summaryrefslogtreecommitdiff
path: root/Master/texmf-dist/doc
diff options
context:
space:
mode:
authorKarl Berry <karl@freefriends.org>2011-01-14 00:18:29 +0000
committerKarl Berry <karl@freefriends.org>2011-01-14 00:18:29 +0000
commit848eb3c2564d5f661827124f0bd174ba0f702d26 (patch)
tree5cc7847a0d640b55d84173eb8d8a1a1d0cba1b1c /Master/texmf-dist/doc
parentafd6d32cfe467cfc3f9379c4130b63990c42ae37 (diff)
xits 1.008 (13jan11)
git-svn-id: svn://tug.org/texlive/trunk@21054 c570f23f-e606-0410-a88d-b1316a301751
Diffstat (limited to 'Master/texmf-dist/doc')
-rw-r--r--Master/texmf-dist/doc/fonts/xits/FONTLOG.txt8
-rw-r--r--Master/texmf-dist/doc/fonts/xits/README2
-rwxr-xr-xMaster/texmf-dist/doc/fonts/xits/build.py50
-rwxr-xr-xMaster/texmf-dist/doc/fonts/xits/fontcoverage.py254
4 files changed, 8 insertions, 306 deletions
diff --git a/Master/texmf-dist/doc/fonts/xits/FONTLOG.txt b/Master/texmf-dist/doc/fonts/xits/FONTLOG.txt
index 45b3487d043..e631191f1f4 100644
--- a/Master/texmf-dist/doc/fonts/xits/FONTLOG.txt
+++ b/Master/texmf-dist/doc/fonts/xits/FONTLOG.txt
@@ -192,11 +192,17 @@ http://github.com/khaledhosny/xits-math
ChangeLog
----------
+12 Jan 2011 (Khaled Hosny) <XITS> Version 1.008
+- Fix sub/supercript positioning for uppercase script family
+- Make \lgroup and \rgroup extensible
+- Mark few more brackets extended shapes
+- Make \lmoustache \rmoustache extensible
+
25 Oct 2010 (Khaled Hosny) <XITS> Version 1.007
- Add TopAccentHorizontal position to U+00B7, U+0131, U+0237, U+203E, U+22C5
- Add spaces and control characters U+0000-001F, U+2000-200F, U+2028-202F
- Tune script positioning for display integrals
-- Disable smaller that text size integrals
+- Disable smaller than text size integrals
- Rename fj ligature glyph to "f_j" for PDF search
- Assign un-encoded glyphs PUA code points
- Use U+23B7 slot for the glyph uni221A.base
diff --git a/Master/texmf-dist/doc/fonts/xits/README b/Master/texmf-dist/doc/fonts/xits/README
index 5d6d5dac8c1..0a121f07216 100644
--- a/Master/texmf-dist/doc/fonts/xits/README
+++ b/Master/texmf-dist/doc/fonts/xits/README
@@ -1,5 +1,5 @@
-The XTIS font project
+The XITS font project
======================
XITS is a Times-like typeface for mathematical and scientific publishing,
diff --git a/Master/texmf-dist/doc/fonts/xits/build.py b/Master/texmf-dist/doc/fonts/xits/build.py
deleted file mode 100755
index 036526e3e06..00000000000
--- a/Master/texmf-dist/doc/fonts/xits/build.py
+++ /dev/null
@@ -1,50 +0,0 @@
-#!/usr/bin/python
-
-import fontforge
-import tempfile
-import os
-import sys
-
-
-family = "xits"
-styles = ("math", "regular", "bold", "italic", "bolditalic")
-flags = ("opentype",)
-source = "sources"
-args = [ ]
-
-def doPUA(font):
- print "Assigning unencoded glyphs to PUA..."
- pua = 0xE000
- for glyph in font.glyphs():
- if glyph.unicode == -1 and glyph.glyphname != ".notdef":
- glyph.unicode = pua
- pua += 1
-
-if len(sys.argv) > 1:
- args = list(sys.argv[1:])
-
-for arg in args:
- if arg == "all":
- args = styles
- elif not arg in styles:
- print "Unknown style requested: %s" %arg
- args.remove(arg)
-
-if len(args) == 0:
- args = styles
-
-for style in args:
- print "Generating %s..." % style
- xits = fontforge.open(os.path.join(source, family+"-"+style+".sfd"))
- if style == "math":
- doPUA(xits)
- # XXX: if we don't save the font and reload it, the generated font
- # contian invalid code points (outside Unicode) instead of PUA
- tmpfont = tempfile.mkstemp()[1]
- xits.save(tmpfont)
- xits.close()
- xits = fontforge.open(tmpfont)
- os.remove(tmpfont)
- xits.mergeFeature (os.path.join(source, family+".fea"))
- xits.generate(family+"-"+style+".otf", flags=flags)
- xits.close()
diff --git a/Master/texmf-dist/doc/fonts/xits/fontcoverage.py b/Master/texmf-dist/doc/fonts/xits/fontcoverage.py
deleted file mode 100755
index c19fed75502..00000000000
--- a/Master/texmf-dist/doc/fonts/xits/fontcoverage.py
+++ /dev/null
@@ -1,254 +0,0 @@
-#!/usr/bin/env python
-from __future__ import division
-import sys
-import fontforge
-import unicodedata
-
-"""
-Unicode blocks list generated from:
- http://www.unicode.org/Public/5.2.0/ucd/Blocks.txt
-
-With the follwing code snippet:
-
- file = open("Blocks.txt", "r")
- blocks = [ ]
- for line in file.readlines():
- if not (line.startswith("#") or line == "\n"):
- start, end = line.split("; ")[0].split("..")
