You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
7045 lines
244 KiB
7045 lines
244 KiB
/* -*- Mode: Java; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */ |
|
/* vim: set shiftwidth=2 tabstop=2 autoindent cindent expandtab: */ |
|
/* Copyright 2012 Mozilla Foundation |
|
* |
|
* Licensed under the Apache License, Version 2.0 (the "License"); |
|
* you may not use this file except in compliance with the License. |
|
* You may obtain a copy of the License at |
|
* |
|
* http://www.apache.org/licenses/LICENSE-2.0 |
|
* |
|
* Unless required by applicable law or agreed to in writing, software |
|
* distributed under the License is distributed on an "AS IS" BASIS, |
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
|
* See the License for the specific language governing permissions and |
|
* limitations under the License. |
|
*/ |
|
/* globals assert, bytesToString, CIDToUnicodeMaps, error, ExpertCharset, |
|
ExpertSubsetCharset, FileReaderSync, globalScope, GlyphsUnicode, |
|
info, isArray, isNum, ISOAdobeCharset, isWorker, PDFJS, Stream, |
|
stringToBytes, TextDecoder, TODO, warn */ |
|
|
|
'use strict'; |
|
|
|
// Unicode Private Use Area |
|
var CMAP_GLYPH_OFFSET = 0xE000; |
|
var GLYPH_AREA_SIZE = 0x1900; |
|
var SYMBOLIC_FONT_GLYPH_OFFSET = 0xF000; |
|
|
|
// PDF Glyph Space Units are one Thousandth of a TextSpace Unit |
|
// except for Type 3 fonts |
|
var PDF_GLYPH_SPACE_UNITS = 1000; |
|
|
|
// Hinting is currently disabled due to unknown problems on windows |
|
// in tracemonkey and various other pdfs with type1 fonts. |
|
var HINTING_ENABLED = false; |
|
|
|
// Accented charactars are not displayed properly on windows, using this flag |
|
// to control analysis of seac charstrings. |
|
var SEAC_ANALYSIS_ENABLED = false; |
|
|
|
var FONT_IDENTITY_MATRIX = [0.001, 0, 0, 0.001, 0, 0]; |
|
|
|
var FontFlags = { |
|
FixedPitch: 1, |
|
Serif: 2, |
|
Symbolic: 4, |
|
Script: 8, |
|
Nonsymbolic: 32, |
|
Italic: 64, |
|
AllCap: 65536, |
|
SmallCap: 131072, |
|
ForceBold: 262144 |
|
}; |
|
|
|
var Encodings = { |
|
ExpertEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'exclamsmall', 'Hungarumlautsmall', '', 'dollaroldstyle', |
|
'dollarsuperior', 'ampersandsmall', 'Acutesmall', 'parenleftsuperior', |
|
'parenrightsuperior', 'twodotenleader', 'onedotenleader', 'comma', |
|
'hyphen', 'period', 'fraction', 'zerooldstyle', 'oneoldstyle', |
|
'twooldstyle', 'threeoldstyle', 'fouroldstyle', 'fiveoldstyle', |
|
'sixoldstyle', 'sevenoldstyle', 'eightoldstyle', 'nineoldstyle', 'colon', |
|
'semicolon', 'commasuperior', 'threequartersemdash', 'periodsuperior', |
|
'questionsmall', '', 'asuperior', 'bsuperior', 'centsuperior', 'dsuperior', |
|
'esuperior', '', '', 'isuperior', '', '', 'lsuperior', 'msuperior', |
|
'nsuperior', 'osuperior', '', '', 'rsuperior', 'ssuperior', 'tsuperior', |
|
'', 'ff', 'fi', 'fl', 'ffi', 'ffl', 'parenleftinferior', '', |
|
'parenrightinferior', 'Circumflexsmall', 'hyphensuperior', 'Gravesmall', |
|
'Asmall', 'Bsmall', 'Csmall', 'Dsmall', 'Esmall', 'Fsmall', 'Gsmall', |
|
'Hsmall', 'Ismall', 'Jsmall', 'Ksmall', 'Lsmall', 'Msmall', 'Nsmall', |
|
'Osmall', 'Psmall', 'Qsmall', 'Rsmall', 'Ssmall', 'Tsmall', 'Usmall', |
|
'Vsmall', 'Wsmall', 'Xsmall', 'Ysmall', 'Zsmall', 'colonmonetary', |
|
'onefitted', 'rupiah', 'Tildesmall', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', 'exclamdownsmall', 'centoldstyle', 'Lslashsmall', |
|
'', '', 'Scaronsmall', 'Zcaronsmall', 'Dieresissmall', 'Brevesmall', |
|
'Caronsmall', '', 'Dotaccentsmall', '', '', 'Macronsmall', '', '', |
|
'figuredash', 'hypheninferior', '', '', 'Ogoneksmall', 'Ringsmall', |
