XINT_expr_var_\XINT_unvar_tmpa\endcsname
       \ifnum\expandafter\xintLength\expandafter{\XINT_unvar_tmpa}=\@ne
         \expandafter\xintnewdummy\XINT_unvar_tmpa
       \else
         \XINT_global\expandafter
           \let\csname XINT_expr_varvalue_\XINT_unvar_tmpa\endcsname\xint_undefined
         \XINT_global\expandafter
           \let\csname XINT_expr_var_\XINT_unvar_tmpa\endcsname\xint_undefined
         \XINT_global\expandafter
           \let\csname XINT_expr_var*_\XINT_unvar_tmpa\endcsname\xint_undefined
         \ifxintverbose\xintMessage {xintexpr}{Info}
           {Variable \XINT_unvar_tmpa\space has been
            \ifxintglobaldefs globally \fi ``unassigned''.}%
         \fi
       \fi
   \else
       \xintMessage {xintexpr}{Warning}
           {Error: there was no such variable \XINT_unvar_tmpa\space to unassign.}%
   \fi
}%
%    \end{macrocode}
% \subsection{Support for dummy variables}
% \localtableofcontents
% \subsubsection{\csh{xintnewdummy}}
% \lverb|&
% Comments under reconstruction.
%
% 1.4 adds multi-letter names as usable dummy variables!
% |
% \changed[2021/06/11]{1.4i}
% \lverb|s/onliteral/var*/ to fix the «seq renaming bug».|
%
%
%    \begin{macrocode}
\catcode`* 11
\def\XINT_expr_makedummy #1%
{%
   \edef\XINT_tmpa{\xint_zapspaces #1 \xint_gobble_i}%
   \ifcsname XINT_expr_var_\XINT_tmpa\endcsname
     \XINT_global
     \expandafter\let\csname XINT_expr_var_\XINT_tmpa/old\expandafter\endcsname
                     \csname XINT_expr_var_\XINT_tmpa\expandafter\endcsname
   \fi
   \ifcsname XINT_expr_var*_\XINT_tmpa\endcsname
     \XINT_global
     \expandafter\let\csname XINT_expr_var*_\XINT_tmpa/old\expandafter\endcsname
                     \csname XINT_expr_var*_\XINT_tmpa\expandafter\endcsname
   \fi
   \expandafter\XINT_global
   \expanded
   {\edef\expandafter\noexpand
         \csname XINT_expr_var_\XINT_tmpa\endcsname ##1\relax !\XINT_tmpa##2}%
      {{##2}##1\relax !\XINT_tmpa{##2}}%
   \expandafter\XINT_global
   \expanded
   {\edef\expandafter\noexpand
         \csname XINT_expr_var*_\XINT_tmpa\endcsname ##1\relax !\XINT_tmpa##2}%
      {\XINT_expr_prec_tacit *{##2}(##1\relax !\XINT_tmpa{##2}}%)
}%
\xintApplyUnbraced \XINT_expr_makedummy {abcdefghijklmnopqrstuvwxyz}%
\xintApplyUnbraced \XINT_expr_makedummy {ABCDEFGHIJKLMNOPQRSTUVWXYZ}%
\def\xintnewdummy #1{%
    \XINT_expr_makedummy{#1}%
    \ifxintverbose\xintMessage {xintexpr}{Info}%
       {\XINT_tmpa\space now
        \ifxintglobaldefs globally \fi usable as dummy variable.}%
    \fi
}%
\catcode`* 12
%    \begin{macrocode}
% The |nil| variable was need in |xint < 1.4| (with some other meaning)
% in places the syntax could not allow emptiness, such as |,,|, and
% other things, but at |1.4| meaning as changed.
%
% The other variables are new with |1.4|.
% Don't use the |None|, it is tentative, and may be input as |[]|.
%
% Refactored at |1.4i| to define them as really genuine variables,
% i.e. also with associated |var*| macros involved in tacit multiplication
% (even though it will be broken with |nil|, and with |None| in |\xintiiexpr|).
% No real reason, because |\XINT_expr_op__| managed them fine even in absence
% of |var*| macros.
%    \begin{macrocode}
\XINT_expr_defvar_one{nil}{}%
\XINT_expr_defvar_one{None}{{}}% ? tentative
\XINT_expr_defvar_one{false}{{0}}% Maple, TeX
\XINT_expr_defvar_one{true}{{1}}%
\XINT_expr_defvar_one{False}{{0}}% Python
\XINT_expr_defvar_one{True}{{1}}%
%    \end{macrocode}
% \subsubsection{\csh{xintensuredummy}, \csh{xintrestorevariable}}
% \lverb|1.3e \xintensuredummy differs from \xintnewdummy only in the informational message...
% Attention that this is not meant to be nested.
%
% 1.4 fixes that the message mentioned non-existent \xintrestoredummy (real
% name was \xintrestorelettervar and renames the latter to
% \xintrestorevariable as it applies also to multi-letter names.|
%    \begin{macrocode}
\def\xintensuredummy #1{%
    \XINT_expr_makedummy{#1}%
    \ifxintverbose\xintMessage {xintexpr}{Info}%
       {\XINT_tmpa\space now
        \ifxintglobaldefs globally \fi usable as dummy variable.&&J
        Issue \string\xintrestorevariable{\XINT_tmpa} to restore former meaning.}%
    \fi
}%
\def\xintrestorevariablesilently #1{%
   \edef\XINT_tmpa{\xint_zapspaces #1 \xint_gobble_i}%
   \ifcsname XINT_expr_var_\XINT_tmpa/old\endcsname
     \XINT_global
     \expandafter\let\csname XINT_expr_var_\XINT_tmpa\expandafter\endcsname
                     \csname XINT_expr_var_\XINT_tmpa/old\expandafter\endcsname
   \fi
   \ifcsname XINT_expr_var*_\XINT_tmpa/old\endcsname
     \XINT_global
     \expandafter\let\csname XINT_expr_var*_\XINT_tmpa\expandafter\endcsname
                     \csname XINT_expr_var*_\XINT_tmpa/old\expandafter\endcsname
   \fi
}%
\def\xintrestorevariable #1{%
    \xintrestorevariablesilently {#1}%
    \ifxintverbose\xintMessage {xintexpr}{Info}%
       {\XINT_tmpa\space
        \ifxintglobaldefs globally \fi restored to its earlier status, if any.}%
    \fi
}%
%    \end{macrocode}
% \subsubsection{Checking (without expansion) that a symbolic expression
% contains correctly nested parentheses}
%
% \lverb|Expands to \xint_c_mone in case a closing ) had no opening ( matching
% it, to \@ne if opening ) had no closing ) matching it, to \z@ if expression
% was balanced. Call it as:
%
% \XINT_isbalanced_a \relax #1(\xint_bye)\xint_bye
%
% This is legacy f-expandable code not using \expanded even at 1.4.
% |
%    \begin{macrocode}
\def\XINT_isbalanced_a #1({\XINT_isbalanced_b #1)\xint_bye }%
\def\XINT_isbalanced_b #1)#2%
   {\xint_bye #2\XINT_isbalanced_c\xint_bye\XINT_isbalanced_error }%
%    \end{macrocode}
% \lverb|if #2 is not \xint_bye, a ) was found, but there was no (. Hence error -> -1|
%    \begin{macrocode}
\def\XINT_isbalanced_error #1)\xint_bye {\xint_c_mone}%
%    \end{macrocode}
% \lverb|#2 was \xint_bye, was there a ) in original #1?|
%    \begin{macrocode}
\def\XINT_isbalanced_c\xint_bye\XINT_isbalanced_error #1%
    {\xint_bye #1\XINT_isbalanced_yes\xint_bye\XINT_isbalanced_d #1}%
%    \end{macrocode}
% \lverb|#1 is \xint_bye, there was never ( nor ) in original #1, hence OK.|
%    \begin{macrocode}
\def\XINT_isbalanced_yes\xint_bye\XINT_isbalanced_d\xint_bye )\xint_bye {\xint_c_ }%
%    \end{macrocode}
% \lverb|#1 is not \xint_bye, there was indeed a ( in original #1. We check if
% we see a ). If we do, we then loop until no ( nor ) is to be found.|
%    \begin{macrocode}
\def\XINT_isbalanced_d #1)#2%
   {\xint_bye #2\XINT_isbalanced_no\xint_bye\XINT_isbalanced_a #1#2}%
%    \end{macrocode}
% \lverb|#2 was \xint_bye, we did not find a closing ) in original #1. Error.|
%    \begin{macrocode}
\def\XINT_isbalanced_no\xint_bye #1\xint_bye\xint_bye {\xint_c_i }%
%    \end{macrocode}
% \subsubsection{Fetching balanced expressions E1, E2 and a variable name Name from E1, Name=E2)}
%
% Multi-letter dummy variables added at |1.4|.
%    \begin{macrocode}
\def\XINT_expr_fetch_E_comma_V_equal_E_a #1#2,%
{%
    \ifcase\XINT_isbalanced_a \relax #1#2(\xint_bye)\xint_bye
           \expandafter\XINT_expr_fetch_E_comma_V_equal_E_c
        \or\expandafter\XINT_expr_fetch_E_comma_V_equal_E_b
      \else\expandafter\xintError:noopening
    \fi {#1#2},%
}%
\def\XINT_expr_fetch_E_comma_V_equal_E_b #1,%
   {\XINT_expr_fetch_E_comma_V_equal_E_a {#1,}}%
\def\XINT_expr_fetch_E_comma_V_equal_E_c #1,#2#3=%
{%
    \expandafter\XINT_expr_fetch_E_comma_V_equal_E_d\expandafter
    {\expanded{{\xint_zapspaces #2#3 \xint_gobble_i}}{#1}}{}%
}%
\def\XINT_expr_fetch_E_comma_V_equal_E_d #1#2#3)%
{%
    \ifcase\XINT_isbalanced_a \relax #2#3(\xint_bye)\xint_bye
        \or\expandafter\XINT_expr_fetch_E_comma_V_equal_E_e
       \else\expandafter\xintError:noopening
    \fi
    {#1}{#2#3}%
}%
\def\XINT_expr_fetch_E_comma_V_equal_E_e #1#2{\XINT_expr_fetch_E_comma_V_equal_E_d {#1}{#2)}}%
%    \end{macrocode}
% \subsubsection{Fetching a balanced expression delimited by a semi-colon}
%
% \lverb|1.4. For subsn() leaner syntax of nested substitutions.
%
% Will also serve to \xintdeffunc, to not have to hide inner semi-colons in
% for example an iter() from \xintdeffunc.
%
% Adding brace removal protection for no serious reason, anyhow the xintexpr
% parsers always removes braces when moving forward, but well.
%
% Trigger by \romannumeral\XINT_expr_fetch_to_semicolon upfront.|
%
%    \begin{macrocode}
\def\XINT_expr_fetch_to_semicolon {\XINT_expr_fetch_to_semicolon_a {}\empty}%
\def\XINT_expr_fetch_to_semicolon_a #1#2;%
{%
    \ifcase\XINT_isbalanced_a \relax #1#2(\xint_bye)\xint_bye
           \xint_dothis{\expandafter\XINT_expr_fetch_to_semicolon_c}%
        \or\xint_dothis{\expandafter\XINT_expr_fetch_to_semicolon_b}%
      \else\expandafter\xintError:noopening
    \fi\xint_orthat{}\expandafter{#2}{#1}%
}%
\def\XINT_expr_fetch_to_semicolon_b #1#2{\XINT_expr_fetch_to_semicolon_a {#2#1;}\empty}%
\def\XINT_expr_fetch_to_semicolon_c #1#2{\xint_c_{#2#1}}%
%    \end{macrocode}
% \subsubsection{Low-level support for omit and abort keywords, the break()
% function, the n++ construct and the semi-colon as used in the syntax of
% seq(), add(), mul(), iter(), rseq(), iterr(), rrseq(), subsm(), subsn(), ndseq(),
% ndmap()}
% \lverb|There is some clever play simply based on setting suitable precedence
% levels combined with special meanings given to op macros.
%
%  The special !? internal operator is a helper for omit and abort keywords in
%  list generators.
%
% Prior to 1.4 support for +[, *[, ..., ]+, ]*,  had some elements here.
% |
% \paragraph{The n++ construct}
% \lverb|1.1 2014/10/29 did \expandafter\.=+\xintiCeil which transformed it into
% \romannumeral0\xinticeil, which seems a bit weird. This exploited the fact
% that dummy variables macros could back then pick braced material (which in the
% case at hand here ended being {\romannumeral0\xinticeil...} and were submitted
% to two expansions. The result of this was to provide a not value which got
% expanded only in the first loop of the :_A and following macros of seq,
% iter, rseq, etc...
%
% Anyhow with 1.2c I have changed the implementation of dummy variables which
% now need to fetch a single locked token, which they do not expand.
%
% The \xintiCeil appears a bit dispendious, but I need the starting value in a
% \numexpr compatible form in the iteration loops.|
%    \begin{macrocode}
\expandafter\def\csname XINT_expr_itself_++\endcsname {++}%
\expandafter\def\csname XINT_expr_itself_++)\endcsname {++)}%
\expandafter\let\csname XINT_expr_precedence_++)\endcsname \xint_c_i
\xintFor #1 in {expr,flexpr,iiexpr} \do {%
    \expandafter\def\csname XINT_#1_op_++)\endcsname ##1##2\relax
     {\expandafter\XINT_expr_foundend
                  \expanded{{+{\XINT:NEhook:f:one:from:one:direct\xintiCeil##1}}}%
     }%
}%
%    \end{macrocode}
% \paragraph{The \cshn{break()} function}
% \lverb|break is a true function, the parsing via expansion of the enclosed
% material proceeds via _oparen macros as with any other function.|
%    \begin{macrocode}
\catcode`? 3
\def\XINT_expr_func_break #1#2#3{#1#2{?#3}}%
\catcode`? 11
\let\XINT_flexpr_func_break \XINT_expr_func_break
\let\XINT_iiexpr_func_break \XINT_expr_func_break
%    \end{macrocode}
% \paragraph{The \cshn{omit} and \cshn{abort} keywords}
% \lverb|&
% Comments are currently undergoing reconstruction.
%
% The mechanism is somewhat complex.  The operator !? will fetch a dummy value
% ! or ^ which is then recognized int the loops implementing the various seq
% etc... construct using dummy variables and implement omit and abort.
%
% In May 2021 I realized that the January 2020 1.4 had broken omit and abort
% if used inside a subs().  The definition
%(\edef\XINT_expr_var_omit  #1\relax !{1\string !?!\relax !}%
%)
% conflicted with the 1.4 refactoring of «subs» and similar things 
% which had replaced formerly clean-up macros (of ! and what's next, as in
% now defunct
% \def\XINT_expr_subx:_end #1!#2#3{#1} which was involved in subs mechanism,
% and by the way would be incompatible with multi-letter dummy variables) by usage
% of an \iffalse as in "\relax\iffalse\relax !" to delimite a sub-expression, which was
% supposed to be clever (the "\relax !" being delimiter for dummy variables).
%
% This \iffalse from subs mechanism ended up being gobbled by omit/abort thus
% inducing breakage.
%
% Grabbing \relax #2! would be a fix but looks a bit dangerous, as there can
% be a subexpression after the omit or abort bringing its own \relax, although
% this is very very unlikely.
%
% I considered to modify the dummy variables delimiter from \relax !  to
% \xint_Bye ! for example but got afraid from the ramifications, as all
% structures handling dummy variables would have needed refactoring.
%
% So finally things here remain unchanged and the refactoring to fix this
% breakage was done in \XINT_allexpr_subsx (and also subsm).
% Done at 1.4h. See \XINT_allexpr_subsx for comments.|
%
%    \begin{macrocode}
\edef\XINT_expr_var_omit  #1\relax !{1\string !?!\relax !}%
\edef\XINT_expr_var_abort #1\relax !{1\string !?^\relax !}%
\def\XINT_expr_itself_!? {!?}%
\def\XINT_expr_op_!? #1#2\relax{\XINT_expr_foundend{#2}}%
\let\XINT_iiexpr_op_!? \XINT_expr_op_!?
\let\XINT_flexpr_op_!? \XINT_expr_op_!?
\let\XINT_expr_precedence_!? \xint_c_iv
%    \end{macrocode}
% \paragraph{The semi-colon}
% 
% \lverb|Obsolete comments undergoing re-construction|
%
%    \begin{macrocode}
\xintFor #1 in {expr,flexpr,iiexpr} \do {%
     \expandafter\def\csname XINT_#1_op_;\endcsname {\xint_c_i ;}%
}%
\expandafter\let\csname XINT_expr_precedence_;\endcsname\xint_c_i
\expandafter\def\csname XINT_expr_itself_;)\endcsname {)}%
\expandafter\let\csname XINT_expr_precedence_;)\endcsname\xint_c_i
%    \end{macrocode}
% \subsubsection{Reserved dummy variables @, @1, @2, @3, @4, @@, @@(1), \dots, @@@,
% @@@(1), \dots{} for recursions}
% \lverb|&
% Comments currently under reconstruction.
%
% 1.4 breaking change: @ and @1 behave differently and one can not use @ in
% place of @1 in iterr() and rrseq(). Formerly @ and @1 had the same
% definition.
%
% Brace stripping in \XINT_expr_func_@@ 
% is prevented by some ending 0 or other token see iterr() and rrseq() code.
%
% For the record, the ~ and ? have catcode 3 in this code.
%
% |
%
%    \begin{macrocode}
\catcode`* 11
\def\XINT_expr_var_@ #1~#2{{#2}#1~{#2}}%
\def\XINT_expr_var*_@ #1~#2{\XINT_expr_prec_tacit *{#2}(#1~{#2}}%
\expandafter
\def\csname XINT_expr_var_@1\endcsname #1~#2{{{#2}}#1~{#2}}%
\expandafter
\def\csname XINT_expr_var_@2\endcsname #1~#2#3{{{#3}}#1~{#2}{#3}}%
\expandafter
\def\csname XINT_expr_var_@3\endcsname #1~#2#3#4{{{#4}}#1~{#2}{#3}{#4}}%
\expandafter
\def\csname XINT_expr_var_@4\endcsname #1~#2#3#4#5{{{#5}}#1~{#2}{#3}{#4}{#5}}%
\expandafter\def\csname XINT_expr_var*_@1\endcsname #1~#2%
               {\XINT_expr_prec_tacit *{{#2}}(#1~{#2}}%
\expandafter\def\csname XINT_expr_var*_@2\endcsname #1~#2#3%
               {\XINT_expr_prec_tacit *{{#3}}(#1~{#2}{#3}}%
\expandafter\def\csname XINT_expr_var*_@3\endcsname #1~#2#3#4%
               {\XINT_expr_prec_tacit *{{#4}}(#1~{#2}{#3}{#4}}%
\expandafter\def\csname XINT_expr_var*_@4\endcsname #1~#2#3#4#5%
               {\XINT_expr_prec_tacit *{{#5}}(#1~{#2}{#3}{#4}{#5}}%
\catcode`* 12
\catcode`? 3
\def\XINT_expr_func_@@ #1#2#3#4~#5?%
{%
   \expandafter#1\expandafter#2\expandafter{\expandafter{%
         \romannumeral0\xintntheltnoexpand{\xintNum#3}{#5}}}#4~#5?%
}%
\def\XINT_expr_func_@@@ #1#2#3#4~#5~#6?%
{%
   \expandafter#1\expandafter#2\expandafter{\expandafter{%
         \romannumeral0\xintntheltnoexpand{\xintNum#3}{#6}}}#4~#5~#6?%
}%
\def\XINT_expr_func_@@@@ #1#2#3#4~#5~#6~#7?%
{%
   \expandafter#1\expandafter#2\expandafter{\expandafter{%
         \romannumeral0\xintntheltnoexpand{\xintNum#3}{#7}}}#4~#5~#6~#7?%
}%
\let\XINT_flexpr_func_@@\XINT_expr_func_@@
\let\XINT_flexpr_func_@@@\XINT_expr_func_@@@
\let\XINT_flexpr_func_@@@@\XINT_expr_func_@@@@
\def\XINT_iiexpr_func_@@ #1#2#3#4~#5?%
{%
   \expandafter#1\expandafter#2\expandafter{\expandafter{%
         \romannumeral0\xintntheltnoexpand{\xint_firstofone#3}{#5}}}#4~#5?%
}%
\def\XINT_iiexpr_func_@@@ #1#2#3#4~#5~#6?%
{%
   \expandafter#1\expandafter#2\expandafter{\expandafter{%
         \romannumeral0\xintntheltnoexpand{\xint_firstofone#3}{#6}}}#4~#5~#6?%
}%
\def\XINT_iiexpr_func_@@@@ #1#2#3#4~#5~#6~#7?%
{%
   \expandafter#1\expandafter#2\expandafter{\expandafter{%
         \romannumeral0\xintntheltnoexpand{\xint_firstofone#3}{#7}}}#4~#5~#6~#7?%
}%
\catcode`? 11
%    \end{macrocode}
% \subsection{Pseudo-functions involving dummy variables and generating scalars or sequences}
% \localtableofcontents
%
% \subsubsection{Comments}
% \lverb|&
% Comments added 2020/01/16.
%
% The mechanism for «seq» is the following. When the parser encounters «seq»,
% which means it parsed these letters and encountered (from expansion) an
% opening parenthesis, the \XINT_expr_func mechanism triggers the «`» operator
% which realizes that «seq» is a pseudo-function (there is no _func_seq) and
% thus spans the \XINT_expr_onliteral_seq macro (currently this means however
% that the knowledge of which parser we are in is lost, see comments of
% \XINT_expr_op_` code). The latter will use delimited macros and parenthesis
% check to fetch (without any expansion), the symbolic expression ExprSeq to
% evaluate, the Name (now possibly multi-letter) of the variable and the
% expression ExprValues to evaluate which will give the values to assign to
% the dummy variable Name. It then positions upstream ExprValues suitably
% terminated (see next) and after it {{Name}{ExprSeq}}. Then it inserts a
% second call to the «`» operator with now «seqx» as argument hence
% the appropriate «{,fl,ii}expr_func_seqx» macros gets executed. The general
% way function macros work is that first all their arguments are evaluated via
% a call not to \xintbare{,float,ii}eval but to the suitable
% \XINT_{expr,flexpr,iiexpr}_oparen core macro which does almost same excepts
% it expects a final closing parenthesis (of course allowing nested
% parenthesis in-between) and stops there. Here, this closing parenthesis got positioned
% deliberately with a \relax after it, so the parser, which always after
% having gathered a value looks ahead to find the next operator, thinks it has
% hit the end of the expression and as result inserts a \xint_c_ (i.e. \z@)
% token for precedence level and a dummy \relax token (place-holder for a
% non-existing operator). Generally speaking «func_foo» macros expect to
% be executed with three parameters #1#2#3, #1 = precedence, #2 = operator, #3
% = values (call it «args») i.e. the fully evaluated list of all its
% arguments. The special «func_seqx» and cousins know that the first two
% tokens are trash and they now proceed forward, having thus lying before them
% upstream the values to loop over, now fully evaluated, and
% {{Name}{ExprSeq}}. It then positions appropriately ExprSeq inside a
% sub-expression and after it, following suitable delimiter, Name and the
% evaluated values to assign to Name.
%
% Dummy variables are essentially simply delimited macros where the delimiter
% is the variable name preceded by a \relax token and a catcode 11 exclamation
% point. Thus the various «subsx», «seqx», «iterx» position the tokens
% appropriately and launch suitable loops.
%
% All of this nests well, inner «seq»'s (or more often in practice «subs»'s)
% being allowed to refer to the dummy variables used by outer «seq»'s because
% the outer «seq»'s have the values to assign to their variables evaluated
% first and their ExprSeq evaluated last. For inner dummy variables to be able
% to refer to outer dummy variables the author must be careful of course to
% not use in the implementation braces { and } which would break dummy
% variables to fetch values beyond the closing brace.
%
% The above «seq» mechanism was done around June 15-25th 2014 at the time of
% the transition from 1.09n to 1.1 but already in October 2014 I made a note
% that I had a hard time to understand it again:
%
% « [START OF YEAR 2014 COMMENTS]
% 
% All of seq, add, mul, rseq, etc... (actually all of the extensive
% changes from xintexpr 1.09n to 1.1) was done around June 15-25th 2014, but the
% problem is that I did not document the code enough, and I had a hard time
% understanding in October what I had done in June. Despite the lesson, again
% being short on time, I do not document enough my current understanding of the
% innards of the beast...
%
% I added subs, and iter in October (also the [:n], [n:] list extractors),
% proving I did at least understand a bit (or rather could imitate) my earlier
% code (but don't ask me to explain \xintNewExpr !)
%
% The \XINT_expr_fetch_E_comma_V_equal_E_a parses: "expression, variable=list)"
% (when it is called the opening ( has been swallowed, and it looks for
% the ending one.) Both expression and list may themselves contain
% parentheses and commas, we allow nesting. For example "x^2,x=1..10)",
% at the end of seq_a we have {variable{expression}}{list}, in this
% example {x{x^2}}{1..10}, or more complicated
% "seq(add(y,y=1..x),x=1..10)" will work too. The variable is a single
% lowercase Latin letter.
%
% The complications with \xint_c_ii^v in seq_f is for the recurrent
% thing that we don't know in what type of expressions we are, hence we
% must move back up, with some loss of efficiency (superfluous check for
% minus sign, etc...). But the code manages simultaneously expr, flexpr
% and iiexpr.
%
% [END OF YEAR 2014 OLD COMMENTS]»
%
% On Jeudi 16 janvier 2020 à 15:13:32 I finally did the documentation as
% above.
%
% The case of «iter», «rseq», «iterr», «rrseq» differs slightly because the
% initial values need evaluation. This is done by genuine functions
% \XINT_<parser>_func_iter etc... (there was no \XINT_<parser>_func_seq). The
% trick is via the semi-colon ; which is a genuine operator having the
% precedence of a closing parenthesis and whose action is only to stop
% expansion. Thus this first step of gathering the initial values is done as
% part of the reguler expansion job of the parser not using delimited macros
% and the ; can be hidden in braces {;} because the three parsers when moving
% forward remove one level of braces always. Thus
% \XINT_<parser>_func_seq simply hand over to \XINT_allexpr_iter which will
% then trigger the fetching without expansion of ExprIter, Name=ExprValues as
% described previously for «seq».
%
% With 1.4, multi-letter names for dummy variables are allowed.
%
% Also there is the additional 1.4 ambition to make the whole thing parsable
% by \xintNewExpr/\xintdeffunc. This is done by checking if all is numerical,
% because the omit, abort and break() mechanisms have no translation into
% macros, and the only solution for symbolic material is to simply keep it as
% is, so that expansion will again activate the xintexpr parsers. At 1.4 this
% approach is fine although the initial goals of \xintNewExpr/\xintdeffunc was
% to completely replace the parsers (whose storage method hit the string pool
% formerly) by macros. Now that 1.4 does not impact the string pool we can
% make \xintdeffunc much more powerful but it will not be a construct using
% only xintfrac macros, it will still be partially the \xintexpr etc...
% parsers in such cases.|
%
% \lverb|Got simpler with 1.2c as now the dummy variable fetches an
% already encapsulated value, which is anyhow the form in which we get
% it.
%
% Refactored at 1.4 using \expanded rather than \csname.
%
% And support for multi-letter variables, which means function declarations
% can now use multi-letter variables !
% |
% \subsubsection{\cshn{subs()}: substitution of one variable}
%    \begin{macrocode}
\def\XINT_expr_onliteral_subs
{%
    \expandafter\XINT_allexpr_subs_f
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_allexpr_subs_f #1#2{\xint_c_ii^v `{subsx}#2)\relax #1}%
\def\XINT_expr_func_subsx   #1#2{\XINT_allexpr_subsx \xintbareeval }%
\def\XINT_flexpr_func_subsx #1#2{\XINT_allexpr_subsx \xintbarefloateval}%
\def\XINT_iiexpr_func_subsx #1#2{\XINT_allexpr_subsx \xintbareiieval }%
%    \end{macrocode}
% \lverb|
% #2 is the value to assign to the dummy variable
% #3 is the dummy variable name (possibly multi-letter), #4 is the expression
% to evaluate
%
% 1.4 was doing something clever to get rid of the ! and tokens following it,
% via an \iffalse...\fi which erased them and propagated the expansion to
% trigger the getopt:
%
%(\expanded\bgroup\romannumeral0#1#4\relax \iffalse\relax !#3{#2}{\fi\expandafter}
%)
%
% But sadly, with a delay of more than one year later (right after having
% released 1.4g) I realized that this had broken omit and abort if inside a
% subs.  As omit and abort would clean all up to \relax !, this meant here
% swallowing in particular the above \iffalse, leaving a dangling \fi.  I had
% the files which show this bug already at time of 1.4 release but did not
% compile them, and they were not included in my test suite.
%
% I hesitated with modifying the delimiter from "\relax !<varname>" (catcode
% 11 !) to "\relax \xint_Bye<varname>" for the dummy variables which would
% have allowed some trickery using \xint_Bye...\xint_bye clean-up but got
% afraid from the breakage potential of such refactoring with many induced
% changes.
%
% A variant like this:
%( \def\XINT_allexpr_subsx #1#2#3#4%
%: {%
%:     \expandafter\XINT_expr_clean_and_put_op_first
%:     \expanded
%:     {\romannumeral0#1#4\relax !#3{#2}\xint:\expandafter}\romannumeral`$&$&\XINT_expr_getop
%: }%
%: \def\XINT_expr_clean_and_put_op_first #1#2\xint:#3#4{#3#4{#1}}%
%)
% breaks nesting: the braces make variables encountered in #4 unable to
% match their definition.  This would work:
%
%( \def\XINT_allexpr_subsx #1#2#3#4%
%: {%
%:     \expandafter\XINT_allexpr_subsx_clean\romannumeral0#1#4\relax !#3{#2}\xint:
%: }%
%: \def\XINT_allexpr_subsx_clean #1#2\xint:
%: {%
%:     \expandafter\XINT_expr_put_op_first
%:     \expanded{\unexpanded{{#1}}\expandafter}\romannumeral`$&$&@\XINT_expr_getop
%: }%
%)
% (not tested).
%
% But in the end I decided to simply fix the first envisioned code above.
% This accepts expansion of supposedly inert #3{#2}.  There is again the
% \iffalse but it is moved to the right.  This change limits possibly hacky
% future developments.  Done at 1.4h (2021/01/27).
%
% No need for the \expandafter's from \XINT_expr_put_op_first in
% \XINT_expr_clean_and_put_op_first.  |
%    \begin{macrocode}
\def\XINT_allexpr_subsx #1#2#3#4%
{%
    \expandafter\XINT_expr_clean_and_put_op_first
    \expanded
    \bgroup\romannumeral0#1#4\relax !#3{#2}\xint:\iffalse{\fi\expandafter}%
    \romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_expr_clean_and_put_op_first #1#2\xint:#3#4{#3#4{#1}}%
%    \end{macrocode}
% \subsubsection{\cshn{subsm()}: simultaneous independent substitutions}
% \lverb|New with 1.4. Globally the var1=expr1; var2=expr2; var2=expr3;...
% part can arise from expansion, except that once a semi-colon has been found
% (from expansion) the varK= thing following it must be there. And as for
% subs() the final parenthesis must be there from the start.
%
% |
%    \begin{macrocode}
\def\XINT_expr_onliteral_subsm
{%
    \expandafter\XINT_allexpr_subsm_f
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_allexpr_subsm_f #1#2{\xint_c_ii^v `{subsmx}#2)\relax #1}%
\def\XINT_expr_func_subsmx
{%
    \expandafter\XINT_allexpr_subsmx\expandafter\xintbareeval
    \expanded\bgroup{\iffalse}\fi\XINT_allexpr_subsm_A\XINT_expr_oparen
}%
\def\XINT_flexpr_func_subsmx
{%
    \expandafter\XINT_allexpr_subsmx\expandafter\xintbarefloateval
    \expanded\bgroup{\iffalse}\fi\XINT_allexpr_subsm_A\XINT_flexpr_oparen
}%
\def\XINT_iiexpr_func_subsmx
{%
    \expandafter\XINT_allexpr_subsmx\expandafter\xintbareiieval
    \expanded\bgroup{\iffalse}\fi\XINT_allexpr_subsm_A\XINT_iiexpr_oparen
}%
\def\XINT_allexpr_subsm_A #1#2#3%
{%
    \ifx#2\xint_c_
       \expandafter\XINT_allexpr_subsm_done
    \else
       \expandafter\XINT_allexpr_subsm_B
    \fi #1%
}%
\def\XINT_allexpr_subsm_B #1#2#3#4=%
{%
    {#2}\relax !\xint_zapspaces#3#4 \xint_gobble_i
    \expandafter\XINT_allexpr_subsm_A\expandafter#1\romannumeral`&&@#1%
}%
%    \end{macrocode}
% \lverb|
%( #1 = \xintbareeval,  or \xintbarefloateval or \xintbareiieval
%: #2 = evaluation of last variable assignment
%)
% |
%    \begin{macrocode}
\def\XINT_allexpr_subsm_done #1#2{{#2}\iffalse{{\fi}}}%
%    \end{macrocode}
% \lverb|
%( #1 = \xintbareeval or \xintbarefloateval or \xintbareiieval
%: #2 = {value1}\relax !var2{value2}....\relax !varN{valueN} (value's may be oples)
%: #3 = {var1}
%: #4 = the expression to evaluate
%)
% Refactored at 1.4h as for \XINT_allexpr_subsx, see comments there related
% to the omit/abort conundrum.
% |
%    \begin{macrocode}
\def\XINT_allexpr_subsmx #1#2#3#4%
{%
    \expandafter\XINT_expr_clean_and_put_op_first
    \expanded
    \bgroup\romannumeral0#1#4\relax !#3#2\xint:\iffalse{\fi\expandafter}%
    \romannumeral`&&@\XINT_expr_getop
}%
%    \end{macrocode}
% \subsubsection{\cshn{subsn()}: leaner syntax for nesting (possibly dependent) substitutions}
% \lverb|New with 1.4. 2020/01/24
%
% |
%    \begin{macrocode}
\def\XINT_expr_onliteral_subsn
{%
    \expandafter\XINT_allexpr_subsn_f
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_allexpr_subsn_f #1{\XINT_allexpr_subsn_g #1}%
%    \end{macrocode}
% \lverb|
%( #1 = Name1
%: #2 = Expression in all variables which is to evaluate
%: #3 = all the stuff after Name1 =  and up to final parenthesis
%)
%
% This one needed no reactoring at 1.4h to fix the omit/abort problem, as
% there was no \iffalse..\fi clean-up: the clean-up is done directly via
% \XINT_allexpr_subsnx_J.
%
% I only added usage of \XINT_expr_put_op_first_noexpand.  There may be other
% locations where it could be used, but I can't afford now reviewing
% usage. For next release after 1.4h bugfix.
% |
%    \begin{macrocode}
\def\XINT_allexpr_subsn_g #1#2#3%
{%
    \expandafter\XINT_allexpr_subsn_h
    \expanded\bgroup{\iffalse}\fi\expandafter\XINT_allexpr_subsn_B
    \romannumeral\XINT_expr_fetch_to_semicolon #1=#3;\hbox=;;^{#2}%
}%
\def\XINT_allexpr_subsn_B #1{\XINT_allexpr_subsn_C #1\vbox}%
\def\XINT_allexpr_subsn_C #1#2=#3\vbox
{%
    \ifx\hbox#1\iffalse{{\fi}\expandafter}\else
    {{\xint_zapspaces #1#2 \xint_gobble_i}};\unexpanded{{{#3}}}%
    \expandafter\XINT_allexpr_subsn_B
    \romannumeral\expandafter\XINT_expr_fetch_to_semicolon\fi
}%
\def\XINT_allexpr_subsn_h
{%
    \xint_c_ii^v `{subsnx}\romannumeral0\xintreverseorder
}%
\def\XINT_expr_func_subsnx #1#2#3#4#5;#6%
{%
    \xint_gob_til_^ #6\XINT_allexpr_subsnx_H ^%
    \expandafter\XINT_allexpr_subsnx\expandafter
    \xintbareeval\romannumeral0\xintbareeval #5\relax !#4{#3}\xintundefined
    {\relax !#4{#3}\relax !#6}%
}%
\def\XINT_iiexpr_func_subsnx #1#2#3#4#5;#6%
{%
    \xint_gob_til_^ #6\XINT_allexpr_subsnx_H ^%
    \expandafter\XINT_allexpr_subsnx\expandafter
    \xintbareiieval\romannumeral0\xintbareiieval #5\relax !#4{#3}\xintundefined
    {\relax !#4{#3}\relax !#6}%
}%
\def\XINT_flexpr_func_subsnx #1#2#3#4#5;#6%
{%
    \xint_gob_til_^ #6\XINT_allexpr_subsnx_H ^%
    \expandafter\XINT_allexpr_subsnx\expandafter
    \xintbarefloateval\romannumeral0\xintbarefloateval #5\relax !#4{#3}\xintundefined
    {\relax !#4{#3}\relax !#6}%
}%
\def\XINT_allexpr_subsnx #1#2!#3\xintundefined#4#5;#6%
{%
    \xint_gob_til_^ #6\XINT_allexpr_subsnx_I ^%
    \expandafter\XINT_allexpr_subsnx\expandafter
    #1\romannumeral0#1#5\relax !#4{#2}\xintundefined
    {\relax !#4{#2}\relax !#6}%
}%
\def\XINT_allexpr_subsnx_H ^#1\romannumeral0#2#3!#4\xintundefined #5#6%
{%
    \expandafter\XINT_allexpr_subsnx_J\romannumeral0#2#6#5%
}%
\def\XINT_allexpr_subsnx_I ^#1\romannumeral0#2#3\xintundefined #4#5%
{%
    \expandafter\XINT_allexpr_subsnx_J\romannumeral0#2#5#4%
}%
\def\XINT_allexpr_subsnx_J #1#2^%
{%
    \expandafter\XINT_expr_put_op_first_noexpand
    \expanded{\unexpanded{{#1}}\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_expr_put_op_first_noexpand#1#2#3{#2#3{#1}}%
%    \end{macrocode}
% \subsubsection{\cshn{seq()}: sequences from assigning values to a
% dummy variable}
% \lverb|
% In seq_f, the #2 is the ExprValues expression which needs evaluation to
% provide the values to the dummy variable and #1 is {Name}{ExprSeq}
% where Name is the name of dummy variable and {ExprSeq} the expression
% which will have to be evaluated.
