/* This is dvipdfmx, an eXtended version of dvipdfm by Mark A. Wicks. Copyright (C) 2002-2014 by Jin-Hwan Cho and Shunsaku Hirata, the dvipdfmx project team. Copyright (C) 1998, 1999 by Mark A. Wicks This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA. */ #ifdef HAVE_CONFIG_H #include #endif #include #include #include "system.h" #include "mem.h" #include "error.h" #include "numbers.h" #include "pdfobj.h" #include "pdffont.h" #include "pdfencoding.h" #include "unicode.h" #include "dpxutil.h" #include "pst_obj.h" #include "pst.h" #include "cff_limits.h" #include "cff_types.h" #include "cff_dict.h" #include "cff.h" #include "t1_char.h" /* * Type 1 -> Type 1C */ /* Charstring decoder/encoder status codes */ #define CS_OP_NOSUPPORT -4 #define CS_BUFFER_ERROR -3 #define CS_STACK_ERROR -2 #define CS_PARSE_ERROR -1 #define CS_PARSE_OK 0 #define CS_PARSE_END 1 #define CS_SUBR_RETURN 2 #define CS_CHAR_END 3 static int status = CS_PARSE_ERROR; #define DST_NEED(a,b) {if ((a) < (b)) { status = CS_BUFFER_ERROR ; return ; }} #define SRC_NEED(a,b) {if ((a) < (b)) { status = CS_PARSE_ERROR ; return ; }} #define NEED(a,b) {if ((a) < (b)) { status = CS_STACK_ERROR ; return ; }} #define T1_CS_PHASE_INIT 0 #define T1_CS_PHASE_HINT 1 #define T1_CS_PHASE_PATH 2 #define T1_CS_PHASE_FLEX 3 static int phase = -1; static int nest = -1; #ifndef CS_STEM_ZONE_MAX #define CS_STEM_ZONE_MAX 96 #endif #ifndef CS_STEM_GROUP_MAX #define CS_STEM_GROUP_MAX CS_STEM_ZONE_MAX #endif #ifndef CS_ARG_STACK_MAX #define CS_ARG_STACK_MAX 48 #endif #ifndef PS_ARG_STACK_MAX /* * Counter control may have CS_STEM_ZONE_MAX*2+2 arguments. */ #define PS_ARG_STACK_MAX (CS_STEM_ZONE_MAX*2+2) #endif typedef struct t1_cpath { int type; int num_args; double args[CS_ARG_STACK_MAX]; struct t1_cpath *next; } t1_cpath; #define HSTEM 0 #define VSTEM 1 typedef struct { int id; int dir; double pos, del; } t1_stem; typedef struct { int num_stems; double stems[CS_STEM_ZONE_MAX]; /* int */ } t1_stemgroup; #define T1_CS_FLAG_NONE 0 #define T1_CS_FLAG_USE_HINTMASK (1 << 0) #define T1_CS_FLAG_USE_CNTRMASK (1 << 1) #define T1_CS_FLAG_USE_SEAC (1 << 2) typedef struct { int flags; struct { double sbx, sby, wx, wy; } sbw; struct { double llx, lly, urx, ury; } bbox; struct { double asb, adx, ady; card8 bchar, achar; } seac; int num_stems; t1_stem stems[CS_STEM_ZONE_MAX]; t1_cpath *charpath; t1_cpath *lastpath; } t1_chardesc; static int cs_stack_top = 0; static int ps_stack_top = 0; /* [vh]stem support require one more stack size. */ static double cs_arg_stack[CS_ARG_STACK_MAX+1]; static double ps_arg_stack[PS_ARG_STACK_MAX]; #define CS_HINT_DECL -1 #define CS_FLEX_CTRL -2 #define CS_CNTR_CTRL -3 /* * Type 1/2 CharString encoding */ /* * 1-byte CharString operaotrs: * cs_escape is first byte of two-byte operator */ /* RESERVED 0 */ #define cs_hstem 1 /* RESERVED 2 */ #define cs_vstem 3 #define cs_vmoveto 4 #define cs_rlineto 5 #define cs_hlineto 6 #define cs_vlineto 7 #define cs_rrcurveto 8 #define cs_closepath 9 #define cs_callsubr 10 #define cs_return 11 #define cs_escape 12 #define cs_hsbw 13 #define cs_endchar 14 /* RESERVED 15 */ /* RESERVED 16 */ /* RESERVED 17 */ #define cs_hstemhm 18 #define cs_hintmask 19 #define cs_cntrmask 20 #define cs_rmoveto 21 #define cs_hmoveto 22 #define cs_vstemhm 23 #define cs_rcurveline 24 #define cs_rlinecurve 25 #define cs_vvcurveto 26 #define cs_hhcurveto 27 /* SHORTINT 28 : first byte of shortint*/ #define cs_callgsubr 29 #define cs_vhcurveto 30 #define cs_hvcurveto 31 /* 2-byte op. flex 34-37 used. */ /* * 2-byte CharString operaotrs: * "dotsection" is obsoleted in Type 2 charstring. */ #define cs_dotsection 0 #define cs_vstem3 1 #define cs_hstem3 2 #define cs_and 3 #define cs_or 4 #define cs_not 5 #define cs_seac 6 #define cs_sbw 7 /* RESERVED 8 */ #define cs_abs 9 #define cs_add 10 #define cs_sub 11 #define cs_div 12 /* RESERVED 13 */ #define cs_neg 14 #define cs_eq 15 #define cs_callothersubr 16 #define cs_pop 17 #define cs_drop 18 /* RESERVED 19 */ #define cs_put 20 #define cs_get 21 #define cs_ifelse 22 #define cs_random 23 #define cs_mul 24 /* RESERVED 25 */ #define cs_sqrt 26 #define cs_dup 27 #define cs_exch 28 #define cs_index 29 #define cs_roll 30 /* RESERVED 31 */ /* RESERVED 32 */ #define