|
'Cedillasmall', '', '', '', 'onequarter', 'onehalf', 'threequarters', |
|
'questiondownsmall', 'oneeighth', 'threeeighths', 'fiveeighths', |
|
'seveneighths', 'onethird', 'twothirds', '', '', 'zerosuperior', |
|
'onesuperior', 'twosuperior', 'threesuperior', 'foursuperior', |
|
'fivesuperior', 'sixsuperior', 'sevensuperior', 'eightsuperior', |
|
'ninesuperior', 'zeroinferior', 'oneinferior', 'twoinferior', |
|
'threeinferior', 'fourinferior', 'fiveinferior', 'sixinferior', |
|
'seveninferior', 'eightinferior', 'nineinferior', 'centinferior', |
|
'dollarinferior', 'periodinferior', 'commainferior', 'Agravesmall', |
|
'Aacutesmall', 'Acircumflexsmall', 'Atildesmall', 'Adieresissmall', |
|
'Aringsmall', 'AEsmall', 'Ccedillasmall', 'Egravesmall', 'Eacutesmall', |
|
'Ecircumflexsmall', 'Edieresissmall', 'Igravesmall', 'Iacutesmall', |
|
'Icircumflexsmall', 'Idieresissmall', 'Ethsmall', 'Ntildesmall', |
|
'Ogravesmall', 'Oacutesmall', 'Ocircumflexsmall', 'Otildesmall', |
|
'Odieresissmall', 'OEsmall', 'Oslashsmall', 'Ugravesmall', 'Uacutesmall', |
|
'Ucircumflexsmall', 'Udieresissmall', 'Yacutesmall', 'Thornsmall', |
|
'Ydieresissmall'], |
|
MacExpertEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'exclamsmall', 'Hungarumlautsmall', 'centoldstyle', |
|
'dollaroldstyle', 'dollarsuperior', 'ampersandsmall', 'Acutesmall', |
|
'parenleftsuperior', 'parenrightsuperior', 'twodotenleader', |
|
'onedotenleader', 'comma', 'hyphen', 'period', 'fraction', 'zerooldstyle', |
|
'oneoldstyle', 'twooldstyle', 'threeoldstyle', 'fouroldstyle', |
|
'fiveoldstyle', 'sixoldstyle', 'sevenoldstyle', 'eightoldstyle', |
|
'nineoldstyle', 'colon', 'semicolon', '', 'threequartersemdash', '', |
|
'questionsmall', '', '', '', '', 'Ethsmall', '', '', 'onequarter', |
|
'onehalf', 'threequarters', 'oneeighth', 'threeeighths', 'fiveeighths', |
|
'seveneighths', 'onethird', 'twothirds', '', '', '', '', '', '', 'ff', |
|
'fi', 'fl', 'ffi', 'ffl', 'parenleftinferior', '', 'parenrightinferior', |
|
'Circumflexsmall', 'hypheninferior', 'Gravesmall', 'Asmall', 'Bsmall', |
|
'Csmall', 'Dsmall', 'Esmall', 'Fsmall', 'Gsmall', 'Hsmall', 'Ismall', |
|
'Jsmall', 'Ksmall', 'Lsmall', 'Msmall', 'Nsmall', 'Osmall', 'Psmall', |
|
'Qsmall', 'Rsmall', 'Ssmall', 'Tsmall', 'Usmall', 'Vsmall', 'Wsmall', |
|
'Xsmall', 'Ysmall', 'Zsmall', 'colonmonetary', 'onefitted', 'rupiah', |
|
'Tildesmall', '', '', 'asuperior', 'centsuperior', '', '', '', '', |
|
'Aacutesmall', 'Agravesmall', 'Acircumflexsmall', 'Adieresissmall', |
|
'Atildesmall', 'Aringsmall', 'Ccedillasmall', 'Eacutesmall', 'Egravesmall', |
|
'Ecircumflexsmall', 'Edieresissmall', 'Iacutesmall', 'Igravesmall', |
|
'Icircumflexsmall', 'Idieresissmall', 'Ntildesmall', 'Oacutesmall', |
|
'Ogravesmall', 'Ocircumflexsmall', 'Odieresissmall', 'Otildesmall', |
|
'Uacutesmall', 'Ugravesmall', 'Ucircumflexsmall', 'Udieresissmall', '', |
|
'eightsuperior', 'fourinferior', 'threeinferior', 'sixinferior', |
|
'eightinferior', 'seveninferior', 'Scaronsmall', '', 'centinferior', |
|
'twoinferior', '', 'Dieresissmall', '', 'Caronsmall', 'osuperior', |
|
'fiveinferior', '', 'commainferior', 'periodinferior', 'Yacutesmall', '', |
|
'dollarinferior', '', 'Thornsmall', '', 'nineinferior', 'zeroinferior', |
|
'Zcaronsmall', 'AEsmall', 'Oslashsmall', 'questiondownsmall', |
|
'oneinferior', 'Lslashsmall', '', '', '', '', '', '', 'Cedillasmall', '', |
|
'', '', '', '', 'OEsmall', 'figuredash', 'hyphensuperior', '', '', '', '', |
|
'exclamdownsmall', '', 'Ydieresissmall', '', 'onesuperior', 'twosuperior', |
|
'threesuperior', 'foursuperior', 'fivesuperior', 'sixsuperior', |