% |
%
%
%    \begin{macrocode}
\def\XINT_allexpr_seq_f #1#2{\xint_c_ii^v `{seqx}#2)\relax #1}%
\def\XINT_expr_onliteral_seq
 {\expandafter\XINT_allexpr_seq_f\romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}}%
\def\XINT_expr_func_seqx   #1#2{\XINT:NEhook:seqx\XINT_allexpr_seqx\xintbareeval }%
\def\XINT_flexpr_func_seqx #1#2{\XINT:NEhook:seqx\XINT_allexpr_seqx\xintbarefloateval}%
\def\XINT_iiexpr_func_seqx #1#2{\XINT:NEhook:seqx\XINT_allexpr_seqx\xintbareiieval }%
\def\XINT_allexpr_seqx #1#2#3#4%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded \bgroup {\iffalse}\fi\XINT_expr_seq:_b {#1#4\relax !#3}#2^%
    \XINT_expr_cb_and_getop
}%
\def\XINT_expr_cb_and_getop{\iffalse{\fi\expandafter}\romannumeral`&&@\XINT_expr_getop}%
%    \end{macrocode}
%
% \lverb|Comments undergoing reconstruction.|
%    \begin{macrocode}
\catcode`? 3
\def\XINT_expr_seq:_b #1#2%
{%
    \ifx +#2\xint_dothis\XINT_expr_seq:_Ca\fi
    \ifx !#2!\xint_dothis\XINT_expr_seq:_noop\fi
    \ifx ^#2\xint_dothis\XINT_expr_seq:_end\fi
    \xint_orthat{\XINT_expr_seq:_c}{#2}{#1}%
}%
\def\XINT_expr_seq:_noop #1{\XINT_expr_seq:_b }%
\def\XINT_expr_seq:_end  #1#2{\iffalse{\fi}}%
\def\XINT_expr_seq:_c #1#2{\expandafter\XINT_expr_seq:_d\romannumeral0#2{{#1}}{#2}}%
\def\XINT_expr_seq:_d #1{\ifx ^#1\xint_dothis\XINT_expr_seq:_abort\fi
                         \ifx ?#1\xint_dothis\XINT_expr_seq:_break\fi
                         \ifx !#1\xint_dothis\XINT_expr_seq:_omit\fi
                         \xint_orthat{\XINT_expr_seq:_goon {#1}}}%
\def\XINT_expr_seq:_abort #1!#2^{\iffalse{\fi}}%
\def\XINT_expr_seq:_break #1!#2^{#1\iffalse{\fi}}%
\def\XINT_expr_seq:_omit  #1!#2#{\expandafter\XINT_expr_seq:_b\xint_gobble_i}%
\def\XINT_expr_seq:_goon  #1!#2#{#1\expandafter\XINT_expr_seq:_b\xint_gobble_i}%
\def\XINT_expr_seq:_Ca #1#2#3{\XINT_expr_seq:_Cc#3.{#2}}%
\def\XINT_expr_seq:_Cb #1{\expandafter\XINT_expr_seq:_Cc\the\numexpr#1+\xint_c_i.}%
\def\XINT_expr_seq:_Cc #1.#2{\expandafter\XINT_expr_seq:_D\romannumeral0#2{{#1}}{#1}{#2}}%
\def\XINT_expr_seq:_D #1{\ifx ^#1\xint_dothis\XINT_expr_seq:_abort\fi
                         \ifx ?#1\xint_dothis\XINT_expr_seq:_break\fi
                         \ifx !#1\xint_dothis\XINT_expr_seq:_Omit\fi
                         \xint_orthat{\XINT_expr_seq:_Goon {#1}}}%
\def\XINT_expr_seq:_Omit #1!#2#{\expandafter\XINT_expr_seq:_Cb\xint_gobble_i}%
\def\XINT_expr_seq:_Goon #1!#2#{#1\expandafter\XINT_expr_seq:_Cb\xint_gobble_i}%
%    \end{macrocode}
% \subsubsection{\cshn{iter()}}
%
% \lverb|Prior to 1.2g, the iter keyword was what is now called iterr,
% analogous with rrseq. Somehow I forgot an iter functioning like rseq
% with the sole difference of printing only the last iteration. Both rseq and
% iter work well with list selectors, as @ refers to the whole comma separated
% sequence of the initial values. I have thus deliberately done the backwards
% incompatible renaming of iter to iterr, and the new iter.
%
% To understand the tokens which are presented to \XINT_allexpr_iter it is
% needed to check elsewhere in the source code how the ; hack is done.
%
% The #2 in \XINT_allexpr_iter is \xint_c_i from the ; hack. Formerly (xint <
% 1.4) there was no such token. The change is motivated to using ; also in
% subsm() syntax.
%
% |
%    \begin{macrocode}
\def\XINT_expr_func_iter   {\XINT_allexpr_iter \xintbareeval      }%
\def\XINT_flexpr_func_iter {\XINT_allexpr_iter \xintbarefloateval }%
\def\XINT_iiexpr_func_iter {\XINT_allexpr_iter \xintbareiieval    }%
\def\XINT_allexpr_iter #1#2#3#4%
{%
    \expandafter\XINT_expr_iterx
    \expandafter#1\expanded{\unexpanded{{#4}}\expandafter}%
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_expr_iterx #1#2#3#4%
{%
    \XINT:NEhook:iter\XINT_expr_itery\romannumeral0#1(#4)\relax {#2}#3#1%
}%
\def\XINT_expr_itery #1#2#3#4#5%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded \bgroup {\iffalse}\fi
    \XINT_expr_iter:_b {#5#4\relax !#3}#1^~{#2}\XINT_expr_cb_and_getop
}%
\def\XINT_expr_iter:_b #1#2%
{%
    \ifx +#2\xint_dothis\XINT_expr_iter:_Ca\fi
    \ifx !#2!\xint_dothis\XINT_expr_iter:_noop\fi
    \ifx ^#2\xint_dothis\XINT_expr_iter:_end\fi
    \xint_orthat{\XINT_expr_iter:_c}{#2}{#1}%
}%
\def\XINT_expr_iter:_noop #1{\XINT_expr_iter:_b }%
\def\XINT_expr_iter:_end  #1#2~#3{#3\iffalse{\fi}}%
\def\XINT_expr_iter:_c #1#2{\expandafter\XINT_expr_iter:_d\romannumeral0#2{{#1}}{#2}}%
\def\XINT_expr_iter:_d #1{\ifx ^#1\xint_dothis\XINT_expr_iter:_abort\fi
                          \ifx ?#1\xint_dothis\XINT_expr_iter:_break\fi
                          \ifx !#1\xint_dothis\XINT_expr_iter:_omit\fi
                          \xint_orthat{\XINT_expr_iter:_goon {#1}}}%
\def\XINT_expr_iter:_abort #1!#2^~#3{#3\iffalse{\fi}}%
\def\XINT_expr_iter:_break #1!#2^~#3{#1\iffalse{\fi}}%
\def\XINT_expr_iter:_omit  #1!#2#{\expandafter\XINT_expr_iter:_b\xint_gobble_i}%
\def\XINT_expr_iter:_goon  #1!#2#{\XINT_expr_iter:_goon_a {#1}}%
\def\XINT_expr_iter:_goon_a #1#2#3~#4{\XINT_expr_iter:_b #3~{#1}}%
\def\XINT_expr_iter:_Ca #1#2#3{\XINT_expr_iter:_Cc#3.{#2}}%
\def\XINT_expr_iter:_Cb #1{\expandafter\XINT_expr_iter:_Cc\the\numexpr#1+\xint_c_i.}%
\def\XINT_expr_iter:_Cc #1.#2{\expandafter\XINT_expr_iter:_D\romannumeral0#2{{#1}}{#1}{#2}}%
\def\XINT_expr_iter:_D #1{\ifx ^#1\xint_dothis\XINT_expr_iter:_abort\fi
                          \ifx ?#1\xint_dothis\XINT_expr_iter:_break\fi
                          \ifx !#1\xint_dothis\XINT_expr_iter:_Omit\fi
                          \xint_orthat{\XINT_expr_iter:_Goon {#1}}}%
\def\XINT_expr_iter:_Omit  #1!#2#{\expandafter\XINT_expr_iter:_Cb\xint_gobble_i}%
\def\XINT_expr_iter:_Goon  #1!#2#{\XINT_expr_iter:_Goon_a {#1}}%
\def\XINT_expr_iter:_Goon_a #1#2#3~#4{\XINT_expr_iter:_Cb #3~{#1}}%
%    \end{macrocode}
% \subsubsection{\cshn{add()}, \cshn{mul()}}
% \lverb|Comments under reconstruction.
%
% These were a bit anomalous as they did not implement omit and abort keyword
% and the break() function (and per force then neither the n++ syntax).
% 
% At 1.4 they are simply mapped to using adequately
% iter(). Thus, there is small loss in efficiency, but supporting omit, abort
% and break is important. Using dedicated macros here would have caused also
% slight efficiency drop. Simpler to remove the old approach.
% |
%
%    \begin{macrocode}
\def\XINT_expr_onliteral_add
 {\expandafter\XINT_allexpr_add_f\romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}}%
\def\XINT_allexpr_add_f #1#2{\xint_c_ii^v `{opx}#2)\relax #1{+}{0}}%
\def\XINT_expr_onliteral_mul
 {\expandafter\XINT_allexpr_mul_f\romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}}%
\def\XINT_allexpr_mul_f #1#2{\xint_c_ii^v `{opx}#2)\relax #1{*}{1}}%
\def\XINT_expr_func_opx   {\XINT:NEhook:opx \XINT_allexpr_opx \xintbareeval     }%
\def\XINT_flexpr_func_opx {\XINT:NEhook:opx \XINT_allexpr_opx \xintbarefloateval}%
\def\XINT_iiexpr_func_opx {\XINT:NEhook:opx \XINT_allexpr_opx \xintbareiieval   }%
%    \end{macrocode}
% \lverb|1.4a In case of usage of omit (did I not test it? obviously
% I didn't as neither omit nor abort could work; and break neither),
% 1.4 code using (#6) syntax caused a
% (somewhat misleading) «missing )» error message which originated in the
% #6. This is non-obvious problem (perhaps explained why prior to 1.4 I had
% not added support for omit and break() to add() and mul()...
%
% Allowing () is not enough as it would have to be 0 or 1 depending on
% whether we are using add() or mul(). Hence the somewhat complicated
% detour (relying on precise way var_omit and var_abort work) via
% \XINT_allexpr_opx_ifnotomitted.
%
% \break() has special meaning here as it is used as last operand, not as last
% value. The code is very unsatisfactory and inefficient but this is hotfix
% for 1.4a.|
%    \begin{macrocode}
\def\XINT_allexpr_opx #1#2#3#4#5#6#7#8%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded \bgroup {\iffalse}\fi
    \XINT_expr_iter:_b {#1%
    \expandafter\XINT_allexpr_opx_ifnotomitted
    \romannumeral0#1#6\relax#7@\relax !#5}#4^~{{#8}}\XINT_expr_cb_and_getop
}%
\def\XINT_allexpr_opx_ifnotomitted #1%
{%
    \ifx !#1\xint_dothis{@\relax}\fi
    \ifx ^#1\xint_dothis{\XINTfstop. ^\relax}\fi
    \if ?\xintFirstItem{#1}\xint_dothis{\XINT_allexpr_opx_break{#1}}\fi
    \xint_orthat{\XINTfstop.{#1}}%
}%
\def\XINT_allexpr_opx_break #1#2\relax
{%
    break(\expandafter\XINTfstop\expandafter.\expandafter{\xint_gobble_i#1}#2)\relax
}%
%    \end{macrocode}
% \subsubsection{\cshn{rseq()}}
%
% \lverb|When func_rseq has its turn, initial segment has been scanned by
% oparen, the ; mimicking the rôle of a closing parenthesis, and stopping
% further expansion (and leaving a \xint_c_i left-over token since 1.4). The ;
% is discovered during standard parsing mode, it may be for example {;} or
% arise from expansion as rseq does not use a delimited macro to locate it.
% |
%    \begin{macrocode}
\def\XINT_expr_func_rseq   {\XINT_allexpr_rseq \xintbareeval      }%
\def\XINT_flexpr_func_rseq {\XINT_allexpr_rseq \xintbarefloateval }%
\def\XINT_iiexpr_func_rseq {\XINT_allexpr_rseq \xintbareiieval    }%
\def\XINT_allexpr_rseq #1#2#3#4%
{%
    \expandafter\XINT_expr_rseqx
    \expandafter #1\expanded{\unexpanded{{#4}}\expandafter}%
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_expr_rseqx #1#2#3#4%
{%
    \XINT:NEhook:rseq \XINT_expr_rseqy\romannumeral0#1(#4)\relax {#2}#3#1%
}%
\def\XINT_expr_rseqy #1#2#3#4#5%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded \bgroup {\iffalse}\fi
    #2%
    \XINT_expr_rseq:_b {#5#4\relax !#3}#1^~{#2}\XINT_expr_cb_and_getop
}%
\def\XINT_expr_rseq:_b #1#2%
{%
    \ifx +#2\xint_dothis\XINT_expr_rseq:_Ca\fi
    \ifx !#2!\xint_dothis\XINT_expr_rseq:_noop\fi
    \ifx ^#2\xint_dothis\XINT_expr_rseq:_end\fi
    \xint_orthat{\XINT_expr_rseq:_c}{#2}{#1}%
}%
\def\XINT_expr_rseq:_noop #1{\XINT_expr_rseq:_b }%
\def\XINT_expr_rseq:_end  #1#2~#3{\iffalse{\fi}}%
\def\XINT_expr_rseq:_c #1#2{\expandafter\XINT_expr_rseq:_d\romannumeral0#2{{#1}}{#2}}%
\def\XINT_expr_rseq:_d #1{\ifx ^#1\xint_dothis\XINT_expr_rseq:_abort\fi
                          \ifx ?#1\xint_dothis\XINT_expr_rseq:_break\fi
                          \ifx !#1\xint_dothis\XINT_expr_rseq:_omit\fi
                          \xint_orthat{\XINT_expr_rseq:_goon {#1}}}%
\def\XINT_expr_rseq:_abort #1!#2^~#3{\iffalse{\fi}}%
\def\XINT_expr_rseq:_break #1!#2^~#3{#1\iffalse{\fi}}%
\def\XINT_expr_rseq:_omit  #1!#2#{\expandafter\XINT_expr_rseq:_b\xint_gobble_i}%
\def\XINT_expr_rseq:_goon  #1!#2#{\XINT_expr_rseq:_goon_a {#1}}%
\def\XINT_expr_rseq:_goon_a #1#2#3~#4{#1\XINT_expr_rseq:_b #3~{#1}}%
\def\XINT_expr_rseq:_Ca #1#2#3{\XINT_expr_rseq:_Cc#3.{#2}}%
\def\XINT_expr_rseq:_Cb #1{\expandafter\XINT_expr_rseq:_Cc\the\numexpr#1+\xint_c_i.}%
\def\XINT_expr_rseq:_Cc #1.#2{\expandafter\XINT_expr_rseq:_D\romannumeral0#2{{#1}}{#1}{#2}}%
\def\XINT_expr_rseq:_D #1{\ifx ^#1\xint_dothis\XINT_expr_rseq:_abort\fi
                          \ifx ?#1\xint_dothis\XINT_expr_rseq:_break\fi
                          \ifx !#1\xint_dothis\XINT_expr_rseq:_Omit\fi
                          \xint_orthat{\XINT_expr_rseq:_Goon {#1}}}%
\def\XINT_expr_rseq:_Omit  #1!#2#{\expandafter\XINT_expr_rseq:_Cb\xint_gobble_i}%
\def\XINT_expr_rseq:_Goon  #1!#2#{\XINT_expr_rseq:_Goon_a {#1}}%
\def\XINT_expr_rseq:_Goon_a #1#2#3~#4{#1\XINT_expr_rseq:_Cb #3~{#1}}%
%    \end{macrocode}
% \subsubsection{\cshn{iterr()}}
% \lverb|ATTENTION! at 1.4 the @ and @1 are not synonymous anymore. One *must* use
% @1 in iterr() context.|
%    \begin{macrocode}
\def\XINT_expr_func_iterr   {\XINT_allexpr_iterr \xintbareeval      }%
\def\XINT_flexpr_func_iterr {\XINT_allexpr_iterr \xintbarefloateval }%
\def\XINT_iiexpr_func_iterr {\XINT_allexpr_iterr \xintbareiieval    }%
\def\XINT_allexpr_iterr #1#2#3#4%
{%
    \expandafter\XINT_expr_iterrx
    \expandafter #1\expanded{{\xintRevWithBraces{#4}}\expandafter}%
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_expr_iterrx #1#2#3#4%
{%
    \XINT:NEhook:iterr\XINT_expr_iterry\romannumeral0#1(#4)\relax {#2}#3#1%
}%
\def\XINT_expr_iterry #1#2#3#4#5%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded \bgroup {\iffalse}\fi
    \XINT_expr_iterr:_b {#5#4\relax !#3}#1^~#20?\XINT_expr_cb_and_getop
}%
\def\XINT_expr_iterr:_b #1#2%
{%
    \ifx +#2\xint_dothis\XINT_expr_iterr:_Ca\fi
    \ifx !#2!\xint_dothis\XINT_expr_iterr:_noop\fi
    \ifx ^#2\xint_dothis\XINT_expr_iterr:_end\fi
    \xint_orthat{\XINT_expr_iterr:_c}{#2}{#1}%
}%
\def\XINT_expr_iterr:_noop #1{\XINT_expr_iterr:_b }%
\def\XINT_expr_iterr:_end  #1#2~#3#4?{{#3}\iffalse{\fi}}%
\def\XINT_expr_iterr:_c #1#2{\expandafter\XINT_expr_iterr:_d\romannumeral0#2{{#1}}{#2}}%
\def\XINT_expr_iterr:_d #1{\ifx ^#1\xint_dothis\XINT_expr_iterr:_abort\fi
                           \ifx ?#1\xint_dothis\XINT_expr_iterr:_break\fi
                           \ifx !#1\xint_dothis\XINT_expr_iterr:_omit\fi
                           \xint_orthat{\XINT_expr_iterr:_goon {#1}}}%
\def\XINT_expr_iterr:_abort #1!#2^~#3?{\iffalse{\fi}}%
\def\XINT_expr_iterr:_break #1!#2^~#3?{#1\iffalse{\fi}}%
\def\XINT_expr_iterr:_omit  #1!#2#{\expandafter\XINT_expr_iterr:_b\xint_gobble_i}%
\def\XINT_expr_iterr:_goon  #1!#2#{\XINT_expr_iterr:_goon_a{#1}}%
\def\XINT_expr_iterr:_goon_a  #1#2#3~#4?%
{%
    \expandafter\XINT_expr_iterr:_b \expanded{\unexpanded{#3~}\xintTrim{-2}{#1#4}}0?%
}%
\def\XINT_expr_iterr:_Ca #1#2#3{\XINT_expr_iterr:_Cc#3.{#2}}%
\def\XINT_expr_iterr:_Cb #1{\expandafter\XINT_expr_iterr:_Cc\the\numexpr#1+\xint_c_i.}%
\def\XINT_expr_iterr:_Cc #1.#2{\expandafter\XINT_expr_iterr:_D\romannumeral0#2{{#1}}{#1}{#2}}%
\def\XINT_expr_iterr:_D #1{\ifx ^#1\xint_dothis\XINT_expr_iterr:_abort\fi
                          \ifx ?#1\xint_dothis\XINT_expr_iterr:_break\fi
                          \ifx !#1\xint_dothis\XINT_expr_iterr:_Omit\fi
                          \xint_orthat{\XINT_expr_iterr:_Goon {#1}}}%
\def\XINT_expr_iterr:_Omit #1!#2#{\expandafter\XINT_expr_iterr:_Cb\xint_gooble_i}%
\def\XINT_expr_iterr:_Goon  #1!#2#{\XINT_expr_iterr:_Goon_a{#1}}%
\def\XINT_expr_iterr:_Goon_a  #1#2#3~#4?%
{%
    \expandafter\XINT_expr_iterr:_Cb \expanded{\unexpanded{#3~}\xintTrim{-2}{#1#4}}0?%
}%
%    \end{macrocode}
% \subsubsection{\cshn{rrseq()}}
%
% \lverb|When func_rrseq has its turn, initial segment has been scanned
% by oparen, the ; mimicking the rôle of a closing parenthesis, and
% stopping further expansion. #2 = \xint_c_i and #3 are left-over trash.|
%    \begin{macrocode}
\def\XINT_expr_func_rrseq   {\XINT_allexpr_rrseq \xintbareeval      }%
\def\XINT_flexpr_func_rrseq {\XINT_allexpr_rrseq \xintbarefloateval }%
\def\XINT_iiexpr_func_rrseq {\XINT_allexpr_rrseq \xintbareiieval    }%
\def\XINT_allexpr_rrseq #1#2#3#4%
{%
    \expandafter\XINT_expr_rrseqx\expandafter#1\expanded
       {\unexpanded{{#4}}{\xintRevWithBraces{#4}}\expandafter}%
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_expr_rrseqx #1#2#3#4#5%
{%
    \XINT:NEhook:rrseq\XINT_expr_rrseqy\romannumeral0#1(#5)\relax {#2}{#3}#4#1%
}%
\def\XINT_expr_rrseqy #1#2#3#4#5#6%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded \bgroup {\iffalse}\fi
    #2\XINT_expr_rrseq:_b {#6#5\relax !#4}#1^~#30?\XINT_expr_cb_and_getop
}%
\def\XINT_expr_rrseq:_b #1#2%
{%
    \ifx +#2\xint_dothis\XINT_expr_rrseq:_Ca\fi
    \ifx !#2!\xint_dothis\XINT_expr_rrseq:_noop\fi
    \ifx ^#2\xint_dothis\XINT_expr_rrseq:_end\fi
    \xint_orthat{\XINT_expr_rrseq:_c}{#2}{#1}%
}%
\def\XINT_expr_rrseq:_noop #1{\XINT_expr_rrseq:_b }%
\def\XINT_expr_rrseq:_end  #1#2~#3?{\iffalse{\fi}}%
\def\XINT_expr_rrseq:_c #1#2{\expandafter\XINT_expr_rrseq:_d\romannumeral0#2{{#1}}{#2}}%
\def\XINT_expr_rrseq:_d #1{\ifx ^#1\xint_dothis\XINT_expr_rrseq:_abort\fi
                           \ifx ?#1\xint_dothis\XINT_expr_rrseq:_break\fi
                           \ifx !#1\xint_dothis\XINT_expr_rrseq:_omit\fi
                           \xint_orthat{\XINT_expr_rrseq:_goon {#1}}}%
\def\XINT_expr_rrseq:_abort #1!#2^~#3?{\iffalse{\fi}}%
\def\XINT_expr_rrseq:_break #1!#2^~#3?{#1\iffalse{\fi}}%
\def\XINT_expr_rrseq:_omit  #1!#2#{\expandafter\XINT_expr_rrseq:_b\xint_gobble_i}%
\def\XINT_expr_rrseq:_goon  #1!#2#{\XINT_expr_rrseq:_goon_a {#1}}%
\def\XINT_expr_rrseq:_goon_a  #1#2#3~#4?%
{%
    #1\expandafter\XINT_expr_rrseq:_b\expanded{\unexpanded{#3~}\xintTrim{-2}{#1#4}}0?%
}%
\def\XINT_expr_rrseq:_Ca #1#2#3{\XINT_expr_rrseq:_Cc#3.{#2}}%
\def\XINT_expr_rrseq:_Cb #1{\expandafter\XINT_expr_rrseq:_Cc\the\numexpr#1+\xint_c_i.}%
\def\XINT_expr_rrseq:_Cc #1.#2{\expandafter\XINT_expr_rrseq:_D\romannumeral0#2{{#1}}{#1}{#2}}%
\def\XINT_expr_rrseq:_D #1{\ifx ^#1\xint_dothis\XINT_expr_rrseq:_abort\fi
                          \ifx ?#1\xint_dothis\XINT_expr_rrseq:_break\fi
                          \ifx !#1\xint_dothis\XINT_expr_rrseq:_Omit\fi
                          \xint_orthat{\XINT_expr_rrseq:_Goon {#1}}}%
\def\XINT_expr_rrseq:_Omit  #1!#2#{\expandafter\XINT_expr_rrseq:_Cb\xint_gobble_i}%
\def\XINT_expr_rrseq:_Goon  #1!#2#{\XINT_expr_rrseq:_Goon_a {#1}}%
\def\XINT_expr_rrseq:_Goon_a  #1#2#3~#4?%
{%
    #1\expandafter\XINT_expr_rrseq:_Cb\expanded{\unexpanded{#3~}\xintTrim{-2}{#1#4}}0?%
}%
\catcode`? 11
%    \end{macrocode}
% \subsection{Pseudo-functions related to N-dimensional hypercubic lists}
% \subsubsection{\cshn{ndseq()}}
% \lverb|New with 1.4. 2020/01/23. It is derived from subsm() but instead of
% evaluating one expression according to one value per variable, it constructs
% a nested bracketed seq... this means the expression is parsed each time !
% Anyway, proof of concept. Nota Bene : omit, abort, break() work !|
%    \begin{macrocode}
\def\XINT_expr_onliteral_ndseq
{%
    \expandafter\XINT_allexpr_ndseq_f
    \romannumeral`&&@\XINT_expr_fetch_E_comma_V_equal_E_a {}%
}%
\def\XINT_allexpr_ndseq_f #1#2{\xint_c_ii^v `{ndseqx}#2)\relax #1}%
\def\XINT_expr_func_ndseqx
{%
    \expandafter\XINT_allexpr_ndseqx\expandafter\xintbareeval
    \expandafter{\romannumeral0\expandafter\xint_gobble_i\string}%
    \expandafter\xintrevwithbraces
    \expanded\bgroup{\iffalse}\fi\XINT_allexpr_ndseq_A\XINT_expr_oparen
}%
\def\XINT_flexpr_func_ndseqx
{%
    \expandafter\XINT_allexpr_ndseqx\expandafter\xintbarefloateval
    \expandafter{\romannumeral0\expandafter\xint_gobble_i\string}%
    \expandafter\xintrevwithbraces
    \expanded\bgroup{\iffalse}\fi\XINT_allexpr_ndseq_A\XINT_flexpr_oparen
}%
\def\XINT_iiexpr_func_ndseqx
{%
    \expandafter\XINT_allexpr_ndseqx\expandafter\xintbareiieval
    \expandafter{\romannumeral0\expandafter\xint_gobble_i\string}%
    \expandafter\xintrevwithbraces
    \expanded\bgroup{\iffalse}\fi\XINT_allexpr_ndseq_A\XINT_iiexpr_oparen
}%
\def\XINT_allexpr_ndseq_A #1#2#3%
{%
    \ifx#2\xint_c_
       \expandafter\XINT_allexpr_ndseq_C
    \else
       \expandafter\XINT_allexpr_ndseq_B
    \fi #1%
}%
\def\XINT_allexpr_ndseq_B #1#2#3#4=%
{%
    {#2}{\xint_zapspaces#3#4 \xint_gobble_i}%
    \expandafter\XINT_allexpr_ndseq_A\expandafter#1\romannumeral`&&@#1%
}%
%    \end{macrocode}
% \lverb|
%  #1 = \xintbareeval,  or \xintbarefloateval or \xintbareiieval
%  #2 = values for last coordinate
% |
%    \begin{macrocode}
\def\XINT_allexpr_ndseq_C #1#2{{#2}\iffalse{{{\fi}}}}%
%    \end{macrocode}
% \lverb|
%  #1 = \xintbareeval or \xintbarefloateval or \xintbareiieval
%  #2 = {valuesN}...{values2}{var2}{values1}
%  #3 = {var1}
%  #4 = the expression to evaluate
% |
%    \begin{macrocode}
\def\XINT_allexpr_ndseqx #1#2#3#4%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded
    \bgroup
      \romannumeral0#1\empty
      \expanded{\xintReplicate{\xintLength{{#3}#2}/2}{[seq(}%
                \unexpanded{#4}%
                \XINT_allexpr_ndseqx_a #2{#3}^^%
                }%
      \relax
    \iffalse{\fi\expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT_allexpr_ndseqx_a #1#2%
{%
    \xint_gob_til_^ #1\XINT_allexpr_ndseqx_e ^%
    \unexpanded{,#2=\XINTfstop.{#1})]}\XINT_allexpr_ndseqx_a
}%
\def\XINT_allexpr_ndseqx_e ^#1\XINT_allexpr_ndseqx_a{}%
%    \end{macrocode}
% \subsubsection{\cshn{ndmap()}}
% \lverb|New with 1.4. 2020/01/24.|
%    \begin{macrocode}
\def\XINT_expr_onliteral_ndmap #1,{\xint_c_ii^v `{ndmapx}\XINTfstop.{#1};}%
\def\XINT_expr_func_ndmapx #1#2#3%
{%
    \expandafter\XINT_allexpr_ndmapx
    \csname XINT_expr_func_\xint_zapspaces #3 \xint_gobble_i\endcsname
    \XINT_expr_oparen
}%
\def\XINT_flexpr_func_ndmapx #1#2#3%
{%
    \expandafter\XINT_allexpr_ndmapx
    \csname XINT_flexpr_func_\xint_zapspaces #3 \xint_gobble_i\endcsname
    \XINT_flexpr_oparen
}%
\def\XINT_iiexpr_func_ndmapx #1#2#3%
{%
    \expandafter\XINT_allexpr_ndmapx
    \csname XINT_iiexpr_func_\xint_zapspaces #3 \xint_gobble_i\endcsname
    \XINT_iiexpr_oparen
}%
\def\XINT_allexpr_ndmapx #1#2%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded\bgroup{\iffalse}\fi
    \expanded
      {\noexpand\XINT:NEhook:x:ndmapx
       \noexpand\XINT_allexpr_ndmapx_a
       \noexpand#1{}\expandafter}%
    \expanded\bgroup\expandafter\XINT_allexpr_ndmap_A
                    \expandafter#2\romannumeral`&&@#2%
}%
\def\XINT_allexpr_ndmap_A #1#2#3%
{%
    \ifx#3;%
       \expandafter\XINT_allexpr_ndmap_B
    \else
       \xint_afterfi{\XINT_allexpr_ndmap_C#2#3}%
    \fi #1%
}%
\def\XINT_allexpr_ndmap_B #1#2%
{%
    {#2}\expandafter\XINT_allexpr_ndmap_A\expandafter#1\romannumeral`&&@#1%
}%
\def\XINT_allexpr_ndmap_C #1#2#3#4%
{%
    {#4}^\relax\iffalse{{{\fi}}}#1#2%
}%
\def\XINT_allexpr_ndmapx_a #1#2#3%
{%
    \xint_gob_til_^ #3\XINT_allexpr_ndmapx_l ^%
    \XINT_allexpr_ndmapx_b #1{#2}{#3}%
}%
\def\XINT_allexpr_ndmapx_l ^#1\XINT_allexpr_ndmapx_b #2#3#4\relax
{%
    #2\empty\xint_firstofone{#3}%
}%
\def\XINT_allexpr_ndmapx_b #1#2#3#4\relax
{%
    {\iffalse}\fi\XINT_allexpr_ndmapx_c {#4\relax}#1{#2}#3^%
}%
\def\XINT_allexpr_ndmapx_c #1#2#3#4%
{%
    \xint_gob_til_^ #4\XINT_allexpr_ndmapx_e ^%
    \XINT_allexpr_ndmapx_a #2{#3{#4}}#1%
    \XINT_allexpr_ndmapx_c  {#1}#2{#3}%
}%
\def\XINT_allexpr_ndmapx_e ^#1\XINT_allexpr_ndmapx_c
   {\iffalse{\fi}\xint_gobble_iii}%
%    \end{macrocode}
% \subsubsection{\cshn{ndfillraw()}}
% \lverb|New with 1.4. 2020/01/24. J'hésite à autoriser un #1 quelconque,
% ou plutôt à le wrapper dans un \xintbareval. Mais il faut alors distinguer
% les trois. De toute façon les variables ne marcheraient pas donc j'hésite
% à mettre un wrapper automatique. Mais ce n'est pas bien d'autoriser l'injection
% de choses quelconques.
%
% Pour des choses comme ndfillraw(\xintRandomBit,[10,10]).
%
% Je n'aime pas le nom !. Le changer. ndconst? Surtout je n'aime pas
% que dans le premier argument il faut rajouter explicitement si nécessaire
% \xintiiexpr wrap.
% |
%    \begin{macrocode}
\def\XINT_expr_onliteral_ndfillraw #1,{\xint_c_ii^v `{ndfillrawx}\XINTfstop.{{#1}},}%
\def\XINT_expr_func_ndfillrawx #1#2#3%
{%
    \expandafter#1\expandafter#2\expanded{{{\XINT_allexpr_ndfillrawx_a #3}}}%
}%
\let\XINT_iiexpr_func_ndfillrawx\XINT_expr_func_ndfillrawx
\let\XINT_flexpr_func_ndfillrawx\XINT_expr_func_ndfillrawx 
\def\XINT_allexpr_ndfillrawx_a #1#2%
{%
    \expandafter\XINT_allexpr_ndfillrawx_b
    \romannumeral0\xintApply{\xintNum}{#2}^\relax {#1}%
}%
\def\XINT_allexpr_ndfillrawx_b #1#2\relax#3%
{%
    \xint_gob_til_^ #1\XINT_allexpr_ndfillrawx_c ^%
    \xintReplicate{#1}{{\XINT_allexpr_ndfillrawx_b #2\relax {#3}}}%
}%
\def\XINT_allexpr_ndfillrawx_c ^\xintReplicate #1#2%
{%
    \expandafter\XINT_allexpr_ndfillrawx_d\xint_firstofone #2%
}%
\def\XINT_allexpr_ndfillrawx_d\XINT_allexpr_ndfillrawx_b \relax #1{#1}%
%    \end{macrocode}
% \subsection{Other pseudo-functions: \cshn{bool()}, \cshn{togl()}, \cshn{protect()},
% \cshn{qraw()}, \cshn{qint()}, \cshn{qfrac()}, \cshn{qfloat()}, \cshn{qrand()},
% \cshn{random()}, \cshn{rbit()}}
%
% \lverb|bool, togl and protect use delimited macros. They are not true
% functions, they turn off the parser to gather their "variable".|
%
% \changed{1.2} Adds |qint()|, |qfrac()|, |qfloat()|.
%
% \changed{1.3c} Adds |qraw()|. Useful to limit impact on \TeX{} memory
% from abuse of |\csname|'s storage when generating many comma separated
% values from a loop.
%
% \changed{1.3e} |qfloat()| keeps a short mantissa if possible.
%
% \lverb|They allow the user to hand over quickly a big number to the parser,
% spaces not immediately removed but should be harmless in general. The qraw()
% does no post-processing at all apart complete expansion, useful for
% comma-separated values, but must be obedient to (non really documented)
% expected format. Each uses a delimited macro, the closing parenthesis can
% not emerge from expansion.|
%
% \lverb|1.3b. random(), qrand()
% Function-like syntax but with no argument currently, so let's
% use fast parsing which requires though the closing parenthesis to be
% explicit.|
%
% \lverb|Attention that qraw()
% which pre-supposes knowledge of internal storage model is fragile
% and may break at any release.
%
% 1.4 adds rbit(). Short for random bit.
% |
%    \begin{macrocode}
\def\XINT_expr_onliteral_bool #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\xintBool{#1}}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_togl #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\xintToggle{#1}}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_protect #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\detokenize{#1}}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_qint #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\xintiNum{#1}}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_qfrac #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\xintRaw{#1}}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_qfloat #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\XINTinFloatSdigits{#1}}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_qraw #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{#1}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_random #1)%
    {\expandafter\XINT_expr_put_op_first\expanded{{{\XINTinRandomFloatSdigits}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_qrand #1)%
   {\expandafter\XINT_expr_put_op_first\expanded{{{\XINTinRandomFloatSixteen}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
\def\XINT_expr_onliteral_rbit #1)%
   {\expandafter\XINT_expr_put_op_first\expanded{{{\xintRandBit}}\expandafter
     }\romannumeral`&&@\XINT_expr_getop}%
%    \end{macrocode}
% \def\auxiliarymacro#1{ \noexpand\cshn{#1()}}
% \edef\zzz{Regular built-in functions: \xintListWithSep{, }{\xintApply\auxiliarymacro
% {{num}{reduce}{preduce}
% {abs}{sgn}{frac}{floor}{ceil}{sqr}
% {?}{!}{not}{odd}{even}{isint}{isone}
% {factorial}{sqrt}{sqrtr}
% {inv}{round}{trunc}
% {float}{sfloat}{ilog10}
% {divmod}{mod}{binomial}{pfactorial}
% {randrange}
% {iquo}{irem}{gcd}{lcm}{max}{min}
% {`+`}{`*`}
% {all}{any}{xor}
% {len}{first}{last}{reversed}
% {if}{ifint}{ifone}{ifsgn}
% {nuple}{unpack}{flat}}}
% and \noexpand\cshn{zip()}}
% \expandafter\subsection\expandafter{\zzz}
%    \begin{macrocode}
\def\XINT:expr:f:one:and:opt #1#2#3!#4#5%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                      \expandafter\xint_secondoftwo\fi
    {#4}{#5[\xintNum{#2}]}{#1}%
}%
\def\XINT:expr:f:tacitzeroifone #1#2#3!#4#5%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                      \expandafter\xint_secondoftwo\fi
    {#4{0}}{#5{\xintNum{#2}}}{#1}%
}%
\def\XINT:expr:f:iitacitzeroifone #1#2#3!#4%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                      \expandafter\xint_secondoftwo\fi
    {#4{0}}{#4{#2}}{#1}%
}%
\def\XINT_expr_func_num #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintNum#3}}%
}%
\let\XINT_flexpr_func_num\XINT_expr_func_num
\let\XINT_iiexpr_func_num\XINT_expr_func_num
\def\XINT_expr_func_reduce #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintIrr#3}}%
}%
\let\XINT_flexpr_func_reduce\XINT_expr_func_reduce
\def\XINT_expr_func_preduce #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintPIrr#3}}%
}%
\let\XINT_flexpr_func_preduce\XINT_expr_func_preduce
\def\XINT_expr_func_abs #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintAbs#3}}%
}%
\let\XINT_flexpr_func_abs\XINT_expr_func_abs
\def\XINT_iiexpr_func_abs #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiAbs#3}}%
}%
\def\XINT_expr_func_sgn #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintSgn#3}}%
}%
\let\XINT_flexpr_func_sgn\XINT_expr_func_sgn
\def\XINT_iiexpr_func_sgn #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiSgn#3}}%
}%
\def\XINT_expr_func_frac #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintTFrac#3}}%
}%
\def\XINT_flexpr_func_frac #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatFrac#3}}%
}%
%    \end{macrocode}
% \lverb|no \XINT_iiexpr_func_frac|
%    \begin{macrocode}
\def\XINT_expr_func_floor #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintFloor#3}}%
}%
\let\XINT_flexpr_func_floor\XINT_expr_func_floor
%    \end{macrocode}
% \lverb|The floor and ceil functions in \xintiiexpr require protect(a/b) or,
% better, \qfrac(a/b); else the / will be executed first and do an integer
% rounded division.|
%    \begin{macrocode}
\def\XINT_iiexpr_func_floor #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiFloor#3}}%
}%
\def\XINT_expr_func_ceil #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintCeil#3}}%
}%
\let\XINT_flexpr_func_ceil\XINT_expr_func_ceil
\def\XINT_iiexpr_func_ceil #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiCeil#3}}%
}%
\def\XINT_expr_func_sqr #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintSqr#3}}%
}%
\def\XINT_flexpr_func_sqr #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatSqr#3}}%
}%
\def\XINT_iiexpr_func_sqr #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiSqr#3}}%
}%
\def\XINT_expr_func_? #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiIsNotZero#3}}%
}%
\let\XINT_flexpr_func_? \XINT_expr_func_?