- name = line.split("; ")[1].strip()
- blocks.append((name, (start, end)))
-"""
-blocks = [
- ('Basic Latin', ('0000', '007F')),
- ('Latin-1 Supplement', ('0080', '00FF')),
- ('Latin Extended-A', ('0100', '017F')),
- ('Latin Extended-B', ('0180', '024F')),
- ('IPA Extensions', ('0250', '02AF')),
- ('Spacing Modifier Letters', ('02B0', '02FF')),
- ('Combining Diacritical Marks', ('0300', '036F')),
- ('Greek and Coptic', ('0370', '03FF')),
- ('Cyrillic', ('0400', '04FF')),
- ('Cyrillic Supplement', ('0500', '052F')),
- ('Armenian', ('0530', '058F')),
- ('Hebrew', ('0590', '05FF')),
- ('Arabic', ('0600', '06FF')),
- ('Syriac', ('0700', '074F')),
- ('Arabic Supplement', ('0750', '077F')),
- ('Thaana', ('0780', '07BF')),
- ('NKo', ('07C0', '07FF')),
- ('Samaritan', ('0800', '083F')),
- ('Devanagari', ('0900', '097F')),
- ('Bengali', ('0980', '09FF')),
- ('Gurmukhi', ('0A00', '0A7F')),
- ('Gujarati', ('0A80', '0AFF')),
- ('Oriya', ('0B00', '0B7F')),
- ('Tamil', ('0B80', '0BFF')),
- ('Telugu', ('0C00', '0C7F')),
- ('Kannada', ('0C80', '0CFF')),
- ('Malayalam', ('0D00', '0D7F')),
- ('Sinhala', ('0D80', '0DFF')),
- ('Thai', ('0E00', '0E7F')),
- ('Lao', ('0E80', '0EFF')),
- ('Tibetan', ('0F00', '0FFF')),
- ('Myanmar', ('1000', '109F')),
- ('Georgian', ('10A0', '10FF')),
- ('Hangul Jamo', ('1100', '11FF')),
- ('Ethiopic', ('1200', '137F')),
- ('Ethiopic Supplement', ('1380', '139F')),
- ('Cherokee', ('13A0', '13FF')),
- ('Unified Canadian Aboriginal Syllabics', ('1400', '167F')),
- ('Ogham', ('1680', '169F')),
- ('Runic', ('16A0', '16FF')),
- ('Tagalog', ('1700', '171F')),
- ('Hanunoo', ('1720', '173F')),
- ('Buhid', ('1740', '175F')),
- ('Tagbanwa', ('1760', '177F')),
- ('Khmer', ('1780', '17FF')),
- ('Mongolian', ('1800', '18AF')),
- ('Unified Canadian Aboriginal Syllabics Extended', ('18B0', '18FF')),
- ('Limbu', ('1900', '194F')),
- ('Tai Le', ('1950', '197F')),
- ('New Tai Lue', ('1980', '19DF')),
- ('Khmer Symbols', ('19E0', '19FF')),
- ('Buginese', ('1A00', '1A1F')),
- ('Tai Tham', ('1A20', '1AAF')),
- ('Balinese', ('1B00', '1B7F')),
- ('Sundanese', ('1B80', '1BBF')),
- ('Lepcha', ('1C00', '1C4F')),
- ('Ol Chiki', ('1C50', '1C7F')),
- ('Vedic Extensions', ('1CD0', '1CFF')),
- ('Phonetic Extensions', ('1D00', '1D7F')),
- ('Phonetic Extensions Supplement', ('1D80', '1DBF')),
- ('Combining Diacritical Marks Supplement', ('1DC0', '1DFF')),
- ('Latin Extended Additional', ('1E00', '1EFF')),
- ('Greek Extended', ('1F00', '1FFF')),
- ('General Punctuation', ('2000', '206F')),
- ('Superscripts and Subscripts', ('2070', '209F')),
- ('Currency Symbols', ('20A0', '20CF')),
- ('Combining Diacritical Marks for Symbols', ('20D0', '20FF')),
- ('Letterlike Symbols', ('2100', '214F')),
- ('Number Forms', ('2150', '218F')),
- ('Arrows', ('2190', '21FF')),
- ('Mathematical Operators', ('2200', '22FF')),
- ('Miscellaneous Technical', ('2300', '23FF')),
- ('Control Pictures', ('2400', '243F')),
- ('Optical Character Recognition', ('2440', '245F')),
- ('Enclosed Alphanumerics', ('2460', '24FF')),
- ('Box Drawing', ('2500', '257F')),