|
'sevensuperior', 'ninesuperior', 'zerosuperior', '', 'esuperior', |
|
'rsuperior', 'tsuperior', '', '', 'isuperior', 'ssuperior', 'dsuperior', |
|
'', '', '', '', '', 'lsuperior', 'Ogoneksmall', 'Brevesmall', |
|
'Macronsmall', 'bsuperior', 'nsuperior', 'msuperior', 'commasuperior', |
|
'periodsuperior', 'Dotaccentsmall', 'Ringsmall'], |
|
MacRomanEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'exclam', 'quotedbl', 'numbersign', 'dollar', 'percent', |
|
'ampersand', 'quotesingle', 'parenleft', 'parenright', 'asterisk', 'plus', |
|
'comma', 'hyphen', 'period', 'slash', 'zero', 'one', 'two', 'three', |
|
'four', 'five', 'six', 'seven', 'eight', 'nine', 'colon', 'semicolon', |
|
'less', 'equal', 'greater', 'question', 'at', 'A', 'B', 'C', 'D', 'E', 'F', |
|
'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', |
|
'V', 'W', 'X', 'Y', 'Z', 'bracketleft', 'backslash', 'bracketright', |
|
'asciicircum', 'underscore', 'grave', 'a', 'b', 'c', 'd', 'e', 'f', 'g', |
|
'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', |
|
'w', 'x', 'y', 'z', 'braceleft', 'bar', 'braceright', 'asciitilde', '', |
|
'Adieresis', 'Aring', 'Ccedilla', 'Eacute', 'Ntilde', 'Odieresis', |
|
'Udieresis', 'aacute', 'agrave', 'acircumflex', 'adieresis', 'atilde', |
|
'aring', 'ccedilla', 'eacute', 'egrave', 'ecircumflex', 'edieresis', |
|
'iacute', 'igrave', 'icircumflex', 'idieresis', 'ntilde', 'oacute', |
|
'ograve', 'ocircumflex', 'odieresis', 'otilde', 'uacute', 'ugrave', |
|
'ucircumflex', 'udieresis', 'dagger', 'degree', 'cent', 'sterling', |
|
'section', 'bullet', 'paragraph', 'germandbls', 'registered', 'copyright', |
|
'trademark', 'acute', 'dieresis', 'notequal', 'AE', 'Oslash', 'infinity', |
|
'plusminus', 'lessequal', 'greaterequal', 'yen', 'mu', 'partialdiff', |
|
'summation', 'product', 'pi', 'integral', 'ordfeminine', 'ordmasculine', |
|
'Omega', 'ae', 'oslash', 'questiondown', 'exclamdown', 'logicalnot', |
|
'radical', 'florin', 'approxequal', 'Delta', 'guillemotleft', |
|
'guillemotright', 'ellipsis', '', 'Agrave', 'Atilde', 'Otilde', 'OE', |
|
'oe', 'endash', 'emdash', 'quotedblleft', 'quotedblright', 'quoteleft', |
|
'quoteright', 'divide', 'lozenge', 'ydieresis', 'Ydieresis', 'fraction', |
|
'currency', 'guilsinglleft', 'guilsinglright', 'fi', 'fl', 'daggerdbl', |
|
'periodcentered', 'quotesinglbase', 'quotedblbase', 'perthousand', |
|
'Acircumflex', 'Ecircumflex', 'Aacute', 'Edieresis', 'Egrave', 'Iacute', |
|
'Icircumflex', 'Idieresis', 'Igrave', 'Oacute', 'Ocircumflex', 'apple', |
|
'Ograve', 'Uacute', 'Ucircumflex', 'Ugrave', 'dotlessi', 'circumflex', |
|
'tilde', 'macron', 'breve', 'dotaccent', 'ring', 'cedilla', 'hungarumlaut', |
|
'ogonek', 'caron'], |
|
StandardEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'exclam', 'quotedbl', 'numbersign', 'dollar', 'percent', |
|
'ampersand', 'quoteright', 'parenleft', 'parenright', 'asterisk', 'plus', |
|
'comma', 'hyphen', 'period', 'slash', 'zero', 'one', 'two', 'three', |
|
'four', 'five', 'six', 'seven', 'eight', 'nine', 'colon', 'semicolon', |
|
'less', 'equal', 'greater', 'question', 'at', 'A', 'B', 'C', 'D', 'E', 'F', |
|
'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', |
|
'V', 'W', 'X', 'Y', 'Z', 'bracketleft', 'backslash', 'bracketright', |
|
'asciicircum', 'underscore', 'quoteleft', 'a', 'b', 'c', 'd', 'e', 'f', |
|
'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', |
|
'v', 'w', 'x', 'y', 'z', 'braceleft', 'bar', 'braceright', 'asciitilde', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', 'exclamdown', |
|
'cent', 'sterling', 'fraction', 'yen', 'florin', 'section', 'currency', |