\let\XINT_iiexpr_func_? \XINT_expr_func_?
\def\XINT_expr_func_! #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiIsZero#3}}%
}%
\let\XINT_flexpr_func_! \XINT_expr_func_!
\let\XINT_iiexpr_func_! \XINT_expr_func_!
\def\XINT_expr_func_not #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiIsZero#3}}%
}%
\let\XINT_flexpr_func_not \XINT_expr_func_not
\let\XINT_iiexpr_func_not \XINT_expr_func_not
\def\XINT_expr_func_odd #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintOdd#3}}%
}%
\let\XINT_flexpr_func_odd\XINT_expr_func_odd
\def\XINT_iiexpr_func_odd #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiOdd#3}}%
}%
\def\XINT_expr_func_even #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintEven#3}}%
}%
\let\XINT_flexpr_func_even\XINT_expr_func_even
\def\XINT_iiexpr_func_even #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiEven#3}}%
}%
\def\XINT_expr_func_isint #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintIsInt#3}}%
}%
\def\XINT_flexpr_func_isint #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintFloatIsInt#3}}%
}%
\let\XINT_iiexpr_func_isint\XINT_expr_func_isint % ? perhaps rather always 1
\def\XINT_expr_func_isone #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintIsOne#3}}%
}%
\let\XINT_flexpr_func_isone\XINT_expr_func_isone
\def\XINT_iiexpr_func_isone #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiIsOne#3}}%
}%
\def\XINT_expr_func_factorial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt #3,!\xintFac\XINTinFloatFac
    }}%
}%
\def\XINT_flexpr_func_factorial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt#3,!\XINTinFloatFacdigits\XINTinFloatFac
    }}%
}%
\def\XINT_iiexpr_func_factorial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiFac#3}}%
}%
\def\XINT_expr_func_sqrt #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt #3,!\XINTinFloatSqrtdigits\XINTinFloatSqrt
    }}%
}%
\let\XINT_flexpr_func_sqrt\XINT_expr_func_sqrt
\def\XINT_iiexpr_func_sqrt #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiSqrt#3}}%
}%
\def\XINT_iiexpr_func_sqrtr #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiSqrtR#3}}%
}%
\def\XINT_expr_func_inv #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintInv#3}}%
}%
\def\XINT_flexpr_func_inv #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatInv#3}}%
}%
\def\XINT_expr_func_round #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:tacitzeroifone:direct
    \XINT:expr:f:tacitzeroifone #3,!\xintiRound\xintRound
    }}%
}%
\let\XINT_flexpr_func_round\XINT_expr_func_round
\def\XINT_iiexpr_func_round #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:iitacitzeroifone:direct
    \XINT:expr:f:iitacitzeroifone #3,!\xintiRound
    }}%
}%
\def\XINT_expr_func_trunc #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:tacitzeroifone:direct
    \XINT:expr:f:tacitzeroifone #3,!\xintiTrunc\xintTrunc
    }}%
}%
\let\XINT_flexpr_func_trunc\XINT_expr_func_trunc
\def\XINT_iiexpr_func_trunc #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:iitacitzeroifone:direct
    \XINT:expr:f:iitacitzeroifone #3,!\xintiTrunc
    }}%
}%
%    \end{macrocode}
% \lverb|Hesitation at 1.3e about using \XINTinFloatSdigits and \XINTinFloatS.
% Finally I add a sfloat() function. It helps for xinttrig.sty.|
%    \begin{macrocode}
\def\XINT_expr_func_float #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt #3,!\XINTinFloatdigits\XINTinFloat
    }}%
}%
\let\XINT_flexpr_func_float\XINT_expr_func_float
%    \end{macrocode}
% \lverb|&
% float_() was added at 1.4, as a shortcut alias to float() skipping the check
% for an optional second argument.  This is useful to transfer function
% definitions between \xintexpr and \xintfloatexpr contexts.
%
%
% No need for a similar shortcut for sfloat() as currently used in
% xinttrig.sty to go from float to expr: as it is used there as sfloat(x) with
% dummy x, it sees there is no optional argument, contrarily to for example
% float(\xintexpr...\relax) which has to allow for the inner expression to
% expand to an ople with two items, so does not know in which branch it is at
% time of definiion.
%
%
% After some hesitation at 1.4e regarding guard digits mechanism the float_()
% got renamed to float_dgt(), but then renamed back to float_() to avoid a
% breaking change and having to document it.
%
% Nevertheless the documentation of 1.4e mentioned float_dgt()... but it was still
% float_()... now changed into float_dgt() for real at 1.4f.
%
% 1.4f also adds private float_dgtormax and sfloat_dgtormax for matters of xinttrig.
%
% |
%    \begin{macrocode}
\def\XINT_expr_func_float_dgt #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatdigits#3}}%
}%
\let\XINT_flexpr_func_float_dgt\XINT_expr_func_float_dgt
% no \XINT_iiexpr_func_float_dgt
\def\XINT_expr_func_float_dgtormax #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatdigitsormax#3}}%
}%
\let\XINT_flexpr_func_float_dgtormax\XINT_expr_func_float_dgtormax
\def\XINT_expr_func_sfloat #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt #3,!\XINTinFloatSdigits\XINTinFloatS
    }}%
}%
\let\XINT_flexpr_func_sfloat\XINT_expr_func_sfloat
% no \XINT_iiexpr_func_sfloat
\def\XINT_expr_func_sfloat_dgtormax #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatSdigitsormax#3}}%
}%
\let\XINT_flexpr_func_sfloat_dgtormax\XINT_expr_func_sfloat_dgtormax
\expandafter\def\csname XINT_expr_func_ilog10\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt #3,!\xintiLogTen\XINTFloatiLogTen
    }}%
}%
\expandafter\def\csname XINT_flexpr_func_ilog10\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:and:opt:direct
    \XINT:expr:f:one:and:opt #3,!\XINTFloatiLogTendigits\XINTFloatiLogTen
    }}%
}%
\expandafter\def\csname XINT_iiexpr_func_ilog10\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\xintiiLogTen#3}}%
}%
\def\XINT_expr_func_divmod #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintDivMod #3}}%
}%
\def\XINT_flexpr_func_divmod #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\XINTinFloatDivMod #3}}%
}%
\def\XINT_iiexpr_func_divmod #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintiiDivMod #3}}%
}%
\def\XINT_expr_func_mod #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintMod#3}}%
}%
\def\XINT_flexpr_func_mod #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\XINTinFloatMod#3}}%
}%
\def\XINT_iiexpr_func_mod #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintiiMod#3}}%
}%
\def\XINT_expr_func_binomial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintBinomial #3}}%
}%
\def\XINT_flexpr_func_binomial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\XINTinFloatBinomial #3}}%
}%
\def\XINT_iiexpr_func_binomial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintiiBinomial #3}}%
}%
\def\XINT_expr_func_pfactorial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintPFactorial #3}}%
}%
\def\XINT_flexpr_func_pfactorial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\XINTinFloatPFactorial #3}}%
}%
\def\XINT_iiexpr_func_pfactorial #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintiiPFactorial #3}}%
}%
\def\XINT_expr_func_randrange #1#2#3%
{%
    \expandafter #1\expandafter #2\expanded{{{%
    \XINT:expr:randrange #3,!%
    }}}%
}%
\let\XINT_flexpr_func_randrange\XINT_expr_func_randrange
\def\XINT_iiexpr_func_randrange #1#2#3%
{%
    \expandafter #1\expandafter #2\expanded{{{%
    \XINT:iiexpr:randrange #3,!%
    }}}%
}%
\def\XINT:expr:randrange #1#2#3!%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                     \expandafter\xint_secondoftwo\fi
    {\xintiiRandRange{\XINT:NEhook:f:one:from:one:direct\xintNum{#1}}}%
    {\xintiiRandRangeAtoB{\XINT:NEhook:f:one:from:one:direct\xintNum{#1}}%
                         {\XINT:NEhook:f:one:from:one:direct\xintNum{#2}}%
     }%
}%
\def\XINT:iiexpr:randrange #1#2#3!%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                     \expandafter\xint_secondoftwo\fi
    {\xintiiRandRange{#1}}%
    {\xintiiRandRangeAtoB{#1}{#2}}%
}%
\def\XINT_iiexpr_func_iquo #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintiiQuo #3}}%
}%
\def\XINT_iiexpr_func_irem #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\romannumeral`&&@%
    \XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintiiRem #3}}%
}%
\def\XINT_expr_func_gcd #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_GCDof#3^}}%
}%
\let\XINT_flexpr_func_gcd\XINT_expr_func_gcd
\def\XINT_iiexpr_func_gcd #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_iiGCDof#3^}}%
}%
\def\XINT_expr_func_lcm #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_LCMof#3^}}%
}%
\let\XINT_flexpr_func_lcm\XINT_expr_func_lcm
\def\XINT_iiexpr_func_lcm #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_iiLCMof#3^}}%
}%
\def\XINT_expr_func_max #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_Maxof#3^}}%
}%
\def\XINT_iiexpr_func_max #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_iiMaxof#3^}}%
}%
\def\XINT_flexpr_func_max #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINTinFloatMaxof#3^}}%
}%
\def\XINT_expr_func_min #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_Minof#3^}}%
}%
\def\XINT_iiexpr_func_min #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_iiMinof#3^}}%
}%
\def\XINT_flexpr_func_min #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINTinFloatMinof#3^}}%
}%
\expandafter
\def\csname XINT_expr_func_+\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_Sum#3^}}%
}%
\expandafter
\def\csname XINT_flexpr_func_+\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINTinFloatSum#3^}}%
}%
\expandafter
\def\csname XINT_iiexpr_func_+\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_iiSum#3^}}%
}%
\expandafter
\def\csname XINT_expr_func_*\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_Prd#3^}}%
}%
\expandafter
\def\csname XINT_flexpr_func_*\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINTinFloatPrd#3^}}%
}%
\expandafter
\def\csname XINT_iiexpr_func_*\endcsname #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_iiPrd#3^}}%
}%
\def\XINT_expr_func_all #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_ANDof#3^}}%
}%
\let\XINT_flexpr_func_all\XINT_expr_func_all
\let\XINT_iiexpr_func_all\XINT_expr_func_all
\def\XINT_expr_func_any #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_ORof#3^}}%
}%
\let\XINT_flexpr_func_any\XINT_expr_func_any
\let\XINT_iiexpr_func_any\XINT_expr_func_any
\def\XINT_expr_func_xor #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{\expandafter
    {\romannumeral`&&@\XINT:NEhook:f:from:delim:u\XINT_XORof#3^}}%
}%
\let\XINT_flexpr_func_xor\XINT_expr_func_xor
\let\XINT_iiexpr_func_xor\XINT_expr_func_xor
\def\XINT_expr_func_len #1#2#3%
{%
    \expandafter#1\expandafter#2\expandafter{\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:LFL\xintLength
    {\romannumeral\XINT:NEhook:r:check#3^}%
    }}%
}%
\let\XINT_flexpr_func_len \XINT_expr_func_len
\let\XINT_iiexpr_func_len \XINT_expr_func_len
\def\XINT_expr_func_first #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:LFL\xintFirstOne
    {\romannumeral\XINT:NEhook:r:check#3^}%
    }%
}%
\let\XINT_flexpr_func_first\XINT_expr_func_first
\let\XINT_iiexpr_func_first\XINT_expr_func_first
\def\XINT_expr_func_last #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:LFL\xintLastOne
    {\romannumeral\XINT:NEhook:r:check#3^}%
    }%
}%
\let\XINT_flexpr_func_last\XINT_expr_func_last
\let\XINT_iiexpr_func_last\XINT_expr_func_last
\def\XINT_expr_func_reversed #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:reverse\XINT_expr_reverse
    #3^^#3\xint:\xint:\xint:\xint:
          \xint:\xint:\xint:\xint:\xint_bye
    }%
}%
\def\XINT_expr_reverse #1#2%
{%
    \if ^\noexpand#2%
        \expandafter\XINT_expr_reverse:_one_or_none\string#1.%
    \else
        \expandafter\XINT_expr_reverse:_at_least_two
    \fi
}%
\def\XINT_expr_reverse:_at_least_two #1^^{\XINT_revwbr_loop {}}%
\def\XINT_expr_reverse:_one_or_none #1%
{%
    \if #1\bgroup\xint_dothis\XINT_expr_reverse:_nutple\fi
    \if #1^\xint_dothis\XINT_expr_reverse:_nil\fi
    \xint_orthat\XINT_expr_reverse:_leaf
}%
\edef\XINT_expr_reverse:_nil #1\xint_bye{\noexpand\fi\space}%
\def\XINT_expr_reverse:_leaf#1\fi #2\xint:#3\xint_bye{\fi\xint_gob_andstop_i#2}%
\def\XINT_expr_reverse:_nutple%
{%
    \expandafter\XINT_expr_reverse:_nutple_a\expandafter{\string}%
}%
\def\XINT_expr_reverse:_nutple_a #1^#2\xint:#3\xint_bye
{%
    \fi\expandafter
    {\romannumeral0\XINT_revwbr_loop{}#2\xint:#3\xint_bye}%
}%
\let\XINT_flexpr_func_reversed\XINT_expr_func_reversed
\let\XINT_iiexpr_func_reversed\XINT_expr_func_reversed
\def\XINT_expr_func_if #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:branch{\romannumeral`&&@\xintiiifNotZero #3}}%
}%
\let\XINT_flexpr_func_if\XINT_expr_func_if
\let\XINT_iiexpr_func_if\XINT_expr_func_if
\def\XINT_expr_func_ifint #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:branch{\romannumeral`&&@\xintifInt #3}}%
}%
\let\XINT_iiexpr_func_ifint\XINT_expr_func_ifint
\def\XINT_flexpr_func_ifint #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:branch{\romannumeral`&&@\xintifFloatInt #3}}%
}%
\def\XINT_expr_func_ifone #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:branch{\romannumeral`&&@\xintifOne #3}}%
}%
\let\XINT_flexpr_func_ifone\XINT_expr_func_ifone
\def\XINT_iiexpr_func_ifone #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:branch{\romannumeral`&&@\xintiiifOne #3}}%
}%
\def\XINT_expr_func_ifsgn #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:branch{\romannumeral`&&@\xintiiifSgn #3}}%
}%
\let\XINT_flexpr_func_ifsgn\XINT_expr_func_ifsgn
\let\XINT_iiexpr_func_ifsgn\XINT_expr_func_ifsgn
\def\XINT_expr_func_nuple #1#2#3{#1#2{{#3}}}%
\let\XINT_flexpr_func_nuple\XINT_expr_func_nuple
\let\XINT_iiexpr_func_nuple\XINT_expr_func_nuple
\def\XINT_expr_func_unpack #1#2%#3%
   {\expandafter#1\expandafter#2\romannumeral0\XINT:NEhook:unpack}%
\let\XINT_flexpr_func_unpack\XINT_expr_func_unpack
\let\XINT_iiexpr_func_unpack\XINT_expr_func_unpack
\def\XINT_expr_func_flat #1#2%#3%
{%
    \expandafter#1\expandafter#2\expanded
    \XINT:NEhook:x:flatten\XINT:expr:flatten
}%
\let\XINT_flexpr_func_flat\XINT_expr_func_flat
\let\XINT_iiexpr_func_flat\XINT_expr_func_flat
\let\XINT:NEhook:x:flatten\empty
\def\XINT_expr_func_zip #1#2%#3%
{%
    \expandafter#1\expandafter#2\romannumeral`&&@%
    \XINT:NEhook:x:zip\XINT:expr:zip
}%
\let\XINT_flexpr_func_zip\XINT_expr_func_zip
\let\XINT_iiexpr_func_zip\XINT_expr_func_zip
\let\XINT:NEhook:x:zip\empty
\def\XINT:expr:zip#1{\expandafter{\expanded\XINT_zip_A#1\xint_bye\xint_bye}}%
%    \end{macrocode}
% \subsection{User declared functions}
% \lverb|&
% It is possible that
% the author actually does understand at this time the
% \xintNewExpr/\xintdeffunc refactored code and mechanisms for the first time
% since 2014: past evolutions such as the 2018 1.3 refactoring were done a bit
% in the fog (although they did accomplish a crucial step).
%
% The 1.4 version of function and macro definitions is much more powerful than
% 1.3 one. But the mechanisms such as «omit», «abort» and «break()» in iter()
% et al. can't be translated into much else than their actual code when they
% potentially have to apply to non-numeric only context. The 1.4 \xintdeffunc
% is thus apparently able to digest them but its pre-parsing benefits are
% limited compared to simply assigning such parts of an expression to a
% mock-function created by \xintNewFunction (which creates simply a TeX macro
% from its substitution expression in macro parameters and add
% syntactic sugar to let it appear to \xintexpr as a genuine «function»
% although nothing of the syntax has really been pre-parsed.)
%
% At 1.4 fetching the expression up to final semi-colon is done using
% \XINT_expr_fetch_to_semicolon, hence semi-colons arising in the syntax do
% not need to be hidden inside braces. 
% |
%
% \localtableofcontents
% 
% \subsubsection{\csh{xintdeffunc}, \csh{xintdefiifunc},
% \csh{xintdeffloatfunc}}
%
% \changed[2015/11/12]{1.2c}
% \lverb|Note: it is possible to have same name assigned both to a variable
% and a function: things such as add(f(f), f=1..10) are possible.|
%
% \changed[2015/11/13]{1.2c}
% \lverb|Function names first expanded then detokenized and cleaned of spaces.|
%
% \changed[2015/11/21]{1.2e}
% \lverb|No \detokenize anymore on the function
% names. And #1(#2)#3=#4 parameter pattern to avoid to have to worry if a : is
% there and it is active.|
%
% \changed[2016/02/22]{1.2f}
% \lverb|La macro associée à la fonction ne débute
% plus par un \romannumeral, car de toute façon elle est pour emploi dans
% \csname..\endcsname.|
%
% \changed[2016/03/08]{1.2f}
% \lverb|Comma separated expressions allowed (formerly this required using
% parenthesis \xintdeffunc foo(x,..):=(.., .., ..);|
%
% \changed[2018/06/17]{1.3c}
% \lverb|Usage of \xintexprSafeCatcodes to be compatible with an active
% semi-colon at time of use; the colon was not a problem (see ##3) already.|
%
% \changed{1.3e}
% \lverb|\xintdefefunc variant added for functions which will expand
% completely if used with numeric arguments in other function definitions.
% They can't be used for recursive definitions.|
%
% \changed[2020/01/10]{1.4}
% \lverb|Multi-letter variables can be used (with no prior declaration)|
%
% \changed[2020/01/11]{1.4}
% \lverb|The new internal data model has caused many worries initially (such
% as whether to allow functions with «ople» outputs in contrast to «numbers»
% or «nutples») but in the end all is simpler again and the refactoring of ?
% and ?? in function definitions allows to fuse inert functions (allowing
% recursive definitions) and expanding functions (expanding completely if with
% numeric arguments) into a single entity.
% 
% Thus the 1.3e \xintdefefunc, \xintdefiiefunc, \xintdeffloatefunc constructors
% of «expanding» functions are kept only as aliases of legacy \xintdeffunc et al. and
% deprecated.
%
% A special situation is with
% functions of no variables. In that case it will be handled as an inert
% entity, else they would not be different from variables.
%
% |
%
% \changed[2020/01/19]{1.4}
% \lverb|Addition de la syntaxe déclarative \xintdeffunc foo(a,b,...,*z) = ...;
%
% |
%
%    \begin{macrocode}
\def\XINT_tmpa #1#2#3#4#5%
{%
  \def #1##1(##2)##3={%
   \edef\XINT_deffunc_tmpa {##1}%
   \edef\XINT_deffunc_tmpa {\xint_zapspaces_o \XINT_deffunc_tmpa}%
   \def\XINT_deffunc_tmpb {0}%
   \edef\XINT_deffunc_tmpd {##2}%
   \edef\XINT_deffunc_tmpd {\xint_zapspaces_o\XINT_deffunc_tmpd}%
   \def\XINT_deffunc_tmpe {0}%
   \expandafter#5\romannumeral\XINT_expr_fetch_to_semicolon
  }%  end of \xintdeffunc_a definition
  \def#5##1{%
   \def\XINT_deffunc_tmpc{(##1)}%
   \ifnum\xintLength:f:csv{\XINT_deffunc_tmpd}>\xint_c_
     \xintFor ####1 in {\XINT_deffunc_tmpd}\do
      {%
       \xintifForFirst{\let\XINT_deffunc_tmpd\empty}{}%
       \def\XINT_deffunc_tmpf{####1}%
       \if*\xintFirstItem{####1}%
         \xintifForLast
         {%
          \def\XINT_deffunc_tmpe{1}%
          \edef\XINT_deffunc_tmpf{\xintTrim{1}{####1}}%
         }%
         {%
          \edef\XINT_deffunc_tmpf{\xintTrim{1}{####1}}%
          \xintMessage{xintexpr}{Error}
          {Only the last positional argument can be variadic. Trimmed ####1 to
          \XINT_deffunc_tmpf}%
         }%
       \fi
       \XINT_expr_makedummy{\XINT_deffunc_tmpf}%
       \edef\XINT_deffunc_tmpd{\XINT_deffunc_tmpd{\XINT_deffunc_tmpf}}%
       \edef\XINT_deffunc_tmpb {\the\numexpr\XINT_deffunc_tmpb+\xint_c_i}%
       \edef\XINT_deffunc_tmpc {subs(\unexpanded\expandafter{\XINT_deffunc_tmpc},%
                             \XINT_deffunc_tmpf=################\XINT_deffunc_tmpb)}%
      }%
   \fi
%    \end{macrocode}
% Place holder for comments. Logic at 1.4 is simplified here compared to
% earlier releases.
%
%    \begin{macrocode}
   \ifcase\XINT_deffunc_tmpb\space
     \expandafter\XINT_expr_defuserfunc_none\csname
   \else
     \expandafter\XINT_expr_defuserfunc\csname
   \fi
           XINT_#2_func_\XINT_deffunc_tmpa\expandafter\endcsname
   \csname XINT_#2_userfunc_\XINT_deffunc_tmpa\expandafter\endcsname
   \expandafter{\XINT_deffunc_tmpa}{#2}%
   \expandafter#3\csname XINT_#2_userfunc_\XINT_deffunc_tmpa\endcsname
                              [\XINT_deffunc_tmpb]{\XINT_deffunc_tmpc}%
   \ifxintverbose\xintMessage {xintexpr}{Info}
        {Function \XINT_deffunc_tmpa\space for \string\xint #4 parser
         associated to \string\XINT_#2_userfunc_\XINT_deffunc_tmpa\space
         with \ifxintglobaldefs global \fi meaning \expandafter\meaning
         \csname XINT_#2_userfunc_\XINT_deffunc_tmpa\endcsname}%
   \fi
   \xintFor* ####1 in {\XINT_deffunc_tmpd}:{\xintrestorevariablesilently{####1}}%
   \xintexprRestoreCatcodes
  }% end of \xintdeffunc_b definition
}%
\def\xintdeffunc      {\xintexprSafeCatcodes\xintdeffunc_a}%
\def\xintdefiifunc    {\xintexprSafeCatcodes\xintdefiifunc_a}%
\def\xintdeffloatfunc {\xintexprSafeCatcodes\xintdeffloatfunc_a}%
\XINT_tmpa\xintdeffunc_a     {expr}  \XINT_NewFunc     {expr}\xintdeffunc_b
\XINT_tmpa\xintdefiifunc_a   {iiexpr}\XINT_NewIIFunc   {iiexpr}\xintdefiifunc_b
\XINT_tmpa\xintdeffloatfunc_a{flexpr}\XINT_NewFloatFunc{floatexpr}\xintdeffloatfunc_b
\def\XINT_expr_defuserfunc_none #1#2#3#4%
{%
    \XINT_global
    \def #1##1##2##3%
    {%
         \expandafter##1\expandafter##2\expanded{%
           {\XINT:NEhook:usernoargfunc\csname XINT_#4_userfunc_#3\endcsname}%
         }%
    }%
}%
\let\XINT:NEhook:usernoargfunc \empty
\def\XINT_expr_defuserfunc #1#2#3#4%
{%
    \if0\XINT_deffunc_tmpe
    \XINT_global
    \def #1##1##2%##3%
    {%
         \expandafter ##1\expandafter##2\expanded\bgroup{\iffalse}\fi
         \XINT:NEhook:userfunc{XINT_#4_userfunc_#3}#2%##3%
    }%
    \else
%    \end{macrocode}
% \lverb|Last argument in the call signature is variadic (was prefixed by *).|
%    \begin{macrocode}
    \def #1##1{%
    \XINT_global\def #1####1####2%####3%
    {%
         \expandafter ####1\expandafter####2\expanded\bgroup{\iffalse}\fi
         \XINT:NEhook:userfunc:argv{##1}{XINT_#4_userfunc_#3}#2%####3%
    }}\expandafter#1\expandafter{\the\numexpr\XINT_deffunc_tmpb-1}%
    \fi
}% 
%    \end{macrocode}
% \lverb|Deliberate brace stripping of #3 to reveal the elements of the ople,
% which may be atoms i.e. numeric data such as {1}, or again oples, which
% means that the corresponding item was a nutple, for example it came from
% input syntax such as foo(1, 2, [1, 2], 3), so (up to details of raw
% encoding) {1}{2}{{1}{2}}{3}, which gives 4 braced arguments to macro #2.|
%    \begin{macrocode}
\def\XINT:NEhook:userfunc #1#2#3{#2#3\iffalse{{\fi}}}%
%    \end{macrocode}
% \lverb|Here #1 indicates the number k-1 of standard positional arguments of
% the call signature, the kth and last one having been declared of variadic
% type. The braces around \xintTrim{#1}{#4} have the effect to gather all
% these remaining elements to provide a single one to the TeX macro.
%
% For example input was foo(1,2,3,4,5) and call signature was foo(a,b,*z).
% Then #4 will fetch {{1}{2}{3}{4}{5}}, with one level of brace removal.
% We will have \xintKeep{2}{{1}{2}{3}{4}{5}} which produces {1}{2}.
% Then {\xintTrim{2}{{1}{2}{3}{4}{5}}} which produces {{3}{4}{5}}.
% So the macro will be used as \macro{1}{2}{{3}{4}{5}} having been
% declared as a macro with 3 arguments.
%
% The above comments were added in June 2021 but the code was done on January
% 19, 2020 for 1.4.
%
% Note on June 10, 2021: at core level \XINT_NewFunc is used which is derived
% from \XINT_NewExpr which has always prepared TeX macros with non-delimited
% parameters.  A refactoring could add a final delimiter, for example \relax.
% The macro with 3 arguments would be defined as \def\macro#1#2#3\relax{...}
% for example.  Then we could transfer to TeX core processing what is achieved
% here via \xintKeep/\xintTrim, of course adding efficiency, via insertion of
% the delimiter.  In the case of foo(1,2,3,4,5) we would have the #3 of
% delimited \macro fetch {3}{4}{5}, no brace removal, which is equivalent to
% current situation fetching {{3}{4}{5}} with brace removal.  But let's see in
% case of foo(1,2,3) then.  This would lead to delimited \macro{1}{2}{3}\relax
% and #3 will fetch {3}, removing one brace pair.  Whereas current
% non-delimited \macro is used as \macro{1}{2}{{3}} from the Keep/Trim, then
% #3 fetches {{3}}, removing one brace pair.  Not the same thing.  So it seems
% there is a stumbling-block here to adopt such an alternative method, in
% relation with brace removal.  Rather relieved in fact, as my head starts
% spinning in ople world.  Seems better to stop thinking about doing something
% like that, and what it would imply as consequences for user declarative
% interface also.  Oples are dangerous to mental health, let's stick with
% one-ples: « named arguments in function body declaration must stand for
% one-ples », even the last one, although a priori it could be envisioned if
% foo has been declared with call signature (x,y,z) and is used with more
% items that z is mapped to the ople of extra elements beyond the first
% two ones.  For my sanity I stick with my January 2020 concept of (x,y,*z)
% which makes z stand for a nutple always and having to be used as such in the
% function body (possibly unpacked there using *z).|
%    \begin{macrocode}
\def\XINT:NEhook:userfunc:argv #1#2#3#4%
   {\expandafter#3\expanded{\xintKeep{#1}{#4}{\xintTrim{#1}{#4}}}\iffalse{{\fi}}}%
\let\xintdefefunc\xintdeffunc
\let\xintdefiifunc\xintdefiifunc
\let\xintdeffloatefunc\xintdeffloatfunc
%    \end{macrocode}
% \subsubsection{\csh{xintdefufunc}, \csh{xintdefiiufunc},
% \csh{xintdeffloatufunc}}
%
% \lverb|1.4|
%
%    \begin{macrocode}
\def\XINT_tmpa #1#2#3#4#5#6%
{%
  \def #1##1(##2)##3={%
   \edef\XINT_defufunc_tmpa {##1}%
   \edef\XINT_defufunc_tmpa {\xint_zapspaces_o \XINT_defufunc_tmpa}%
   \edef\XINT_defufunc_tmpd {##2}%
   \edef\XINT_defufunc_tmpd {\xint_zapspaces_o\XINT_defufunc_tmpd}%
   \expandafter#5\romannumeral\XINT_expr_fetch_to_semicolon
  }% end of \xint_defufunc_a
  \def#5##1{%
   \def\XINT_defufunc_tmpc{(##1)}%
   \ifnum\xintLength:f:csv{\XINT_defufunc_tmpd}=\xint_c_i
     \expandafter#6%
   \else
   \xintMessage {xintexpr}{ERROR}
        {Universal functions must be functions of one argument only,
         but the declaration of \XINT_defufunc_tmpa\space
         has \xintLength:f:csv{\XINT_defufunc_tmpd} of them. Cancelled.}%
   \xintexprRestoreCatcodes
   \fi
   }% end of \xint_defufunc_b
  \def #6{%
   \XINT_expr_makedummy{\XINT_defufunc_tmpd}%
   \edef\XINT_defufunc_tmpc {subs(\unexpanded\expandafter{\XINT_defufunc_tmpc},%
                             \XINT_defufunc_tmpd=########1)}%
   \expandafter\XINT_expr_defuserufunc
   \csname XINT_#2_func_\XINT_defufunc_tmpa\expandafter\endcsname
   \csname XINT_#2_userufunc_\XINT_defufunc_tmpa\expandafter\endcsname
   \expandafter{\XINT_defufunc_tmpa}{#2}%
   \expandafter#3\csname XINT_#2_userufunc_\XINT_defufunc_tmpa\endcsname
                              [1]{\XINT_defufunc_tmpc}%
   \ifxintverbose\xintMessage {xintexpr}{Info}
        {Universal function \XINT_defufunc_tmpa\space for \string\xint #4 parser
         associated to \string\XINT_#2_userufunc_\XINT_defufunc_tmpa\space
         with \ifxintglobaldefs global \fi meaning \expandafter\meaning
         \csname XINT_#2_userufunc_\XINT_defufunc_tmpa\endcsname}%
   \fi
  }% end of \xint_defufunc_c
}%
\def\xintdefufunc      {\xintexprSafeCatcodes\xintdefufunc_a}%
\def\xintdefiiufunc    {\xintexprSafeCatcodes\xintdefiiufunc_a}%
\def\xintdeffloatufunc {\xintexprSafeCatcodes\xintdeffloatufunc_a}%
\XINT_tmpa\xintdefufunc_a     {expr}  \XINT_NewFunc     {expr}%
          \xintdefufunc_b\xintdefufunc_c
\XINT_tmpa\xintdefiiufunc_a   {iiexpr}\XINT_NewIIFunc   {iiexpr}%
          \xintdefiiufunc_b\xintdefiiufunc_c
\XINT_tmpa\xintdeffloatufunc_a{flexpr}\XINT_NewFloatFunc{floatexpr}%
          \xintdeffloatufunc_b\xintdeffloatufunc_c
\def\XINT_expr_defuserufunc #1#2#3#4%
{%
    \XINT_global
    \def #1##1##2%##3%
    {%
         \expandafter ##1\expandafter##2\expanded
         \XINT:NEhook:userufunc{XINT_#4_userufunc_#3}#2%##3%
    }%
}%
\def\XINT:NEhook:userufunc #1{\XINT:expr:mapwithin}%
%    \end{macrocode}
% 
% \subsubsection{\csh{xintunassignexprfunc}, \csh{xintunassigniiexprfunc}, \csh{xintunassignfloatexprfunc}}
% See the \csbxint{unassignvar} for the embarrassing explanations why I had
% not done that earlier. A bit lazy here, no warning if undefining something
% not defined, and attention no precaution respective built-in functions.
%    \begin{macrocode}
\def\XINT_tmpa #1{\expandafter\def\csname xintunassign#1func\endcsname ##1{%
   \edef\XINT_unfunc_tmpa{##1}%
   \edef\XINT_unfunc_tmpa {\xint_zapspaces_o\XINT_unfunc_tmpa}%
   \XINT_global\expandafter
     \let\csname XINT_#1_func_\XINT_unfunc_tmpa\endcsname\xint_undefined
   \XINT_global\expandafter
     \let\csname XINT_#1_userfunc_\XINT_unfunc_tmpa\endcsname\xint_undefined
   \XINT_global\expandafter
     \let\csname XINT_#1_userufunc_\XINT_unfunc_tmpa\endcsname\xint_undefined
   \ifxintverbose\xintMessage {xintexpr}{Info}
    {Function \XINT_unfunc_tmpa\space for \string\xint #1 parser now
     \ifxintglobaldefs globally \fi undefined.}%
   \fi}}%
\XINT_tmpa{expr}\XINT_tmpa{iiexpr}\XINT_tmpa{floatexpr}%
%    \end{macrocode}
% \subsubsection{\csh{xintNewFunction}}
% \lverb|1.2h (2016/11/20). Syntax is \xintNewFunction{<name>}[nb of
% arguments]{expression with #1, #2,... as in \xintNewExpr}. This defines
% a function for all three parsers but the expression parsing is delayed until
% function execution. Hence the expression admits all constructs, contrarily
% to \xintNewExpr or \xintdeffunc.
%
% As the letters used for variables in \xintdeffunc, #1, #2, etc... can not
% stand for non numeric «oples», because at time of function call f(a, b, c,
% ...) how to decide if #1 stands for a or a, b etc... ? Or course «a» can be
% packed and thus the macro function can handle #1 as a «nutple» and for this
% be defined with the * unpacking operator being applied to it.
% |
%    \begin{macrocode}
\def\xintNewFunction #1#2[#3]#4%
{%
  \edef\XINT_newfunc_tmpa {#1}%
  \edef\XINT_newfunc_tmpa {\xint_zapspaces_o \XINT_newfunc_tmpa}%
  \def\XINT_newfunc_tmpb ##1##2##3##4##5##6##7##8##9{#4}%
  \begingroup
    \ifcase #3\relax
        \toks0{}%
    \or \toks0{##1}%
    \or \toks0{##1##2}%
    \or \toks0{##1##2##3}%
    \or \toks0{##1##2##3##4}%
    \or \toks0{##1##2##3##4##5}%
    \or \toks0{##1##2##3##4##5##6}%
    \or \toks0{##1##2##3##4##5##6##7}%
    \or \toks0{##1##2##3##4##5##6##7##8}%
    \else \toks0{##1##2##3##4##5##6##7##8##9}%
    \fi
    \expandafter
  \endgroup\expandafter
  \XINT_global\expandafter
  \def\csname XINT_expr_macrofunc_\XINT_newfunc_tmpa\expandafter\endcsname
  \the\toks0\expandafter{\XINT_newfunc_tmpb
    {\XINTfstop.{{##1}}}{\XINTfstop.{{##2}}}{\XINTfstop.{{##3}}}%
    {\XINTfstop.{{##4}}}{\XINTfstop.{{##5}}}{\XINTfstop.{{##6}}}%
    {\XINTfstop.{{##7}}}{\XINTfstop.{{##8}}}{\XINTfstop.{{##9}}}}%
  \expandafter\XINT_expr_newfunction
    \csname XINT_expr_func_\XINT_newfunc_tmpa\expandafter\endcsname
    \expandafter{\XINT_newfunc_tmpa}\xintbareeval
  \expandafter\XINT_expr_newfunction
    \csname XINT_iiexpr_func_\XINT_newfunc_tmpa\expandafter\endcsname
    \expandafter{\XINT_newfunc_tmpa}\xintbareiieval
  \expandafter\XINT_expr_newfunction
    \csname XINT_flexpr_func_\XINT_newfunc_tmpa\expandafter\endcsname
    \expandafter{\XINT_newfunc_tmpa}\xintbarefloateval
  \ifxintverbose
    \xintMessage {xintexpr}{Info}
        {Function \XINT_newfunc_tmpa\space for the expression parsers is
         associated to \string\XINT_expr_macrofunc_\XINT_newfunc_tmpa\space
         with \ifxintglobaldefs global \fi meaning \expandafter\meaning
         \csname XINT_expr_macrofunc_\XINT_newfunc_tmpa\endcsname}%
  \fi
}%
\def\XINT_expr_newfunction #1#2#3%
{%
    \XINT_global
    \def#1##1##2##3%
      {\expandafter ##1\expandafter ##2%
       \romannumeral0\XINT:NEhook:macrofunc
       #3{\csname XINT_expr_macrofunc_#2\endcsname##3}\relax
      }%
}%
\let\XINT:NEhook:macrofunc\empty
%    \end{macrocode}
% \subsubsection{Mysterious stuff}
%
% There was an |\xintNewExpr| already in 1.07 from May 2013, which was
% modified in September 2013 to work with the \# macro parameter character,
% and then refactored into a more powerful version in June 2014 for 1.1
% release of 2014/10/28.