cs_setcurrentpoint 33 #define cs_hflex 34 #define cs_flex 35 #define cs_hflex1 36 #define cs_flex1 37 #define IS_PATH_OPERATOR(o) (((o) >= cs_vmoveto && (o) <= cs_closepath) || \ ((o) >= cs_rmoveto && (o) <= cs_hvcurveto && \ (o) != cs_vstemhm && (o) != cs_callgsubr && (o) != 28)\ ) /* * Stem: * * 1. Stems must be sorted in the increasing bottom/left edge order. * 2. The encoded values are all relative; The value x(y) of the first * stem is relative to 0 for Type 2 charstring and is relative to * the left(bottom) side-bearing for Type 1 charstring. * 3. A width of -20(-21) specifies the top/right(bottom/left) edge * of an edge hint in Type 2 charstring. But the width of 'ghost' * hint in Type 1 charstring is positive with value 20 or 21. * 4. The h(v)stemhm MUST be used instead of h(v)stem if charstring * contains hintmask operator. * * TODO: * * Convert ghost hint to edge hint, Counter control for hstem3/vstem3. */ static int CDECL stem_compare (const void *v1, const void *v2) { int cmp = 0; const t1_stem *s1, *s2; s1 = (const t1_stem *) v1; s2 = (const t1_stem *) v2; if (s1->dir == s2->dir) { if (s1->pos == s2->pos) { if (s1->del == s2->del) cmp = 0; else cmp = (s1->del < s2->del) ? -1 : 1; } else { cmp = (s1->pos < s2->pos) ? -1 : 1; } } else { cmp = (s1->dir == HSTEM) ? -1 : 1; } return cmp; } #define SORT_STEMS(cd) qsort((cd)->stems,(cd)->num_stems,sizeof(t1_stem),stem_compare) static int get_stem (t1_chardesc *cd, int stem_id) { int i; for (i = 0; i < cd->num_stems; i++) { if (cd->stems[i].id == stem_id) break; } return ((i < cd->num_stems) ? i : -1); } static int add_stem (t1_chardesc *cd, double pos, double del, int dir) { int i; ASSERT(cd); pos += (dir == HSTEM) ? cd->sbw.sby : cd->sbw.sbx; for (i = 0; i < cd->num_stems; i++) { if (cd->stems[i].dir == dir && cd->stems[i].pos == pos && cd->stems[i].del == del) break; } if (i == cd->num_stems) { if (cd->num_stems == CS_STEM_ZONE_MAX) return -1; cd->stems[i].dir = dir; cd->stems[i].pos = pos; cd->stems[i].del = del; cd->stems[i].id = cd->num_stems; (cd->num_stems)++; } return cd->stems[i].id; } static void copy_args (double *args1, double *args2, int count) { while (count-- > 0) { *args1 = *args2; args1++; args2++; } } /* * Stack: */ #define LIMITCHECK(n) do {\ if (cs_stack_top+(n) > CS_ARG_STACK_MAX) {\ status = CS_STACK_ERROR;\ return;\ }\ } while (0) #define CHECKSTACK(n) do {\ if (cs_stack_top < (n)) {\ status = CS_STACK_ERROR;\ return;\ }\ } while (0) #define CLEARSTACK() do {\ cs_stack_top = 0;\ } while (0) /* * Path construction: */ /* Get operands from cs_arg_stack[] */ static void add_charpath (t1_chardesc *cd, int type, double *argv, int argn) { t1_cpath *p; ASSERT(cd); ASSERT(argn <= CS_ARG_STACK_MAX); p = NEW(1, t1_cpath); p->type = type; p->num_args = argn; p->next = NULL; while (argn-- > 0) p->args[argn] = argv[argn]; if (!cd->charpath) cd->charpath = p; if (cd->lastpath) cd->lastpath->next = p; cd->lastpath = p; if (type >= 0 && phase != T1_CS_PHASE_FLEX && IS_PATH_OPERATOR(type)) phase = T1_CS_PHASE_PATH; } static void init_charpath (t1_chardesc *cd) { cd->flags = T1_CS_FLAG_NONE; cd->num_stems = 0; cd->sbw.wx = cd->sbw.wy = 0.0; cd->sbw.sbx = cd->sbw.sby = 0.0; cd->bbox.llx = cd->bbox.lly = cd->bbox.urx = cd->bbox.ury = 0.0; cd->charpath = cd->lastpath = NULL; } static void release_charpath (t1_chardesc *cd) { t1_cpath *curr, *next; ASSERT(cd); curr = cd->charpath; while (curr != NULL) { next = curr->next; RELEASE(curr); curr = next; } cd->charpath = cd->lastpath = NULL; } /* * Type 1 charstring operators: */ #define ADD_PATH(p,t,n) add_charpath((p),(t),&(cs_arg_stack[cs_stack_top-(n)]),(n)) /* * Single byte operators: */ static void do_operator1 (t1_chardesc *cd, card8 **data, card8 *endptr) { card8 op = **data; *data += 1; switch (op) { case cs_closepath: /* * From T1 spec.: * Note that, unlike the closepath command in the PostScript language, * this command does not reposition the current point. Any subsequent * rmoveto must be relative to the current point in force before the * Type 1 font format closepath command was given. */ /* noop */ CLEARSTACK(); break; case cs_hsbw: CHECKSTACK(2); cd->sbw.wx = cs_arg_stack[--cs_stack_top]; cd->sbw.wy = 0; cd->sbw.sbx = cs_arg_stack[--cs_stack_top]; cd->sbw.sby = 0; CLEARSTACK(); /* hsbw does NOT set currentpoint. */ break; case cs_hstem: case cs_vstem: CHECKSTACK(2); { int stem_id; stem_id = add_stem(cd, cs_arg_stack[cs_stack_top-2], cs_arg_stack[cs_stack_top-1], ((op == cs_hstem) ? HSTEM : VSTEM)); if (stem_id < 0) { WARN("Too many hints..."); status = CS_PARSE_ERROR; return; } /* Put stem_id onto the stack... */ cs_arg_stack[cs_stack_top++] = stem_id; ADD_PATH(cd, CS_HINT_DECL, 1); } CLEARSTACK(); break; case cs_rmoveto: /* * Reference point is (0, 0) in Type 2 charstring. */ CHECKSTACK(2); { if (phase < T1_CS_PHASE_PATH) { cs_arg_stack[cs_stack_top-2] += cd->sbw.sbx; cs_arg_stack[cs_stack_top-1] += cd->sbw.sby; } ADD_PATH(cd, op, 2); } CLEARSTACK(); break; case cs_hmoveto: case cs_vmoveto: CHECKSTACK(1); { int argn = 1; if (phase < T1_CS_PHASE_PATH) { /* * The reference point for the first moveto operator is diferrent * between Type 1 charstring and Type 2 charstring. We compensate it. */ if (op == cs_hmoveto) { cs_arg_stack[cs_stack_top-1] += cd->sbw.sbx; if (cd->sbw.sby != 0.0) { cs_arg_stack[cs_stack_top++] = cd->sbw.sby; argn = 2; op = cs_rmoveto; } } else { cs_arg_stack[cs_stack_top-1] += cd->sbw.sby; if (cd->sbw.sbx != 0.0) { cs_arg_stack[cs_stack_top] = cs_arg_stack[cs_stack_top-1]; cs_arg_stack[cs_stack_top-1] = cd->sbw.sbx; cs_stack_top++; argn = 2; op = cs_rmoveto; } } } ADD_PATH(cd, op, argn); } CLEARSTACK(); break; case cs_endchar: status = CS_CHAR_END; CLEARSTACK(); break; /* above oprators are candidate for first stack-clearing operator */ case cs_rlineto: CHECKSTACK(2); ADD_PATH(cd, op, 2); CLEARSTACK(); break; case cs_hlineto: case cs_vlineto: CHECKSTACK(1); ADD_PATH(cd, op, 1); CLEARSTACK(); break; case cs_rrcurveto: CHECKSTACK(6); ADD_PATH(cd, op, 6); CLEARSTACK(); break; case cs_vhcurveto: case cs_hvcurveto: CHECKSTACK(4); ADD_PATH(cd, op, 4); CLEARSTACK(); break; /* all operotors above are stack-clearing operator */ /* no output */ case cs_return: break; case cs_callsubr: ERROR("Unexpected callsubr."); break; default: /* no-op ? */ WARN("Unknown charstring operator: 0x%02x", op); status = CS_PARSE_ERROR; break; } return; } /* * OtherSubrs: * * arg0 arg1 ... argn n othersubr# callothersubr * * 0: Build flex: * fd x y 3 0 callothersubr * Othersubr #0 pushes x and y to PostScript interpreter operand stack. * 1: Start flex * 0 1 callothersubr * 2: Mark flex control points * 0 2 callothersubr * 3: Discard hint * subr# 1 3 callothersubr pop callsubr * Subroutine subr# (only) contains stem declaration. * Othersubr #3 pushes subr# to PostScript interpreter operand stack. * 12: Counter control * A subr to avoid stack overflow. * 13: Counter control */ /* * Convert six control points marked as CS_FLEX_CTRL to a flex path. */ static void do_othersubr0 (t1_chardesc *cd) { t1_cpath *flex, *cur, *next; if (ps_stack_top < 1) { status = CS_PARSE_ERROR; return; } /* Seek first CS_FLEX_CTRL mark */ for (cur = cd->charpath; cur != NULL && cur->type != CS_FLEX_CTRL; cur = cur->next); flex = cur; { int i; cur = cur->next; for (i = 1; i < 7; i++) { if (cur == NULL || cur->type != CS_FLEX_CTRL || cur->num_args != 2) { status = CS_PARSE_ERROR; return; } if (i == 1) { flex->args[0] += cur->args[0]; flex->args[1] += cur->args[1]; } else { copy_args(&(flex->args[2*i-2]), cur->args, 2); } next = cur->next; RELEASE(cur); cur = next; } } if (cur != NULL) { status = CS_PARSE_ERROR; return; } /* * Now 'flex' have all six control points, the first pair is relative * from starting point. */ flex->type = cs_flex; flex->args[12] = ps_arg_stack[--ps_stack_top]; /* flex depth */ flex->num_args = 13; flex->next = NULL; cd->lastpath = flex; phase = T1_CS_PHASE_PATH; } /* Start flex */ static void do_othersubr1 (t1_chardesc *cd) { phase = T1_CS_PHASE_FLEX; } /* Mark flex control point */ static void do_othersubr2 (t1_chardesc *cd) { if (phase != T1_CS_PHASE_FLEX || !cd->lastpath) { status = CS_PARSE_ERROR; return; } switch (cd->lastpath->type) { case cs_rmoveto: break; case cs_hmoveto: cd->lastpath->num_args = 2; cd->lastpath->args[1] = 0.0; break; case cs_vmoveto: cd->lastpath->num_args = 2; cd->lastpath->args[1] = cd->lastpath->args[0]; cd->lastpath->args[0] = 0.0; break; default: status = CS_PARSE_ERROR; return; } cd->lastpath->type = CS_FLEX_CTRL; } /* * Hint Replacement: * "Adobe Type 1 Font Format", Chapter 8. */ static void do_othersubr3 (t1_chardesc *cd) { cd->flags |= T1_CS_FLAG_USE_HINTMASK; } static void do_othersubr12 (t1_chardesc *cd) { /* Othersubr12 call must immediately follow the hsbw or sbw. */ if (phase != T1_CS_PHASE_INIT) { status = CS_PARSE_ERROR; return; } /* noop */ } static void do_othersubr13 (t1_chardesc *cd) { t1_stemgroup stemgroups[CS_STEM_GROUP_MAX]; int num_hgroups, num_vgroups, n, stem_id; double pos, del; /* After #12 callothersubr or hsbw or sbw. */ if (phase != T1_CS_PHASE_INIT) { status = CS_PARSE_ERROR; return; } for (n = 0; n < CS_STEM_GROUP_MAX; n++) { stemgroups[n].num_stems = 0; } num_hgroups = (int) ps_arg_stack[--ps_stack_top]; if (num_hgroups < 0 || num_hgroups > CS_STEM_GROUP_MAX) { status = CS_PARSE_ERROR; return; } n = 0; pos = 0.0; while (ps_stack_top >= 2 && n < num_hgroups) { /* add_stem() add sidebearing */ pos += ps_arg_stack[--ps_stack_top]; del = ps_arg_stack[--ps_stack_top]; stem_id = add_stem(cd, (del < 0.0) ? pos + del : pos, (del < 0.0) ? -del : del, HSTEM); stemgroups[n].stems[stemgroups[n].num_stems] = stem_id; stemgroups[n].num_stems += 1; pos += del; if (del < 0.0) { pos = 0.0; n++; } } if (n != num_hgroups) { status = CS_STACK_ERROR; return; } num_vgroups = (int) ps_arg_stack[--ps_stack_top]; if (num_vgroups < 0 || num_vgroups > CS_STEM_GROUP_MAX) { status = CS_PARSE_ERROR; return; } n = 0; pos = 0.0; while (ps_stack_top >= 2 && n < num_vgroups) { /* add_stem() add sidebearing */ pos += ps_arg_stack[--ps_stack_top]; del = ps_arg_stack[--ps_stack_top]; stem_id = add_stem(cd, (del < 0.0) ? pos + del : pos, (del < 0.0) ? -del : del, VSTEM); stemgroups[n].stems[stemgroups[n].num_stems] = stem_id; stemgroups[n].num_stems += 1; pos += del; if (del < 0.0) { pos = 0.0; n++; } } if (n != num_vgroups) { status = CS_STACK_ERROR; return; } for (n = 0; n < MAX(num_hgroups, num_vgroups); n++) { add_charpath(cd, cs_cntrmask, stemgroups[n].stems, stemgroups[n].num_stems); } cd->flags |= T1_CS_FLAG_USE_CNTRMASK; } static void do_callothersubr (t1_chardesc *cd) { int argn, subrno; CHECKSTACK(2); subrno = (int) cs_arg_stack[--cs_stack_top]; argn = (int) cs_arg_stack[--cs_stack_top]; CHECKSTACK(argn); if (ps_stack_top+argn > PS_ARG_STACK_MAX) { status = CS_PARSE_ERROR; return; } while (argn-- > 0) ps_arg_stack[ps_stack_top++] = cs_arg_stack[--cs_stack_top]; switch (subrno) { case 0: do_othersubr0(cd) ; break; case 1: do_othersubr1(cd) ; break; case 2: do_othersubr2(cd) ; break; case 3: do_othersubr3(cd) ; break; case 12: do_othersubr12(cd); break; case 13: do_othersubr13(cd); break; default: ERROR("Unknown othersubr #%ld.", subrno); break; } } /* * Double byte operators: */ static void do_operator2 (t1_chardesc *cd, card8 **data, card8 *endptr) { card8 op; *data += 1; SRC_NEED(endptr, *data + 1); op = **data; *data += 1; switch(op) { case cs_sbw: CHECKSTACK(4); cd->sbw.wy = cs_arg_stack[--cs_stack_top]; cd->sbw.wx = cs_arg_stack[--cs_stack_top]; cd->sbw.sby = cs_arg_stack[--cs_stack_top]; cd->sbw.sbx = cs_arg_stack[--cs_stack_top]; CLEARSTACK(); break; case cs_hstem3: case cs_vstem3: /* * TODO: * The counter control can be used for hstem3 and vstem3 * operator if LanguageGroup is not equal to 1. */ CHECKSTACK(6); { int i; for (i = 2; i >= 0; i--) { int stem_id; stem_id = add_stem(cd, cs_arg_stack[cs_stack_top-2*i-2], cs_arg_stack[cs_stack_top-2*i-1], ((op == cs_hstem3) ? HSTEM : VSTEM)); if (stem_id < 0) { WARN("Too many hints..."); status = CS_PARSE_ERROR; return; } /* Put stem_id onto the stack... */ cs_arg_stack[cs_stack_top++] = stem_id; ADD_PATH(cd, CS_HINT_DECL, 1); cs_stack_top--; } } CLEARSTACK(); break; case cs_setcurrentpoint: CHECKSTACK(2); /* noop */ CLEARSTACK(); break; /* all operator above are stack-clearing */ case cs_pop: /* * Transfer a operand from PS interpreter operand stack to BuildChar * operand stack. */ if (ps_stack_top < 1) { status = CS_PARSE_ERROR; return; } LIMITCHECK(1); cs_arg_stack[cs_stack_top++] = ps_arg_stack[--ps_stack_top]; break; case cs_dotsection: #if 0 /* * If the hint replacement feature is used in the font, the * "dotsection" operator exist only for compatibility to older * (more than 10 years old) Type 1 font rasterizer which can't * perform hint replacement. In this case, we silently ignore * the "dotsection" operator. * * The following code will wrongly warn about "dotsection" when * the charstring only contains dot (e.g., "bullet") where the * hint replacement is not necessary. * * Adobe ATM renderers always treat this operator as a no-op. * (See, Adobe Technical Note #5177, Appendix C) */ if (!