|
'quotesingle', 'quotedblleft', 'guillemotleft', 'guilsinglleft', |
|
'guilsinglright', 'fi', 'fl', '', 'endash', 'dagger', 'daggerdbl', |
|
'periodcentered', '', 'paragraph', 'bullet', 'quotesinglbase', |
|
'quotedblbase', 'quotedblright', 'guillemotright', 'ellipsis', |
|
'perthousand', '', 'questiondown', '', 'grave', 'acute', 'circumflex', |
|
'tilde', 'macron', 'breve', 'dotaccent', 'dieresis', '', 'ring', 'cedilla', |
|
'', 'hungarumlaut', 'ogonek', 'caron', 'emdash', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', 'AE', '', 'ordfeminine', '', '', |
|
'', '', 'Lslash', 'Oslash', 'OE', 'ordmasculine', '', '', '', '', '', 'ae', |
|
'', '', '', 'dotlessi', '', '', 'lslash', 'oslash', 'oe', 'germandbls'], |
|
WinAnsiEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'exclam', 'quotedbl', 'numbersign', 'dollar', 'percent', |
|
'ampersand', 'quotesingle', 'parenleft', 'parenright', 'asterisk', 'plus', |
|
'comma', 'hyphen', 'period', 'slash', 'zero', 'one', 'two', 'three', |
|
'four', 'five', 'six', 'seven', 'eight', 'nine', 'colon', 'semicolon', |
|
'less', 'equal', 'greater', 'question', 'at', 'A', 'B', 'C', 'D', 'E', 'F', |
|
'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', |
|
'V', 'W', 'X', 'Y', 'Z', 'bracketleft', 'backslash', 'bracketright', |
|
'asciicircum', 'underscore', 'grave', 'a', 'b', 'c', 'd', 'e', 'f', 'g', |
|
'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', |
|
'w', 'x', 'y', 'z', 'braceleft', 'bar', 'braceright', 'asciitilde', |
|
'bullet', 'Euro', 'bullet', 'quotesinglbase', 'florin', 'quotedblbase', |
|
'ellipsis', 'dagger', 'daggerdbl', 'circumflex', 'perthousand', 'Scaron', |
|
'guilsinglleft', 'OE', 'bullet', 'Zcaron', 'bullet', 'bullet', 'quoteleft', |
|
'quoteright', 'quotedblleft', 'quotedblright', 'bullet', 'endash', |
|
'emdash', 'tilde', 'trademark', 'scaron', 'guilsinglright', 'oe', 'bullet', |
|
'zcaron', 'Ydieresis', '', 'exclamdown', 'cent', 'sterling', |
|
'currency', 'yen', 'brokenbar', 'section', 'dieresis', 'copyright', |
|
'ordfeminine', 'guillemotleft', 'logicalnot', 'hyphen', 'registered', |
|
'macron', 'degree', 'plusminus', 'twosuperior', 'threesuperior', 'acute', |
|
'mu', 'paragraph', 'periodcentered', 'cedilla', 'onesuperior', |
|
'ordmasculine', 'guillemotright', 'onequarter', 'onehalf', 'threequarters', |
|
'questiondown', 'Agrave', 'Aacute', 'Acircumflex', 'Atilde', 'Adieresis', |
|
'Aring', 'AE', 'Ccedilla', 'Egrave', 'Eacute', 'Ecircumflex', 'Edieresis', |
|
'Igrave', 'Iacute', 'Icircumflex', 'Idieresis', 'Eth', 'Ntilde', 'Ograve', |
|
'Oacute', 'Ocircumflex', 'Otilde', 'Odieresis', 'multiply', 'Oslash', |
|
'Ugrave', 'Uacute', 'Ucircumflex', 'Udieresis', 'Yacute', 'Thorn', |
|
'germandbls', 'agrave', 'aacute', 'acircumflex', 'atilde', 'adieresis', |
|
'aring', 'ae', 'ccedilla', 'egrave', 'eacute', 'ecircumflex', 'edieresis', |
|
'igrave', 'iacute', 'icircumflex', 'idieresis', 'eth', 'ntilde', 'ograve', |
|
'oacute', 'ocircumflex', 'otilde', 'odieresis', 'divide', 'oslash', |
|
'ugrave', 'uacute', 'ucircumflex', 'udieresis', 'yacute', 'thorn', |
|
'ydieresis'], |
|
symbolsEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'exclam', 'universal', 'numbersign', 'existential', 'percent', |
|
'ampersand', 'suchthat', 'parenleft', 'parenright', 'asteriskmath', 'plus', |
|
'comma', 'minus', 'period', 'slash', 'zero', 'one', 'two', 'three', 'four', |
|
'five', 'six', 'seven', 'eight', 'nine', 'colon', 'semicolon', 'less', |
|
'equal', 'greater', 'question', 'congruent', 'Alpha', 'Beta', 'Chi', |
|
'Delta', 'Epsilon', 'Phi', 'Gamma', 'Eta', 'Iota', 'theta1', 'Kappa', |