%
% It is always too soon to try to comment and explain. In brief, this attempts
% to hack into the \emph{purely numeric} |\xintexpr| parsers to transform them
% into \emph{symbolic} parsers, allowing to do once and for all the parsing
% job and inherit a gigantic nested macro. Originally only f-expandable
% nesting. The initial motivation was that the |\csname| encapsulation impacted
% the string pool memory. Later this work proved to be the basis to provide
% support for implementing
% user-defined functions and it is now its main purpose.
%
% Deep refactorings happened at 1.3 and 1.4.
%
% At 1.3 the crucial idea of the «hook» macros was introduced, reducing
% considerably the preparatory work done by |\xintNewExpr|.
%
% At 1.4 further considerable simplifications happened, and it is possible
% that the author currently does at long last understand the code!
%
% The 1.3 code had serious complications with trying
% to identify would-be «list» arguments, distinguishing them from «single»
% arguments (things like parsing |#2+[[#1..[#3]..#4][#5:#6]]*#7| and convert
% it to a single nested f-exandable macro...)
%
% The conversion at 1.4 is both more powerful and simpler, due in part to the
% new storage model which from |\csname| encapsulated comma separated values
% up to 1.3f became simply a braced list of braced values, and also crucially
% due to the possibilities opened up by usage of |\expanded| primitive.
%
%    \begin{macrocode}
\catcode`~ 12
\def\XINT:NE:hastilde#1~#2#3\relax{\unless\if !#21\fi}%
\def\XINT:NE:hashash#1{%
\def\XINT:NE:hashash##1#1##2##3\relax{\unless\if !##21\fi}%
}\expandafter\XINT:NE:hashash\string#%
\def\XINT:NE:unpack #1{%
\def\XINT:NE:unpack ##1%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0\else
        \expandafter\XINT:NE:unpack:p\fi
    \xint_stop_atfirstofone{##1}%
}}\expandafter\XINT:NE:unpack\string#%
\def\XINT:NE:unpack:p#1#2%
    {{~romannumeral0~expandafter~xint_stop_atfirstofone~expanded{#2}}}%
\def\XINT:NE:f:one:from:one #1{%
\def\XINT:NE:f:one:from:one ##1%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0\else
        \xint_dothis\XINT:NE:f:one:from:one_a\fi
    \xint_orthat\XINT:NE:f:one:from:one_b
    ##1&&A%
}}\expandafter\XINT:NE:f:one:from:one\string#%
\def\XINT:NE:f:one:from:one_a\romannumeral`&&@#1#2&&A%
{%
    \expandafter{\detokenize{\expandafter#1}#2}%
}%
\def\XINT:NE:f:one:from:one_b#1{%
\def\XINT:NE:f:one:from:one_b\romannumeral`&&@##1##2&&A%
{%
    \expandafter{\romannumeral`&&@%
        \if0\XINT:NE:hastilde ##2~!\relax
            \XINT:NE:hashash  ##2#1!\relax 0\else
        \expandafter\string\fi
    ##1{##2}}%
}}\expandafter\XINT:NE:f:one:from:one_b\string#%
\def\XINT:NE:f:one:from:one:direct #1#2{\XINT:NE:f:one:from:one:direct_a #2&&A{#1}}%
\def\XINT:NE:f:one:from:one:direct_a #1#2&&A#3%
{%
    \if ###1\xint_dothis {\detokenize{#3}}\fi
    \if  ~#1\xint_dothis {\detokenize{#3}}\fi
    \xint_orthat {#3}{#1#2}%
}%
\def\XINT:NE:f:one:from:two #1{%
\def\XINT:NE:f:one:from:two ##1%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0\else
        \xint_dothis\XINT:NE:f:one:from:two_a\fi
    \xint_orthat\XINT:NE:f:one:from:two_b ##1&&A%
}}\expandafter\XINT:NE:f:one:from:two\string#%
\def\XINT:NE:f:one:from:two_a\romannumeral`&&@#1#2&&A%
{%
    \expandafter{\detokenize{\expandafter#1\expanded}{#2}}%
}%
\def\XINT:NE:f:one:from:two_b#1{%
\def\XINT:NE:f:one:from:two_b\romannumeral`&&@##1##2##3&&A%
{%
    \expandafter{\romannumeral`&&@%
        \if0\XINT:NE:hastilde ##2##3~!\relax
            \XINT:NE:hashash  ##2##3#1!\relax 0\else
        \expandafter\string\fi
    ##1{##2}{##3}}%
}}\expandafter\XINT:NE:f:one:from:two_b\string#%
\def\XINT:NE:f:one:from:two:direct #1#2#3{\XINT:NE:two_fork #2&&A#3&&A#1{#2}{#3}}%
\def\XINT:NE:two_fork #1#2&&A#3#4&&A{\XINT:NE:two_fork_nn#1#3}%
\def\XINT:NE:two_fork_nn #1#2%
{%
    \if #1##\xint_dothis\string\fi
    \if  #1~\xint_dothis\string\fi
    \if #2##\xint_dothis\string\fi
    \if  #2~\xint_dothis\string\fi
    \xint_orthat{}%
}%
\def\XINT:NE:f:one:and:opt:direct#1{%
\def\XINT:NE:f:one:and:opt:direct##1!%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0\else
        \xint_dothis\XINT:NE:f:one:and:opt_a\fi
    \xint_orthat\XINT:NE:f:one:and:opt_b ##1&&A%
}}\expandafter\XINT:NE:f:one:and:opt:direct\string#%
\def\XINT:NE:f:one:and:opt_a #1#2&&A#3#4%
{%
    \detokenize{\romannumeral-`0\expandafter#1\expanded{#2}$XINT_expr_exclam#3#4}%$
}%
\def\XINT:NE:f:one:and:opt_b\XINT:expr:f:one:and:opt #1#2#3&&A#4#5%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                     \expandafter\xint_secondoftwo\fi
    {\XINT:NE:f:one:from:one:direct#4}%
    {\expandafter\XINT:NE:f:onewithopttoone\expandafter#5%
        \expanded{{\XINT:NE:f:one:from:one:direct\xintNum{#2}}}}%
    {#1}%
}%
\def\XINT:NE:f:onewithopttoone#1#2#3{\XINT:NE:two_fork #2&&A#3&&A#1[#2]{#3}}%
\def\XINT:NE:f:tacitzeroifone:direct#1{%
\def\XINT:NE:f:tacitzeroifone:direct##1!%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0\else
        \xint_dothis\XINT:NE:f:one:and:opt_a\fi
    \xint_orthat\XINT:NE:f:tacitzeroifone_b ##1&&A%
}}\expandafter\XINT:NE:f:tacitzeroifone:direct\string#%
\def\XINT:NE:f:tacitzeroifone_b\XINT:expr:f:tacitzeroifone #1#2#3&&A#4#5%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                      \expandafter\xint_secondoftwo\fi
    {\XINT:NE:f:one:from:two:direct#4{0}}%
    {\expandafter\XINT:NE:f:one:from:two:direct\expandafter#5%
        \expanded{{\XINT:NE:f:one:from:one:direct\xintNum{#2}}}}%
    {#1}%
}%
\def\XINT:NE:f:iitacitzeroifone:direct#1{%
\def\XINT:NE:f:iitacitzeroifone:direct##1!%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0\else
        \xint_dothis\XINT:NE:f:iitacitzeroifone_a\fi
    \xint_orthat\XINT:NE:f:iitacitzeroifone_b ##1&&A%
}}\expandafter\XINT:NE:f:iitacitzeroifone:direct\string#%
\def\XINT:NE:f:iitacitzeroifone_a #1#2&&A#3%
{%
    \detokenize{\romannumeral`$XINT_expr_null\expandafter#1\expanded{#2}$XINT_expr_exclam#3}%
}%
\def\XINT:NE:f:iitacitzeroifone_b\XINT:expr:f:iitacitzeroifone #1#2#3&&A#4%
{%
    \if\relax#3\relax\expandafter\xint_firstoftwo\else
                      \expandafter\xint_secondoftwo\fi
    {\XINT:NE:f:one:from:two:direct#4{0}}%
    {\XINT:NE:f:one:from:two:direct#4{#2}}%
    {#1}%
}%
\def\XINT:NE:x:one:from:two #1#2#3{\XINT:NE:x:one:from:two_fork #2&&A#3&&A#1{#2}{#3}}%
\def\XINT:NE:x:one:from:two_fork #1{%
\def\XINT:NE:x:one:from:two_fork ##1##2&&A##3##4&&A%
{%
    \if0\XINT:NE:hastilde ##1##3~!\relax\XINT:NE:hashash ##1##3#1!\relax 0%
    \else
        \expandafter\XINT:NE:x:one:from:two:p
    \fi
}}\expandafter\XINT:NE:x:one:from:two_fork\string#%
\def\XINT:NE:x:one:from:two:p #1#2#3%
   {~expanded{\detokenize{\expandafter#1}~expanded{{#2}{#3}}}}%
\def\XINT:NE:x:listsel #1{%
\def\XINT:NE:x:listsel ##1##2&%
{%
    \if0\expandafter\XINT:NE:hastilde\detokenize{##2}~!\relax
        \expandafter\XINT:NE:hashash\detokenize{##2}#1!\relax 0%
    \else
        \expandafter\XINT:NE:x:listsel:p
    \fi
    ##1##2&%
}}\expandafter\XINT:NE:x:listsel\string#%
\def\XINT:NE:x:listsel:p #1#2_#3&(#4%
{%
    \detokenize{\expanded\XINT:expr:ListSel{{#3}{#4}}}%
}%
\def\XINT:expr:ListSel{\expandafter\XINT:expr:ListSel_i\expanded}%
\def\XINT:expr:ListSel_i #1#2{{\XINT_ListSel_top #2_#1&({#2}}}%
\def\XINT:NE:f:reverse #1{%
\def\XINT:NE:f:reverse ##1^%
{%
    \if0\expandafter\XINT:NE:hastilde\detokenize\expandafter{\xint_gobble_i##1}~!\relax
        \expandafter\XINT:NE:hashash\detokenize{##1}#1!\relax 0%
    \else
        \expandafter\XINT:NE:f:reverse:p
    \fi
    ##1^%
}}\expandafter\XINT:NE:f:reverse\string#%
\def\XINT:NE:f:reverse:p #1^#2\xint_bye
{%
    \expandafter\XINT:NE:f:reverse:p_i\expandafter{\xint_gobble_i#1}%
}%
\def\XINT:NE:f:reverse:p_i #1%
{%
    \detokenize{\romannumeral0\XINT:expr:f:reverse{{#1}}}%
}%
\def\XINT:expr:f:reverse{\expandafter\XINT:expr:f:reverse_i\expanded}%
\def\XINT:expr:f:reverse_i #1%
{%
    \XINT_expr_reverse #1^^#1\xint:\xint:\xint:\xint:
                         \xint:\xint:\xint:\xint:\xint_bye
}%
\def\XINT:NE:f:from:delim:u #1{%
\def\XINT:NE:f:from:delim:u ##1##2^%
{%
    \if0\expandafter\XINT:NE:hastilde\detokenize{##2}~!\relax
        \expandafter\XINT:NE:hashash\detokenize{##2}#1!\relax 0%
        \xint_afterfi{\expandafter\XINT_fooof_checkifnumber\expandafter##1\string}%
    \else
        \xint_afterfi{\XINT:NE:f:from:delim:u:p##1\empty}%
    \fi
    ##2^%
}}\expandafter\XINT:NE:f:from:delim:u\string#%
\def\XINT:NE:f:from:delim:u:p #1#2^%
{%
    \detokenize{\expandafter\XINT:fooof:checkifnumber\expandafter#1}~expanded{#2}$XINT_expr_caret%$
}%
\def\XINT:fooof:checkifnumber#1{\expandafter\XINT_fooof_checkifnumber\expandafter#1\string}%
\def\XINT:NE:f:LFL#1#2{\expandafter\XINT:NE:f:LFL_a\expandafter#1#2\XINT:NE:f:LFL_a}%
\def\XINT:NE:f:LFL_a#1#2%
{%
    \if#2i\else\expandafter\XINT:NE:f:LFL_p
    \fi #1%
}%
\def\XINT:NE:r:check#1{%
\def\XINT:NE:r:check##1\XINT:NE:f:LFL_a
{%
    \if0\expandafter\XINT:NE:hastilde\detokenize{##1}~!\relax%
        \expandafter\XINT:NE:hashash\detokenize{##1}#1!\relax 0%
    \else
        \expandafter\XINT:NE:r:check:p
    \fi
    1\expandafter{\romannumeral\XINT:NEsaved:r:check##1}%
}}\expandafter\XINT:NE:r:check\string#%
\def\XINT:NE:r:check:p 1\expandafter#1{\XINT:NE:r:check:p_i#1}%
\def\XINT:NE:r:check:p_i\romannumeral\XINT:NEsaved:r:check{\XINT:NE:r:check:p_ii\empty}%
\def\XINT:NE:r:check:p_ii#1^%
{%
    5~expanded{{~romannumeral~XINT:NEsaved:r:check#1$XINT_expr_caret}}%$
}%
\def\XINT:NE:f:LFL_p#1%
{%
    \detokenize{\romannumeral`$XINT_expr_null\expandafter#1}%$
}%
\catcode`- 11
\def\XINT:NE:exec_? #1#2%
{%
    \XINT:NE:exec_?_b #2&&A#1{#2}%
}%
\def\XINT:NE:exec_?_b #1{%
\def\XINT:NE:exec_?_b ##1&&A%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0%
    \xint_dothis\XINT:NE:exec_?:x\fi
    \xint_orthat\XINT:NE:exec_?:p
}}\expandafter\XINT:NE:exec_?_b\string#%
\def\XINT:NE:exec_?:x #1#2#3%
{%
    \expandafter\XINT_expr_check-_after?\expandafter#1%
    \romannumeral`&&@\expandafter\XINT_expr_getnext\romannumeral0\xintiiifnotzero#3%
}%
\def\XINT:NE:exec_?:p #1#2#3#4#5%
{%
    \csname XINT_expr_func_*If\expandafter\endcsname
    \romannumeral`&&@#2\XINTfstop.{#3},[#4],[#5])%
}%
\expandafter\def\csname XINT_expr_func_*If\endcsname #1#2#3%
{%
    #1#2{~expanded{~xintiiifNotZero#3}}%
}%
\def\XINT:NE:exec_?? #1#2#3%
{%
    \XINT:NE:exec_??_b #2&&A#1{#2}%
}%
\def\XINT:NE:exec_??_b #1{%
\def\XINT:NE:exec_??_b ##1&&A%
{%
    \if0\XINT:NE:hastilde ##1~!\relax
        \XINT:NE:hashash  ##1#1!\relax 0%
    \xint_dothis\XINT:NE:exec_??:x\fi
    \xint_orthat\XINT:NE:exec_??:p
}}\expandafter\XINT:NE:exec_??_b\string#%
\def\XINT:NE:exec_??:x #1#2#3%
{%
    \expandafter\XINT_expr_check-_after?\expandafter#1%
    \romannumeral`&&@\expandafter\XINT_expr_getnext\romannumeral0\xintiiifsgn#3%
}%
\def\XINT:NE:exec_??:p #1#2#3#4#5#6%
{%
    \csname XINT_expr_func_*IfSgn\expandafter\endcsname
    \romannumeral`&&@#2\XINTfstop.{#3},[#4],[#5],[#6])%
}%
\expandafter\def\csname XINT_expr_func_*IfSgn\endcsname #1#2#3%
{%
    #1#2{~expanded{~xintiiifSgn#3}}%
}%
\catcode`- 12
\def\XINT:NE:branch #1%
{%
    \if0\XINT:NE:hastilde #1~!\relax 0\else
        \xint_dothis\XINT:NE:branch_a\fi
    \xint_orthat\XINT:NE:branch_b #1&&A%
}%
\def\XINT:NE:branch_a\romannumeral`&&@#1#2&&A%
{%
    \expandafter{\detokenize{\expandafter#1\expanded}{#2}}%
}%
\def\XINT:NE:branch_b#1{%
\def\XINT:NE:branch_b\romannumeral`&&@##1##2##3&&A%
{%
    \expandafter{\romannumeral`&&@%
        \if0\XINT:NE:hastilde ##2~!\relax
            \XINT:NE:hashash  ##2#1!\relax 0\else
        \expandafter\string\fi
    ##1{##2}##3}%
}}\expandafter\XINT:NE:branch_b\string#%
\def\XINT:NE:seqx#1{%
\def\XINT:NE:seqx\XINT_allexpr_seqx##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:seqx:p
    \fi \XINT_allexpr_seqx{##1}{##2}%
}}\expandafter\XINT:NE:seqx\string#%
\def\XINT:NE:seqx:p\XINT_allexpr_seqx #1#2#3#4%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded {%
    {%
        \detokenize
        {%
           \expanded\bgroup
           \expanded
           {\unexpanded{\XINT_expr_seq:_b{#1#4\relax $XINT_expr_exclam #3}}%
             #2$XINT_expr_caret}%
        }%
    }%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT:NE:opx#1{%
\def\XINT:NE:opx\XINT_allexpr_opx ##1##2##3##4%##5##6##7##8%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##4}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##4}#1!\relax 0%
    \else
      \expandafter\XINT:NE:opx:p
    \fi \XINT_allexpr_opx ##1{##2}{##3}{##4}% en fait ##2 = \xint_c_, ##3 = \relax
}}\expandafter\XINT:NE:opx\string#%
\def\XINT:NE:opx:p\XINT_allexpr_opx #1#2#3#4#5#6#7#8%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded {%
    {%
        \detokenize
        {%
            \expanded\bgroup
            \expanded{\unexpanded{\XINT_expr_iter:_b
               {#1\expandafter\XINT_allexpr_opx_ifnotomitted
                  \romannumeral0#1#6\relax#7@\relax $XINT_expr_exclam #5}}%
                      #4$XINT_expr_caret$XINT_expr_tilde{{#8}}}%$
        }%
    }%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT:NE:iter{\expandafter\XINT:NE:itery\expandafter}%
\def\XINT:NE:itery#1{%
\def\XINT:NE:itery\XINT_expr_itery##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##1##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##1##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:itery:p
    \fi \XINT_expr_itery{##1}{##2}%
}}\expandafter\XINT:NE:itery\string#%
\def\XINT:NE:itery:p\XINT_expr_itery #1#2#3#4#5%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded {%
    {%
        \detokenize
        {%
          \expanded\bgroup
          \expanded{\unexpanded{\XINT_expr_iter:_b {#5#4\relax $XINT_expr_exclam #3}}%
                    #1$XINT_expr_caret$XINT_expr_tilde{#2}}%$
        }%
    }%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT:NE:rseq{\expandafter\XINT:NE:rseqy\expandafter}%
\def\XINT:NE:rseqy#1{%
\def\XINT:NE:rseqy\XINT_expr_rseqy##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##1##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##1##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:rseqy:p
    \fi \XINT_expr_rseqy{##1}{##2}%
}}\expandafter\XINT:NE:rseqy\string#%
\def\XINT:NE:rseqy:p\XINT_expr_rseqy #1#2#3#4#5%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded {%
    {%
        \detokenize
        {%
          \expanded\bgroup
          \expanded{#2\unexpanded{\XINT_expr_rseq:_b {#5#4\relax $XINT_expr_exclam #3}}%
                      #1$XINT_expr_caret$XINT_expr_tilde{#2}}%$
         }%
    }%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT:NE:iterr{\expandafter\XINT:NE:iterry\expandafter}%
\def\XINT:NE:iterry#1{%
\def\XINT:NE:iterry\XINT_expr_iterry##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##1##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##1##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:iterry:p
    \fi \XINT_expr_iterry{##1}{##2}%
}}\expandafter\XINT:NE:iterry\string#%
\def\XINT:NE:iterry:p\XINT_expr_iterry #1#2#3#4#5%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded {%
    {%
        \detokenize
        {%
          \expanded\bgroup
          \expanded{\unexpanded{\XINT_expr_iterr:_b {#5#4\relax $XINT_expr_exclam #3}}%
                    #1$XINT_expr_caret$XINT_expr_tilde #20$XINT_expr_qmark}%
        }%
    }%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT:NE:rrseq{\expandafter\XINT:NE:rrseqy\expandafter}%
\def\XINT:NE:rrseqy#1{%
\def\XINT:NE:rrseqy\XINT_expr_rrseqy##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##1##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##1##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:rrseqy:p
    \fi \XINT_expr_rrseqy{##1}{##2}%
}}\expandafter\XINT:NE:rrseqy\string#%
\def\XINT:NE:rrseqy:p\XINT_expr_rrseqy #1#2#3#4#5#6%
{%
    \expandafter\XINT_expr_put_op_first
    \expanded {%
    {%
        \detokenize
        {%
           \expanded\bgroup
           \expanded{#2\unexpanded{\XINT_expr_rrseq:_b {#6#5\relax $XINT_expr_exclam #4}}%
                        #1$XINT_expr_caret$XINT_expr_tilde #30$XINT_expr_qmark}%
         }%
    }%
    \expandafter}\romannumeral`&&@\XINT_expr_getop
}%
\def\XINT:NE:x:toblist#1{%
\def\XINT:NE:x:toblist\XINT:expr:toblistwith##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:x:toblist:p
    \fi \XINT:expr:toblistwith{##1}{##2}%
}}\expandafter\XINT:NE:x:toblist\string#%
\def\XINT:NE:x:toblist:p\XINT:expr:toblistwith #1#2{{\XINTfstop.{#2}}}%
\def\XINT:NE:x:flatten#1{%
\def\XINT:NE:x:flatten\XINT:expr:flatten##1%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##1}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##1}#1!\relax 0%
    \else
      \expandafter\XINT:NE:x:flatten:p
    \fi \XINT:expr:flatten{##1}%
}}\expandafter\XINT:NE:x:flatten\string#%
\def\XINT:NE:x:flatten:p\XINT:expr:flatten #1%
{%
    {{%
     \detokenize
     {%
         \expandafter\XINT:expr:flatten_checkempty
         \detokenize\expandafter{\expanded{#1}}$XINT_expr_caret%$
     }%
    }}%
}%
\def\XINT:NE:x:zip#1{%
\def\XINT:NE:x:zip\XINT:expr:zip##1%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##1}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##1}#1!\relax 0%
    \else
      \expandafter\XINT:NE:x:zip:p
    \fi \XINT:expr:zip{##1}%
}}\expandafter\XINT:NE:x:zip\string#%
\def\XINT:NE:x:zip:p\XINT:expr:zip #1%
{%
    \expandafter{%
     \detokenize
     {%
        \expanded\expandafter\XINT_zip_A\expanded{#1}\xint_bye\xint_bye
     }%
    }%
}%
\def\XINT:NE:x:mapwithin#1{%
\def\XINT:NE:x:mapwithin\XINT:expr:mapwithin ##1##2%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:x:mapwithin:p
    \fi \XINT:expr:mapwithin {##1}{##2}%
}}\expandafter\XINT:NE:x:mapwithin\string#%
\def\XINT:NE:x:mapwithin:p \XINT:expr:mapwithin  #1#2%
{%
    {{%
     \detokenize
     {%
%%      \expanded
%%      {%
         \expandafter\XINT:expr:mapwithin_checkempty
         \expanded{\noexpand#1$XINT_expr_exclam\expandafter}%$
         \detokenize\expandafter{\expanded{#2}}$XINT_expr_caret%$
%%      }%
     }%
    }}%
}%
\def\XINT:NE:x:ndmapx#1{%
\def\XINT:NE:x:ndmapx\XINT_allexpr_ndmapx_a ##1##2^%
{%
    \if 0\expandafter\XINT:NE:hastilde\detokenize{##2}~!\relax
         \expandafter\XINT:NE:hashash \detokenize{##2}#1!\relax 0%
    \else
      \expandafter\XINT:NE:x:ndmapx:p
    \fi \XINT_allexpr_ndmapx_a ##1##2^%
}}\expandafter\XINT:NE:x:ndmapx\string#%
\def\XINT:NE:x:ndmapx:p #1#2#3^\relax
{%
    \detokenize
    {%
      \expanded{%
      \expandafter#1\expandafter#2\expanded{#3}$XINT_expr_caret\relax %$
               }%
    }%
}%
%    \end{macrocode}
% \lverb|&
% Attention here that user function
% names may contain digits, so we don't use a \detokenize or ~ approach.
%
% This syntax means that a function defined by \xintdeffunc never expands
% when used in another definition, so it can implement recursive definitions.
%
% \XINT:NE:userefunc et al. added at 1.3e.
%
% I added at \xintdefefunc, \xintdefiiefunc, \xintdeffloatefunc at 1.3e to on
% the contrary expand if possible (i.e. if used only with numeric arguments)
% in another definition.
%
% The \XINTusefunc
% uses \expanded. Its ancestor \xintExpandArgs (xinttools 1.3)
% had some more primitive f-expansion technique.
% |
%    \begin{macrocode}
\def\XINTusenoargfunc #1%
{%
    0\csname #1\endcsname
}%
\def\XINT:NE:usernoargfunc\csname #1\endcsname
{%
    ~romannumeral~XINTusenoargfunc{#1}%
}%
\def\XINTusefunc #1%
{%
    0\csname #1\expandafter\endcsname\expanded
}%
\def\XINT:NE:usefunc #1#2#3%
{%
    ~romannumeral~XINTusefunc{#1}{#3}\iffalse{{\fi}}%
}%
\def\XINTuseufunc #1%
{%
    \expanded\expandafter\XINT:expr:mapwithin\csname #1\expandafter\endcsname\expanded
}%
\def\XINT:NE:useufunc #1#2#3%
{%
    {{~expanded~XINTuseufunc{#1}{#3}}}%
}%
\def\XINT:NE:userfunc #1{%
\def\XINT:NE:userfunc ##1##2##3%
{%
    \if0\expandafter\XINT:NE:hastilde\detokenize{##3}~!\relax
        \expandafter\XINT:NE:hashash\detokenize{##3}#1!\relax 0%
      \expandafter\XINT:NE:userfunc_x
    \else
      \expandafter\XINT:NE:usefunc
    \fi {##1}{##2}{##3}%
}}\expandafter\XINT:NE:userfunc\string#%
\def\XINT:NE:userfunc_x #1#2#3{#2#3\iffalse{{\fi}}}%
\def\XINT:NE:userufunc #1{%
\def\XINT:NE:userufunc ##1##2##3%
{%
    \if0\expandafter\XINT:NE:hastilde\detokenize{##3}~!\relax
        \expandafter\XINT:NE:hashash\detokenize{##3}#1!\relax 0%
      \expandafter\XINT:NE:userufunc_x
    \else
      \expandafter\XINT:NE:useufunc
    \fi {##1}{##2}{##3}%
}}\expandafter\XINT:NE:userufunc\string#%
\def\XINT:NE:userufunc_x #1{\XINT:expr:mapwithin}%
\def\XINT:NE:macrofunc #1#2%
   {\expandafter\XINT:NE:macrofunc:a\string#1#2\empty&}%
\def\XINT:NE:macrofunc:a#1\csname #2\endcsname#3&%
   {{~XINTusemacrofunc{#1}{#2}{#3}}}%
\def\XINTusemacrofunc #1#2#3%
{%
    \romannumeral0\expandafter\xint_stop_atfirstofone
    \romannumeral0#1\csname #2\endcsname#3\relax
}%
%    \end{macrocode}
% \subsubsection{\csh{XINT_expr_redefinemacros}}
% \lverb|Completely refactored at 1.3.
%
% Again refactored at 1.4. The availability of \expanded allows more powerful
% mechanisms and more importantly I better thought out the root problems
% caused by the handling of list operations in this context and this helped
% simplify considerably the code.
%
% |
%    \begin{macrocode}
\catcode`- 11
\def\XINT_expr_redefinemacros {%
  \let\XINT:NEhook:unpack           \XINT:NE:unpack
  \let\XINT:NEhook:f:one:from:one        \XINT:NE:f:one:from:one
  \let\XINT:NEhook:f:one:from:one:direct \XINT:NE:f:one:from:one:direct
  \let\XINT:NEhook:f:one:from:two        \XINT:NE:f:one:from:two
  \let\XINT:NEhook:f:one:from:two:direct \XINT:NE:f:one:from:two:direct
  \let\XINT:NEhook:x:one:from:two        \XINT:NE:x:one:from:two
  \let\XINT:NEhook:f:one:and:opt:direct      \XINT:NE:f:one:and:opt:direct
  \let\XINT:NEhook:f:tacitzeroifone:direct   \XINT:NE:f:tacitzeroifone:direct
  \let\XINT:NEhook:f:iitacitzeroifone:direct \XINT:NE:f:iitacitzeroifone:direct
  \let\XINT:NEhook:x:listsel             \XINT:NE:x:listsel
  \let\XINT:NEhook:f:reverse             \XINT:NE:f:reverse
  \let\XINT:NEhook:f:from:delim:u        \XINT:NE:f:from:delim:u
  \let\XINT:NEhook:f:LFL            \XINT:NE:f:LFL
  \let\XINT:NEhook:r:check          \XINT:NE:r:check
  \let\XINT:NEhook:branch           \XINT:NE:branch
  \let\XINT:NEhook:seqx             \XINT:NE:seqx
  \let\XINT:NEhook:opx              \XINT:NE:opx
  \let\XINT:NEhook:rseq             \XINT:NE:rseq
  \let\XINT:NEhook:iter             \XINT:NE:iter
  \let\XINT:NEhook:rrseq            \XINT:NE:rrseq
  \let\XINT:NEhook:iterr            \XINT:NE:iterr
  \let\XINT:NEhook:x:toblist        \XINT:NE:x:toblist
  \let\XINT:NEhook:x:flatten        \XINT:NE:x:flatten
  \let\XINT:NEhook:x:zip            \XINT:NE:x:zip
  \let\XINT:NEhook:x:mapwithin      \XINT:NE:x:mapwithin
  \let\XINT:NEhook:x:ndmapx         \XINT:NE:x:ndmapx
  \let\XINT:NEhook:userfunc         \XINT:NE:userfunc
  \let\XINT:NEhook:userufunc        \XINT:NE:userufunc
  \let\XINT:NEhook:usernoargfunc    \XINT:NE:usernoargfunc
  \let\XINT:NEhook:macrofunc        \XINT:NE:macrofunc
  \def\XINTinRandomFloatSdigits{~XINTinRandomFloatSdigits }%
  \def\XINTinRandomFloatSixteen{~XINTinRandomFloatSixteen }%
  \def\xintiiRandRange{~xintiiRandRange }%
  \def\xintiiRandRangeAtoB{~xintiiRandRangeAtoB }%
  \def\xintRandBit{~xintRandBit }%
  \let\XINT_expr_exec_?  \XINT:NE:exec_?
  \let\XINT_expr_exec_?? \XINT:NE:exec_??
  \def\XINT_expr_op_?  {\XINT_expr_op__?{\XINT_expr_op_-xii\XINT_expr_oparen}}%
  \def\XINT_flexpr_op_?{\XINT_expr_op__?{\XINT_flexpr_op_-xii\XINT_flexpr_oparen}}%
  \def\XINT_iiexpr_op_?{\XINT_expr_op__?{\XINT_iiexpr_op_-xii\XINT_iiexpr_oparen}}%
}%
\catcode`- 12
%    \end{macrocode}
% \subsubsection{\csh{xintNewExpr}, \csh{xintNewIExpr}, \csh{xintNewFloatExpr},
% \csh{xintNewIIExpr}}
% \lverb|&
% 1.2c modifications to accomodate \XINT_expr_deffunc_newexpr etc..
%
% 1.2f adds token \XINT_newexpr_clean to be able to have a different
% \XINT_newfunc_clean.
%
% As \XINT_NewExpr always execute \XINT_expr_redefineprints since 1.3e whether
% with \xintNewExpr or \XINT_NewFunc, it has been moved from argument to
% hardcoded in replacement text.
%
% NO MORE \XINT_expr_redefineprints at 1.4 ! This allows better support for
% \xinteval, \xinttheexpr as sub-entities inside an \xintNewExpr. And the
% «cleaning» will remove the new \XINTfstop (detokenized from \meaning
% output), to maintain backwards compatibility with former behaviour
% that created macros expand to explicit digits and not an encapsulated
% result.
%
% The #2#3 in clean stands for \noexpand\XINTfstop (where the actual
% scantoken-ized input uses $$ originally with catcode letter as the escape
% character).
% |
%    \begin{macrocode}
\def\xintNewExpr     {\XINT_NewExpr\xint_firstofone\xintexpr     \XINT_newexpr_clean}%
\def\xintNewFloatExpr{\XINT_NewExpr\xint_firstofone\xintfloatexpr\XINT_newexpr_clean}%
\def\xintNewIExpr    {\XINT_NewExpr\xint_firstofone\xintiexpr    \XINT_newexpr_clean}%
\def\xintNewIIExpr   {\XINT_NewExpr\xint_firstofone\xintiiexpr   \XINT_newexpr_clean}%
\def\xintNewBoolExpr {\XINT_NewExpr\xint_firstofone\xintboolexpr \XINT_newexpr_clean}%
\def\XINT_newexpr_clean #1>#2#3{\noexpand\expanded\noexpand\xintNEprinthook}%
\def\xintNEprinthook#1.#2{\expanded{\unexpanded{#1.}{#2}}}%
%    \end{macrocode}
% \lverb|1.2c for \xintdeffunc, \xintdefiifunc, \xintdeffloatfunc.
%
% At 1.3, NewFunc does not use anymore a comma delimited pattern for the
% arguments to the macro being defined.
%
% At 1.4 we use \xintthebareeval, whose meaning  now does not mean unlock
% from csname but firstofone to remove a level of braces
% This is involved in functioning of expr:userfunc and expr:userefunc
% |
%    \begin{macrocode}
\def\XINT_NewFunc     {\XINT_NewExpr\xint_gobble_i\xintthebareeval\XINT_newfunc_clean}%
\def\XINT_NewFloatFunc{\XINT_NewExpr\xint_gobble_i\xintthebarefloateval\XINT_newfunc_clean}%
\def\XINT_NewIIFunc   {\XINT_NewExpr\xint_gobble_i\xintthebareiieval\XINT_newfunc_clean}%
\def\XINT_newfunc_clean #1>{}%
%    \end{macrocode}
% \lverb|1.2c adds optional logging. For this needed to pass to _NewExpr_a the
% macro name as parameter.
%
% Up to and including 1.2c the definition was global. Starting with 1.2d it is
% done locally.
%
% Modified at 1.3c so that \XINT_NewFunc et al. do not execute the
% \xintexprSafeCatcodes, as it is now already done earlier by \xintdeffunc.
%
% |
%    \begin{macrocode}
\def\XINT_NewExpr #1#2#3#4#5[#6]%
{%
 \begingroup
    \ifcase #6\relax
        \toks0 {\endgroup\XINT_global\def#4}%
    \or \toks0 {\endgroup\XINT_global\def#4##1}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3##4}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3##4##5}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3##4##5##6}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3##4##5##6##7}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3##4##5##6##7##8}%
    \or \toks0 {\endgroup\XINT_global\def#4##1##2##3##4##5##6##7##8##9}%
    \fi
    #1\xintexprSafeCatcodes
    \XINT_expr_redefinemacros
    \XINT_NewExpr_a #1#2#3#4%
}%
%    \end{macrocode}
% \lverb|& attention que & est de catcode 14
%
% 1.2d's \xintNewExpr makes a local definition. In earlier releases, the
% definition was global.
%
% \the\toks0 inserts the \endgroup, but this will happen
% after \XINT_tmpa has already been expanded...
%
% The $%1 is \xint_firstofone for \xintNewExpr, \xint_gobble_i
% for \xintdeffunc.
%
% Attention that at 1.4, there might be entire sub-xintexpressions embedded in
% detokenized form. They are re-tokenized and the main thing is that the
% parser should not mis-interpret catcode 11 characters as starting variable
% names. As some macros use : in their names, the retokenization must be done
% with : having catcode 11. To not break embedded non-evaluated
% sub-expressions, the \XINT_expr_getop was extended to intercept the :
% (alternative would have been to never inject any macro with : in its name...
% too late now). On the other hand the ! is not used in the macro names
% potentially kept as is non expanded by the \xintNewExpr/\xintdeffunc
% process; it can thus be retokenized with catcode 12. But the «hooks» of
% seq(), iter(), etc... if deciding they can't evaluate immediately will
% inject a full sub-expression (possibly arbitrarily complicated) and append
% to it for its delayed expansion a catcode 11 ! character (as well as
% possibly catcode 3 ~ and ? and catcode 11 caret ^ and even catcode 7 $&).
% The macros \XINT_expr_tilde etc... below serve for this injection (there are
% *two* successive \scantokens using different catcode regimes and these
% macros remain detokenized during the first pass!) and as consequence the
% final meaning may have characters such as ! or & present with both standard
% and special catcodes depending on where they are located. It may thus not be
% possible to (easily) retokenize the meaning as printed in the log file if
% \xintverbosetrue was issued.