(cd->flags & T1_CS_FLAG_USE_HINTMASK)) { if (__verbose > 1) WARN("Obsolete Type 1 charstring operator \"dotsection\" not supported."); } #endif /* noop */ break; case cs_div: /* TODO: check overflow */ CHECKSTACK(2); cs_arg_stack[cs_stack_top-2] /= cs_arg_stack[cs_stack_top-1]; cs_stack_top--; break; case cs_callothersubr: do_callothersubr(cd); break; case cs_seac: CHECKSTACK(5); cd->flags |= T1_CS_FLAG_USE_SEAC; cd->seac.achar = (card8) cs_arg_stack[--cs_stack_top]; cd->seac.bchar = (card8) cs_arg_stack[--cs_stack_top]; cd->seac.ady = cs_arg_stack[--cs_stack_top]; cd->seac.adx = cs_arg_stack[--cs_stack_top]; /* We must compensate the difference of the glyph origin. */ cd->seac.ady += cd->sbw.sby; cd->seac.adx += cd->sbw.sbx - cs_arg_stack[--cs_stack_top]; CLEARSTACK(); break; default: /* no-op ? */ WARN("Unknown charstring operator: 0x0c%02x", op); status = CS_PARSE_ERROR; break; } return; } /* * Charstring encoding: * Copied from cs_type2.c * Note: * The Type 2 interpretation of a number encoded in five-bytes (those with * an initial byte value of 255) differs from how it is interpreted in the * Type 1 format. */ /* Type 2 5-bytes encoding used. */ static void put_numbers (double *argv, int argn, card8 **dest, card8 *limit) { int i; for (i = 0; i < argn; i++) { double value; long ivalue; value = argv[i]; /* Nearest integer value */ ivalue = (long) floor(value+0.5); if (value >= 0x8000L || value <= (-0x8000L - 1)) { /* * This number cannot be represented as a single operand. * We must use `a b mul ...' or `a c div' to represent large values. */ ERROR("Argument value too large. (This is bug)"); } else if (fabs(value - ivalue) > 3.0e-5) { /* 16.16-bit signed fixed value */ DST_NEED(limit, *dest + 5); *(*dest)++ = 255; ivalue = (long) floor(value); /* mantissa */ *(*dest)++ = (ivalue >> 8) & 0xff; *(*dest)++ = ivalue & 0xff; ivalue = (long)((value - ivalue) * 0x10000l); /* fraction */ *(*dest)++ = (ivalue >> 8) & 0xff; *(*dest)++ = ivalue & 0xff; /* Everything else are integers. */ } else if (ivalue >= -107 && ivalue <= 107) { DST_NEED(limit, *dest + 1); *(*dest)++ = ivalue + 139; } else if (ivalue >= 108 && ivalue <= 1131) { DST_NEED(limit, *dest + 2); ivalue = 0xf700u + ivalue - 108; *(*dest)++ = (ivalue >> 8) & 0xff; *(*dest)++ = ivalue & 0xff; } else if (ivalue >= -1131 && ivalue <= -108) { DST_NEED(limit, *dest + 2); ivalue = 0xfb00u - ivalue - 108; *(*dest)++ = (ivalue >> 8) & 0xff; *(*dest)++ = ivalue & 0xff; } else if (ivalue >= -32768 && ivalue <= 32767) { /* shortint */ DST_NEED(limit, *dest + 3); *(*dest)++ = 28; *(*dest)++ = (ivalue >> 8) & 0xff; *(*dest)++ = (ivalue) & 0xff; } else { /* Shouldn't come here */ ERROR("Unexpected error."); } } return; } static void get_integer (card8 **data, card8 *endptr) { long result = 0; card8 b0 = **data, b1, b2; *data += 1; if (b0 == 28) { /* shortint */ SRC_NEED(endptr, *data + 2); b1 = **data; b2 = *(*data+1); result = b1*256+b2; if (result > 0x7fff) result -= 0x10000L; *data += 2; } else if (b0 >= 32 && b0 <= 246) { /* int (1) */ result = b0 - 139; } else if (b0 >= 247 && b0 <= 250) { /* int (2) */ SRC_NEED(endptr, *data + 1); b1 = **data; result = (b0-247)*256+b1+108; *data += 1; } else if (b0 >= 251 && b0 <= 254) { SRC_NEED(endptr, *data + 1); b1 = **data; result = -(b0-251)*256-b1-108; *data += 1; } else { status = CS_PARSE_ERROR; return; } LIMITCHECK(1); cs_arg_stack[cs_stack_top++] = (double) result; return; } /* Type 1 */ static void get_longint (card8 **data, card8 *endptr) { long result = 0; int i; *data += 1; SRC_NEED(endptr, *data + 4); result = **data; if (result >= 0x80L) result -= 0x100L; *data += 1; for (i = 1; i < 4; i++) { result = result*256 + (**data); *data += 1; } LIMITCHECK(1); cs_arg_stack[cs_stack_top++] = (double) result; return; } /* * TODO: * Check "seac" * We cannot do backword parsing due to subroutine, div