- ('Block Elements', ('2580', '259F')),
- ('Geometric Shapes', ('25A0', '25FF')),
- ('Miscellaneous Symbols', ('2600', '26FF')),
- ('Dingbats', ('2700', '27BF')),
- ('Miscellaneous Mathematical Symbols-A', ('27C0', '27EF')),
- ('Supplemental Arrows-A', ('27F0', '27FF')),
- ('Braille Patterns', ('2800', '28FF')),
- ('Supplemental Arrows-B', ('2900', '297F')),
- ('Miscellaneous Mathematical Symbols-B', ('2980', '29FF')),
- ('Supplemental Mathematical Operators', ('2A00', '2AFF')),
- ('Miscellaneous Symbols and Arrows', ('2B00', '2BFF')),
- ('Glagolitic', ('2C00', '2C5F')),
- ('Latin Extended-C', ('2C60', '2C7F')),
- ('Coptic', ('2C80', '2CFF')),
- ('Georgian Supplement', ('2D00', '2D2F')),
- ('Tifinagh', ('2D30', '2D7F')),
- ('Ethiopic Extended', ('2D80', '2DDF')),
- ('Cyrillic Extended-A', ('2DE0', '2DFF')),
- ('Supplemental Punctuation', ('2E00', '2E7F')),
- ('CJK Radicals Supplement', ('2E80', '2EFF')),
- ('Kangxi Radicals', ('2F00', '2FDF')),
- ('Ideographic Description Characters', ('2FF0', '2FFF')),
- ('CJK Symbols and Punctuation', ('3000', '303F')),
- ('Hiragana', ('3040', '309F')),
- ('Katakana', ('30A0', '30FF')),
- ('Bopomofo', ('3100', '312F')),
- ('Hangul Compatibility Jamo', ('3130', '318F')),
- ('Kanbun', ('3190', '319F')),
- ('Bopomofo Extended', ('31A0', '31BF')),
- ('CJK Strokes', ('31C0', '31EF')),
- ('Katakana Phonetic Extensions', ('31F0', '31FF')),
- ('Enclosed CJK Letters and Months', ('3200', '32FF')),
- ('CJK Compatibility', ('3300', '33FF')),
- ('CJK Unified Ideographs Extension A', ('3400', '4DBF')),
- ('Yijing Hexagram Symbols', ('4DC0', '4DFF')),
- ('CJK Unified Ideographs', ('4E00', '9FFF')),
- ('Yi Syllables', ('A000', 'A48F')),
- ('Yi Radicals', ('A490', 'A4CF')),
- ('Lisu', ('A4D0', 'A4FF')),
- ('Vai', ('A500', 'A63F')),
- ('Cyrillic Extended-B', ('A640', 'A69F')),
- ('Bamum', ('A6A0', 'A6FF')),
- ('Modifier Tone Letters', ('A700', 'A71F')),
- ('Latin Extended-D', ('A720', 'A7FF')),
- ('Syloti Nagri', ('A800', 'A82F')),
- ('Common Indic Number Forms', ('A830', 'A83F')),
- ('Phags-pa', ('A840', 'A87F')),
- ('Saurashtra', ('A880', 'A8DF')),
- ('Devanagari Extended', ('A8E0', 'A8FF')),
- ('Kayah Li', ('A900', 'A92F')),
- ('Rejang', ('A930', 'A95F')),
- ('Hangul Jamo Extended-A', ('A960', 'A97F')),
- ('Javanese', ('A980', 'A9DF')),
- ('Cham', ('AA00', 'AA5F')),
- ('Myanmar Extended-A', ('AA60', 'AA7F')),
- ('Tai Viet', ('AA80', 'AADF')),
- ('Meetei Mayek', ('ABC0', 'ABFF')),
- ('Hangul Syllables', ('AC00', 'D7AF')),
- ('Hangul Jamo Extended-B', ('D7B0', 'D7FF')),
- ('High Surrogates', ('D800', 'DB7F')),
- ('High Private Use Surrogates', ('DB80', 'DBFF')),
- ('Low Surrogates', ('DC00', 'DFFF')),
- ('Private Use Area', ('E000', 'F8FF')),
- ('CJK Compatibility Ideographs', ('F900', 'FAFF')),
- ('Alphabetic Presentation Forms', ('FB00', 'FB4F')),
- ('Arabic Presentation Forms-A', ('FB50', 'FDFF')),
- ('Variation Selectors', ('FE00', 'FE0F')),
- ('Vertical Forms', ('FE10', 'FE1F')),
- ('Combining Half Marks', ('FE20', 'FE2F')),
- ('CJK Compatibility Forms', ('FE30', 'FE4F')),