%
% If a defined function is used in another expression it would thus break
% things if its meaning was included pre-expanded ; a mechanism exists which
% keeps only the name of the macro associated to the function (this name may
% contain digits by the way), when the macro can not be immediately fully
% expanded. Thus its meaning (with its possibly funny catcodes) is not
% exposed. And this gives opportunity to pre-expand its arguments before
% actually expanding the macro. |
%    \begin{macrocode}
\catcode`~ 3 \catcode`? 3
\def\XINT_expr_tilde{~}\def\XINT_expr_qmark{?}% catcode 3
\def\XINT_expr_caret{^}\def\XINT_expr_exclam{!}% catcode 11
\def\XINT_expr_tab{&}% catcode 7
\def\XINT_expr_null{&&@}%
\catcode`~ 13 \catcode`@ 14 \catcode`\% 6 \catcode`# 12 \catcode`$ 11 @ $
\def\XINT_NewExpr_a %1%2%3%4%5@
{@
    \def\XINT_tmpa %%1%%2%%3%%4%%5%%6%%7%%8%%9{%5}@
    \def~{$noexpand$}@
    \catcode`: 11 \catcode`_ 11 \catcode`\@ 11
    \catcode`# 12 \catcode`~ 13 \escapechar 126
    \endlinechar -1 \everyeof {\noexpand }@
    \edef\XINT_tmpb
    {\scantokens\expandafter{\romannumeral`&&@\expandafter
     %2\XINT_tmpa{#1}{#2}{#3}{#4}{#5}{#6}{#7}{#8}{#9}\relax}@
    }@
    \escapechar 92 \catcode`# 6 \catcode`$ 0 @ $
    \edef\XINT_tmpa %%1%%2%%3%%4%%5%%6%%7%%8%%9@
      {\scantokens\expandafter{\expandafter%3\meaning\XINT_tmpb}}@
    \the\toks0\expandafter
      {\XINT_tmpa{%%1}{%%2}{%%3}{%%4}{%%5}{%%6}{%%7}{%%8}{%%9}}@
    %1{\ifxintverbose
        \xintMessage{xintexpr}{Info}@
                    {\string%4\space now with @
                     \ifxintglobaldefs global \fi meaning \meaning%4}@
       \fi}@
}@
\catcode`% 14
\XINTsetcatcodes % clean up to avoid surprises if something changes
%    \end{macrocode}
% \subsubsection{\csh{ifxintexprsafecatcodes}, \csh{xintexprSafeCatcodes}, \csh{xintexprRestoreCatcodes}}
% \changed[2018/06/17]{1.3c}
% \lverb|Added \ifxintexprsafecatcodes to allow nesting|
%    \begin{macrocode}
\newif\ifxintexprsafecatcodes
\let\xintexprRestoreCatcodes\empty
\def\xintexprSafeCatcodes
{%
  \unless\ifxintexprsafecatcodes
    \edef\xintexprRestoreCatcodes  {%
        \endlinechar=\the\endlinechar
        \catcode59=\the\catcode59   % ;
        \catcode34=\the\catcode34   % "
        \catcode63=\the\catcode63   % ?
        \catcode124=\the\catcode124 % |
        \catcode38=\the\catcode38   % &
        \catcode33=\the\catcode33   % !
        \catcode93=\the\catcode93   % ]
        \catcode91=\the\catcode91   % [
        \catcode94=\the\catcode94   % ^
        \catcode95=\the\catcode95   % _
        \catcode47=\the\catcode47   % /
        \catcode41=\the\catcode41   % )
        \catcode40=\the\catcode40   % (
        \catcode42=\the\catcode42   % *
        \catcode43=\the\catcode43   % +
        \catcode62=\the\catcode62   % >
        \catcode60=\the\catcode60   % <
        \catcode58=\the\catcode58   % :
        \catcode46=\the\catcode46   % .
        \catcode45=\the\catcode45   % -
        \catcode44=\the\catcode44   % ,
        \catcode61=\the\catcode61   % =
        \catcode96=\the\catcode96   % `
        \catcode32=\the\catcode32\relax % space
        \noexpand\xintexprsafecatcodesfalse
    }%
  \fi
  \xintexprsafecatcodestrue
        \endlinechar=13 %
        \catcode59=12  % ;
        \catcode34=12  % "
        \catcode63=12  % ?
        \catcode124=12 % |
        \catcode38=4   % &
        \catcode33=12  % !
        \catcode93=12  % ]
        \catcode91=12  % [
        \catcode94=7   % ^
        \catcode95=8   % _
        \catcode47=12  % /
        \catcode41=12  % )
        \catcode40=12  % (
        \catcode42=12  % *
        \catcode43=12  % +
        \catcode62=12  % >
        \catcode60=12  % <
        \catcode58=12  % :
        \catcode46=12  % .
        \catcode45=12  % -
        \catcode44=12  % ,
        \catcode61=12  % =
        \catcode96=12  % `
        \catcode32=10  % space
}%
\let\XINT_tmpa\undefined \let\XINT_tmpb\undefined \let\XINT_tmpc\undefined
\let\XINT_tmpd\undefined \let\XINT_tmpe\undefined
\ifdefined\RequirePackage\expandafter\xint_firstoftwo\else\expandafter\xint_secondoftwo\fi
{\RequirePackage{xinttrig}%
\RequirePackage{xintlog}}%
{\input xinttrig.sty
\input xintlog.sty
}%
\XINTrestorecatcodesendinput%
%    \end{macrocode}
% \StoreCodelineNo {xintexpr}
% \cleardoublepage\let\xintexprnameUp\undefined
%\gardesactifs
%\let</xintexpr>\relax
%\let<*xinttrig>\gardesinactifs
%</xintexpr>^^A--------------------------------------------------
%<*xinttrig>^^A---------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xinttrignameUp\endcsname
% \section{Package \xinttrignameimp implementation}
% \RaisedLabel{sec:trigimp}
% \etocarticlestylenomarks
% \etocsetnexttocdepth {subsubsection}
%
% \localtableofcontents
%
% A preliminary implementation was done only late in the development of 
% \cs{xintexpr}, as an example of the high level user interface, in January
% 2019.  In March and April 2019 I improved the algorithm for the inverse
% trigonometrical functions and included the whole as a new \csbxint{expr}
% module.  But, as the high level interface provided no way to have
% intermediate steps executed with guard digits, the whole scheme could only
% target say P-2 digits where P is the prevailing precision, and only with a
% moderate requirement on what it means to have P-2 digits about correct.
%
% Finally in April 2021, after having at long last added exponential and
% logarithm up to 62 digits and at a rather strong precision requirement
% (something like, say with inputs in normal ranges: targeting at most 0.505ulp
% distance to exact result), I revisited the code here.
%
% We keep most of the high level usage of \csbxint{deffloatfunc}, but hack
% into its process in order to let it map the 4 operations and some functions
% such as square-root to macros using 4 extra digits. This hack is enough to
% support the used syntax here, but is not usable generally. All functions and
% their auxiliaries defined during the time the hack applies are named with
% |@| as first letter.
%
% Later the public functions, without the |@|, are defined as wrappers of the
% |@|-named ones, which float-round to P digits on output.
% 
% Apart from that the sine and cosine series were implemented at macro level,
% bypassing the \csbxint{deffloatfunc} interface. This is done mainly for
% handling Digits at high value (24 or more) as it then becomes beneficial to
% float-round the variable to less and less digits, the deeper one goes into
% the series.
%
% And regarding the arcsine I modified a bit my original idea in order to
% execute the first step in a single \cs{numexpr}.  It turns out that that for
% 16 digits the algorithm then ``only'' needs one sine and one cosine
% evaluation (and a square-root), and there is no need for an arcsine series
% auxiliary then.  I am aware this is by far not the ``best'' approach but the
% problem is that I am a bit enamored into the idea of the algorithm even
% though it is at least twice as costly than a sine evaluation!  Actually, for
% many digits, it turns out the arcsine is less costly than two random sine
% evaluations, probably because the latter have the overhead of range
% reduction.
%
% Speaking of this, the range reduction is rather naive and not extremely
% ambitious.  I wrote it initially having only |sind()| and |cosd()| in mind,
% and in 2019 reduced degrees to radians in the most naive way possible.  I
% have only slighly improved this for this |1.4e| 2021 release, the announced
% precision for inputs less than say |1e6|, but at |1e8| and higher, one will
% start feeling the gradual loss of precision compared to the task of
% computing the exact mathematical result correctly rounded. Also, I do not
% worry here about what happens when the input is very near a big multiple of
% $\pi$, and one computes a sine for example. Maybe I will improve in future
% this aspect but I decided I was seriously running out of steam for the
% |1.4e| release.
%
% As commented in \xintlognameimp regarding exponential and logarithms, even
% though we have instilled here some dose of lower level coding, the whole
% suffers from \xintfracnameimp not yet having made floating point numbers a
% native type.  Thus inefficiencies accumulate...
%
% At 8 digits, the gain was only about 40\% compared to 16 digits. So at the
% last minute, on the day I was going to do the release I decided to implement
% a poorman way for sine and cosine, for "speed". I transferred the idea from
% the arcsine numexpr to sine and cosine. Indeed there is an interesting speed
% again of about 4X compared to applying the same approach as for higher
% values of Digits. Correct rounding during random testing is still obtained
% reasonably often (at any rate more than 95\% of cases near 45 degrees and
% always faithful rounding), although at less than the 99\% reached for the
% main branch handling Digits up to 62. But the precision is more than enough
% for usage in plots for example. I am keeping the guard digits, as removing
% then would add a further speed gain of about 20\% to 40\% but the precision
% then would drop dramatically, and this is not acceptable at the time of our
% 2021 standards (not a period of enlightenment generally speaking, though).
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \catcode94=7    % ^
  \def\z{\endgroup}%
  \def\empty{}\def\space{ }\newlinechar10
  \expandafter\let\expandafter\w\csname ver@xintexpr.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info:^^J%
                 \space\space\space\space#2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xinttrig}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\w\relax % xintexpr.sty not yet loaded.
        \y{xinttrig}%
          {Loading should be via \ifx\x\empty\string\usepackage{xintexpr.sty}
                          \else\string\input\space xintexpr.sty \fi
           rather, aborting}%
         \aftergroup\endinput
    \fi
  \fi
\z%
\edef\XINTtrigendinput{\XINTrestorecatcodes\noexpand\endinput}\XINTsetcatcodes%
\catcode`? 12
%    \end{macrocode}
% \subsection{Library identification}
%    \begin{macrocode}
\ifcsname xintlibver@trig\endcsname
  \expandafter\xint_firstoftwo
\else
  \expandafter\xint_secondoftwo
\fi
{\immediate\write-1{Reloading xinttrig library using Digits=\xinttheDigits.}}%
{\expandafter\gdef\csname xintlibver@trig\endcsname{2021/07/13 v1.4j}%
\XINT_providespackage
\ProvidesPackage{xinttrig}%
[2021/07/13 v1.4j Trigonometrical functions for xintexpr (JFB)]%
}%
%    \end{macrocode}
% \subsection{Ensure used letters are dummy letters}
%    \begin{macrocode}
\xintFor* #1 in {iDTVtuwxyzX}\do{\xintensuredummy{#1}}%
%    \end{macrocode}
% \subsection{\csh{xintreloadxinttrig}}
% \lverb|Much simplified at 1.4e, from a modified catcode regime management.|
%    \begin{macrocode}
\def\xintreloadxinttrig{\input xinttrig.sty }%
%    \end{macrocode}
% \subsection{Auxiliary variables}
% \lverb|&
% The variables with private names have extra digits. Whether private or
% public, the variables can all be redefined without impacting the defined
% functions, whose meanings will contain already the variable values.
%
% Formerly variables holding the 1/n! were defined, but this got removed at 1.4e.
% |
% \subsubsection{\cshn{@twoPi}, \cshn{@threePiover2}, \cshn{@Pi}, \cshn{@Piover2}}
% \lverb|&
% At 1.4e we need more digits, also \xintdeffloatvar changed and always rounds
% to P=Digits precision so we use another path to store values with extra digits.
% |
%    \begin{macrocode}
\xintdefvar @twoPi :=
    float(
6.2831853071795864769252867665590057683943387987502116419498891846156328125724180
    ,\XINTdigitsormax+4);%
\xintdefvar @threePiover2 :=
    float(
4.7123889803846898576939650749192543262957540990626587314624168884617246094293135
    ,\XINTdigitsormax+4);%
\xintdefvar @Pi :=
    float(
3.1415926535897932384626433832795028841971693993751058209749445923078164062862090
    ,\XINTdigitsormax+4);%
\xintdefvar @Piover2 :=
    float(
1.5707963267948966192313216916397514420985846996875529104874722961539082031431045
    ,\XINTdigitsormax+4);%
%    \end{macrocode}
% \subsubsection{\cshn{@oneDegree}, \cshn{@oneRadian}}
% \lverb|&
% Those are needed for range reduction, particularly @oneRadian.  We define
% it with 12 extra digits.  But the whole process of range reduction in radians
% is very naive one.|
%    \begin{macrocode}
\xintdefvar @oneDegree :=
    float(
0.017453292519943295769236907684886127134428718885417254560971914401710091146034494
    ,\XINTdigitsormax+4);%
\xintdefvar @oneRadian :=
    float(
57.295779513082320876798154814105170332405472466564321549160243861202847148321553
    ,\XINTdigitsormax+12);%
%    \end{macrocode}
% \subsection{Hack \cs{xintdeffloatfunc} for inserting usage of guard digits}
% \lverb|1.4e. This is not a general approach, but it sufficient for the
% limited use case done here of \xintdeffloatfunc. What it does is to let
% \xintdeffloatfunc hardcode usage of macros which will execute computations
% with an elevated number of digits.  But for example if 5/3 is encountered in
% a float expression it will remain unevaluated so one would have to use
% alternate input syntax for efficiency (\xintexpr
% float(5/3,\xinttheDigits+4)\relax as a subexpression, for example).|
%    \begin{macrocode}
\catcode`~ 12
\def\XINT_tmpa#1#2#3.#4.%
{%
  \let #1#2%
  \def #2##1##2##3##4{##2##3{{~expanded{~unexpanded{#4[#3]}~expandafter}~expanded{##1##4}}}}%
}%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_exec_+_\expandafter\endcsname
    \csname XINT_flexpr_exec_+\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatAdd_wopt.%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_exec_-_\expandafter\endcsname
    \csname XINT_flexpr_exec_-\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatSub_wopt.%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_exec_*_\expandafter\endcsname
    \csname XINT_flexpr_exec_*\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatMul_wopt.%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_exec_/_\expandafter\endcsname
    \csname XINT_flexpr_exec_/\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatDiv_wopt.%
\def\XINT_tmpa#1#2#3.#4.%
{%
  \let #1#2%
  \def #2##1##2##3{##1##2{{~expanded{~unexpanded{#4[#3]}~expandafter}##3}}}%
}%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_sqrfunc\expandafter\endcsname
    \csname XINT_flexpr_func_sqr\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatSqr_wopt.%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_sqrtfunc\expandafter\endcsname
    \csname XINT_flexpr_func_sqrt\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatSqrt.%
\expandafter\XINT_tmpa
    \csname XINT_flexpr_invfunc\expandafter\endcsname
    \csname XINT_flexpr_func_inv\expandafter\endcsname
    \the\numexpr\XINTdigitsormax+4.~XINTinFloatInv_wopt.%
\catcode`~ 3
%    \end{macrocode}
% \subsection{The sine and cosine series}
% \lverb|&
% Old pending question: should I rather use successive divisions by (2n+1)(2n), or rather
% multiplication by their precomputed inverses, in a modified Horner scheme ?
% The \ifnum tests are executed at time of definition.
%
% Update at last minute: this is actually exactly what I do if Digits is at
% most 8.
%
% Small values of the variable are very badly handled here because a much
% shorter truncation of the series should be used.
%
% At 1.4e the original \xintdeffloatfunc was converted into macros, whose
% principle can be seen also at work in xintlog.sty. We prepare the input
% variables with shorter and shorter mantissas for usage deep in the series.
%
% This divided by about 3 the execution cost of the series for P about 60.
%
% Originally, the thresholds were computed a priori with 0.79 as upper bound
% of the variable, but then for 1.4e I developped enough test files to try to
% adjust heuristically with a target of say
% 99,5$% of correct rounding, and always at most
% 1ulp error. The numerical analysis is not easy due to the complications of
% the implementation...
%
% Also, random testing never explores the weak spots...
%
% The 0.79 (a bit more than Pi/4) upper bound induces a costly check of
% variable on input, if Digits is big. Much faster would be to check if input
% is less than 10 degrees or 1 radian as done in xfp. But using enough
% coefficients for allowing up to 1 radian, which is without pain for
% Digits=16 starts being annoying for higher values such as Digits=48.
%
% But the main reason I don't do it now is that I spend too much time
% fine-tuning the table of thresholds... maybe in next release.
% |
% \subsubsection{Support macros for the sine and cosine series}
%
% \lverb|Computing the 1/n! from n! then inverting would require
% costly divisions and significantly increase the loading time.
%
% So a method is employed to simply divide by 2k(2k-1) or (2k+1)(2k)
% step by step, with what we hope are enough 8 security digits, and
% reducing the sizes of the mantissas at each step.
%
% This whole section is conditional on Digits being at least nine.
% |
%
%    \begin{macrocode}
\ifnum\XINTdigits>8
\edef\XINT_tmpG % 1/3!
  {1\xintReplicate{\XINTdigitsormax+2}{6}7[\the\numexpr-\XINTdigitsormax-4]}%
\edef\XINT_tmpH % 1/5!
  {8\xintReplicate{\XINTdigitsormax+1}{3}[\the\numexpr-\XINTdigitsormax-4]}%
\edef\XINT_tmpd % 1/5!
  {8\xintReplicate{\XINTdigitsormax+9}{3}[\the\numexpr-\XINTdigitsormax-12]}%
\def\XINT_tmpe#1.#2.#3.#4.#5#6#7%
{%
\def#5##1\xint:
{%
    \expandafter#6\romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def#6##1\xint:
{%
    \expandafter#7\romannumeral0\xintsub{#4}{\XINTinFloat[#2]{\xintMul{#3}{##1}}}\xint:
}%
\def#7##1\xint:##2\xint:
{%
    \xintSub{1/1[0]}{\XINTinFloat[#1]{\xintMul{##1}{##2}}}%
}%
}%
\expandafter\XINT_tmpe
  \the\numexpr\XINTdigitsormax+4\expandafter.%
  \the\numexpr\XINTdigitsormax+2\expandafter.\expanded{%
  \XINT_tmpH.% 1/5!
  \XINT_tmpG.% 1/3!
  \expandafter}%
  \csname XINT_SinAux_series_a_iii\expandafter\endcsname
  \csname XINT_SinAux_series_b\expandafter\endcsname
  \csname XINT_SinAux_series_c_i\endcsname
\def\XINT_tmpa #1 #2 #3 #4 #5 #6 #7 #8 %
{%
 \def\XINT_tmpb ##1##2##3##4##5%
 {%
  \def\XINT_tmpc####1.####2.####3.####4.####5.%
  {%
   \def##1########1\xint:
   {%
       \expandafter##2%
       \romannumeral0\XINTinfloatS[####1]{########1}\xint:########1\xint:
   }%
   \def##2########1\xint:
   {%
       \expandafter##3%
       \romannumeral0\XINTinfloatS[####2]{########1}\xint:########1\xint:
   }%
   \def##3########1\xint:
   {%
       \expandafter##4%
       \romannumeral0\xintsub{####4}{\XINTinFloat[####2]{\xintMul{####3}{########1}}}\xint:
   }%
   \def##4########1\xint:########2\xint:
   {%
       \expandafter##5%
       \romannumeral0\xintsub{####5}{\XINTinFloat[####1]{\xintMul{########1}{########2}}}\xint:
   }%
  }%
 }%
 \expandafter\XINT_tmpb
 \csname XINT_#8Aux_series_a_\romannumeral\numexpr#1-1\expandafter\endcsname
 \csname XINT_#8Aux_series_a_\romannumeral\numexpr#1\expandafter\endcsname
 \csname XINT_#8Aux_series_b\expandafter\endcsname
 \csname XINT_#8Aux_series_c_\romannumeral\numexpr#1-2\expandafter\endcsname
 \csname XINT_#8Aux_series_c_\romannumeral\numexpr#1-3\endcsname
 \edef\XINT_tmpd
   {\XINTinFloat[\XINTdigitsormax-#2+8]{\xintDiv{\XINT_tmpd}{\the\numexpr#5*(#5-1)\relax}}}%
 \let\XINT_tmpF\XINT_tmpG
 \let\XINT_tmpG\XINT_tmpH
 \edef\XINT_tmpH{\XINTinFloat[\XINTdigitsormax-#2]{\XINT_tmpd}}%
 \expandafter\XINT_tmpc
 \the\numexpr\XINTdigitsormax-#3\expandafter.%
 \the\numexpr\XINTdigitsormax-#2\expandafter.\expanded{%
 \XINT_tmpH.%
 \XINT_tmpG.%
 \XINT_tmpF.%
 }%
}%
\XINT_tmpa 4 -1 -2 -4 7 5 3 Sin %
\ifnum\XINTdigits>3  \XINT_tmpa  5  1 -1 -2  9  7  5 Sin \fi
\ifnum\XINTdigits>5  \XINT_tmpa  6  3  1 -1 11  9  7 Sin \fi
\ifnum\XINTdigits>8  \XINT_tmpa  7  6  3  1 13 11  9 Sin \fi
\ifnum\XINTdigits>11 \XINT_tmpa  8  9  6  3 15 13 11 Sin \fi
\ifnum\XINTdigits>14 \XINT_tmpa  9 12  9  6 17 15 13 Sin \fi
\ifnum\XINTdigits>16 \XINT_tmpa 10 14 12  9 19 17 15 Sin \fi
\ifnum\XINTdigits>19 \XINT_tmpa 11 17 14 12 21 19 17 Sin \fi
\ifnum\XINTdigits>22 \XINT_tmpa 12 20 17 14 23 21 19 Sin \fi
\ifnum\XINTdigits>25 \XINT_tmpa 13 23 20 17 25 23 21 Sin \fi
\ifnum\XINTdigits>28 \XINT_tmpa 14 26 23 20 27 25 23 Sin \fi
\ifnum\XINTdigits>31 \XINT_tmpa 15 29 26 23 29 27 25 Sin \fi
\ifnum\XINTdigits>34 \XINT_tmpa 16 32 29 26 31 29 27 Sin \fi
\ifnum\XINTdigits>37 \XINT_tmpa 17 35 32 29 33 31 29 Sin \fi
\ifnum\XINTdigits>40 \XINT_tmpa 18 38 35 32 35 33 31 Sin \fi
\ifnum\XINTdigits>44 \XINT_tmpa 19 42 38 35 37 35 33 Sin \fi
\ifnum\XINTdigits>47 \XINT_tmpa 20 45 42 38 39 37 35 Sin \fi
\ifnum\XINTdigits>51 \XINT_tmpa 21 49 45 42 41 39 37 Sin \fi
\ifnum\XINTdigits>55 \XINT_tmpa 22 53 49 45 43 41 39 Sin \fi
\ifnum\XINTdigits>58 \XINT_tmpa 23 56 53 49 45 43 41 Sin \fi
\edef\XINT_tmpd % 1/4!
  {41\xintReplicate{\XINTdigitsormax+8}{6}7[\the\numexpr-\XINTdigitsormax-12]}%
\edef\XINT_tmpH % 1/4!
  {41\xintReplicate{\XINTdigitsormax}{6}7[\the\numexpr-\XINTdigitsormax-4]}%
\def\XINT_tmpG{5[-1]}% 1/2!
\expandafter\XINT_tmpe
  \the\numexpr\XINTdigitsormax+4\expandafter.%
  \the\numexpr\XINTdigitsormax+3\expandafter.\expanded{%
  \XINT_tmpH.%
  \XINT_tmpG.%
  \expandafter}%
  \csname XINT_CosAux_series_a_iii\expandafter\endcsname
  \csname XINT_CosAux_series_b\expandafter\endcsname
  \csname XINT_CosAux_series_c_i\endcsname
\XINT_tmpa 4 -2 -3 -4 6 4 2 Cos %
\ifnum\XINTdigits>2  \XINT_tmpa  5  0 -2 -3  8  6  4 Cos \fi
\ifnum\XINTdigits>4  \XINT_tmpa  6  2  0 -2 10  8  6 Cos \fi
\ifnum\XINTdigits>7  \XINT_tmpa  7  5  2  0 12 10  8 Cos \fi
\ifnum\XINTdigits>9  \XINT_tmpa  8  7  5  2 14 12 10 Cos \fi
\ifnum\XINTdigits>12 \XINT_tmpa  9 10  7  5 16 14 12 Cos \fi
\ifnum\XINTdigits>15 \XINT_tmpa 10 13 10  7 18 16 14 Cos \fi
\ifnum\XINTdigits>18 \XINT_tmpa 11 16 13 10 20 18 16 Cos \fi
\ifnum\XINTdigits>20 \XINT_tmpa 12 18 16 13 22 20 18 Cos \fi
\ifnum\XINTdigits>24 \XINT_tmpa 13 22 18 16 24 22 20 Cos \fi
\ifnum\XINTdigits>27 \XINT_tmpa 14 25 22 18 26 24 22 Cos \fi
\ifnum\XINTdigits>30 \XINT_tmpa 15 28 25 22 28 26 24 Cos \fi
\ifnum\XINTdigits>33 \XINT_tmpa 16 31 28 25 30 28 26 Cos \fi
\ifnum\XINTdigits>36 \XINT_tmpa 17 34 31 28 32 30 28 Cos \fi
\ifnum\XINTdigits>39 \XINT_tmpa 18 37 34 31 34 32 30 Cos \fi
\ifnum\XINTdigits>42 \XINT_tmpa 19 40 37 34 36 34 32 Cos \fi
\ifnum\XINTdigits>45 \XINT_tmpa 20 43 40 37 38 36 34 Cos \fi
\ifnum\XINTdigits>49 \XINT_tmpa 21 47 43 40 40 38 36 Cos \fi
\ifnum\XINTdigits>53 \XINT_tmpa 22 51 47 43 42 40 38 Cos \fi
\ifnum\XINTdigits>57 \XINT_tmpa 23 55 51 47 44 42 40 Cos \fi
\ifnum\XINTdigits>60 \XINT_tmpa 24 58 55 51 46 44 42 Cos \fi
\let\XINT_tmpH\xint_undefined\let\XINT_tmpG\xint_undefined\let\XINT_tmpF\xint_undefined
\let\XINT_tmpd\xint_undefined\let\XINT_tmpe\xint_undefined
\def\XINT_SinAux_series#1%
{%
    \expandafter\XINT_SinAux_series_a_iii
    \romannumeral0\XINTinfloatS[\XINTdigitsormax+4]{#1}\xint:
}%
\def\XINT_CosAux_series#1%
{%
    \expandafter\XINT_CosAux_series_a_iii
    \romannumeral0\XINTinfloatS[\XINTdigitsormax+4]{#1}\xint:
}%
\fi % end of \XINTdigits>8
%    \end{macrocode}
% \subsubsection{The poor man approximate but speedier approach for Digits at most 8}
%    \begin{macrocode}
\ifnum\XINTdigits<9
\def\XINT_SinAux_series#1%
{%
    \the\numexpr\expandafter\XINT_SinAux_b\romannumeral0\xintiround9{#1}.[-9]%
}%
\def\XINT_SinAux_b#1.%
{%
  ((((((((((((%(\xint_c_x^ix/-210)
  -4761905*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -156)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -110)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -72)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -42)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -20)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -6)*#1/\xint_c_x^ix+\xint_c_x^ix
}%
\def\XINT_CosAux_series#1%
{%
    \the\numexpr\expandafter\XINT_CosAux_b\romannumeral0\xintiround9{#1}.[-9]%
}%
\def\XINT_CosAux_b#1.%
{%
  ((((((((((((((%(\xint_c_x^ix/-240)
  -4166667*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -182)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -132)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -90)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -56)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -30)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -12)*#1/\xint_c_x^ix+\xint_c_x^ix)/%
  -2)*#1/\xint_c_x^ix+\xint_c_x^ix
}%
\fi
%    \end{macrocode}
% \subsubsection{Declarations of the \cshn{@sin_aux()} and \cshn{@cos_aux()} functions}
%    \begin{macrocode}
\def\XINT_flexpr_func_@sin_aux#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINT_SinAux_series#3}}%
}%
\def\XINT_flexpr_func_@cos_aux#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINT_CosAux_series#3}}%
}%
%    \end{macrocode}
% \subsubsection{\cshn{@sin_series()}, \cshn{@cos_series()}}
%    \begin{macrocode}
\xintdeffloatfunc @sin_series(x) := x * @sin_aux(sqr(x));%
\xintdeffloatfunc @cos_series(x) := @cos_aux(sqr(x));%
%    \end{macrocode}
% \subsection{Range reduction for sine and cosine using degrees}
%
% \lverb|&
% As commented in the package introduction, Range reduction is a demanding
% domain and we handle it semi-satisfactorily.  The main problem is that in
% January 2019 I had done only support for degrees, and when I added radians I
% used the most naive approach.  But one can find worse: in 2019 I was
% surprised to observe important divergences with Maple's results at 16 digits
% near -π. Turns out that Maple probably adds π
% in the floating point sense causing catastrophic loss of digits when one is
% near -π. On the other hand even though the approach here is still naive, it
% behaves much better.
%
% The @sind_rr() and @cosd_rr() sine and cosine "doing range reduction" are
% coded directly at macro level via \xintSind and \xintCosd which will
% dispatch to usage of the sine or cosine series, depending on case.
%
% Old note from 2019: attention that \xintSind and \xintCosd must be used
% with a positive argument.
%
% We start with an auxiliary macro to reduce modulo 360 quickly.
% |
% \subsubsection{Low level modulo 360 helper macro \csh{XINT_mod_ccclx_i}}
% \lverb|&
% input:   \the\numexpr\XINT_mod_ccclx_i k.N. (delimited by dots)
%
% output: (N times 10^k) modulo 360. (with a final dot)
%
% Attention that N must be non-negative (I could make it accept negative
% but the fact that numexpr / is not periodical in numerator
% adds overhead).
%
% 360 divides 9000 hence 10^{k} is 280 for k at least 3 and the additive
% group generated by it modulo 360 is the set of multiples of 40.
% |
%    \begin{macrocode}
\def\XINT_mod_ccclx_i #1.%
{%
    \expandafter\XINT_mod_ccclx_e\the\numexpr
    \expandafter\XINT_mod_ccclx_j\the\numexpr1\ifcase#1 \or0\or00\else000\fi.%
}%
\def\XINT_mod_ccclx_j 1#1.#2.%
{%
    (\XINT_mod_ccclx_ja {++}#2#1\XINT_mod_ccclx_jb 0000000\relax
}%
\def\XINT_mod_ccclx_ja #1#2#3#4#5#6#7#8#9%
{%
    #9+#8+#7+#6+#5+#4+#3+#2\xint_firstoftwo{+\XINT_mod_ccclx_ja{+#9+#8+#7}}{#1}%
}%
\def\XINT_mod_ccclx_jb #1\xint_firstoftwo#2#3{#1+0)*280\XINT_mod_ccclx_jc #1#3}%
%    \end{macrocode}
% \lverb|&
% Attention that \XINT_cclcx_e wants non negative input because \numexpr
% division is not periodical ...
% |
%    \begin{macrocode}
\def\XINT_mod_ccclx_jc  +#1+#2+#3#4\relax{+80*(#3+#2+#1)+#3#2#1.}%
\def\XINT_mod_ccclx_e#1.{\expandafter\XINT_mod_ccclx_z\the\numexpr(#1+180)/360-1.#1.}%
\def\XINT_mod_ccclx_z#1.#2.{#2-360*#1.}%
%    \end{macrocode}
% \subsubsection{\cshn{@sind_rr()} function and its support macro \csh{xintSind}}
%    \begin{macrocode}
\def\XINT_flexpr_func_@sind_rr #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one{\romannumeral`&&@\xintSind#3}}%
}%
%    \end{macrocode}
% \lverb|&
% old comment: Must be f-expandable for nesting macros from \xintNewExpr
%
% This is where the prize of using the same macros for two distinct use cases
% has serious disadvantages. The reason of Digits+12 is only to support an
% input which contains a multiplication by @oneRadian with its extended
% digits.
%
% Then we do a somewhat strange truncation to a fixed point of fractional
% digits, which is ok in the "Degrees" case, but causes issues of its own in
% the "Radians" case. Please consider this whole thing as marked for future
% improvement, when times allows.
%
% ATTENTION \xintSind ONLY FOR POSITIVE ARGUMENTS
% |
%    \begin{macrocode}
\def\XINT_tmpa #1.{%
\def\xintSind##1%
{%
    \romannumeral`&&@\expandafter\xintsind\romannumeral0\XINTinfloatS[#1]{##1}}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+12.%
\def\xintsind #1[#2#3]%
{%
    \xint_UDsignfork 
      #2\XINT_sind
       -\XINT_sind_int
    \krof#2#3.#1..%
}%
\def\XINT_tmpa #1.{%
\def\XINT_sind ##1.##2.%
{%
    \expandafter\XINT_sind_a
    \romannumeral0\xinttrunc{#1}{##2[##1]}%
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+5.%
\def\XINT_sind_a{\expandafter\XINT_sind_i\the\numexpr\XINT_mod_ccclx_i0.}%
\def\XINT_sind_int
{%
    \expandafter\XINT_sind_i\the\numexpr\expandafter\XINT_mod_ccclx_i
}%
\def\XINT_sind_i #1.%
{%
    \ifcase\numexpr#1/90\relax
       \expandafter\XINT_sind_A
    \or\expandafter\XINT_sind_B\the\numexpr-90+%
    \or\expandafter\XINT_sind_C\the\numexpr-180+%
    \or\expandafter\XINT_sind_D\the\numexpr-270+%
    \else\expandafter\XINT_sind_E\the\numexpr-360+%
    \fi#1.%
}%
%    \end{macrocode}
%    \begin{macrocode}
\def\XINT_tmpa #1.#2.{%
\def\XINT_sind_A##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_sind_B_n-##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
\def\XINT_sind_B_p##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_sind_C_n-##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
\def\XINT_sind_C_p##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_sind_D_n-##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
\def\XINT_sind_D_p##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_sind_E-##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
}\expandafter\XINT_tmpa
    \the\numexpr\XINTdigitsormax+4\expandafter.%
    \romannumeral`&&@\xintbarefloateval @oneDegree\relax.%
\def\XINT_sind_B#1{\xint_UDsignfork#1\XINT_sind_B_n-\XINT_sind_B_p\krof #1}%
\def\XINT_sind_C#1{\xint_UDsignfork#1\XINT_sind_C_n-\XINT_sind_C_p\krof #1}%
\def\XINT_sind_D#1{\xint_UDsignfork#1\XINT_sind_D_n-\XINT_sind_D_p\krof #1}%
%    \end{macrocode}
% \subsubsection{\cshn{@cosd_rr()} function and its support macro \csh{xintCosd}}
%    \begin{macrocode}
\def\XINT_flexpr_func_@cosd_rr #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one{\romannumeral`&&@\xintCosd#3}}%
}%
%    \end{macrocode}
% \lverb|&
% ATTENTION ONLY FOR POSITIVE ARGUMENTS
% |
%    \begin{macrocode}
\def\XINT_tmpa #1.{%
\def\xintCosd##1%
{%
    \romannumeral`&&@\expandafter\xintcosd\romannumeral0\XINTinfloatS[#1]{##1}}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+12.%
\def\xintcosd #1[#2#3]%
{%
    \xint_UDsignfork 
      #2\XINT_cosd
       -\XINT_cosd_int
    \krof#2#3.#1..%
}%
\def\XINT_tmpa #1.{%
\def\XINT_cosd ##1.##2.%
{%
    \expandafter\XINT_cosd_a
    \romannumeral0\xinttrunc{#1}{##2[##1]}%
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+5.%
\def\XINT_cosd_a{\expandafter\XINT_cosd_i\the\numexpr\XINT_mod_ccclx_i0.}%
\def\XINT_cosd_int
{%
    \expandafter\XINT_cosd_i\the\numexpr\expandafter\XINT_mod_ccclx_i
}%
\def\XINT_cosd_i #1.%
{%
    \ifcase\numexpr#1/90\relax
       \expandafter\XINT_cosd_A
    \or\expandafter\XINT_cosd_B\the\numexpr-90+%
    \or\expandafter\XINT_cosd_C\the\numexpr-180+%
    \or\expandafter\XINT_cosd_D\the\numexpr-270+%
    \else\expandafter\XINT_cosd_E\the\numexpr-360+%
    \fi#1.%
}%
%    \end{macrocode}
% \lverb|#2 will be empty in the "integer" branch, but attention in general
% branch to handling of negative integer part after the subtraction of 90,
% 180, 270, or 360.
% |
%    \begin{macrocode}
\def\XINT_tmpa#1.#2.{%
\def\XINT_cosd_A##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_cosd_B_n-##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
\def\XINT_cosd_B_p##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_cosd_C_n-##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
\def\XINT_cosd_C_p##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_cosd_D_n-##1.##2.%
{%
    \xintiiopp\XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
\def\XINT_cosd_D_p##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@sin_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{##1.##2}#2}}%
}%
\def\XINT_cosd_E-##1.##2.%
{%
    \XINT_expr_unlock\expandafter\XINT_flexpr_userfunc_@cos_series\expandafter
         {\romannumeral0\XINTinfloat[#1]{\xintMul{\xintSub{##1[0]}{.##2}}#2}}%
}%
}\expandafter\XINT_tmpa
    \the\numexpr\XINTdigitsormax+4\expandafter.%
    \romannumeral`&&@\xintbarefloateval @oneDegree\relax.%
\def\XINT_cosd_B#1{\xint_UDsignfork#1\XINT_cosd_B_n-\XINT_cosd_B_p\krof #1}%
\def\XINT_cosd_C#1{\xint_UDsignfork#1\XINT_cosd_C_n-\XINT_cosd_C_p\krof #1}%
\def\XINT_cosd_D#1{\xint_UDsignfork#1\XINT_cosd_D_n-\XINT_cosd_D_p\krof #1}%
%    \end{macrocode}
% \subsection{\cshn{@sind()}, \cshn{@cosd()}}
% \lverb|&
% The -45 is stored internally as -45/1[0] from the action of the unary minus
% operator, which float macros then parse faster. The 45e0 is to let it become
% 45[0] and not simply 45.
%
% Here and below the \ifnum\XINTdigits>8 45\else60\fi will all be resolved
% at time of definition. This is the charm and power of expandable parsers!
% |
%    \begin{macrocode}
\xintdeffloatfunc @sind(x) := (x)??
                              {(x>=-\ifnum\XINTdigits>8 45\else60\fi)?
                                {@sin_series(x*@oneDegree)}
                                {-@sind_rr(-x)}
                              }
                              {0e0}
                              {(x<=\ifnum\XINTdigits>8 45\else60\fi e0)?
                                {@sin_series(x*@oneDegree)}
                                {@sind_rr(x)}
                              }
                              ;%
\xintdeffloatfunc @cosd(x) := (x)??