|
'Lambda', 'Mu', 'Nu', 'Omicron', 'Pi', 'Theta', 'Rho', 'Sigma', 'Tau', |
|
'Upsilon', 'sigma1', 'Omega', 'Xi', 'Psi', 'Zeta', 'bracketleft', |
|
'therefore', 'bracketright', 'perpendicular', 'underscore', 'radicalex', |
|
'alpha', 'beta', 'chi', 'delta', 'epsilon', 'phi', 'gamma', 'eta', 'iota', |
|
'phi1', 'kappa', 'lambda', 'mu', 'nu', 'omicron', 'pi', 'theta', 'rho', |
|
'sigma', 'tau', 'upsilon', 'omega1', 'omega', 'xi', 'psi', 'zeta', |
|
'braceleft', 'bar', 'braceright', 'similar', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', 'Euro', 'Upsilon1', 'minute', 'lessequal', |
|
'fraction', 'infinity', 'florin', 'club', 'diamond', 'heart', 'spade', |
|
'arrowboth', 'arrowleft', 'arrowup', 'arrowright', 'arrowdown', 'degree', |
|
'plusminus', 'second', 'greaterequal', 'multiply', 'proportional', |
|
'partialdiff', 'bullet', 'divide', 'notequal', 'equivalence', |
|
'approxequal', 'ellipsis', 'arrowvertex', 'arrowhorizex', 'carriagereturn', |
|
'aleph', 'Ifraktur', 'Rfraktur', 'weierstrass', 'circlemultiply', |
|
'circleplus', 'emptyset', 'intersection', 'union', 'propersuperset', |
|
'reflexsuperset', 'notsubset', 'propersubset', 'reflexsubset', 'element', |
|
'notelement', 'angle', 'gradient', 'registerserif', 'copyrightserif', |
|
'trademarkserif', 'product', 'radical', 'dotmath', 'logicalnot', |
|
'logicaland', 'logicalor', 'arrowdblboth', 'arrowdblleft', 'arrowdblup', |
|
'arrowdblright', 'arrowdbldown', 'lozenge', 'angleleft', 'registersans', |
|
'copyrightsans', 'trademarksans', 'summation', 'parenlefttp', |
|
'parenleftex', 'parenleftbt', 'bracketlefttp', 'bracketleftex', |
|
'bracketleftbt', 'bracelefttp', 'braceleftmid', 'braceleftbt', 'braceex', |
|
'', 'angleright', 'integral', 'integraltp', 'integralex', 'integralbt', |
|
'parenrighttp', 'parenrightex', 'parenrightbt', 'bracketrighttp', |
|
'bracketrightex', 'bracketrightbt', 'bracerighttp', 'bracerightmid', |
|
'bracerightbt'], |
|
zapfDingbatsEncoding: ['', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'space', 'a1', 'a2', 'a202', 'a3', 'a4', 'a5', 'a119', 'a118', 'a117', |
|
'a11', 'a12', 'a13', 'a14', 'a15', 'a16', 'a105', 'a17', 'a18', 'a19', |
|
'a20', 'a21', 'a22', 'a23', 'a24', 'a25', 'a26', 'a27', 'a28', 'a6', 'a7', |
|
'a8', 'a9', 'a10', 'a29', 'a30', 'a31', 'a32', 'a33', 'a34', 'a35', 'a36', |
|
'a37', 'a38', 'a39', 'a40', 'a41', 'a42', 'a43', 'a44', 'a45', 'a46', |
|
'a47', 'a48', 'a49', 'a50', 'a51', 'a52', 'a53', 'a54', 'a55', 'a56', |
|
'a57', 'a58', 'a59', 'a60', 'a61', 'a62', 'a63', 'a64', 'a65', 'a66', |
|
'a67', 'a68', 'a69', 'a70', 'a71', 'a72', 'a73', 'a74', 'a203', 'a75', |
|
'a204', 'a76', 'a77', 'a78', 'a79', 'a81', 'a82', 'a83', 'a84', 'a97', |
|
'a98', 'a99', 'a100', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', '', |
|
'', '', 'a101', 'a102', 'a103', 'a104', 'a106', 'a107', 'a108', 'a112', |
|
'a111', 'a110', 'a109', 'a120', 'a121', 'a122', 'a123', 'a124', 'a125', |
|
'a126', 'a127', 'a128', 'a129', 'a130', 'a131', 'a132', 'a133', 'a134', |
|
'a135', 'a136', 'a137', 'a138', 'a139', 'a140', 'a141', 'a142', 'a143', |
|
'a144', 'a145', 'a146', 'a147', 'a148', 'a149', 'a150', 'a151', 'a152', |
|
'a153', 'a154', 'a155', 'a156', 'a157', 'a158', 'a159', 'a160', 'a161', |
|
'a163', 'a164', 'a196', 'a165', 'a192', 'a166', 'a167', 'a168', 'a169', |
|
'a170', 'a171', 'a172', 'a173', 'a162', 'a174', 'a175', 'a176', 'a177', |
|
'a178', 'a179', 'a193', 'a180', 'a199', 'a181', 'a200', 'a182', '', 'a201', |
|
'a183', 'a184', 'a197', 'a185', 'a194', 'a198', 'a186', 'a195', 'a187', |
|
'a188', 'a189', 'a190', 'a191'] |
|
}; |
|
|
|
/** |
|