- ('Small Form Variants', ('FE50', 'FE6F')),
- ('Arabic Presentation Forms-B', ('FE70', 'FEFF')),
- ('Halfwidth and Fullwidth Forms', ('FF00', 'FFEF')),
- ('Specials', ('FFF0', 'FFFF')),
- ('Linear B Syllabary', ('10000', '1007F')),
- ('Linear B Ideograms', ('10080', '100FF')),
- ('Aegean Numbers', ('10100', '1013F')),
- ('Ancient Greek Numbers', ('10140', '1018F')),
- ('Ancient Symbols', ('10190', '101CF')),
- ('Phaistos Disc', ('101D0', '101FF')),
- ('Lycian', ('10280', '1029F')),
- ('Carian', ('102A0', '102DF')),
- ('Old Italic', ('10300', '1032F')),
- ('Gothic', ('10330', '1034F')),
- ('Ugaritic', ('10380', '1039F')),
- ('Old Persian', ('103A0', '103DF')),
- ('Deseret', ('10400', '1044F')),
- ('Shavian', ('10450', '1047F')),
- ('Osmanya', ('10480', '104AF')),
- ('Cypriot Syllabary', ('10800', '1083F')),
- ('Imperial Aramaic', ('10840', '1085F')),
- ('Phoenician', ('10900', '1091F')),
- ('Lydian', ('10920', '1093F')),
- ('Kharoshthi', ('10A00', '10A5F')),
- ('Old South Arabian', ('10A60', '10A7F')),
- ('Avestan', ('10B00', '10B3F')),
- ('Inscriptional Parthian', ('10B40', '10B5F')),
- ('Inscriptional Pahlavi', ('10B60', '10B7F')),
- ('Old Turkic', ('10C00', '10C4F')),
- ('Rumi Numeral Symbols', ('10E60', '10E7F')),
- ('Kaithi', ('11080', '110CF')),
- ('Cuneiform', ('12000', '123FF')),
- ('Cuneiform Numbers and Punctuation', ('12400', '1247F')),
- ('Egyptian Hieroglyphs', ('13000', '1342F')),
- ('Byzantine Musical Symbols', ('1D000', '1D0FF')),
- ('Musical Symbols', ('1D100', '1D1FF')),
- ('Ancient Greek Musical Notation', ('1D200', '1D24F')),
- ('Tai Xuan Jing Symbols', ('1D300', '1D35F')),
- ('Counting Rod Numerals', ('1D360', '1D37F')),
- ('Mathematical Alphanumeric Symbols', ('1D400', '1D7FF')),
- ('Mahjong Tiles', ('1F000', '1F02F')),
- ('Domino Tiles', ('1F030', '1F09F')),
- ('Enclosed Alphanumeric Supplement', ('1F100', '1F1FF')),
- ('Enclosed Ideographic Supplement', ('1F200', '1F2FF')),
- ('CJK Unified Ideographs Extension B', ('20000', '2A6DF')),
- ('CJK Unified Ideographs Extension C', ('2A700', '2B73F')),
- ('CJK Compatibility Ideographs Supplement', ('2F800', '2FA1F')),
- ('Tags', ('E0000', 'E007F')),
- ('Variation Selectors Supplement', ('E0100', 'E01EF')),
- ('Supplementary Private Use Area-A', ('F0000', 'FFFFF')),
- ('Supplementary Private Use Area-B', ('100000', '10FFFF'))
-]
-
-font = fontforge.open(sys.argv[1])
-font.encoding = "UnicodeFull"
-found = [ ]
-
-for block in blocks:
- t = f = 0
- name = block[0]
- start, end = int(block[1][0], 16), int(block[1][1], 16)
- i = start
- while (i <= end):
- category = unicodedata.category(unichr(i))
- if category != "Cc" and category!= "Cn":
- try:
- glyph = font[i]
- f += 1
- except TypeError:
- pass
- t += 1
- i += 1
- if f:
- found.append((name, (t,f)))
-
-
-print "%s:" %font.fullname
-for f in found:
- for b in blocks:
- if b[0] == f[0]:
- name = f[0]
- start, end = b[1]
- total, present = f[1]
- percent = present/total*100
- print "%s (U+%s-%s): %s/%s (%.2f%%)" %(
- name, start, end, present, total, percent)
-