                              {(x>=-\ifnum\XINTdigits>8 45\else60\fi)?
                                {@cos_series(x*@oneDegree)}
                                {@cosd_rr(-x)}
                              }
                              {1e0}
                              {(x<=\ifnum\XINTdigits>8 45\else60\fi e0)?
                                {@cos_series(x*@oneDegree)}
                                {@cosd_rr(x)}
                              }
                              ;%
%    \end{macrocode}
% \subsection{\cshn{@sin()}, \cshn{@cos()}}
% \lverb|&
% For some reason I did not define sin() and cos() in January 2019 ??
%
% The sub \xintexpr x*@oneRadian\relax means that the multiplication will be
% done exactly @oneRadian having its 12 extra digits (and x its 4 extra
% digits), before being rounded in entrance of \xintSind, respectively
% \xintCosd, to P+12 mantissa.
%
% The strange 79e-2 could be 0.79 which would give 79[-2] internally too.
% |
%    \begin{macrocode}
\xintdeffloatfunc @sin(x):= (abs(x)<\ifnum\XINTdigits>8 79e-2\else1e0\fi)?
                            {@sin_series(x)}
                            {(x)??
                              {-@sind_rr(-\xintexpr x*@oneRadian\relax)}
                              {0}
                              {@sind_rr(\xintexpr x*@oneRadian\relax)}
                            }
                            ;%
\xintdeffloatfunc @cos(x):= (abs(x)<\ifnum\XINTdigits>8 79e-2\else1e0\fi)?
                             {@cos_series(x)}
                             {@cosd_rr(abs(\xintexpr x*@oneRadian\relax))}
                             ;%
%    \end{macrocode}
% \subsection{\cshn{@sinc()}}
% \lverb|&
% Should I also consider adding (1-cos(x))/(x^2/2) ? it is sinc^2(x/2) but
% avoids a square.
% |
%    \begin{macrocode}
\xintdeffloatfunc @sinc(x):= (abs(x)<\ifnum\XINTdigits>8 79e-2\else1e0\fi) ?
                             {@sin_aux(sqr(x))}
                             {@sind_rr(\xintexpr abs(x)*@oneRadian\relax)/abs(x)}
                             ;%
%    \end{macrocode}
% \subsection{\cshn{@tan()}, \cshn{@tand()}, \cshn{@cot()}, \cshn{@cotd()}}
% \lverb|The 0 in cot(x) is a dummy place holder. We don't have a notion of
% Inf yet.|
%    \begin{macrocode}
\xintdeffloatfunc @tand(x):= @sind(x)/@cosd(x);%
\xintdeffloatfunc @cotd(x):= @cosd(x)/@sind(x);%
\xintdeffloatfunc @tan(x) := (x)??
                             {(x>-\ifnum\XINTdigits>8 79e-2\else1e0\fi)?
                               {@sin(x)/@cos(x)}
                               {-@cotd(\xintexpr9e1+x*@oneRadian\relax)
                               }
                             }
                             {0e0}
                             {(x<\ifnum\XINTdigits>8 79e-2\else1e0\fi)?
                               {@sin(x)/@cos(x)}
                               {@cotd(\xintexpr9e1-x*@oneRadian\relax)}
                             }
                             ;%
\xintdeffloatfunc @cot(x) := (abs(x)<\ifnum\XINTdigits>8 79e-2\else1e0\fi)?
                              {@cos(x)/@sin(x)}
                              {(x)??
                                {-@tand(\xintexpr9e1+x*@oneRadian\relax)}
                                {0}
                                {@tand(\xintexpr9e1-x*@oneRadian\relax)}
                              };%
%    \end{macrocode}
% \subsection{\cshn{@sec()}, \cshn{@secd()}, \cshn{@csc()}, \cshn{@cscd()}}
%    \begin{macrocode}
\xintdeffloatfunc @sec(x) := inv(@cos(x));%
\xintdeffloatfunc @csc(x) := inv(@sin(x));%
\xintdeffloatfunc @secd(x):= inv(@cosd(x));%
\xintdeffloatfunc @cscd(x):= inv(@sind(x));%
%    \end{macrocode}
% \subsection{Core routine for inverse trigonometry}
%
% \lverb|&
% I always liked very much the general algorithm whose idea I found
% in 2019. But it costs a square root plus a sine plus a cosine all
% at target precision. For the arctangent the square root will be
% avoided by a trick. (memo: it is replaced by a division and I am not so sure
% now this is advantageous in fact)
%
% And now I like it even more as I have re-done the first step entirely
% in a single \numexpr... Thus the inverse trigonometry got a serious
% improvement at 1.4e...
%
% Here is the idea. We have 0<t<sqrt(2)/2 and we want a = Arcsin t.
%
% Imagine we have some very good approximation b = a - h. We know b,
% and don't know yet h. No problem h is a-b so sin(h)=sin(a)cos(b)-
% cos(a)sin(b). And we know everything here: sin(a) is t, cos(a) is 
% u  = sqrt(1-t^2), and we can compute cos(b) and sin(b).
%
% I said h was small so the computation of sin(a)cos(b)-cos(a)sin(b) will
% involve a lot of cancellation, no problem with xint, as it knows how to
% compute exactly... and if we wanted to go very low level we could do
% cos(a)sin(b) paying attention only on least significant digits.
%
% Ok, so we have sin(h), but h is small, so the series of Arcsine can be used
% with few terms!
%
% In fact h will be at most of the order of 1e-9, so it is no problem to
% simply replace sin(h) with h if the target precision is 16 !
%
% Ok, so how do we obtain b, the good approximation to Arcsin t ?  Simply by
% using its Taylor series, embedded in a single \numexpr working with nine
% digits numbers... I like this one! Notice that it reminisces with my
% questioning about how to best do Horner like for sine and cosine.  Here in
% \numexpr we can only manipulate whole integers and simply can't do things
% such as ...)*x + 5/112)*x + 3/40)*x + 1/6)*x +1 .... But I found another
% way, see the code, which uses extensively the "scaling" operations in
% \numexpr.
%
% I have not proven rigorously that b-a is always less or equal in absolute
% value than 1e-9, but it is possible for example in Python to program it and
% go through all possible (less than) 1e9 inputs and check what happens.
%
% Very small inputs will give b=0 (first step is a fixed point rounding of t to
% nine fractional digits, so this rounding gives zero for input <0.5e-9,
% others will give b=t, because the arcsine numexpr will end up with
% 1000000000 (last time I checked that was for t a bit less than 5e-5,
% the latter gives 1000000001).
% All seems to work perfectly fine, in practice...
%
%
% |
%
% \lverb|First we let the @sin_aux() and @cos_aux() functions be usable in exact
% \xintexpr context.
%
% The @asin_II() function will be used only for Digits>16.|
%    \begin{macrocode}
\expandafter\let\csname XINT_expr_func_@sin_aux\expandafter\endcsname
                \csname XINT_flexpr_func_@sin_aux\endcsname
\expandafter\let\csname XINT_expr_func_@cos_aux\expandafter\endcsname
                \csname XINT_flexpr_func_@cos_aux\endcsname
\ifnum\XINTdigits>16
\def\XINT_flexpr_func_@asin_II#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINT_Arcsin_II_a#3}}%
}%
\def\XINT_tmpc#1.%
{%
\def\XINT_Arcsin_II_a##1%
{%
    \expandafter\XINT_Arcsin_II_c_i\romannumeral0\XINTinfloatS[#1]{##1}%
}%
\def\XINT_Arcsin_II_c_i##1[##2]%
{%
    \xintAdd{1/1[0]}{##1/6[##2]}%
}%
}%
\expandafter\XINT_tmpc\the\numexpr\XINTdigitsormax-14.%
\fi
\ifnum\XINTdigits>34
\def\XINT_tmpc#1.#2.#3.#4.%
{%
\def\XINT_Arcsin_II_a##1%
{%
    \expandafter\XINT_Arcsin_II_a_iii\romannumeral0\XINTinfloatS[#1]{##1}\xint:
}%
\def\XINT_Arcsin_II_a_iii##1\xint:
{%
    \expandafter\XINT_Arcsin_II_b\romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_Arcsin_II_b##1\xint:
{%
    \expandafter\XINT_Arcsin_II_c_i\romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{#3}{##1}}}\xint:
}%
\def\XINT_Arcsin_II_c_i##1\xint:##2\xint:
{%
    \xintAdd{1/1[0]}{\XINTinFloat[#1]{\xintMul{##1}{##2}}}%
}%
}%
\expandafter\XINT_tmpc
  \the\numexpr\XINTdigitsormax-14\expandafter.%
  \the\numexpr\XINTdigitsormax-32\expandafter.\expanded{%
  \XINTinFloat[\XINTdigitsormax-32]{3/40[0]}.%
  \XINTinFloat[\XINTdigitsormax-14]{1/6[0]}.%
  }%
\fi
\ifnum\XINTdigits>52
\def\XINT_tmpc#1.#2.#3.#4.#5.%
{%
\def\XINT_Arcsin_II_a_iii##1\xint:
{%
    \expandafter\XINT_Arcsin_II_a_iv\romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_Arcsin_II_a_iv##1\xint:
{%
    \expandafter\XINT_Arcsin_II_b\romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_Arcsin_II_b##1\xint:
{%
    \expandafter\XINT_Arcsin_II_c_ii
    \romannumeral0\xintadd{#4}{\XINTinfloat[#2]{\xintMul{#3}{##1}}}\xint:
}%
\def\XINT_Arcsin_II_c_ii##1\xint:##2\xint:
{%
    \expandafter\XINT_Arcsin_II_c_i
    \romannumeral0\xintadd{#5}{\XINTinFloat[#1]{\xintMul{##1}{##2}}}\xint:
}%
}%
\expandafter\XINT_tmpc
  \the\numexpr\XINTdigitsormax-32\expandafter.%
  \the\numexpr\XINTdigitsormax-50\expandafter.\expanded{%
  \XINTinFloat[\XINTdigitsormax-50]{5/112[0]}.%
  \XINTinFloat[\XINTdigitsormax-32]{3/40[0]}.%
  \XINTinFloat[\XINTdigitsormax-14]{1/6[0]}.%
  }%
\fi
\def\XINT_flexpr_func_@asin_I#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINT_Arcsin_I#3}}%
}%
\def\XINT_Arcsin_I#1{\the\numexpr\expandafter\XINT_Arcsin_Ia\romannumeral0\xintiround9{#1}.}%
\def\XINT_Arcsin_Ia#1.%
{%
  (\expandafter\XINT_Arcsin_Ib\the\numexpr#1*#1/\xint_c_x^ix.)*%
  #1/\xint_c_x^ix[-9]%
}%
\def\XINT_Arcsin_Ib#1.%
{%((((((((((((((((
  % 3481/3660)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 3249/3422)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 3025/3192)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 2809/2970)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 2601/2756)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 2401/2550)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 2209/2352)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  % 2025/2162)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
((((((((((((((((((((((((((((((((((((((((((%
  %(\xint_c_x^ix*1849/1980)*%
  933838384*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  1681/1806)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  1521/1640)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  1369/1482)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  1225/1332)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  1089/1190)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  961/1056)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  841/930)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  729/812)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  625/702)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  529/600)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  441/506)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  361/420)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  289/342)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  225/272)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  169/210)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  121/156)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  81/110)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  49/72)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  25/42)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  9/20)*#1/\xint_c_x^ix+\xint_c_x^ix)*%
  1/6)*#1/\xint_c_x^ix+\xint_c_x^ix
}%
\ifnum\XINTdigits>16
  \xintdeffloatfunc @asin_o(D, T) := T + D*@asin_II(sqr(D));%
  \xintdeffloatfunc @asin_n(V, T, t, u) :=%
           @asin_o(\xintexpr t*@cos_aux(V) - u*T*@sin_aux(V)\relax, T);%
\else
  \xintdeffloatfunc @asin_n(V, T, t, u) :=%
                   \xintexpr t*@cos_aux(V) - u*T*@sin_aux(V)\relax + T;%
\fi
\xintdeffloatfunc @asin_m(T, t, u) := @asin_n(sqr(T), T, t, u);%
\xintdeffloatfunc @asin_l(t, u)    := @asin_m(@asin_I(t), t, u);%
%    \end{macrocode}
% \subsection{\cshn{@asin()}, \cshn{@asind()}}
% \lverb|&
% Only non-negative arguments t and u for asin_a(t,u), and asind_a(t,u).
%  |
%    \begin{macrocode}
\xintdeffloatfunc @asin_a(t, u) := (t<u)?
                                   {@asin_l(t, u)}
                                   {@Piover2 - @asin_l(u, t)}
                                   ;%
\xintdeffloatfunc @asind_a(t, u):= (t<u)?
                                   {@asin_l(t, u) * @oneRadian}
                                   {9e1 - @asin_l(u, t) * @oneRadian}
                                   ;%
\xintdeffloatfunc @asin(t) := (t)??
                              {-@asin_a(-t, sqrt(1e0-sqr(t)))}
                              {0e0}
                              {@asin_a(t, sqrt(1e0-sqr(t)))}
                              ;%
\xintdeffloatfunc @asind(t) := (t)??
                               {-@asind_a(-t, sqrt(1e0-sqr(t)))}
                               {0e0}
                               {@asind_a(t, sqrt(1e0-sqr(t)))}
                               ;%
%    \end{macrocode}
% \subsection{\cshn{@acos()}, \cshn{@acosd()}}
%    \begin{macrocode}
\xintdeffloatfunc @acos(t) := @Piover2 - @asin(t);%
\xintdeffloatfunc @acosd(t):= 9e1 - @asind(t);%
%    \end{macrocode}
% \subsection{\cshn{@atan()}, \cshn{@atand()}}
% \lverb|&
% Uses same core routine asin_l() as for asin(), but avoiding a square-root
% extraction in preparing its arguments (to the cost of computing an inverse,
% rather).
%
% radians
% |
%    \begin{macrocode}
\xintdeffloatfunc @atan_b(t, w, z):= 5e-1 * (w< 0)?
                                           {@Pi - @asin_a(2e0*z * t, -w*z)}
                                           {@asin_a(2e0*z * t, w*z)}
                                           ;%
\xintdeffloatfunc @atan_a(t, T) := @atan_b(t, 1e0-T, inv(1e0+T));%
\xintdeffloatfunc @atan(t):= (t)??
                             {-@atan_a(-t, sqr(t))}
                             {0}
                             {@atan_a(t, sqr(t))}
                             ;%
%    \end{macrocode}
% \lverb|&
% degrees
% |
%    \begin{macrocode}
\xintdeffloatfunc @atand_b(t, w, z) := 5e-1 * (w< 0)?
                                             {18e1 - @asind_a(2e0*z * t, -w*z)}
                                             {@asind_a(2e0*z * t, w*z)}
                                             ;%
\xintdeffloatfunc @atand_a(t, T) := @atand_b(t, 1e0-T, inv(1e0+T));%
\xintdeffloatfunc @atand(t) := (t)??
                               {-@atand_a(-t, sqr(t))}
                               {0}
                               {@atand_a(t, sqr(t))}
                               ;%
%    \end{macrocode}
% \subsection{\cshn{@Arg()}, \cshn{@atan2()}, \cshn{@Argd()}, \cshn{@atan2d()}, \cshn{@pArg()}, \cshn{@pArgd()}}
% \lverb|&
% Arg(x,y) function from -π (excluded) to +π (included)
% |
%    \begin{macrocode}
\xintdeffloatfunc @Arg(x, y):= (y>x)?
                               {(y>-x)?
                                 {@Piover2 - @atan(x/y)}
                                 {(y<0)?
                                   {-@Pi + @atan(y/x)}
                                   {@Pi + @atan(y/x)}
                                 }
                               }
                               {(y>-x)?
                                 {@atan(y/x)}
                                 {-@Piover2 + @atan(x/-y)}
                               }
                               ;%
%    \end{macrocode}
% \lverb|&
% atan2(y,x) = Arg(x,y) ... (some people have atan2 with arguments reversed
% but the convention here seems the most often encountered)
% |
%    \begin{macrocode}
\xintdeffloatfunc @atan2(y,x) := @Arg(x, y);%
%    \end{macrocode}
% \lverb|&
% Argd(x,y) function from -180 (excluded) to +180 (included)
% |
%    \begin{macrocode}
\xintdeffloatfunc @Argd(x, y):= (y>x)?
                                {(y>-x)?
                                  {9e1 - @atand(x/y)}
                                  {(y<0)?
                                    {-18e1 + @atand(y/x)}
                                    {18e1 + @atand(y/x)}
                                  }
                                }
                                {(y>-x)?
                                  {@atand(y/x)}
                                  {-9e1 + @atand(x/-y)}
                                }
                                ;%
%    \end{macrocode}
% \lverb|&
% atan2d(y,x) = Argd(x,y)
% |
%    \begin{macrocode}
\xintdeffloatfunc @atan2d(y,x) := @Argd(x, y);%
%    \end{macrocode}
% \lverb|&
% pArg(x,y) function from 0 (included) to 2π (excluded)
% I hesitated between pArg, Argpos, and Argplus. Opting for pArg in the end.
% |
%    \begin{macrocode}
\xintdeffloatfunc @pArg(x, y):= (y>x)?
                                {(y>-x)?
                                  {@Piover2 - @atan(x/y)}
                                  {@Pi + @atan(y/x)}
                                }
                                {(y>-x)?
                                  {(y<0)?
                                    {@twoPi + @atan(y/x)}
                                    {@atan(y/x)}
                                  }
                                  {@threePiover2 + @atan(x/-y)}
                                }
                                ;%
%    \end{macrocode}
% \lverb|&
% pArgd(x,y) function from 0 (included) to 360 (excluded)
% |
%    \begin{macrocode}
\xintdeffloatfunc @pArgd(x, y):=(y>x)?
                                {(y>-x)?
                                  {9e1 - @atan(x/y)*@oneRadian}
                                  {18e1 + @atan(y/x)*@oneRadian}
                                }
                                {(y>-x)?
                                  {(y<0e0)?
                                    {36e1 + @atan(y/x)*@oneRadian}
                                    {@atan(y/x)*@oneRadian}
                                  }
                                  {27e1 + @atan(x/-y)*@oneRadian}
                                }
                                ;%
%    \end{macrocode}
% \subsection{Restore \cs{xintdeffloatfunc} to its normal state, with no extra
% digits}
%    \begin{macrocode}
\expandafter\let
    \csname XINT_flexpr_exec_+\expandafter\endcsname
    \csname XINT_flexpr_exec_+_\endcsname
\expandafter\let
    \csname XINT_flexpr_exec_-\expandafter\endcsname
    \csname XINT_flexpr_exec_-_\endcsname
\expandafter\let
    \csname XINT_flexpr_exec_*\expandafter\endcsname
    \csname XINT_flexpr_exec_*_\endcsname
\expandafter\let
    \csname XINT_flexpr_exec_/\expandafter\endcsname
    \csname XINT_flexpr_exec_/_\endcsname
\expandafter\let
    \csname XINT_flexpr_func_sqr\expandafter\endcsname
    \csname XINT_flexpr_sqrfunc\endcsname
\expandafter\let
    \csname XINT_flexpr_func_sqrt\expandafter\endcsname
    \csname XINT_flexpr_sqrtfunc\endcsname
\expandafter\let
    \csname XINT_flexpr_func_inv\expandafter\endcsname
    \csname XINT_flexpr_invfunc\endcsname
%    \end{macrocode}
% \subsection{Let the functions be known to the \cshnolabel{xintexpr} parser}
% \lverb|&
% We use here float_dgtormax which uses the smaller of Digits and 64.
% |
%    \begin{macrocode}
\edef\XINTinFloatdigitsormax{\noexpand\XINTinFloat[\the\numexpr\XINTdigitsormax]}%
\edef\XINTinFloatSdigitsormax{\noexpand\XINTinFloatS[\the\numexpr\XINTdigitsormax]}%
\xintFor #1 in {sin, cos, tan, sec, csc, cot,
                asin, acos, atan}\do
{%
    \xintdeffloatfunc #1(x) := float_dgtormax(@#1(x));%
    \xintdeffloatfunc #1d(x) := float_dgtormax(@#1d(x));%
    \xintdeffunc #1(x) := float_dgtormax(\xintfloatexpr @#1(sfloat_dgtormax(x))\relax);%
    \xintdeffunc #1d(x):= float_dgtormax(\xintfloatexpr @#1d(sfloat_dgtormax(x))\relax);%
}%
\xintFor #1 in {Arg, pArg, atan2}\do
{%
    \xintdeffloatfunc #1(x, y)  := float_dgtormax(@#1(x, y));%
    \xintdeffloatfunc #1d(x, y) := float_dgtormax(@#1d(x, y));%
    \xintdeffunc #1(x, y) := float_dgtormax(\xintfloatexpr @#1(sfloat_dgtormax(x), sfloat_dgtormax(y))\relax);%
    \xintdeffunc #1d(x, y):= float_dgtormax(\xintfloatexpr @#1d(sfloat_dgtormax(x), sfloat_dgtormax(y))\relax);%
}%
\xintdeffloatfunc sinc(x):= float_dgtormax(@sinc(x));%
\xintdeffunc      sinc(x):= float_dgtormax(\xintfloatexpr @sinc(sfloat_dgtormax(x))\relax);%
%    \end{macrocode}
% \subsection{Synonyms: \cshn{@tg()}, \cshn{@cotg()}}
% \lverb|These are my childhood notations and I am attached to them. In
% radians only, and for \xintfloateval only. We skip some overhead here by using a \let at core level.|
%    \begin{macrocode}
\expandafter\let\csname XINT_flexpr_func_tg\expandafter\endcsname
                \csname XINT_flexpr_func_tan\endcsname
\expandafter\let\csname XINT_flexpr_func_cotg\expandafter\endcsname
                \csname XINT_flexpr_func_cot\endcsname
%    \end{macrocode}
% \subsection{Final clean-up}
% \lverb|Restore used dummy variables to their status prior to the package reloading.
% On first loading this is not needed, but I have not added a way to check
% here whether this a first loading or a re-loading.|
%    \begin{macrocode}
\xintdefvar twoPi := float_dgtormax(@twoPi);%
\xintdefvar threePiover2 := float_dgtormax(@threePiover2);%
\xintdefvar Pi := float_dgtormax(@Pi);%
\xintdefvar Piover2 := float_dgtormax(@Piover2);%
\xintdefvar oneDegree := float_dgtormax(@oneDegree);%
\xintdefvar oneRadian := float_dgtormax(@oneRadian);%
\xintunassignvar{@twoPi}\xintunassignvar{@threePiover2}%
\xintunassignvar{@Pi}\xintunassignvar{@Piover2}%
\xintunassignvar{@oneRadian}\xintunassignvar{@oneDegree}%
\xintFor* #1 in {iDTVtuwxyzX}\do{\xintrestorevariable{#1}}%
\XINTtrigendinput%
%    \end{macrocode}
% \StoreCodelineNo {xinttrig}
% \cleardoublepage\let\xinttrignameUp\undefined
%\gardesactifs
%\let</xinttrig>\relax
%\let<*xintlog>\gardesinactifs
%</xinttrig>^^A--------------------------------------------------
%<*xintlog>^^A---------------------------------------------------
%^^A -*- coding: utf-8; mode: doctex; fill-column: 78; sentence-end-double-space: t; -*-
% \clearpage\csname xintlognameUp\endcsname
% \section{Package \xintlognameimp implementation}
% \RaisedLabel{sec:logimp}
% \etocarticlestylenomarks
% \etocsetnexttocdepth {subsubsection}
%
% \localtableofcontents
%
% \etocsettocstyle{}{}
%
% In 2019, at 1.3e release I almost included extended precision for log() and
% exp() but the time I could devote to xint expired. Finally, at long last,
% (and I had procrastinated far more than the two years since 2019) the 1.4e
% release in April 2021 brings log10(), pow10(), log(), pow() to P=Digits
% precision: up to 62 digits with at least (said roughly) 99\% chances of
% correct rounding (the design is targeting less than about 0.005ulp distance
% to mathematical value, before rounding).
%
% Implementation is EXPERIMENTAL.
%
% For up to Digits=8, it is simply based upon the poormanlog package. The
% probability of correct rounding will be less than for Digits>8, especially
% in the cases of Digits=8 and to a lesser extent Digits=7. And, for all
% Digits<=8, there is a systematic loss of rounding precision in the floating
% point sense in the case of log10(x) for inputs close to 1:
%
% \lverb|&
%
% Summary of limitations of log10() and pow10() in the case of Digits<=8:
%( - For log10(x) with x near 1, the precision of output as floating point&
%   will be mechanically reduced from the fact that this is based on a fixed&
%   point result, for example log10(1.0011871) is produced as 5.15245e-4,&
%   which stands for 0.000515145 having indeed 9 correct fractional digits,&
%   but only 6 correct digits in the floating point sense.&
%
%   This feature affects the entire range Digits<=8.
%: - Even if limiting to inputs x with 1.26<x<10 (1.26 is a bit more than&
%   10^0.1 hence its choice as lower bound), the poormanlog documentation&
%   mentions an absolute error possibly up to about 1e-9. In practice a&
%   test of 10000 random inputs 1.26<x<10 revealed 9490 correctly rounded&
%   log10(x) at 8 digits (and the 510 non-correctly rounded ones with an error&
%   of 1 in last digit compared to correct rounding).  So correct rounding&
%   achieved only in about 95$% of cases here.&
%
%   At 7 digits the same 10000 random&
%   inputs are correctly rounded in 99.4$% of cases, and at 6 digits it is&
%   99.94$% of cases.&
%
%   Againd with Digits=8, the  log10(i) for i in 1..1000 are all correctly&
%   rounded to 8 digits with two exceptions: log10(3) and log10(297) with a&
%   1ulp error.
%: - Regarding the computation of 10^x, I obtained for -1<x<1 the following&
%   with 10000 random inputs: 518/10000 errors at 1ulp, 60/10000, and 8/10000,&
%   at respectively Digits = 8, 7, 6 so chances of correct rounding are&
%   respectively about 95$%, 99.4$% and more than 99.9$%.
%)
% |
%
% Despite its limitations the poormanlog based approach used for Digits up to
% 8 has the advantage of speed (at least 8X compared to working with 16
% digits) and is largely precise enough for plots.
%
% For 9 digits or more, the observed precision in some random tests appears to
% be at least of 99.9\% chances of correct rounding, and the log10(x) with x
% near 1 are correctly (if not really efficiently) handled in the floating
% point sense for the output. The poormanlog approximate log10() is still used
% to boot-strap the process, generally.  The pow10() at Digits=9 or more is
% done independently of poormanlog.
%
% All of this is done on top of my 2013 structures for floating point
% computations which have always been marked as provisory and rudimentary and
% instills intrinsic non-efficiency:
%
%\lverb|&
%( - no internal data format for a ``floating point number at P digits'',
%: - mantissa lengths are again and again computed,
%: - digits are not pre-organized say in blocks of 4 by 4 or 8 by 8,
%: - floating point multiplication is done via an *exact* multiplication, then&
%    rounding to P digits!
%)
% |
%
% This is legacy of the fact that the project was initially devoted to big
% integers only, but in the weeks that followed its inception in March 2013 I
% added more and more functionalities without a well laid out preliminary
% plan.
%
% Anyway, for years I have felt a better foundation would help achieve at
% least something such as 2X gain (perhaps the last item by itself, if
% improved upon, would bring most of such 2X gain?)
%
% I did not try to optimize for the default 16 digits, the goal being more of
% having a general scalable structure in place and there is no difficulty to
% go up to 100 digits precision if one stores extended pre-computed constants
% and increases the length of the ``series'' support.
%
% Apart from log(10) and its inverse, no other logarithms are stored or
% pre-computed: the rest of the stored data is the same for pow10() and
% log10() and consists of the fractional powers 10\string^±0.i,
% 10\string^±0.0i, ..., 10\string^±0.00000i at P+5 and also at P+10 digits.
%
% In order to reduce the loading time of the package the inverses are not
% computed internally (as this would require costly divisions) but simply
% hard-coded with enough digits to cover the allowed Digits range.
%
%
% \subsection{Catcodes, \protect\eTeX{} and reload detection}
%    \begin{macrocode}
\begingroup\catcode61\catcode48\catcode32=10\relax%
  \catcode13=5    % ^^M
  \endlinechar=13 %
  \catcode123=1   % {
  \catcode125=2   % }
  \catcode64=11   % @
  \catcode35=6    % #
  \catcode44=12   % ,
  \catcode45=12   % -
  \catcode46=12   % .
  \catcode58=12   % :
  \catcode94=7    % ^
  \def\z{\endgroup}%
  \def\empty{}\def\space{ }\newlinechar10
  \expandafter\let\expandafter\w\csname ver@xintexpr.sty\endcsname
  \expandafter
    \ifx\csname PackageInfo\endcsname\relax
      \def\y#1#2{\immediate\write-1{Package #1 Info:^^J%
                 \space\space\space\space#2.}}%
    \else
      \def\y#1#2{\PackageInfo{#1}{#2}}%
    \fi
  \expandafter
  \ifx\csname numexpr\endcsname\relax
     \y{xintlog}{\numexpr not available, aborting input}%
     \aftergroup\endinput
  \else
    \ifx\w\relax % xintexpr.sty not yet loaded.
        \y{xintlog}%
          {Loading should be via \ifx\x\empty\string\usepackage{xintexpr.sty}
                          \else\string\input\space xintexpr.sty \fi
           rather, aborting}%
        \aftergroup\endinput
    \fi
  \fi
\z%
\edef\XINTendxintloginput{\XINTrestorecatcodes\noexpand\endinput}\XINTsetcatcodes%
%    \end{macrocode}
% \subsection{Library identification}
%    \begin{macrocode}
\ifcsname xintlibver@log\endcsname
  \expandafter\xint_firstoftwo
\else
  \expandafter\xint_secondoftwo
\fi
{\immediate\write-1{Reloading xintlog library using Digits=\xinttheDigits.}}%
{\expandafter\gdef\csname xintlibver@log\endcsname{2021/07/13 v1.4j}%
\XINT_providespackage
\ProvidesPackage{xintlog}%
[2021/07/13 v1.4j Logarithms and exponentials for xintexpr (JFB)]%
}%
%    \end{macrocode}
% \subsection{\csh{xintreloadxintlog}}
% \lverb|Now needed at 1.4e.
%
% |
%    \begin{macrocode}
\def\xintreloadxintlog{\input xintlog.sty }%
%    \end{macrocode}
% \subsection{Loading the \cshn{poormanlog} package}
% \lverb|&
% Attention to the catcode regime when loading poormanlog.
%
% As I learned the hard way (I never use my user macros), at the worst moment
% when wrapping up the final things for 1.3e release,
% \xintexprSafeCatcodes MUST be followed by some \xintexprRestoreCatcodes
% quickly, else next time it is used (for example by \xintdefvar) the
% \xintexprRestoreCatcodes will restore an obsolete catcode regime...
%
% Also, for xintlog.sty to be multiple-times loadable, we need to avoid using
% LaTeX's \RequirePackage twice.|
%    \begin{macrocode}
\xintexprSafeCatcodes\catcode`_ 11
\unless\ifdefined\XINTinFloatPowTen
\ifdefined\RequirePackage
  \RequirePackage{poormanlog}%
\else
  \input poormanlog.tex
\fi\fi
\xintexprRestoreCatcodes\XINTsetcatcodes
%    \end{macrocode}
% \subsection{Macro layer on top of the poormanlog package}
% \lverb|This was moved here with some macro renames from xintfrac on occasion
% of 1.4e release.
%
% Breaking changes at 1.4e:
%
%( - \poormanloghack now a no-op,
%: - \xintLog was used for \xinteval and differed slightly from its&
%    counterpart used for \xintfloateval, the latter float-rounded&
%    to P = Digits, the former did not and kept completly meaning-less&
%    digits in output. Both macros now replaced by a \PoorManLog&
%    which will always  float round the output to P = Digits. Because xint&
%    does not really implement a fixed point interface anyhow.
%: - \xintExp (used in \xinteval) and another macro (used in \xintfloateval)&
%    did not use a sufficiently long approximation to 1/log(10) to&
%    support precisely enough exp(x) if output of the order of 10^10000 for&
%    example, (last two digits wrong then)&
%    and situation became worse for very high values such as exp(1e8)&
%    which had only 4 digits correct.
%    The new \PoorManExp which replaces them is more careful... and for&
%    example exp(12345678) obtains correct rounding (Digits=8).
%: - \XINTinFloatxintLog and \XINTinFloatxintExp were removed; they were&
%    used for log() and exp() in \xintfloateval, and differed from \xintLog&
%    and \xintExp a bit, now renamed to \PoorManLog and \PoorManExp.
%: - \PoorManPower has simply disappeared, see \XINTinFloatSciPow and \xintPow.
%)
%
% See the general xintlog introduction for some comments on the achieved
% precision and probabilities of correct rounding.
%
% |
%
% \subsubsection{\csh{PoorManLogBaseTen}, \csh{PoorManLog}}
% \lverb|1.3f. Code originally in poormanlog v0.04 got transferred here. It
% produces the logarithm in base 10 with an error (believed to be at most) of
% the order of 1 unit in the 9th (i.e. last, fixed point) fractional
% digit. Testing seems to indicate the error is never exceeding 2 units in the
% 9th place, in worst cases.
%
% These macros will still be the support macros for \xintfloatexpr log10(),
% pow10(), etc... up to Digits=8 and the poormanlog logarithm is used as
% starting point for higher precision if Digits is at least 9.
%
% Notice that \PML@999999999. expands (in \numexpr) to 1000000000 (ten
% digits), which is the only case with the output having ten digits. But there
% is no need here to treat this case especially, it works fine in
% \PML@logbaseten.
%
% Breaking change at 1.4e: for consistency with various considerations
% on floats, the output will be float rounded to P=Digits.
%
% One could envision the \xinteval variant to keep 9 fractional digits
% (it is known the last one may very well be off by 1 unit). But this
% creates complications of principles.
%
% All of this is very strange because the logarithm clearly shows the
% deficiencies of the whole idea of floating point arithmetic, logarithm goes
% from floating point to fixed point, and coercing it into pure floating point
% has moral costs. Anyway, I shall obide.
%
% |
%    \begin{macrocode}
\def\PoorManLogBaseTen{\romannumeral0\poormanlogbaseten}%
\def\poormanlogbaseten #1%
{%
    \XINTinfloat[\XINTdigits]%
    {\romannumeral0\expandafter\PML@logbaseten\romannumeral0\XINTinfloat[9]{#1}}%
}%
\def\PoorManLogBaseTen_raw%#1
{%
    \romannumeral0\expandafter\PML@logbaseten\romannumeral0\XINTinfloat[9]%{#1}%
}%
\def\PML@logbaseten#1[#2]%
{%
    \xintiiadd{\xintDSx{-9}{\the\numexpr#2+8\relax}}{\the\numexpr\PML@#1.}[-9]%
}%
\def\PoorManLog#1%
{%
    \XINTinFloat[\XINTdigits]{\xintMul{\PoorManLogBaseTen_raw{#1}}{23025850923[-10]}}%
}%
%    \end{macrocode}
% \subsubsection{\csh{PoorManPowerOfTen}, \csh{PoorManExp}}
% \lverb|Originally in poormanlog v0.04, got transferred into xintfrac.sty at
% 1.3f, then here into xintlog.sty at 1.4e.
%
% Produces 10^x with 9 digits of float precision, with an error (believed to
% be) at most 2 units in the last place, when 0<x<1. Of course for this the
% input must be precise enough to have 9 fractional digits of **fixed point**
% precision.
%
% Breaking change at 1.4e: output always float-rounded at P=Digits.
%
% The 1.3f definition for \xintExp (now \PoorManExp) was not careful enough
% (see comments above) for very large exponents. This has been corrected at
% 1.4e. Formerly exp(12345678) produced shameful 6.3095734e5361659 where only
% the first digit (and exponent...) is correct!  Now, with \xintDigits*:=8;,
% exp(12345678) will produce 6.7725836e5361659 which is correct rounding to 8
% digits. Sorry if your rover expedition to Mars ended in failure due to using
% my software. I was not expecting anyone to use it so I did back then in 2019
% a bit too expeditively the \xintExp thing on top of 10^x.
%
% The 1.4e \PoorManExp replaces and amends deceased \xintExp.
%
% Before using \xintRound we screen out the case of zero as \xintRound in this
% case outputs no fractional digits.
% |
%    \begin{macrocode}
\def\PoorManPowerOfTen{\romannumeral0\poormanpoweroften}%
\def\poormanpoweroften #1%
{%
    \expandafter\PML@powoften@out
    \the\numexpr\expandafter\PML@powoften\romannumeral0\xintraw{#1}%
}%
\def\PML@powoften@out#1[#2]{\XINTinfloat[\XINTdigits]{#1[#2]}}%
\def\PML@powoften#1%
{%
    \xint_UDzerominusfork
      #1-\PML@powoften@zero
      0#1\PML@powoften@neg
       0-\PML@powoften@pos
    \krof #1%
}%
\def\PML@powoften@zero 0/1[0]{1\relax/1[0]}%
\def\PML@powoften@pos#1[#2]%
{%
    \expandafter\PML@powoften@pos@a\romannumeral0\xintround{9}{#1[#2]}.%
}%
\def\PML@powoften@pos@a#1.#2.{\PML@Pa#2.\expandafter[\the\numexpr-8+#1]}%
\def\PML@powoften@neg#1[#2]%
{%
    \expandafter\PML@powoften@neg@a\romannumeral0\xintround{9}{#1[#2]}.%
}%
\def\PML@powoften@neg@a#1.#2.%
{%
   \ifnum#2=\xint_c_ \xint_afterfi{1\relax/1[#1]}\else
   \expandafter\expandafter\expandafter
   \PML@Pa\expandafter\xint_gobble_i\the\numexpr2000000000-#2.%
   \expandafter[\the\numexpr-9+#1\expandafter]\fi
}%
\def\PoorManExp#1{\PoorManPowerOfTen{\xintMul{#1}{43429448190325182765[-20]}}}%
%    \end{macrocode}
% \subsubsection{Removed: \csh{PoorManPower}, see \cshnolabel{XINTinFloatSciPow}}
% \lverb|Removed at  1.4e. See \XINTinFloatSciPow.|
% \subsubsection{Made a no-op: \csh{poormanloghack}}
% \lverb|Made a no-op at 1.4e.|
%    \begin{macrocode}
\def\poormanloghack#1%
{%
    \xintMessage{xintexpr}{Warning}%
    {\string\poormanloghack\space is a no-op since 1.4e and will be removed at next major release}%
}%
%    \end{macrocode}
% \subsection{Macro support for powers}
%
% \subsubsection{\csh{XINTinFloatSciPow}}
%
%
% \lverb|&
%
% This is the new name and extension of \XINTinFloatPowerH which was
% a non user-documented macro used for a^b previously, and previously
% was located in $xintfracnameimp.