* Hold a map of decoded fonts and of the standard fourteen Type1 |
|
* fonts and their acronyms. |
|
*/ |
|
var stdFontMap = { |
|
'ArialNarrow': 'Helvetica', |
|
'ArialNarrow-Bold': 'Helvetica-Bold', |
|
'ArialNarrow-BoldItalic': 'Helvetica-BoldOblique', |
|
'ArialNarrow-Italic': 'Helvetica-Oblique', |
|
'ArialBlack': 'Helvetica', |
|
'ArialBlack-Bold': 'Helvetica-Bold', |
|
'ArialBlack-BoldItalic': 'Helvetica-BoldOblique', |
|
'ArialBlack-Italic': 'Helvetica-Oblique', |
|
'Arial': 'Helvetica', |
|
'Arial-Bold': 'Helvetica-Bold', |
|
'Arial-BoldItalic': 'Helvetica-BoldOblique', |
|
'Arial-Italic': 'Helvetica-Oblique', |
|
'Arial-BoldItalicMT': 'Helvetica-BoldOblique', |
|
'Arial-BoldMT': 'Helvetica-Bold', |
|
'Arial-ItalicMT': 'Helvetica-Oblique', |
|
'ArialMT': 'Helvetica', |
|
'Courier-Bold': 'Courier-Bold', |
|
'Courier-BoldItalic': 'Courier-BoldOblique', |
|
'Courier-Italic': 'Courier-Oblique', |
|
'CourierNew': 'Courier', |
|
'CourierNew-Bold': 'Courier-Bold', |
|
'CourierNew-BoldItalic': 'Courier-BoldOblique', |
|
'CourierNew-Italic': 'Courier-Oblique', |
|
'CourierNewPS-BoldItalicMT': 'Courier-BoldOblique', |
|
'CourierNewPS-BoldMT': 'Courier-Bold', |
|
'CourierNewPS-ItalicMT': 'Courier-Oblique', |
|
'CourierNewPSMT': 'Courier', |
|
'Helvetica-Bold': 'Helvetica-Bold', |
|
'Helvetica-BoldItalic': 'Helvetica-BoldOblique', |
|
'Helvetica-Italic': 'Helvetica-Oblique', |
|
'Symbol-Bold': 'Symbol', |
|
'Symbol-BoldItalic': 'Symbol', |
|
'Symbol-Italic': 'Symbol', |
|
'TimesNewRoman': 'Times-Roman', |
|
'TimesNewRoman-Bold': 'Times-Bold', |
|
'TimesNewRoman-BoldItalic': 'Times-BoldItalic', |
|
'TimesNewRoman-Italic': 'Times-Italic', |
|
'TimesNewRomanPS': 'Times-Roman', |
|
'TimesNewRomanPS-Bold': 'Times-Bold', |
|
'TimesNewRomanPS-BoldItalic': 'Times-BoldItalic', |
|
'TimesNewRomanPS-BoldItalicMT': 'Times-BoldItalic', |
|
'TimesNewRomanPS-BoldMT': 'Times-Bold', |
|
'TimesNewRomanPS-Italic': 'Times-Italic', |
|
'TimesNewRomanPS-ItalicMT': 'Times-Italic', |
|
'TimesNewRomanPSMT': 'Times-Roman', |
|
'TimesNewRomanPSMT-Bold': 'Times-Bold', |
|
'TimesNewRomanPSMT-BoldItalic': 'Times-BoldItalic', |
|
'TimesNewRomanPSMT-Italic': 'Times-Italic' |
|
}; |
|
|
|
/** |
|
* Holds the map of the non-standard fonts that might be included as a standard |
|
* fonts without glyph data. |
|
*/ |
|
var nonStdFontMap = { |
|
'ComicSansMS': 'Comic Sans MS', |
|
'ComicSansMS-Bold': 'Comic Sans MS-Bold', |
|
'ComicSansMS-BoldItalic': 'Comic Sans MS-BoldItalic', |
|
'ComicSansMS-Italic': 'Comic Sans MS-Italic', |
|
'LucidaConsole': 'Courier', |
|
'LucidaConsole-Bold': 'Courier-Bold', |
|
'LucidaConsole-BoldItalic': 'Courier-BoldOblique', |
|
'LucidaConsole-Italic': 'Courier-Oblique' |
|
}; |
|
|
|
var serifFonts = { |
|
'Adobe Jenson': true, 'Adobe Text': true, 'Albertus': true, |
|
'Aldus': true, 'Alexandria': true, 'Algerian': true, |
|
'American Typewriter': true, 'Antiqua': true, 'Apex': true, |
|
'Arno': true, 'Aster': true, 'Aurora': true, |
|
'Baskerville': true, 'Bell': true, 'Bembo': true, |
|
'Bembo Schoolbook': true, 'Benguiat': true, 'Berkeley Old Style': true, |
|
'Bernhard Modern': true, 'Berthold City': true, 'Bodoni': true, |
|
'Bauer Bodoni': true, 'Book Antiqua': true, 'Bookman': true, |
|
'Bordeaux Roman': true, 'Californian FB': true, 'Calisto': true, |
|
'Calvert': true, 'Capitals': true, 'Cambria': true, |
|
'Cartier': true, 'Caslon': true, 'Catull': true, |
|
'Centaur': true, 'Century Old Style': true, 'Century Schoolbook': true, |
|
'Chaparral': true, 'Charis SIL': true, 'Cheltenham': true, |
|
'Cholla Slab': true, 'Clarendon': true, 'Clearface': true, |
|