etc. */ /* Parse charstring and build charpath. */ static void t1char_build_charpath (t1_chardesc *cd, card8 **data, card8 *endptr, cff_index *subrs) { card8 b0 = 0, *subr; long len; if (nest > CS_SUBR_NEST_MAX) ERROR("Subroutine nested too deeply."); nest++; while (*data < endptr && status == CS_PARSE_OK) { b0 = **data; if (b0 == 255) { get_longint(data, endptr); /* Type 1 */ } else if (b0 == cs_return) { status = CS_SUBR_RETURN; } else if (b0 == cs_callsubr) { if (cs_stack_top < 1) { status = CS_STACK_ERROR; } else { int idx; idx = cs_arg_stack[--cs_stack_top]; if (!subrs || idx >= subrs->count) ERROR("Invalid Subr#."); subr = subrs->data + subrs->offset[idx] - 1; len = subrs->offset[idx+1] - subrs->offset[idx]; t1char_build_charpath(cd, &subr, subr+len, subrs); *data += 1; } } else if (b0 == cs_escape) { do_operator2(cd, data, endptr); } else if (b0 < 32 && b0 != 28) { /* 19, 20 need mask */ do_operator1(cd, data, endptr); } else if ((b0 <= 22 && b0 >= 27) || b0 == 31) { /* reserved */ status = CS_PARSE_ERROR; /* not an error ? */ } else { /* integer */ get_integer(data, endptr); } } if (status == CS_SUBR_RETURN) { status = CS_PARSE_OK; } else if (status == CS_CHAR_END && *data < endptr) { if (!(*data == endptr - 1 && **data == cs_return)) WARN("Garbage after endchar. (%ld bytes)", (long) (endptr - *data)); } else if (status < CS_PARSE_OK) { /* error */ ERROR("Parsing charstring failed: (status=%d, stack=%d)", status, cs_stack_top); } nest--; return; } /* * Calculate BoundingBox and compress path. * The essentials of PDF size reduction is not Type 2 charstring compression * but Type 1 charstring encryption. Encryption makes lossless compression * useless. We will only do very simple charstring compression. */ static void do_postproc (t1_chardesc *cd) { int i; t1_cpath *cur, *prev, *next; double x, y; if (!cd->charpath) return; /* Set dummy large value. */ cd->bbox.llx = cd->bbox.lly = 100000.0; cd->bbox.urx = cd->bbox.ury = -100000.0; cur = cd->charpath; prev = NULL; x = y = 0.0; #define UPDATE_BBOX(b,x,y) do {\ if ((b).llx > (x)) (b).llx = (x);\ if ((b).urx < (x)) (b).urx = (x);\ if ((b).lly > (y)) (b).lly = (y);\ if ((b).ury < (y)) (b).ury = (y);\ } while (0) #define TRY_COMPACT (prev && cur && ((prev->num_args + cur->num_args) <= CS_ARG_STACK_MAX)) while (cur != NULL) { next = cur->next; switch (cur->type) { case cs_rmoveto: x += cur->args[0]; y += cur->args[1]; UPDATE_BBOX(cd->bbox, x, y); break; case cs_rlineto: x += cur->args[0]; y += cur->args[1]; UPDATE_BBOX(cd->bbox, x, y); if (TRY_COMPACT) { if (prev->type == cs_rlineto) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->next = next; RELEASE(cur); cur = NULL; } else if (prev->type == cs_rrcurveto) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->type = cs_rcurveline; prev->next = next; RELEASE(cur); cur = NULL; } } break; case cs_hmoveto: x += cur->args[0]; UPDATE_BBOX(cd->bbox, x, y); break; case cs_hlineto: x += cur->args[0]; UPDATE_BBOX(cd->bbox, x, y); if (TRY_COMPACT) { if ((prev->type == cs_vlineto && (prev->num_args % 2) == 1) || (prev->type == cs_hlineto && (prev->num_args % 2) == 0)) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->next = next; RELEASE(cur); cur = NULL; } } break; case cs_vmoveto: y += cur->args[0]; UPDATE_BBOX(cd->bbox, x, y); break; case cs_vlineto: y += cur->args[0]; UPDATE_BBOX(cd->bbox, x, y); if (TRY_COMPACT) { if ((prev->type == cs_hlineto && (prev->num_args % 2) == 1) || (prev->type == cs_vlineto && (prev->num_args % 2) == 0)) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->next = next; RELEASE(cur); cur = NULL; } } break; case cs_rrcurveto: for (i = 0; i < 3; i++) { x += cur->args[2*i]; y += cur->args[2*i+1]; UPDATE_BBOX(cd->bbox, x, y); } if (TRY_COMPACT) { if (prev->type == cs_rrcurveto) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->next = next; RELEASE(cur); cur = NULL; } else if (prev->type == cs_rlineto) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->type = cs_rlinecurve; prev->next = next; RELEASE(cur); cur = NULL; } } break; case cs_vhcurveto: y += cur->args[0]; UPDATE_BBOX(cd->bbox, x, y); x += cur->args[1]; y += cur->args[2]; UPDATE_BBOX(cd->bbox, x, y); x += cur->args[3]; UPDATE_BBOX(cd->bbox, x, y); if (TRY_COMPACT) { if ((prev->type == cs_hvcurveto && ((prev->num_args / 4) % 2) == 1) || (prev->type == cs_vhcurveto && ((prev->num_args / 4) % 2) == 0)) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->next = next; RELEASE(cur); cur = NULL; } } break; case cs_hvcurveto: x += cur->args[0]; UPDATE_BBOX(cd->bbox, x, y); x += cur->args[1]; y += cur->args[2]; UPDATE_BBOX(cd->bbox, x, y); y += cur->args[3]; UPDATE_BBOX(cd->bbox, x, y); if (TRY_COMPACT) { if ((prev->type == cs_vhcurveto && ((prev->num_args / 4) % 2) == 1) || (prev->type == cs_hvcurveto && ((prev->num_args / 4) % 2) == 0)) { copy_args(prev->args+prev->num_args, cur->args, cur->num_args); prev->num_args += cur->num_args; prev->next = next; RELEASE(cur); cur = NULL; } } break; case cs_flex: for (i = 0; i < 6; i++) { x += cur->args[2*i]; y += cur->args[2*1+1]; UPDATE_BBOX(cd->bbox, x, y); } if (cur->args[12] == 50.0) { if (cur->args[1] == 0.0 && cur->args[11] == 0.0 && cur->args[5] == 0.0 && cur->args[7] == 0.0 && cur->args[3] + cur->args[9] == 0.0) { /* cur->args[0] = cur->args[0]; dx1 */ cur->args[1] = cur->args[2]; /* dx2 */ cur->args[2] = cur->args[3]; /* dy2 */ cur->args[3] = cur->args[4]; /* dx3 */ cur->args[4] = cur->args[6]; /* dx4 */ cur->args[5] = cur->args[8]; /* dx5 */ cur->args[6] = cur->args[10]; /* dx6 */ cur->num_args = 7; cur->type = cs_hflex; } else if (cur->args[5] == 0.0 && cur->args[7] == 0.0 && (cur->args[1] + cur->args[3] + cur->args[9] + cur->args[11]) == 0) { /* cur->args[0] = cur->args[0]; dx1 */ /* cur->args[1] = cur->args[1]; dy1 */ /* cur->args[2] = cur->args[2]; dx2 */ /* cur->args[3] = cur->args[3]; dy2 */ /* cur->args[4] = cur->args[4]; dx3 */ cur->args[5] = cur->args[6]; /* dx4 */ cur->args[6] = cur->args[8]; /* dx5 */ cur->args[7] = cur->args[9]; /* dy5 */ cur->args[8] = cur->args[10]; /* dx6 */ cur->num_args = 9; cur->type = cs_hflex1; } } break; case CS_HINT_DECL: case cs_cntrmask: /* noop */ break; default: ERROR("Unexpected Type 2 charstring command %d.", cur->type); break; } if (cur != NULL) prev = cur; cur = next; } /* Had no path. Fix lower-left point. */ if (cd->bbox.llx > cd->bbox.urx) cd->bbox.llx = cd->bbox.urx = cd->sbw.wx; if (cd->bbox.lly > cd->bbox.ury) cd->bbox.lly = cd->bbox.ury = cd->sbw.wy; return; } #define RESET_STATE() do {\ status = CS_PARSE_OK;\ phase = T1_CS_PHASE_INIT;\ nest = 0;\ ps_stack_top = 0;\ } while (0) int t1char_get_metrics (card8 *src, long srclen, cff_index *subrs, t1_ginfo *ginfo) { t1_chardesc t1char, *cd; cd = &t1char; init_charpath(cd); RESET_STATE(); CLEARSTACK(); t1char_build_charpath(cd, &src, src+srclen, subrs); if (cs_stack_top != 0 || ps_stack_top != 0) WARN("Stack not empty. (%d, %d)", cs_stack_top, ps_stack_top); do_postproc(cd); if (ginfo) { ginfo->wx = cd->sbw.wx; ginfo->wy = cd->sbw.wy; ginfo->bbox.llx = cd->bbox.llx; ginfo->bbox.lly = cd->bbox.lly; ginfo->bbox.urx = cd->bbox.urx; ginfo->bbox.ury = cd->bbox.ury; if (cd->flags & T1_CS_FLAG_USE_SEAC) { ginfo->use_seac = 1; ginfo->seac.adx = cd->seac.adx; ginfo->seac.ady = cd->seac.ady; ginfo->seac.bchar = cd->seac.bchar; ginfo->seac.achar = cd->seac.achar; } else { ginfo->use_seac = 0; } } release_charpath(cd); return 0; } #define CHECK_BUFFER(n) if (dst+(n) >= endptr) {\ ERROR("Buffer overflow.");\ } #define CHECK_STATUS() if (status != CS_PARSE_OK) {\ ERROR("Charstring encoder error: %d", status);\ } /* * Encode Charpath as a Type 2 Charstring */ static long t1char_encode_charpath (t1_chardesc *cd, double default_width, double nominal_width, card8 *dst, card8 *endptr) { card8 *save; t1_cpath *curr; ASSERT(cd); save = dst; curr = cd->charpath; RESET_STATE(); CLEARSTACK(); /* * Advance Width */ if (cd->sbw.wx != default_width) { double wx = cd->sbw.wx - nominal_width; put_numbers(&wx, 1, &dst, endptr); CHECK_STATUS(); } /* * Hint Declaration */ { int num_hstems = 0, num_vstems = 0; int i, reset = 1; double stem[2]; for (i = 0; i < cd->num_stems && cd->stems[i].dir == HSTEM; i++) { num_hstems++; stem[0] = (reset ? (cd->stems[i].pos) : (cd->stems[i].pos - (cd->stems[i-1].pos + cd->stems[i-1].del))); stem[1] = cd->stems[i].del; put_numbers(stem, 2, &dst, endptr); CHECK_STATUS(); reset = 0; if (2*num_hstems == CS_ARG_STACK_MAX) { CHECK_BUFFER(1); *dst++ = (card8) ((cd->flags & T1_CS_FLAG_USE_HINTMASK) ? cs_hstemhm : cs_hstem); reset = 1; } } if (reset == 0) { CHECK_BUFFER(1); *dst++ = (card8) ((cd->flags & T1_CS_FLAG_USE_HINTMASK) ? cs_hstemhm : cs_hstem); } reset = 1; if (cd->num_stems - num_hstems > 0) { for (i = num_hstems; i < cd->num_stems; i++) { num_vstems++; stem[0] = (reset ? (cd->stems[i].pos) : (cd->stems[i].pos - (cd->stems[i-1].pos + cd->stems[i-1].del))); stem[1] = cd->stems[i].del; put_numbers(stem, 2, &dst, endptr); CHECK_STATUS(); reset = 0; if (2*num_vstems == CS_ARG_STACK_MAX) { CHECK_BUFFER(1); *dst++ = (card8) ((cd->flags & T1_CS_FLAG_USE_HINTMASK) ? cs_vstemhm : cs_vstem); reset = 1; } } if (reset == 0) { CHECK_BUFFER(1); if ((cd->flags & T1_CS_FLAG_USE_HINTMASK) || (cd->flags & T1_CS_FLAG_USE_CNTRMASK)) { /* * The vstem hint operator can be ommited if hstem and vstem hints * are both declared at the beginning of a charstring, and is * followed directly by the hintmask or cntrmask operators. */ if (curr->type != CS_HINT_DECL && curr->type != cs_cntrmask) { *dst++ = (card8) cs_vstemhm; } } else { *dst++ = (card8) cs_vstem; } } } } /* * Path Construction and Hint Replacement */ while (curr != NULL && curr->type != cs_endchar) { switch (curr->type) { case CS_HINT_DECL: { card8 hintmask[(CS_STEM_ZONE_MAX+7)/8]; memset(hintmask, 0, (cd->num_stems+7)/8); while (curr != NULL && curr->type == CS_HINT_DECL) { int stem_idx; stem_idx = get_stem(cd, (int) curr->args[0]); ASSERT(stem_idx < cd->num_stems); hintmask[stem_idx/8] |= (1 << (7 - (stem_idx % 8))); curr = curr->next; } if (cd->flags & T1_CS_FLAG_USE_HINTMASK) { CHECK_BUFFER((cd->num_stems+7)/8 + 1); *dst++ = (card8) cs_hintmask; memcpy(dst, hintmask, (cd->num_stems+7)/8); dst += (cd->num_stems+7)/8; } } break; case cs_cntrmask: { card8 cntrmask[(CS_STEM_ZONE_MAX+7)/8]; int stem_idx, i; memset(cntrmask, 0, (cd->num_stems+7)/8); for (i = 0; i < curr->num_args; i++) { stem_idx = get_stem(cd, (int) curr->args[i]); ASSERT(stem_idx < cd->num_stems); cntrmask[stem_idx/8] |= (1 << (7 - (stem_idx % 8))); } CHECK_BUFFER((cd->num_stems+7)/8 + 1); *dst++ = (card8) cs_cntrmask; memcpy(dst, cntrmask, (cd->num_stems+7)/8); dst += (cd->num_stems+7)/8; curr = curr->next; } break; case cs_rmoveto: case cs_hmoveto: case cs_vmoveto: case cs_rlineto: case cs_hlineto: case cs_vlineto: case cs_rrcurveto: case cs_hvcurveto: case cs_vhcurveto: case cs_rlinecurve: case cs_rcurveline: { put_numbers(curr->args, curr->num_args, &dst, endptr); CHECK_STATUS(); CHECK_BUFFER(1); *dst++ = (card8) curr->type; curr = curr->next; } break; case cs_flex: case cs_hflex: case cs_hflex1: { put_numbers(curr->args, curr->num_args, &dst, endptr); CHECK_STATUS(); CHECK_BUFFER(2); *dst++ = (card8) cs_escape; *dst++ = (card8) curr->type; curr = curr->next; } break; default: ERROR("Unknown Type 2 charstring command: %d", curr->type); break; } } /* * (adx ady bchar achar) endchar */ if (cd->flags & T1_CS_FLAG_USE_SEAC) { double seac[4]; seac[0] = cd->seac.adx; seac[1] = cd->seac.ady; seac[2] = cd->seac.bchar; seac[3] = cd->seac.achar; put_numbers(seac, 4, &dst, endptr); CHECK_STATUS(); CHECK_BUFFER(2); WARN("Obsolete four arguments of \"endchar\" will be used for Type 1 \"seac\" operator."); } CHECK_BUFFER(1); *dst++ = (card8) cs_endchar; return (long) (dst - save); } long t1char_convert_charstring (card8 *dst, long dstlen, card8 *src, long srclen, cff_index *subrs, double default_width, double nominal_width, t1_ginfo *ginfo) { long length; t1_chardesc t1char, *cd; cd = &t1char; init_charpath(cd); RESET_STATE(); CLEARSTACK(); t1char_build_charpath(cd, &src, src+srclen, subrs); if (cs_stack_top != 0 || ps_stack_top != 0) WARN("Stack not empty. (%d, %d)", cs_stack_top, ps_stack_top); do_postproc(cd); SORT_STEMS(cd); length = t1char_encode_charpath(cd, default_width, nominal_width, dst, dst+dstlen); if (ginfo) { ginfo->wx = cd->sbw.wx; ginfo->wy = cd->sbw.wy; ginfo->bbox.llx = cd->bbox.llx; ginfo->bbox.lly = cd->bbox.lly; ginfo->bbox.urx = cd->bbox.urx; ginfo->bbox.ury = cd->bbox.ury; if (cd->flags & T1_CS_FLAG_USE_SEAC) { ginfo->use_seac = 1; ginfo->seac.adx = cd->seac.adx; ginfo->seac.ady = cd->seac.ady; ginfo->seac.bchar = cd->seac.bchar; ginfo->seac.achar = cd->seac.achar; } else { ginfo->use_seac = 0; } } release_charpath(cd); return length; }