%
% A check is done whether the exponent is integer or half-integer, and if
% positive, the legacy \xintFloatPower/\xintFloatSqrt macros are used. The
% rationale is that:
%
%( - they give faster evaluations for integer exponent b < 10000 (and beyond)
%: - they operate at any value of Digits
%: - they keep accuracy even with gigantic exponents, whereas the pow10()/log10()&
%  path starts losing accuracy for b about 1e8. In fact at 1.4e it was even&
%  for b about 1000, as log10(A) was not computed with enough fractional&
%  digits, except for 0.8<A<1.26 (roughly), for this usage. At the 1.4f&
%  bugfix we compute log10(A) with enough accuracy for A^b to be safe with&
%  b as large as 1e7, and show visible degradation only for b about 1e9.
%)
%
% The user documentation of \xintFloatPower mentions a 0.52 ulp(Z) error where
% Z is the computed result, which seems not as good as the kind of accuracy we
% target for pow10() (for -1<x<1) and log10() (for 1<x<10) which is more like
% about 0.505ulp.  Perhaps in future I will examine if I need to increase a
% bit the theoretical accuracy of \xintFloatPower but at time of 1.4e/1.4f
% release I have left it standing as is.
%
% The check whether exponent is integer or half-integer is not on the value
% but on the representation. Even in \xintfloatexpr, input such
% 10^\xintexpr4/2\relax is possible, and 4/2 will not be recognized as integer
% to avoid costly overhead. 3/2 will not be recognized as half-integer.
% But 2.0 will be recognized as integer, 25e-1 as half-integer.
%
% In the computation of A^b, A will be float-rounded to Digits, but the
% exponent b will be handled "as is" until last minute. Recall that the
% \xintfloatexpr parser does not automatically float round isolated inputs,
% this happens only once involved in computations.
%
% In the Digits<=8 branch we do the same as for Digits>8 since 1.4f. At 1.4e I
% had strangely chosen (for "speed", but that was anyhow questionable for
% integer exponents less than 10 for example) to always use log10()/pow10()...
% But with only 9 fractional digits for the logarithms, exponents such as 1000
% naturally led to last 2 or 3 digits being wrong and let's not even mention
% when the exponent was of the order or 1e6... now A^1000 and A^1000.5 are
% accurately computed and one can handle a^1000.1 as a^1000*a^0.1
%
% I wrote the code during 1.4e to 1.4f transition for doing this split of
% exponent automatically, but it induced a very significant time penalty down
% the line for fractional exponents, whereas currently a^b is computed at
% Digits=8 with perfectly acceptable accuracy for fractional abs(b)<10, and at
% high speed, and accuracy for big exponents can be obtained by manually
% splitting as above (although the above has no user interface for keeping
% each contribution with its extra digits; a single one for a^h, -1<h<1).|
%    \begin{macrocode}
\def\XINTinFloatSciPow{\romannumeral0\XINTinfloatscipow}%
\def\XINTinfloatscipow#1#2%
{%
    \expandafter\XINT_scipow_a\romannumeral0\xintrez{#2}\XINT_scipow_int{#1}%
}%
\def\XINT_scipow_a #1%
{%
    \xint_gob_til_zero#1\XINT_scipow_Biszero0\XINT_scipow_b#1%
}%
\def\XINT_scipow_Biszero#1]#2#3{ 1[0]}%
\def\XINT_scipow_b #1#2/#3[#4]#5%
{%
    \unless\if1\XINT_is_One#3XY\xint_dothis\XINT_scipow_c\fi
    \ifnum#4<\xint_c_mone\xint_dothis\XINT_scipow_c\fi
    \ifnum#4=\xint_c_mone
      \if5\xintLDg{#1#2} %
       \xint_afterfi{\xint_dothis\XINT_scipow_halfint}\else
       \xint_afterfi{\xint_dothis\XINT_scipow_c}%
      \fi
    \fi
    \xint_orthat#5#1#2/#3[#4]%
}%
\def\XINT_scipow_int #1/1[#2]#3%
{%
    \expandafter\XINT_flpower_checkB_a
    \romannumeral0\XINT_dsx_addzeros{#2}#1;.\XINTdigits.{#3}{\XINTinfloatS[\XINTdigits]}%
}%
%    \end{macrocode}
% \lverb|The \XINT_flpowerh_finish is the sole remnant of \XINTinFloatPowerH
% which was formerly stitched to \xintFloatPower and checked for half-integer
% exponent.|
%    \begin{macrocode}
\def\XINT_scipow_halfint#1/1[#2]#3%
{%
    \expandafter\XINT_flpower_checkB_a
    \romannumeral0\xintdsr{\xintDouble{#1}}.\XINTdigits.{#3}\XINT_flpowerh_finish
}%
\def\XINT_flpowerh_finish #1%
{%
    \XINTinfloatS[\XINTdigits]{\XINTinFloatSqrt[\XINTdigits+\xint_c_iii]{#1}}%
}%
\def\XINT_tmpa#1.{%
\def\XINT_scipow_c ##1[##2]##3%
{%
    \expandafter\XINT_scipow_d\romannumeral0\XINTinfloatS[#1]{##3}\xint:##1[##2]\xint:
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigits.%
\def\XINT_scipow_d #1%
{%
    \xint_UDzerominusfork
       #1-\XINT_scipow_Aiszero
       0#1\XINT_scipow_Aisneg
        0-\XINT_scipow_Aispos
    \krof #1%
}%
\def\XINT_scipow_Aiszero #1\xint:#2#3\xint:
{%
%    \end{macrocode}
% \lverb|Missing NaN and Infinity causes problems. Inserting something
% like 1["7FFF8000] is risky as certain macros convert [N] into N zeros...
% so the run can appear to stall and will crash possibly badly if we do that.
% There is some usage in relation to ilog10 in xint.sty and xintfrac.sty
% of "7FFF8000 but here I will stay prudent and insert the usual 0 value
% (changed at 1.4g)
% |
%    \begin{macrocode}
    \if-#2\xint_dothis
      {\XINT_signalcondition{InvalidOperation}{0 raised to power #2#3.}{}{ 0[0]}}\fi
    \xint_orthat{ 0[0]}%
}%
\def\XINT_scipow_Aispos #1\xint:#2\xint:
{%
    \XINTinfloatpowten{\xintMul{#2}{\XINTinFloatLogTen_xdgout#1}}%
}%
%    \end{macrocode}
% \lverb|If a^b with a<0, we arrive here only if b was not considered to be
% an integer exponent. So let's raise an error.|
%    \begin{macrocode}
\def\XINT_scipow_Aisneg #1#2\xint:#3\xint:
{%
   \XINT_signalcondition{InvalidOperation}%
        {Fractional power #3 of negative #1#2.}{}{ 0[0]}%
}%
\ifnum\XINTdigits<9
%    \end{macrocode}
% \lverb|At 1.4f we only need for Digits up to 8 to insert usage of poormanlog
% for non integer, non half-integer exponents. At 1.4e the code was more
% complicated because I had strangely opted for using always the log10()
% path. However we have to be careful to use \PML@logbaseten with 9 digits
% always.
%
% As the legacy macros used for integer and half-integer exponents float-round
% the input to Digits digits, we must do the same here for coherence. Which
% induces some small complications here.
% |
%    \begin{macrocode}
  \def\XINT_tmpa#1.#2.#3.{%
  \def\XINT_scipow_c ##1[##2]##3%
  {%
      \expandafter\XINT_scipow_d
      \romannumeral0\expandafter\XINT_scipow_c_i
      \romannumeral0\XINTinfloat[#1]{##3}\xint:##1[##2]\xint:
  }%
  \def\XINT_scipow_c_i##1[##2]{ ##1#3[##2-#2]}%
  }\expandafter\XINT_tmpa\the\numexpr\XINTdigits\expandafter.%
   \the\numexpr9-\XINTdigits\expandafter.%
   \romannumeral\xintreplicate{9-\XINTdigits}0.%
  \def\XINT_scipow_Aispos #1\xint:#2\xint:
  {%
      \poormanpoweroften{\xintMul{#2}{\romannumeral0\expandafter\PML@logbaseten#1}}%
  }%
\fi
%    \end{macrocode}
% \subsubsection{\cshnolabel{xintPow}}
% \lverb|&
%
% Support macro for a^b in \xinteval. This overloads the original
% $xintfracnameimp macro, keeping its original meaning only for integer
% exponents, which are not too big: for exact evaluation of A^b, we want the
% output to not have more than about 10000 digits (separately for numerator
% and denominator). For this we limit b depending on the length of A, simply
% we want b to be smaller than the rounded value of 10000 divided by the
% length of A. For one-digit A, this would give 10000 as maximal exponent but
% due to organization of code related to avoir arithmetic overflow (we can't
% immediately operate in \numexpr with b as it is authorized to be beyond TeX
% bound), the maximal exponent is 9999.
%
% The criterion, which guarantees output (numerator and
% denominator separately) does not exceed by much 10000 digits if at all is
% that the exponent should be less than the (rounded in the sense of \numexpr)
% quotient of 10000 by the number of digits of a (considering separately
% numerator and denominator).
%
% The decision whether to compute A^b exactly depends on the length of
% internal representation of A. So 9^9999 is evaluated exactly (in \xinteval)
% but for 9.0 it is 9.0^5000 the maximal power.  This may change in future.
%
% 1.4e had the following bug (for Digits>8): big integer exponents used the
% log10()/pow10() based approach rather than the legacy macro path which goes
% via \xintFloatPower, as done by \xintfloateval! As a result powers with very
% large integer exponents were more precise in \xintfloateval than in \xinteval!
%
% 1.4f fixes this. Also, it handles Digits<=8 as Digits>8, bringing much
% simplification here.
% |
%    \begin{macrocode}
\def\xintPow{\romannumeral0\xintpow}%
\def\xintpow#1#2%
{%
    \expandafter\XINT_scipow_a\romannumeral0\xintrez{#2}\XINT_pow_int{#1}%
}%
%    \end{macrocode}
% \lverb|In case of half-integer exponent the \XINT_scipow_a will have
% triggered usage of the (new incarnation) of \XINTinFloatPowerH which combines
% \xintFloatPower and square root extraction. So we only have to handle here
% the case of integer exponents which will trigger execution of this
% \XINT_pow_int macro passed as parameter to \xintpow.|
%    \begin{macrocode}
\def\XINT_pow_int #1/1[#2]%
{%
  \expandafter\XINT_pow_int_a\romannumeral0\XINT_dsx_addzeros{#2}#1;.%
}%
%    \end{macrocode}
% \lverb|1.4e had a bug here for integer exponents >= 10000: they triggered
% going back to the floating point routine but at a late location where
% the log10()/pow10() approach is used.|
%    \begin{macrocode}
\def\XINT_pow_int_a #1#2.%
{%
  \ifnum\if-#1\xintLength{#2}\else\xintLength{#1#2}\fi>\xint_c_iv
       \expandafter\XINT_pow_bigint
  \else\expandafter\XINT_pow_int_b
  \fi #1#2.%
}%
%    \end{macrocode}
% \lverb|At 1.4f we correctly jump to the appropriate entry point into the
% \xintFloatPower routine of $xintfracnameimp, in case of a big integer
% exponent.|
%    \begin{macrocode}
\def\XINT_pow_bigint #1.#2%
{%
    \XINT_flpower_checkB_a#1.\XINTdigits.{#2}{\XINTinfloatS[\XINTdigits]}%
}%
\def\XINT_pow_int_b #1.#2%
{%
%    \end{macrocode}
% \lverb|We now check if the output will not be too bulky. We use here (on the
% a of a^b) \xintraw, not \xintrez, on purpose so that for example 9.0^9999
% is computed in floating point sense but 9^9999 is computed exactly. However
% 9.0^5000 will be computed exactly.  And if I used \xintrez here \xinteval{100^2}
% would print 10000.0 and \xinteval{100^3} would print 1.0e6.  Thus situation
% is complex.
%
% By the way I am happy to see that 9.0*9.0 in
% \xinteval does print 81.0 but the truth is that internally it does have the
% more bulky 8100/1[-2] maybe I should make some revision of this, i.e. use
% rather systematically \xintREZ on input rather than \xintRaw (note taken on
% 2021/05/08 at time of doing 1.4f bugfix release).|
%    \begin{macrocode}
    \expandafter\XINT_pow_int_c\romannumeral0\xintraw{#2}\xint:#1\xint:
}%
%    \end{macrocode}
% \lverb|The \XINT_fpow_fork is (quasi top level) entry point we have found
% into the legacy \xintPow routine of $xintfracnameimp. Its interface is a bit
% weird, but let's not worry about this now.
%
%|
%    \begin{macrocode}
\def\XINT_pow_int_c#1#2/#3[#4]\xint:#5\xint:
{%
    \if0\ifnum\numexpr\xint_c_x^iv/%
           (\xintLength{#1#2}\if-#1-\xint_c_i\fi)<\XINT_Abs#5 %
       1\else
        \ifnum\numexpr\xint_c_x^iv/\xintLength{#3}<\XINT_Abs#5 %
       1\else
    0\fi\fi
      \expandafter\XINT_fpow_fork\else\expandafter\XINT_pow_bigint_i
    \fi
    #5\Z{#4}{#1#2}{#3}%
}%
%    \end{macrocode}
% \lverb|\XINT_pow_bigint_i is like \XINT_pow_bigint but has its parameters
% organized differently.|
%    \begin{macrocode}
\def\XINT_pow_bigint_i#1\Z#2#3#4%
{%
    \XINT_flpower_checkB_a#1.\XINTdigits.{#3/#4[#2]}{\XINTinfloatS[\XINTdigits]}%
}%
%    \end{macrocode}
% \subsection{Macro support for \cshnolabel{xintexpr} and \cshnolabel{xintfloatexpr} syntax}
% \subsubsection{The \cshn{log10()} and \cshn{pow10()} functions}
% \lverb|&
% Up to 8 digits included we use the poormanlog based ones.
% |
%    \begin{macrocode}
\ifnum\XINTdigits<9
\expandafter\def\csname XINT_expr_func_log10\endcsname#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\PoorManLogBaseTen#3}}%
}%
\expandafter\def\csname XINT_expr_func_pow10\endcsname#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\PoorManPowerOfTen#3}}%
}%
\else
\expandafter\def\csname XINT_expr_func_log10\endcsname#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatLogTen#3}}%
}%
\expandafter\def\csname XINT_expr_func_pow10\endcsname#1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatPowTen#3}}%
}%
\fi
\expandafter\let\csname XINT_flexpr_func_log10\expandafter\endcsname
                \csname XINT_expr_func_log10\endcsname
\expandafter\let\csname XINT_flexpr_func_pow10\expandafter\endcsname
                \csname XINT_expr_func_pow10\endcsname
%    \end{macrocode}
% \subsubsection{The \cshn{log()}, \cshn{exp()} functions}
%    \begin{macrocode}
\ifnum\XINTdigits<9
\def\XINT_expr_func_log #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\PoorManLog#3}}%
}%
\def\XINT_expr_func_exp #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\PoorManExp#3}}%
}%
\let\XINT_flexpr_func_log\XINT_expr_func_log
\let\XINT_flexpr_func_exp\XINT_expr_func_exp
\else
\def\XINT_expr_func_log #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatLog#3}}%
}%
\def\XINT_expr_func_exp #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:one
    {\romannumeral`&&@\XINTinFloatExp#3}}%
}%
\let\XINT_flexpr_func_log\XINT_expr_func_log
\let\XINT_flexpr_func_exp\XINT_expr_func_exp
\fi
%    \end{macrocode}
% \subsubsection{The \cshn{pow()} function}
% \lverb|The mapping of ** and ^ to \XINTinFloatSciPow (in \xintfloatexpr
% context) and \xintPow (in \xintexpr context),
% is done in $xintexprnameimp.|
%    \begin{macrocode}
\def\XINT_expr_func_pow #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\xintPow#3}}%
}%
\def\XINT_flexpr_func_pow #1#2#3%
{%
    \expandafter #1\expandafter #2\expandafter{%
    \romannumeral`&&@\XINT:NEhook:f:one:from:two
    {\romannumeral`&&@\XINTinFloatSciPow#3}}%
}%
%    \end{macrocode}
% \subsection{End of package loading for low Digits}
%    \begin{macrocode}
\ifnum\XINTdigits<9 \expandafter\XINTendxintloginput\fi%
%    \end{macrocode}
% \subsection{Stored constants}
% \lverb|The constants were obtained from Maple at 80 digits: fractional power
% of 10, but only one logarithm log(10).
%
% Currently the code whether for exponential or logarihm will not screen out 0
% digits and even will do silly multiplication by 10^0 = 1 in that case, and
% we need to store such silly values.
%
% We add the data for the 10^-0.i etc... because pre-computing them on the fly
% significantly adds overhead to the package loading.
%
% The fractional powers of ten with D+5 digits are used to compute pow10()
% function, those with D+10 digits are used to compute log10() function. This
% is done with
% an elevated precision for two reasons:
% (- handling of inputs near 1,
% :- in order for a^b = pow10(b*log10(a)) to keep accuracy&
%    even with large exponents, say in absolute value up to 1e7,&
%    degradation beginning to show-up at 1e8.
% )
% |
%    \begin{macrocode}
\def\XINT_tmpa{1[0]}%
\expandafter\let\csname XINT_c_1_0\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_2_0\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_3_0\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_4_0\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_5_0\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_6_0\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_1_0_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_2_0_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_3_0_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_4_0_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_5_0_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_6_0_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_1_0_inv\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_2_0_inv\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_3_0_inv\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_4_0_inv\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_5_0_inv\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_6_0_inv\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_1_0_inv_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_2_0_inv_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_3_0_inv_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_4_0_inv_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_5_0_inv_x\endcsname\XINT_tmpa
\expandafter\let\csname XINT_c_6_0_inv_x\endcsname\XINT_tmpa
\def\XINT_tmpa#1#2#3#4;%
   {\expandafter\edef\csname XINT_c_#1_#2\endcsname{\XINTinFloat[\XINTdigitsormax+5]{#3#4[-79]}}%
    \expandafter\edef\csname XINT_c_#1_#2_x\endcsname{\XINTinFloat[\XINTdigitsormax+10]{#3#4[-79]}}%
   }%
% 10^0.i
\XINT_tmpa 1 1 12589254117941672104239541063958006060936174094669310691079230195266476157825020;%
\XINT_tmpa 1 2 15848931924611134852021013733915070132694421338250390683162968123166568636684540;%
\XINT_tmpa 1 3 19952623149688796013524553967395355579862743154053460992299136670049309106980490;%
\XINT_tmpa 1 4 25118864315095801110850320677993273941585181007824754286798884209082432477235613;%
\XINT_tmpa 1 5 31622776601683793319988935444327185337195551393252168268575048527925944386392382;%
\XINT_tmpa 1 6 39810717055349725077025230508775204348767703729738044686528414806022485386945804;%
\XINT_tmpa 1 7 50118723362727228500155418688494576806047198983281926392969745588901125568883069;%
\XINT_tmpa 1 8 63095734448019324943436013662234386467294525718822872452772952883349494329768681;%
\XINT_tmpa 1 9 79432823472428150206591828283638793258896063175548433209232392931695569719148754;%
% 10^0.0i
\XINT_tmpa 2 1 10232929922807541309662751748198778273411640572379813085994255856738296458625172;%
\XINT_tmpa 2 2 10471285480508995334645020315281400790567914715039292120056525299012577641023719;%
\XINT_tmpa 2 3 10715193052376064174083022246945087339158659633422172707894501914136771607653870;%
\XINT_tmpa 2 4 10964781961431850131437136061411270464271158762483023169080841607885740984711300;%
\XINT_tmpa 2 5 11220184543019634355910389464779057367223085073605529624450744481701033026862244;%
\XINT_tmpa 2 6 11481536214968827515462246116628360182562102373996119340874991068894793593040890;%
\XINT_tmpa 2 7 11748975549395295417220677651268442278134317971793124791953875805007912852226246;%
\XINT_tmpa 2 8 12022644346174129058326127151935204486942664354881189151104892745683155052368222;%
\XINT_tmpa 2 9 12302687708123815342415404364750907389955639574572144413097319170011637639124482;%
% 10^0.00i
\XINT_tmpa 3 1 10023052380778996719154048893281105540536684535421606464116348523047431367720401;%
\XINT_tmpa 3 2 10046157902783951424046519858132787392010166060319618489538315083825599423438638;%
\XINT_tmpa 3 3 10069316688518041699296607872661381368099438247964820601930206419324524707606686;%
\XINT_tmpa 3 4 10092528860766844119155277641202580844111492027373621434478800545314309618714957;%
\XINT_tmpa 3 5 10115794542598985244409323144543146957419235215102899054703546688078254946034250;%
\XINT_tmpa 3 6 10139113857366794119988279023017296985954042032867436525450889437280417044987125;%
\XINT_tmpa 3 7 10162486928706956276733661150135543062420167220622552197768982666050994284378619;%
\XINT_tmpa 3 8 10185913880541169240797988673338257820431768224957171297560936579346433061037662;%
\XINT_tmpa 3 9 10209394837076799554149033101487543990018213667630072574873723356334069913329713;%
% 10^0.000i
\XINT_tmpa 4 1 10002302850208247526835942556719413318678216124626534526963475845228205382579041;%
\XINT_tmpa 4 2 10004606230728403216239656646745503559081482371024284871882409614422496765669196;%
\XINT_tmpa 4 3 10006910141682589957025973521996241909035914023642264228577379693841345823180462;%
\XINT_tmpa 4 4 10009214583192958761081718336761022426385537997384755843291864010938378093197023;%
\XINT_tmpa 4 5 10011519555381688769842032367472488618040778885656970999331288116685029387850446;%
\XINT_tmpa 4 6 10013825058370987260768186632475607982636715641432550952229573271596547716373358;%
\XINT_tmpa 4 7 10016131092283089653826887255241073941084503769368844606021481400409002185558343;%
\XINT_tmpa 4 8 10018437657240259517971072914549205297136779497498835020699531587537662833033174;%
\XINT_tmpa 4 9 10020744753364788577622204725249622301332888222801030351604197113557132455165040;%
% 10^0.0000i
\XINT_tmpa 5 1 10000230261160268806710649793464495797824846841503180050673957122443571394978721;%
\XINT_tmpa 5 2 10000460527622557806255008596155855743730116854295068547616656160734125748005947;%
\XINT_tmpa 5 3 10000690799386989083565213461287219981856579552059660369243804541364501659468630;%
\XINT_tmpa 5 4 10000921076453684726384543254593368743049141124080210677706489564626675960578367;%
\XINT_tmpa 5 5 10001151358822766825267483384008265483772370538793312970508590203623535763866465;%
\XINT_tmpa 5 6 10001381646494357473579790530833073090516914490540536234536867917078761046656260;%
\XINT_tmpa 5 7 10001611939468578767498557382394677469502542123237272447312733350028467607076918;%
\XINT_tmpa 5 8 10001842237745552806012277366194752842273812293689190856411757410911882303011468;%
\XINT_tmpa 5 9 10002072541325401690920909385549403068574626162727745910217443397959031898734024;%
% 10^0.00000i
\XINT_tmpa 6 1 10000023025877439451356029805459000097926504781151663770980171880313737943886754;%
\XINT_tmpa 6 2 10000046051807898005897723104514851394069452605882077809669546315010724085277647;%
\XINT_tmpa 6 3 10000069077791375785706217087438809625967243923218032821061587553353589726808164;%
\XINT_tmpa 6 4 10000092103827872912862930047032391734439796534302560512742030066798473305401477;%
\XINT_tmpa 6 5 10000115129917389509449561379274639104559958866285946533811801963402821672829477;%
\XINT_tmpa 6 6 10000138156059925697548091583969382297005329013199894805417325991907389143667949;%
\XINT_tmpa 6 7 10000161182255481599240782265392507269793911275470978276390154932321984777772469;%
\XINT_tmpa 6 8 10000184208504057336610176132939223090407041937631374389422968832433217547184883;%
\XINT_tmpa 6 9 10000207234805653031739097001771331138303016031686764989867510425362339583809842;%
\def\XINT_tmpa#1#2#3#4;%
   {\expandafter\edef\csname XINT_c_#1_#2_inv\endcsname{\XINTinFloat[\XINTdigitsormax+5]{#3#4[-80]}}%
    \expandafter\edef\csname XINT_c_#1_#2_inv_x\endcsname{\XINTinFloat[\XINTdigitsormax+10]{#3#4[-80]}}%
   }%
% 10^-0.i
\XINT_tmpa 1 1 79432823472428150206591828283638793258896063175548433209232392931695569719148754;%
\XINT_tmpa 1 2 63095734448019324943436013662234386467294525718822872452772952883349494329768681;%
\XINT_tmpa 1 3 50118723362727228500155418688494576806047198983281926392969745588901125568883069;%
\XINT_tmpa 1 4 39810717055349725077025230508775204348767703729738044686528414806022485386945804;%
\XINT_tmpa 1 5 31622776601683793319988935444327185337195551393252168268575048527925944386392382;%
\XINT_tmpa 1 6 25118864315095801110850320677993273941585181007824754286798884209082432477235613;%
\XINT_tmpa 1 7 19952623149688796013524553967395355579862743154053460992299136670049309106980490;%
\XINT_tmpa 1 8 15848931924611134852021013733915070132694421338250390683162968123166568636684540;%
\XINT_tmpa 1 9 12589254117941672104239541063958006060936174094669310691079230195266476157825020;%
% 10^-0.0i
\XINT_tmpa 2 1 97723722095581068269707600696156123863427170069897801526639004097175507042084888;%
\XINT_tmpa 2 2 95499258602143594972395937950148401513087269708053320302465127242741421479104601;%
\XINT_tmpa 2 3 93325430079699104353209661168364840720225485199736026149257155811788093771138272;%
\XINT_tmpa 2 4 91201083935590974212095940791872333509323858755696109214760361851771695487999100;%
\XINT_tmpa 2 5 89125093813374552995310868107829696398587478293004836994794349506746891059190135;%
\XINT_tmpa 2 6 87096358995608063751082742520877054774747128501284704090761796673224328569285177;%
\XINT_tmpa 2 7 85113803820237646781712631859248682794521725442067093899553745086385146367436049;%
\XINT_tmpa 2 8 83176377110267100616669140273840405263880767161887438462740286611379995442629360;%
\XINT_tmpa 2 9 81283051616409924654127879773132980187568851100062454636602325121954484722491710;%
% 10^-0.00i
\XINT_tmpa 3 1 99770006382255331719442194285376231055211861394573154624878230890945476532432225;%
\XINT_tmpa 3 2 99540541735152696244806147089510943107144177264574823668081299845609359857038344;%
\XINT_tmpa 3 3 99311604842093377157642607688515474663519162181123336122073822476734517364853150;%
\XINT_tmpa 3 4 99083194489276757440828314388392035249938006860819409201135652190410238171119287;%
\XINT_tmpa 3 5 98855309465693884028524792978202683686410726723055209558576898759166522286083202;%
\XINT_tmpa 3 6 98627948563121047157261523093421290951784086730437722805070296627452491731402556;%
\XINT_tmpa 3 7 98401110576113374484101831088824192144756194053451911515003663381199842081528019;%
\XINT_tmpa 3 8 98174794301998439937928161622872240632362817134775142288598128693131032909278350;%
\XINT_tmpa 3 9 97948998540869887269961493687844910565420716785032030061251916654655049965062649;%
% 10^-0.000i
\XINT_tmpa 4 1 99976976799815658635141604638981297541396466984477711459083930684685186989697929;%
\XINT_tmpa 4 2 99953958900308784552845777251512089759003230012954649234748668826546533498169555;%
\XINT_tmpa 4 3 99930946300258992168693777702512591351888960684418033717545524043693899420866954;%
\XINT_tmpa 4 4 99907938998446176870082987427724649318531547584410414997787083472394558389284098;%
\XINT_tmpa 4 5 99884936993650514951538205746462968844845952521633937925370747725933629958238429;%
\XINT_tmpa 4 6 99861940284652463550037839584112909891259691850983307437097305856727153967481065;%
\XINT_tmpa 4 7 99838948870232760580354983175435314251655958968480344701699631967048474751069525;%
\XINT_tmpa 4 8 99815962749172424670413384320528274471550942114263604264788586703624513163664479;%
\XINT_tmpa 4 9 99792981920252755096658293766085025870392854106037465990011216356523334125368417;%
% 10^-0.0000i
\XINT_tmpa 5 1 99997697441416293040019992468837639003787989306240470048763511538639048400765328;%
\XINT_tmpa 5 2 99995394935850346394065999228750187791584034668237852053859761641089829514536011;%
\XINT_tmpa 5 3 99993092483300939297147020491645017932348508508297743745039515152378182676736684;%
\XINT_tmpa 5 4 99990790083766851012380885556584619169980753943113396677545915245611923361705686;%
\XINT_tmpa 5 5 99988487737246860830993605587529673614422529030613405900998412734419982883669223;%
\XINT_tmpa 5 6 99986185443739748072318726405984801565268578044798475766025647187221659622450651;%
\XINT_tmpa 5 7 99983883203244292083796681298546635825139453823571398432959235283529730820181019;%
\XINT_tmpa 5 8 99981581015759272240974143839353881367972777961073357987943600347058023396510672;%
\XINT_tmpa 5 9 99979278881283467947503380727439017235290006415950636109257677645557027950744160;%
% 10^-0.00000i
\XINT_tmpa 6 1 99999769741755795297487775997495948154386159348543852707438213487494386559762090;%
\XINT_tmpa 6 2 99999539484041779185217876175552674518572114763104546143049036309870762496098218;%
\XINT_tmpa 6 3 99999309226857950442387361668529812394860404492721699528707852590634886516924591;%
\XINT_tmpa 6 4 99999078970204307848196104610199226516866442484686906173860803560254163287393673;%
\XINT_tmpa 6 5 99998848714080850181846788127272455158309917012010320554498356105168896062430977;%
\XINT_tmpa 6 6 99998618458487576222544906332928167145404344730731751204389698696345970645201375;%
\XINT_tmpa 6 7 99998388203424484749498764320339633772810463403640242228131015918494067456365331;%
\XINT_tmpa 6 8 99998157948891574541919478156202215623119146605983303201215215949834619332550929;%
\XINT_tmpa 6 9 99997927694888844379020974874260864289829523807763942234420930258187873904191138;%
% log(10)
\edef\XINT_c_logten
 {\XINTinFloat[\XINTdigitsormax+4]
  {23025850929940456840179914546843642076011014886287729760333279009675726096773525[-79]}}%
\edef\XINT_c_oneoverlogten
 {\XINTinFloat[\XINTdigitsormax+4]
  {43429448190325182765112891891660508229439700580366656611445378316586464920887077[-80]}}%
\edef\XINT_c_oneoverlogten_xx
 {\XINTinFloat[\XINTdigitsormax+14]
  {43429448190325182765112891891660508229439700580366656611445378316586464920887077[-80]}}%
%    \end{macrocode}
% \subsection{April 2021: at last, \csh{XINTinFloatPowTen}, \csh{XINTinFloatExp}}
% \lverb|Done April 2021. I have procrastinated (or did not have time to
% devote to this) at least 5 years, even more.
%
% Speed improvements will have to wait to long delayed refactoring of core
% floating point support which is still in the 2013 primitive state !
%
% I did not try to optimize for say 16 digits, as I was more focused on
% reaching 60 digits in a reasonably efficient manner (trigonometric functions
% achieved this since 2019) in the same coding framework.  Finally, up to 62 digits.
%
% The stored constants are log(10) at P+4 digits and the powers 10^0.d,
% 10^0.0d, etc, up to 10^0.00000d for d=1..9, as well as their inverses, at
% P+5 and P+10 digits. The constants were obtained from Maple at 80 digits.
%
% Initially I constructed the exponential series exp(h) as one big unique nested
% macro. It contained pre-rounded values of the 1/i! but would float-round h
% to various numbers of digits, with always the full initial h as input.
%
% After having experimented with the logarithm, I redid exp(h) = 1 + h(1 +
% h(1/2 + ...)) with many macros in order to have more readable code, and to
% dynamically cut-off more and more digits from h the deeper it is used. See
% the logarithm code for (perhaps) more comments.
%
% The thresholds have been obtained from considerations including an hmax
% (a bit more than 0.5 log(10) 10^-6). Here is the table:
%
%( - maximal value of P:  8, 15, 21,  28,  35,  42,  48,  55,  62
%: - last included term: /1, /2, /6, /4!, /5!, /6!, /7!, /8!, /9!
%)
% 
% Computations are done morally targeting P+4 fractional fixed point digits,
% with a stopping criteria at say about 5e(-P-4), which was used for the table
% above using only the worst case.  As the used macros are a mix of exact
% operations and floating point reductions this is in practice a bit
% different. The h will be initially float rounded to P-1 digits. It is
% cut-off more and more, the deeper nested it is used.
%
% The code for this evaluation of 10^x is very poor with x very near zero: it
% does silly multiplication by 1, and uses more terms of exponential series
% than would then be necessary.
%
% For the computation of exp(x) as 10^(c*x) with c=log(10)^-1, we need more
% precise c the larger abs(x) is. For abs(x)<1 (or 2), the c with P+4
% fractional digits is sufficient. But decimal exponents are more or less
% allowed to be near the TeX maximum 2^31-1, which means that abs(x) could be
% as big  as 0.5e10, and we then need c with P+14 digits to cover that range.
%
% I am hesitating whether to first examine integral part of abs(x) and for
% example to use c with either P+4, P+9 or P+14 digits, and also take this
% opportunity to inject an error message if x is too big before TeX arithmetic
% overflow happens later on. For time being I will use overhead of
% oneoverlogten having ample enough digits...
%
% The exponent received as input is float rounded to P + 14 digits. In
% practice the input will be already a P-digits float. The motivation here is
% for low Digits situation: but this done so that for example with Digits=4,
% we want exp(12345) not to be evaluated as exp(12350) which would have no
% meaning at all. The +14 is because we have prepared 1/log(10) with that many
% significant digits. This conundrum is due to the inadequation of the world
% of floating point numbers with exp() and log(): clearly exp() goes from
% fixed point to floating point and log() goes from floating point to fixed
% point, and coercing them to work inside the sole floating point domain is
% not mathematically natural.  Although admittedly it does create interesting
% mathematical questions!  A similar situation applies to functions such as
% cos() and sin(), what sense is there in the expression cos(exp(50)) for
% example with 16 digits precision? My opinion is that it does not make ANY
% sense.  Anyway, I shall obide.
%
% As \XINTinFloatS will not add unnecessarily trailing zeros, the
% \XINTdigits+14 is not really an enormous overhead for integer exponents,
% such as in the example above the 12345, or more realistically small integer
% exponents, and if the input is already float rounded to P digits, the
% overhead is also not enormous (float-rounding is costly when the input is a
% fraction).
%
% \XINTinfloatpowten will receive an input with at least P+14 and up to 2P+28
% digits... fortunaltely with no fraction part and will start rounding it in
% the fixed point sense of its input to P+4 digits after decimal point, which
% is not enormously costly.
%
% Of course all these things pile up...
% |
%    \begin{macrocode}
\def\XINTinFloatExp{\romannumeral0\XINTinfloatexp}%
\def\XINT_tmpa#1.{%
\def\XINTinfloatexp##1%
{%
    \XINTinfloatpowten
    {\xintMul{\XINT_c_oneoverlogten_xx}{\XINTinFloatS[#1]{##1}}}%
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+14.%
%    \end{macrocode}
% \lverb|&
% Here is how the reduction to computations of an exp(h) via series is done.
%
% Starting from x, after initial argument normalization, it is fixed-point
% rounded to 6 fractional digits giving x'' = ±n.d_1...d_6 (which may be 0).
%
% I have to resist temptation using very low level routines here and wisely
% will employ the available user-level stuff.  One computes then the
% difference x-x'' which gives some eta, and the h will be log(10).eta.  The
% subtraction and multiplication are done exactly then float rounded to P-1
% digits to obtain the h.
%
% Then exp(h) is computed. And to finish it is multiplied with the stored
% 10^±0.d_1, 10^±0.0d_2, etc...., constants and its decimal exponent is
% increased by ±n. These operations are done at P+5 floating point digits. The
% final result is then float-rounded to the target P digits.
%
% Currently I may use nested macros for some operations but will perhaps
% revise in future (it makes tracing very complicated if one does not have
% intermediate macros). The exponential series itself was initially only one
% single macro, but as commented above I have now modified it.|
%    \begin{macrocode}
\def\XINTinFloatPowTen{\romannumeral0\XINTinfloatpowten}%
\def\XINT_tmpa#1.{%
\def\XINTinfloatpowten##1%
{%
    \expandafter\XINT_powten_fork
    \romannumeral0\xintiround{#1}{##1}[-#1]%
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+4.%
\def\XINT_powten_fork#1%
{%
    \xint_UDzerominusfork
       #1-\XINT_powten_zero
       0#1\XINT_powten_neg
        0-\XINT_powten_pos
    \krof #1%
}%
\def\XINT_powten_zero #1[#2]{ 1[0]}%
%    \end{macrocode}
% \lverb|This rounding may produce 0.000000 but will always have 6 exactly
% fractional digits, because the special case of a zero input was filtered out
% preventively.