'Cochin': true, 'Colonna': true, 'Computer Modern': true, |
|
'Concrete Roman': true, 'Constantia': true, 'Cooper Black': true, |
|
'Corona': true, 'Ecotype': true, 'Egyptienne': true, |
|
'Elephant': true, 'Excelsior': true, 'Fairfield': true, |
|
'FF Scala': true, 'Folkard': true, 'Footlight': true, |
|
'FreeSerif': true, 'Friz Quadrata': true, 'Garamond': true, |
|
'Gentium': true, 'Georgia': true, 'Gloucester': true, |
|
'Goudy Old Style': true, 'Goudy Schoolbook': true, 'Goudy Pro Font': true, |
|
'Granjon': true, 'Guardian Egyptian': true, 'Heather': true, |
|
'Hercules': true, 'High Tower Text': true, 'Hiroshige': true, |
|
'Hoefler Text': true, 'Humana Serif': true, 'Imprint': true, |
|
'Ionic No. 5': true, 'Janson': true, 'Joanna': true, |
|
'Korinna': true, 'Lexicon': true, 'Liberation Serif': true, |
|
'Linux Libertine': true, 'Literaturnaya': true, 'Lucida': true, |
|
'Lucida Bright': true, 'Melior': true, 'Memphis': true, |
|
'Miller': true, 'Minion': true, 'Modern': true, |
|
'Mona Lisa': true, 'Mrs Eaves': true, 'MS Serif': true, |
|
'Museo Slab': true, 'New York': true, 'Nimbus Roman': true, |
|
'NPS Rawlinson Roadway': true, 'Palatino': true, 'Perpetua': true, |
|
'Plantin': true, 'Plantin Schoolbook': true, 'Playbill': true, |
|
'Poor Richard': true, 'Rawlinson Roadway': true, 'Renault': true, |
|
'Requiem': true, 'Rockwell': true, 'Roman': true, |
|
'Rotis Serif': true, 'Sabon': true, 'Scala': true, |
|
'Seagull': true, 'Sistina': true, 'Souvenir': true, |
|
'STIX': true, 'Stone Informal': true, 'Stone Serif': true, |
|
'Sylfaen': true, 'Times': true, 'Trajan': true, |
|
'Trinité': true, 'Trump Mediaeval': true, 'Utopia': true, |
|
'Vale Type': true, 'Bitstream Vera': true, 'Vera Serif': true, |
|
'Versailles': true, 'Wanted': true, 'Weiss': true, |
|
'Wide Latin': true, 'Windsor': true, 'XITS': true |
|
}; |
|
|
|
var symbolsFonts = { |
|
'Dingbats': true, 'Symbol': true, 'ZapfDingbats': true |
|
}; |
|
|
|
var CMapConverterList = { |
|
'H': jis7ToUnicode, |
|
'V': jis7ToUnicode, |
|
'EUC-H': eucjpToUnicode, |
|
'EUC-V': eucjpToUnicode, |
|
'83pv-RKSJ-H': sjis83pvToUnicode, |
|
'90pv-RKSJ-H': sjis90pvToUnicode, |
|
'90ms-RKSJ-H': sjisToUnicode, |
|
'90ms-RKSJ-V': sjisToUnicode, |
|
'90msp-RKSJ-H': sjisToUnicode, |
|
'90msp-RKSJ-V': sjisToUnicode, |
|
'GBK-EUC-H': gbkToUnicode, |
|
'ETenms-B5-H': big5ToUnicode, |
|
'ETenms-B5-V': big5ToUnicode, |
|
}; |
|
|
|
// CMaps using Hankaku (Halfwidth) Latin glyphs instead of proportional one. |
|
// We need to distinguish them to get correct widths from CIDFont dicts. |
|
var HalfwidthCMaps = { |
|
'H': true, |
|
'V': true, |
|
'EUC-H': true, |
|
'EUC-V': true, |
|
'90ms-RKSJ-H': true, |
|
'90ms-RKSJ-V': true, |
|
'UniJIS-UCS2-HW-H': true, |
|
'UniJIS-UCS2-HW-V': true |
|
}; |
|
|
|
var decodeBytes; |
|
if (typeof TextDecoder !== 'undefined') { |
|
// The encodings supported by TextDecoder can be found at: |
|
// http://encoding.spec.whatwg.org/#concept-encoding-get |
|
decodeBytes = function(bytes, encoding, fatal) { |
|
return new TextDecoder(encoding, {fatal: !!fatal}).decode(bytes); |
|
}; |
|
} else if (typeof FileReaderSync !== 'undefined') { |
|
decodeBytes = function(bytes, encoding) { |
|
return new FileReaderSync().readAsText(new Blob([bytes]), encoding); |
|
}; |
|
} else { |
|
// Clear the list so that decodeBytes will never be called. |
|
CMapConverterList = {}; |
|
} |
|
|
|
function jis7ToUnicode(str) { |
|
var bytes = stringToBytes(str); |
|
var length = bytes.length; |
|
for (var i = 0; i < length; ++i) { |
|
bytes[i] |= 0x80; |
|
} |
|
return decodeBytes(bytes, 'euc-jp'); |
|
} |
|
|
|