% |
%    \begin{macrocode}
\def\XINT_powten_pos#1[#2]%
{%
    \expandafter\XINT_powten_pos_a\romannumeral0\xintround{6}{#1[#2]}#1[#2]%
}%
\def\XINT_tmpa #1.#2.#3.{%
\def\XINT_powten_pos_a ##1.##2##3##4##5##6##7##8[##9]%
{%
   \expandafter\XINT_infloate
    \romannumeral0\XINTinfloat[#3]{%
     \xintMul{\csname XINT_c_1_##2\endcsname}{%
      \XINTinFloat[#1]{%
       \xintMul{\csname XINT_c_2_##3\endcsname}{%
        \XINTinFloat[#1]{%
         \xintMul{\csname XINT_c_3_##4\endcsname}{%
          \XINTinFloat[#1]{%
           \xintMul{\csname XINT_c_4_##5\endcsname}{%
            \XINTinFloat[#1]{%
             \xintMul{\csname XINT_c_5_##6\endcsname}{%
              \XINTinFloat[#1]{%
               \xintMul{\csname XINT_c_6_##7\endcsname}{%
                \xintAdd{1[0]}{%
                 \expandafter\XINT_Exp_series_a_ii
                  \romannumeral0\XINTinfloat[#2]{%
                   \xintMul{\XINT_c_logten}%
                           {\xintAdd{-##1.##2##3##4##5##6##7}{##8[##9]}}%
                   }%
                 \xint:
                }%
               }}}}}}}}}}}}{##1}%
}}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax+5\expandafter.%
  \the\numexpr\XINTdigitsormax-1\expandafter.%
  \the\numexpr\XINTdigitsormax.%
%    \end{macrocode}
% \lverb|This rounding may produce -0.000000 but will always have 6 exactly
% fractional digits and a leading minus sign.|
%    \begin{macrocode}
\def\XINT_powten_neg#1[#2]%
{%
    \expandafter\XINT_powten_neg_a\romannumeral0\xintround{6}{#1[#2]}#1[#2]%
}%
\def\XINT_tmpa #1.#2.#3.{%
\def\XINT_powten_neg_a -##1.##2##3##4##5##6##7##8[##9]%
{%
   \expandafter\XINT_infloate
    \romannumeral0\XINTinfloat[#3]{%
     \xintMul{\csname XINT_c_1_##2_inv\endcsname}{%
      \XINTinFloat[#1]{%
       \xintMul{\csname XINT_c_2_##3_inv\endcsname}{%
        \XINTinFloat[#1]{%
         \xintMul{\csname XINT_c_3_##4_inv\endcsname}{%
          \XINTinFloat[#1]{%
           \xintMul{\csname XINT_c_4_##5_inv\endcsname}{%
            \XINTinFloat[#1]{%
             \xintMul{\csname XINT_c_5_##6_inv\endcsname}{%
              \XINTinFloat[#1]{%
               \xintMul{\csname XINT_c_6_##7_inv\endcsname}{%
                \xintAdd{1[0]}{%
                 \expandafter\XINT_Exp_series_a_ii
                  \romannumeral0\XINTinfloat[#2]{%
                   \xintMul{\XINT_c_logten}%
                           {\xintAdd{##1.##2##3##4##5##6##7}{##8[##9]}}%
                   }%
                 \xint:
                }%
               }}}}}}}}}}}}{-##1}%
}}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax+5\expandafter.%
  \the\numexpr\XINTdigitsormax-1\expandafter.%
  \the\numexpr\XINTdigitsormax.%
%    \end{macrocode}
% \subsubsection{Exponential series}
% \lverb|Or rather here h(1 + h(1/2 + h (1/6 + ....))). Upto at most h^9/9!
% term.
%
% The used initial h has been float rounded to P-1 digits.|
%    \begin{macrocode}
\def\XINT_tmpa#1.#2.{%
\def\XINT_Exp_series_a_ii##1\xint:
{%
    \expandafter\XINT_Exp_series_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_Exp_series_b##1[##2]\xint:
{%
    \expandafter\XINT_Exp_series_c_
    \romannumeral0\xintadd{1}{\xintHalf{##10}[##2-1]}\xint:
}%
\def\XINT_Exp_series_c_##1\xint:##2\xint:
{%
    \XINTinFloat[#2]{\xintMul{##1}{##2}}%
}%
}%
\expandafter\XINT_tmpa
            \the\numexpr\XINTdigitsormax-6\expandafter.%
            \the\numexpr\XINTdigitsormax-1.% 
\ifnum\XINTdigits>15
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_Exp_series_a_ii##1\xint:
{%
    \expandafter\XINT_Exp_series_a_iii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_Exp_series_a_iii##1\xint:
{%
    \expandafter\XINT_Exp_series_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_Exp_series_b##1[##2]\xint:
{%
    \expandafter\XINT_Exp_series_c_i
    \romannumeral0\xintadd{#3}{##1/6[##2]}\xint:
}%
\def\XINT_Exp_series_c_i##1\xint:##2\xint:
{%
    \expandafter\XINT_Exp_series_c_
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-13\expandafter.%
  \the\numexpr\XINTdigitsormax-6.%
  {5[-1]}.%
  {1[0]}.%
\fi
\ifnum\XINTdigits>21
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_Exp_series_a_iii##1\xint:
{%
    \expandafter\XINT_Exp_series_a_iv
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_Exp_series_a_iv##1\xint:
{%
    \expandafter\XINT_Exp_series_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_Exp_series_b##1[##2]\xint:
{%
    \expandafter\XINT_Exp_series_c_ii
    \romannumeral0\xintadd{#3}{##1/24[##2]}\xint:
}%
\def\XINT_Exp_series_c_ii##1\xint:##2\xint:
{%
    \expandafter\XINT_Exp_series_c_i
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-19\expandafter.%
  \the\numexpr\XINTdigitsormax-13\expandafter.%
  \romannumeral0\XINTinfloat[\XINTdigitsormax-13]{1/6[0]}.%
  {5[-1]}.%
\fi
\ifnum\XINTdigits>28 
\def\XINT_tmpa #1 #2 #3 #4 #5 #6 #7 %
{%
 \def\XINT_tmpb ##1##2##3##4%
 {%
  \def\XINT_tmpc####1.####2.####3.####4.%
  {%
   \def##2########1\xint:
   {%
       \expandafter##1%
       \romannumeral0\XINTinfloatS[####2]{########1}\xint:########1\xint:
   }%
   \def##1########1\xint:
   {%
       \expandafter\XINT_Exp_series_b
       \romannumeral0\XINTinfloatS[####1]{########1}\xint:########1\xint:
   }%
   \def\XINT_Exp_series_b########1[########2]\xint:
   {%
       \expandafter##3%
       \romannumeral0\xintadd{####3}{########1/#5[########2]}\xint:
   }%
   \def##3########1\xint:########2\xint:
   {%
       \expandafter##4%
       \romannumeral0\xintadd{####4}{\XINTinFloat[####2]{\xintMul{########1}{########2}}}\xint:
   }%
  }%
 }%
 \expandafter\XINT_tmpb
 \csname XINT_Exp_series_a_\romannumeral\numexpr#1\expandafter\endcsname
 \csname XINT_Exp_series_a_\romannumeral\numexpr#1-1\expandafter\endcsname
 \csname XINT_Exp_series_c_\romannumeral\numexpr#1-2\expandafter\endcsname
 \csname XINT_Exp_series_c_\romannumeral\numexpr#1-3\endcsname
 \expandafter\XINT_tmpc
 \the\numexpr\XINTdigitsormax-#2\expandafter.%
 \the\numexpr\XINTdigitsormax-#3\expandafter.\expanded{%
 \XINTinFloat[\XINTdigitsormax-#3]{1/#6[0]}.%
 \XINTinFloat[\XINTdigitsormax-#4]{1/#7[0]}.%
 }%
}%
\XINT_tmpa 5 26 19 13 120 24 6 %<-- keep space
\ifnum\XINTdigits>35 \XINT_tmpa 6 33 26 19 720 120 24 \fi
\ifnum\XINTdigits>42 \XINT_tmpa 7 40 33 26 5040 720 120 \fi
\ifnum\XINTdigits>48 \XINT_tmpa 8 46 40 33 40320 5040 720 \fi
\ifnum\XINTdigits>55 \XINT_tmpa 9 53 46 40 362880 40320 5040 \fi
\fi
%    \end{macrocode}
% \subsection{April 2021: at last \csh{XINTinFloagLogTen}, \csh{XINTinFloatLog}}
% \lverb?&
% Attention that this is not supposed to be used with \XINTdigits at
% 8 or less, it will crash if that is the case. The log10() and log()
% functions in case \XINTdigits is at most 8 are mapped to \PoormanLogBaseTen
% respectively \PoormanLog macros.
%
% In the explications here I use the function names rather than the macro
% names.
%
% Both log(x) and log10(x) are on top of an underlying macro which will
% produce z and h such that x is about 10^z e^h (with h being small is
% obtained via a log series). Then log(x) computes log(10)z+h whereas log10(x)
% computes as z+h/log(10).
%
% There will be three branches [NO FINALLY ONLY TWO BRANCHES SINCE 1.4f]
% according to situation of x relative to 1.  Let y be the math value log10(x)
% that we want to approximate to target precision P digits. P is assumed at
% least 9.
%
% I will describe the algorithm roughly, but skip its underlying support
% analysis; at some point I mention "fixed point calculations", but in
% practice it is not done exactly that way, but describing it would be
% complicated so look at the code which is very readable (by the author, at
% the present time).
%
% First we compute z = ±n.d_1d_2...d_6 as the rounded to 6 fractional digits
% approximation of y=log10(x) obtained by first using the poormanlog macros on x
% (float rounded to 9 digits) then rounding as above.
%
% Warning: this description is not in sync with the code, now the case where
% d_1d_2...d_6 is 000000 is filtered out and one jumps directly either to case
% I if n≠0 or to case III if n=0. The case when rounding produces a z equal
% to zero is also handled especially.
%
% WARNING: at 1.4f, the CASE I was REMOVED. Everything is handled as CASE II
% or exceptionally case III. Indeed this removal was observed to simply cost
% about 10$% extra time at D=16 digits, which was deemed an acceptable cost.
% The cost is certainly higher at D=9 but also relatively lower at high
% D's. It means that logarithms are always computed with 9, not 4, safety
% **fractional** digits, and this allows to compute powers accurately with
% exponents say up to 1e7, degradation starting to show at 1e8 and for sure at
% 1e9. However for integer and half-integer exponents the old routine
% \xintFloatPower will still be used, and perhaps it will need some increased
% precision update as the documented 0.52ulp error bound is higher than our
% more stringent standards of 2021.
%
% CASE I: [removed at 1.4f!]
% either n is NOT zero or d_1d_2....d_6 is at least 100001. Then we
% compute X = 10^(-z)*x which is near 1, by using the table of powers of
% 10, using P+5 digits significands. Then we compute (exactly) eta = X-1,
% (which is in absolute value less than 0.0000012)
% and obtain
% y as z + log(10)^(-1) times log(1+eta)
% where log(1+eta) = eta - eta^2/2 + eta^3/3- ... is "computed with
% P+4 fractional fixed point digits" [1]_ according to the following table:
%
%( - maximal value of P:  9, 15, 21, 27, 33, 39, 45, 51, 57,  63  
%: - last included term: /1, /2, /3, /4, /5, /6, /7, /8, /9, /10
%)
%
% .. [1] this "P+4" includes leading fractional zeroes so in practice it will
% rather be done as eta(1 - eta(1/2 + eta(1/3-...))), and the inner sums will
% be done in various precisions, the top level (external) eta probably at P-1
% digits, the first inner eta at P-7 digits, the next at P-13, something in
% this style. The heuristics is simple: at P=9 we don't need the first inner
% eta, so let's use there P-9 or rather P-7 digits by security. Similarly at
% P=3 we would not need at all the eta, so let's use the top level one rounded
% at P-3+2 = P-1 digits. And there is a shift by 6 less digits at each inner
% level. RÉFLÉCHIR SI C'EST PAS PLUTÔT P-2 ICI, suffisant au regard de la
% précision par ailleurs pour la réduction près de 1.
%
% The sequence of maximal P's is simply an arithmetic progression.
%
% The addition of z will trigger the final rounding to P digits.
% The inverse of log(10) is precomputed with P+4 digits.
%
% This case I essentially handles x such as max(x,1/x)>10^0.1=1.2589...
%
% CASE II: n is zero and d_1d_2....d_6 is not zero. We operate as in CASE I,
% up to the following differences:
%( - the table of fractional powers of 10 is used with P+10 significands.
%: - the X is also computed with P+10 digits, i.e. eta = X-1 (which obeys&
%  the given estimate) is estimated with P+9 [2]_ fractional fixed points digits&
%  and the log series will be evaluated in this sense.
%: - the constant log(10)^(-1) is still used with only P+4 digits
%)
%
% The log series is terminated according to the following table:
%( - maximal value of P:  4, 10, 16, 22, 28, 34, 40, 46, 52,  58, 64  
%: - last included term: /1, /2, /3, /4, /5, /6, /7, /8, /9, /10
%)
%
% Again the P's are in arithmetic progression, the same as before shifted by
% 5.
%
% .. [2] same remark as above. The top level eta in eta(1 - eta(1/2 -
% eta(...))) will use P+4 significant digits, but the first inner eta will be
% used with only P-2 digits, the next inner one with P-8 digits etc...
%
% This case II handles the x which are near 1, but not as close as 10^±0.000001.
%
% CASE III: z=0. In this case X = x = 1+eta and we use the log series in
% this sense : log(10)^(-1)*eta*(1 - eta/2 + eta^2/3-....)
% where again log(10)^(-1) has been precomputed with P+4 digits
% and morally the series uses P+4 fractional digits (P+3 would probably
% be enough for the precision I want, need to check my notes)
% and the thresholds table is:
%( - maximal value of P:  3,  9, 15, 21, 27, 33, 39, 45, 51,  57,  63  
%: - last included term: /1, /2, /3, /4, /5, /6, /7, /8, /9, /10, /11
%)
%
% This is same progression but shifted by one.
% 
% To summarize some relevant aspects:
%( - this algorithm uses only log(10)^(-1) as precomputed logarithm
%: - in particular the logarithms of small integers 2, 3, 5,... are&
%    not pre-computed. Added note: I have now tested at 16, 32, 48 and 62&
%    digits that all of the log10(n), for n = 1..1000, are computed with&
%    correct rounding. In fact, generally speaking, random testing of a&
%    about 20000 inputs has failed to reveal a single non-correct&
%    rounding. Naturally, randomly testing is not the way to corner&
%    the software into its weak points...
%: - it uses two tables of fractional powers of ten: one with P+5 digits and&
%    another one with extended precision at P+10 digits.
%: - it needs three distinct implementations of the log series.
%: - it does not use the well-known trick reducing to using only odd powers&
%    in the log series (somehow I have come to dread divisions, even though&
%    here as is well-known it could be replaced with some product, my&
%    impression was that what is gained on one side is lost on the other,&
%    for the range of P I am targeting, i.e. P up to about 60.)
%: - all of this is experimental (in particular the previous item was not&
%    done perhaps out of sheer laziness)
%)
%
% Absolutely no error check is done whether the input x is really positive.
% As seen above the maximal target precision is 63 (not 64).
%
% Update for 1.4f: when the logarithm is computed via case I, i.e. basically
% always except roughly for 0.8<a<1.26, its fractional part has only about 4
% safety digits. This is barely enough for a^b with b near 1000 and certainly
% not enough for a^b with b of the order 10000.
%
% I hesitated with the option to always handle b as N+h with N integer for
% which we can use old \xintFloatPower (which perhaps I will have to update to
% ensure better than the 0.52ulp it mentions in its documentation).  But in
% the end, I decided to simply add a variant where case I is handled as case
% II, i.e. with 9 not 4 safety fractional digits for the logarithm.  This
% variant will be the one used by the power function for fractional exponents
% (non integer, non half-integer).
% 
% ?
%    \begin{macrocode}
\def\XINT_tmpa#1.{%
\def\XINTinFloatLog{\romannumeral0\XINTinfloatlog}%
\def\XINTinfloatlog
{%
    \expandafter\XINT_log_out
    \romannumeral0\expandafter\XINT_logtenxdg_a
    \romannumeral0\XINTinfloat[#1]%{##1}
}%
\def\XINT_log_out ##1\xint:##2\xint:
{%
    \XINTinfloat[#1]%
     {\xintAdd{\xintMul{\XINT_c_logten}{##1}}{##2}}%
}%
\def\XINTinFloatLogTen{\romannumeral0\XINTinfloatlogten}%
\def\XINTinfloatlogten
{%
    \expandafter\XINT_logten_out
    \romannumeral0\expandafter\XINT_logtenxdg_a
    \romannumeral0\XINTinfloat[#1]%{##1}
}%
\def\XINT_logten_out ##1\xint:##2\xint:
{%
    \XINTinfloat[#1]%
     {\xintAdd{##1}{\xintMul{\XINT_c_oneoverlogten}{##2}}}%
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax.%
\def\XINTinFloatLogTen_xdgout%#1[#2]
{%
    \romannumeral0\expandafter\XINT_logten_xdgout\romannumeral0\XINT_logtenxdg_a
}%
\def\XINT_logten_xdgout #1\xint:#2\xint:
{%
    \xintadd{#1}{\xintMul{\XINT_c_oneoverlogten_xx}{#2}}%
}%
%    \end{macrocode}
% \lverb|No check is done whether input is negative or vanishes. We apply
% \XINTinfloat[9] which if input is not zero always produces 9 digits (and
% perhaps a minus sign) the first digit is non-zero. This is the expected
% input to \numexpr\PML@<digits><dot>.\relax
%
% The variants xdg_a, xdg_b, xdg_c, xdg_d were added at 1.4f to always go via
% II or III, ensuring more fractional digits to the logarithm for accuracy of
% fractional powers with big exponents. "Old" 1.4e routines were removed.|
%    \begin{macrocode}
\def\XINT_logtenxdg_a#1[#2]%
{%
    \expandafter\XINT_logtenxdg_b
    \romannumeral0\XINTinfloat[9]{#1[#2]}#1[#2]%
}%
\def\XINT_logtenxdg_b#1[#2]%
{%
    \expandafter\XINT_logtenxdg_c
      \romannumeral0\xintround{6}%
      {\xintiiAdd{\xintDSx{-9}{\the\numexpr#2+8\relax}}%
                           {\the\numexpr\PML@#1.\relax}%
       [-9]}%
    \xint:
}%
%    \end{macrocode}
% \lverb|If we were either in 100000000[0] or 999999999[-1] for the #1[#2]
% \XINT_logten_b input, and only in those cases, the \xintRound{6} produced
% "0". We are very near 1 and will treat this as case III, but this is
% sub-optimal.|
%    \begin{macrocode}
\def\XINT_logtenxdg_c #1#2%
{%
    \xint_gob_til_xint:#2\XINT_logten_IV\xint:
    \XINT_logtenxdg_d #1#2%
}%
\def\XINT_logten_IV\xint:\XINT_logtenxdg_d0{\XINT_logten_f_III}%
%    \end{macrocode}
% \lverb|Here we are certain that \xintRound{6} produced a decimal point and
% 6 fractional digit tokens #2, but they can be zeros and also -0.000000 is possible.
%
% If #1 vanishes and #2>100000 we are in case I.
%
% If #1 vanishes and 100000>=#2>0 we are in case II.
%
% If #1 and #2 vanish we are in case III.
%
% If #1 does not vanish we are in case I with a direct quicker access if #2 vanishes.
% 
% Attention to the sign of #1, it is checked later on.
%
% At 1.4f, we handle the case I with as many digits as case II (and exceptionnally case III).
%
% |
%    \begin{macrocode}
\def\XINT_logtenxdg_d #1.#2\xint:
{%
    \ifcase
      \ifnum#1=\xint_c_
        \ifnum #2=\xint_c_ \xint_c_iii\else \xint_c_ii\fi
      \else
        \ifnum#2>\xint_c_ \xint_c_ii\else \xint_c_\fi
      \fi
       \expandafter\XINT_logten_f_Isp
    \or% never
    \or\expandafter\XINT_logten_f_IorII
    \else\expandafter\XINT_logten_f_III
    \fi
    #1.#2\xint:
}%
\def\XINT_logten_f_IorII#1%
{%
    \xint_UDsignfork
       #1\XINT_logten_f_IorII_neg
        -\XINT_logten_f_IorII_pos
    \krof #1%
}%
%    \end{macrocode}
% \lverb|We are here only with a non-zero ##1, so no risk of a -0[0] which
% would be illegal usage of A[N] raw format. A negative ##1 is no trouble in ##3-##1.|
%    \begin{macrocode}
\def\XINT_tmpa#1.{%
\def\XINT_logten_f_Isp##1.000000\xint:##2[##3]%
{%
    {##1[0]}\xint:
    {\expandafter\XINT_LogTen_serII_a_ii
        \romannumeral0\XINTinfloatS[#1]{\xintAdd{##2[##3-##1]}{-1[0]}}%
     \xint:
    }\xint:
}%
}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax.%
\def\XINT_tmpa#1.{%
\def\XINT_logten_f_III##1\xint:##2[##3]%
{%
    {0[0]}\xint:
    {\expandafter\XINT_LogTen_serIII_a_ii
        \romannumeral0\XINTinfloatS[#1]{\xintAdd{##2[##3]}{-1[0]}}%
     \xint:
    }\xint:
}}\expandafter\XINT_tmpa\the\numexpr\XINTdigitsormax+4.%
\def\XINT_tmpa#1.#2.{%
\def\XINT_logten_f_IorII_pos##1.##2##3##4##5##6##7\xint:##8[##9]%
{%
    {\the\numexpr##1##2##3##4##5##6##7[-6]}\xint:
    {\expandafter\XINT_LogTen_serII_a_ii
     \romannumeral0\XINTinfloat[#2]%
     {\xintAdd{-1[0]}%
     {\xintMul{\csname XINT_c_1_##2_inv_x\endcsname}{%
      \XINTinFloat[#1]{%
       \xintMul{\csname XINT_c_2_##3_inv_x\endcsname}{%
        \XINTinFloat[#1]{%
         \xintMul{\csname XINT_c_3_##4_inv_x\endcsname}{%
          \XINTinFloat[#1]{%
           \xintMul{\csname XINT_c_4_##5_inv_x\endcsname}{%
            \XINTinFloat[#1]{%
             \xintMul{\csname XINT_c_5_##6_inv_x\endcsname}{%
              \XINTinFloat[#1]{%
               \xintMul{\csname XINT_c_6_##7_inv_x\endcsname}
                       {##8[##9-##1]}%
               }}}}}}}}}}%
     }%
     }\xint:
    }\xint:
}%
\def\XINT_logten_f_IorII_neg##1.##2##3##4##5##6##7\xint:##8[##9]%
{%
    {\the\numexpr##1##2##3##4##5##6##7[-6]}\xint:
    {\expandafter\XINT_LogTen_serII_a_ii
     \romannumeral0\XINTinfloat[#2]%
     {\xintAdd{-1[0]}%
     {\xintMul{\csname XINT_c_1_##2_x\endcsname}{%
      \XINTinFloat[#1]{%
       \xintMul{\csname XINT_c_2_##3_x\endcsname}{%
        \XINTinFloat[#1]{%
         \xintMul{\csname XINT_c_3_##4_x\endcsname}{%
          \XINTinFloat[#1]{%
           \xintMul{\csname XINT_c_4_##5_x\endcsname}{%
            \XINTinFloat[#1]{%
             \xintMul{\csname XINT_c_5_##6_x\endcsname}{%
              \XINTinFloat[#1]{%
               \xintMul{\csname XINT_c_6_##7_x\endcsname}
                       {##8[##9-##1]}%
               }}}}}}}}}}%
     }%
     }\xint:
    }\xint:
}%
}\expandafter\XINT_tmpa
 \the\numexpr\XINTdigitsormax+10\expandafter.\the\numexpr\XINTdigitsormax+4.%
%    \end{macrocode}
% \lverb|Initially all of this was done in a single big nested macro but the
% float-rounding of argument to less digits worked again each time from
% initial long input; the advantage on the other hand was that the 1/i
% constants were all pre-computed and rounded.
%
% Pre-coding the successive rounding to six digits less at each stage could be
% done via a single loop which would then walk back up inserting coeffs like
% 1/#1 having no special optimizing tricks. Pre-computing the 1/#1 too is
% possible but then one would have to copy the full set of such constants
% (which would be pre-computed depending on P), and this will add grabbing
% overhead in the loop expansion. Or one defines macros to hold the
% pre-rounded constants.
%
% Finally I do define macros, not only to hold the constants but to hold the
% whole build-up. Sacrificing brevity of code to benefit of expansion "speed".
%
% Firts one prepares eta, with P+4 digits for mantissa, and then hands it over
% to the log series. This will proceed via first preparing eta\xint: eta\xint:
% .... eta\xint:, the leftmost ones being more and more reduced in number of
% digits. Finally one goes back up to the right, the hard-coded number of
% steps depending on value of P=\XINTdigits at time of reloading of
% package. This number of steps is hard-coded in the number of macros which
% get defined.
%
% Descending (leftwards) chain: _a, Turning point: _b, Ascending: _c.
%
% As it is very easy to make silly typing mistakes in the numerous macros I
% have refactored a number of times the set-up to make manual verification
% straightforward. Automatization is possible but the _b macros complicate
% things, each one is its own special case.  In the end the set-up will define
% then redefine some _a and the (finally unique) _b macro, this allows easier
% to read code, with no nesting of conditionals or else branches.
%
% Actually series III and series II differ by only a shift by and we could use
% always the slightly more costly series III in place of series II. But that
% would add one un-needed term and a bit overhead to the default P which is
% 16...
%
% (1.4f: hesitation on 2021/05/09 after removal or case I log series should
% I not follow the simplifying logic and use always the slightly more costly III?)
% |
%
% \subsubsection{Log series, case II}
%    \begin{macrocode}
\def\XINT_tmpa#1.#2.{%
\def\XINT_LogTen_serII_a_ii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b#1[#2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_
    \romannumeral0\xintadd{1}{\xintiiOpp\xintHalf{#10}[#2-1]}\xint:
}%
\def\XINT_LogTen_serII_c_##1\xint:##2\xint:
{%
    \XINTinFloat[#2]{\xintMul{##1}{##2}}%
}%
}%
\expandafter\XINT_tmpa
            \the\numexpr\XINTdigitsormax-2\expandafter.%
            \the\numexpr\XINTdigitsormax+4.% 
\ifnum\XINTdigits>10
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_ii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_iii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_iii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_i
    \romannumeral0\xintadd{#3}{##1/3[##2]}\xint:
}%
\def\XINT_LogTen_serII_c_i##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-8\expandafter.%
  \the\numexpr\XINTdigitsormax-2.%
  {-5[-1]}.%
  {1[0]}.%
\fi
\ifnum\XINTdigits>16
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_iii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_iv
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_iv##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_ii
    \romannumeral0\xintadd{#3}{\xintiiMul{-25}{##1}[##2-2]}\xint:
}%
\def\XINT_LogTen_serII_c_ii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_i
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-14\expandafter.%
  \the\numexpr\XINTdigitsormax-8\expandafter.%
  \romannumeral0\XINTinfloat[\XINTdigitsormax-8]{1/3[0]}.%
  {-5[-1]}.%
\fi
\ifnum\XINTdigits>22
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_iv##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_v
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_v##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_iii
    \romannumeral0\xintadd{#3}{\xintDouble{##1}[##2-1]}\xint:
}%
\def\XINT_LogTen_serII_c_iii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_ii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-20\expandafter.%
  \the\numexpr\XINTdigitsormax-14\expandafter.\expanded{%
  {-25[-2]}.%
  \XINTinFloat[\XINTdigitsormax-8]{1/3[0]}.%
  }%
\fi
\ifnum\XINTdigits>28
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_v##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_vi
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_vi##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_iv
    \romannumeral0\xintadd{#3}{\xintiiOpp##1/6[##2]}\xint:
}%
\def\XINT_LogTen_serII_c_iv##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_iii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-26\expandafter.%
  \the\numexpr\XINTdigitsormax-20.%
  {2[-1]}.%
  {-25[-2]}.%
\fi
\ifnum\XINTdigits>34
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_vi##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_vii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_vii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_v
    \romannumeral0\xintadd{#3}{##1/7[##2]}\xint:
}%
\def\XINT_LogTen_serII_c_v##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_iv
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-32\expandafter.%
  \the\numexpr\XINTdigitsormax-26\expandafter.%
  \romannumeral0\XINTinfloatS[\XINTdigitsormax-26]{-1/6[0]}.%
  {2[-1]}.%
\fi
\ifnum\XINTdigits>40
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_vii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_viii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_viii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_vi
    \romannumeral0\xintadd{#3}{\xintiiMul{-125}{##1}[##2-3]}\xint:
}%
\def\XINT_LogTen_serII_c_vi##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_v
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-38\expandafter.%
  \the\numexpr\XINTdigitsormax-32\expandafter.\expanded{%
  \XINTinFloat[\XINTdigitsormax-32]{1/7[0]}.%
  \XINTinFloat[\XINTdigitsormax-26]{-1/6[0]}.%
  }%
\fi
\ifnum\XINTdigits>46
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_viii##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_ix
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_ix##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_vii
    \romannumeral0\xintadd{#3}{##1/9[##2]}\xint:
}%
\def\XINT_LogTen_serII_c_vii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_vi
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-44\expandafter.%
  \the\numexpr\XINTdigitsormax-38\expandafter.\expanded{%
  {-125[-3]}.%
  \XINTinFloat[\XINTdigitsormax-32]{1/7[0]}.%
  }%
\fi
\ifnum\XINTdigits>52
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_ix##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_x
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_x##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_viii
    \romannumeral0\xintadd{#3}{\xintiiOpp##1[##2-1]}\xint:
}%
\def\XINT_LogTen_serII_c_viii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_vii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-50\expandafter.%
  \the\numexpr\XINTdigitsormax-44\expandafter.%
  \romannumeral0\XINTinfloat[\XINTdigitsormax-44]{1/9[0]}.%
  {-125[-3]}.%
\fi
\ifnum\XINTdigits>58
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serII_a_x##1\xint:
{%
    \expandafter\XINT_LogTen_serII_a_xi
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_a_xi##1\xint:
{%
    \expandafter\XINT_LogTen_serII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serII_c_ix
    \romannumeral0\xintadd{#3}{##1/11[##2]}\xint:
}%
\def\XINT_LogTen_serII_c_ix##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serII_c_viii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-56\expandafter.%
  \the\numexpr\XINTdigitsormax-50\expandafter.\expanded{%
  {-1[-1]}.%
  \XINTinFloat[\XINTdigitsormax-44]{1/9[0]}.%
  }%
\fi
%    \end{macrocode}
% \subsubsection{Log series, case III}
%    \begin{macrocode}
\def\XINT_tmpa#1.#2.{%
\def\XINT_LogTen_serIII_a_ii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b#1[#2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_
    \romannumeral0\xintadd{1}{\xintiiOpp\xintHalf{#10}[#2-1]}\xint:
}%
\def\XINT_LogTen_serIII_c_##1\xint:##2\xint:
{%
    \XINTinFloat[#2]{\xintMul{##1}{##2}}%
}%
}%
\expandafter\XINT_tmpa
            \the\numexpr\XINTdigitsormax-1\expandafter.%
            \the\numexpr\XINTdigitsormax+4.% 
\ifnum\XINTdigits>9
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_ii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_iii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_iii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_i
    \romannumeral0\xintadd{#3}{##1/3[##2]}\xint:
}%
\def\XINT_LogTen_serIII_c_i##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-7\expandafter.%
  \the\numexpr\XINTdigitsormax-1.%
  {-5[-1]}.%
  {1[0]}.%
\fi
\ifnum\XINTdigits>15
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_iii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_iv
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_iv##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_ii
    \romannumeral0\xintadd{#3}{\xintiiMul{-25}{##1}[##2-2]}\xint:
}%
\def\XINT_LogTen_serIII_c_ii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_i
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-13\expandafter.%
  \the\numexpr\XINTdigitsormax-7\expandafter.%
  \romannumeral0\XINTinfloat[\XINTdigitsormax-7]{1/3[0]}.%
  {-5[-1]}.%
\fi
\ifnum\XINTdigits>21
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_iv##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_v
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_v##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_iii
    \romannumeral0\xintadd{#3}{\xintDouble{##1}[##2-1]}\xint:
}%
\def\XINT_LogTen_serIII_c_iii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_ii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-19\expandafter.%
  \the\numexpr\XINTdigitsormax-13\expandafter.\expanded{%
  {-25[-2]}.%
  \XINTinFloat[\XINTdigitsormax-7]{1/3[0]}.%
  }%
\fi
\ifnum\XINTdigits>27
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_v##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_vi
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_vi##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_iv
    \romannumeral0\xintadd{#3}{\xintiiOpp##1/6[##2]}\xint:
}%
\def\XINT_LogTen_serIII_c_iv##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_iii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-25\expandafter.%
  \the\numexpr\XINTdigitsormax-19.%
  {2[-1]}.%
  {-25[-2]}.%
\fi
\ifnum\XINTdigits>33
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_vi##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_vii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_vii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_v
    \romannumeral0\xintadd{#3}{##1/7[##2]}\xint:
}%
\def\XINT_LogTen_serIII_c_v##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_iv
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-31\expandafter.%
  \the\numexpr\XINTdigitsormax-25\expandafter.%
  \romannumeral0\XINTinfloatS[\XINTdigitsormax-25]{-1/6[0]}.%
  {2[-1]}.%
\fi
\ifnum\XINTdigits>39
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_vii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_viii
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_viii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_vi
    \romannumeral0\xintadd{#3}{\xintiiMul{-125}{##1}[##2-3]}\xint:
}%
\def\XINT_LogTen_serIII_c_vi##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_v
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-37\expandafter.%
  \the\numexpr\XINTdigitsormax-31\expandafter.\expanded{%
  \XINTinFloat[\XINTdigitsormax-31]{1/7[0]}.%
  \XINTinFloat[\XINTdigitsormax-25]{-1/6[0]}.%
  }%
\fi
\ifnum\XINTdigits>45
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_viii##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_ix
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_ix##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_vii
    \romannumeral0\xintadd{#3}{##1/9[##2]}\xint:
}%
\def\XINT_LogTen_serIII_c_vii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_vi
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-43\expandafter.%
  \the\numexpr\XINTdigitsormax-37\expandafter.\expanded{%
  {-125[-3]}.%
  \XINTinFloat[\XINTdigitsormax-31]{1/7[0]}.%
  }%
\fi
\ifnum\XINTdigits>51
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_ix##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_x
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_x##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_viii
    \romannumeral0\xintadd{#3}{\xintiiOpp##1[##2-1]}\xint:
}%
\def\XINT_LogTen_serIII_c_viii##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_vii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-49\expandafter.%
  \the\numexpr\XINTdigitsormax-43\expandafter.%
  \romannumeral0\XINTinfloat[\XINTdigitsormax-43]{1/9[0]}.%
  {-125[-3]}.%
\fi
\ifnum\XINTdigits>57
\def\XINT_tmpa#1.#2.#3.#4.{%
\def\XINT_LogTen_serIII_a_x##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_a_xi
    \romannumeral0\XINTinfloatS[#2]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_a_xi##1\xint:
{%
    \expandafter\XINT_LogTen_serIII_b
    \romannumeral0\XINTinfloatS[#1]{##1}\xint:##1\xint:
}%
\def\XINT_LogTen_serIII_b##1[##2]\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_ix
    \romannumeral0\xintadd{#3}{##1/11[##2]}\xint:
}%
\def\XINT_LogTen_serIII_c_ix##1\xint:##2\xint:
{%
    \expandafter\XINT_LogTen_serIII_c_viii
    \romannumeral0\xintadd{#4}{\XINTinFloat[#2]{\xintMul{##1}{##2}}}\xint:
}%
}\expandafter\XINT_tmpa
  \the\numexpr\XINTdigitsormax-55\expandafter.%
  \the\numexpr\XINTdigitsormax-49\expandafter.\expanded{%
  {-1[-1]}.%
  \XINTinFloat[\XINTdigitsormax-43]{1/9[0]}.%
  }%
\fi
\XINTendxintloginput%
%    \end{macrocode}
% \StoreCodelineNo {xintlog}
% \cleardoublepage\let\xintlognameUp\undefined
% \MakePercentComment
%</xintlog>------------------------------------------------------
%<*dtx>-----------------------------------------------------------
\iffalse
% grep -c -e "^{%" xint*sty
xint.sty:205
xintbinhex.sty:53
xintcfrac.sty:183
xintcore.sty:272
xintexpr.sty:453
xintfrac.sty:511
xintgcd.sty:41
xintkernel.sty:17
xintlog.sty:150
xintseries.sty:48
xinttools.sty:157
xinttrig.sty:65
\fi
% grep -o "^{%" xint*sty | wc -l
\def\totala{    2155}
\iffalse
% grep -c -e "^}%" xint*sty
xint.sty:204
xintbinhex.sty:52
xintcfrac.sty:183
xintcore.sty:269
xintexpr.sty:436
xintfrac.sty:515
xintgcd.sty:43
xintkernel.sty:17
xintlog.sty:151
xintseries.sty:48
xinttools.sty:156
xinttrig.sty:64
\fi
% grep -o "^}%" xint*sty | wc -l
\def\totalb{    2138}
\cleardoublepage
\section{Cumulative line count}

\def\mymacro #1{\mymacroaux #1}
\def\mymacroaux #1#2{\strut \csname #1nameimp\endcsname:& \dtt{ #2.}\tabularnewline }
\indent
\begin{tabular}[t]{r@{}r}
\xintApplyInline\mymacro\storedlinecounts
\end{tabular}
\def\mymacroaux #1#2{#2}%
%
\parbox[t]{10cm}{Total number of code lines:
    \dtt{\the\numexpr
    \xintListWithSep+{\xintApply\mymacro\storedlinecounts}\relax }.
    \ifdefined\totala
     (but \dtt{\the\numexpr \totala+\totalb\relax} lines among them 
      start either with \{\% or with \}\%.)\fi

    Each package starts with circa \dtt{50} lines dealing with catcodes,
    package identification and reloading management, also for Plain
    \TeX\strut. Version {\xintbndlversion} of {\xintbndldate}.\par
}

\CheckSum {38590}% 1.4j
% 38591 pour 1.4i, 38427 pour 1.4h
% 38423 pour 1.4g, 38212 pour 1.4f, 38813 pour 1.4e, 35184 pour 1.4d
% 35109 pour 1.4c, 35103 pour 1.4b, 34648 pour 1.4a, 34575 pour 1.4
% 33497 pour 1.3f, 33274 pour 1.3e, 31601 pour 1.3d, 31122 pour 1.3c
% 31069 pour 1.3b, 30482 pour 1.3a, 30621 pour 1.3, 30988 pour 1.2q,
% 30982 pour 1.2p, 30524 pour 1.2o, 30303 pour 1.2h, 30403 pour 1.2i,
% 30750 pour 1.2j, 30677 pour 1.2k, 30931 pour 1.2l, 30439 pour 1.2m,
% 30253 pour 1.2n
\makeatletter\check@checksum
\Finale
%% End of file xint.dtx