function eucjpToUnicode(str) { |
|
return decodeBytes(stringToBytes(str), 'euc-jp'); |
|
} |
|
|
|
function sjisToUnicode(str) { |
|
return decodeBytes(stringToBytes(str), 'shift_jis'); |
|
} |
|
|
|
function sjis83pvToUnicode(str) { |
|
var bytes = stringToBytes(str); |
|
try { |
|
// TODO: 83pv has incompatible mappings in ed40..ee9c range. |
|
return decodeBytes(bytes, 'shift_jis', true); |
|
} catch (e) { |
|
TODO('Unsupported 83pv character found'); |
|
// Just retry without checking errors for now. |
|
return decodeBytes(bytes, 'shift_jis'); |
|
} |
|
} |
|
|
|
function sjis90pvToUnicode(str) { |
|
var bytes = stringToBytes(str); |
|
try { |
|
// TODO: 90pv has incompatible mappings in 8740..879c and eb41..ee9c. |
|
return decodeBytes(bytes, 'shift_jis', true); |
|
} catch (e) { |
|
TODO('Unsupported 90pv character found'); |
|
// Just retry without checking errors for now. |
|
return decodeBytes(bytes, 'shift_jis'); |
|
} |
|
} |
|
|
|
function gbkToUnicode(str) { |
|
return decodeBytes(stringToBytes(str), 'gbk'); |
|
} |
|
|
|
function big5ToUnicode(str) { |
|
return decodeBytes(stringToBytes(str), 'big5'); |
|
} |
|
|
|
// Some characters, e.g. copyrightserif, mapped to the private use area and |
|
// might not be displayed using standard fonts. Mapping/hacking well-known chars |
|
// to the similar equivalents in the normal characters range. |
|
function mapPrivateUseChars(code) { |
|
switch (code) { |
|
case 0xF8E9: // copyrightsans |
|
case 0xF6D9: // copyrightserif |
|
return 0x00A9; // copyright |
|
default: |
|
return code; |
|
} |
|
} |
|
|
|
var FontLoader = { |
|
//#if !(MOZCENTRAL) |
|
loadingContext: { |
|
requests: [], |
|
nextRequestId: 0 |
|
}, |
|
|
|
isSyncFontLoadingSupported: (function detectSyncFontLoadingSupport() { |
|
if (isWorker) |
|
return false; |
|
|
|
// User agent string sniffing is bad, but there is no reliable way to tell |
|
// if font is fully loaded and ready to be used with canvas. |
|
var userAgent = window.navigator.userAgent; |
|
var m = /Mozilla\/5.0.*?rv:(\d+).*? Gecko/.exec(userAgent); |
|
if (m && m[1] >= 14) |
|
return true; |
|
// TODO other browsers |
|
return false; |
|
})(), |
|
|
|
bind: function fontLoaderBind(fonts, callback) { |
|
assert(!isWorker, 'bind() shall be called from main thread'); |
|
|
|
var rules = [], fontsToLoad = []; |
|
for (var i = 0, ii = fonts.length; i < ii; i++) { |
|
var font = fonts[i]; |
|
|
|
// Add the font to the DOM only once or skip if the font |
|
// is already loaded. |
|
if (font.attached || font.loading === false) { |
|
continue; |
|
} |
|
font.attached = true; |
|
|
|
var rule = font.bindDOM(); |
|
if (rule) { |
|
rules.push(rule); |
|
fontsToLoad.push(font); |
|
} |
|
} |
|
|
|
var request = FontLoader.queueLoadingCallback(callback); |
|
if (rules.length > 0 && !this.isSyncFontLoadingSupported) { |
|
FontLoader.prepareFontLoadEvent(rules, fontsToLoad, request); |
|
} else { |
|
request.complete(); |
|
} |
|
}, |
|
|
|
queueLoadingCallback: function FontLoader_queueLoadingCallback(callback) { |
|
function LoadLoader_completeRequest() { |
|
assert(!request.end, 'completeRequest() cannot be called twice'); |
|
request.end = Date.now(); |
|
|
|
// sending all completed requests in order how they were queued |
|
while (context.requests.length > 0 && context.requests[0].end) { |
|
var otherRequest = context.requests.shift(); |
|
setTimeout(otherRequest.callback, 0); |
|
} |
|
} |
|
|
|
var context = FontLoader.loadingContext; |
|
var requestId = 'pdfjs-font-loading-' + (context.nextRequestId++); |
|
var request = { |
|
id: requestId, |
|
complete: LoadLoader_completeRequest, |
|
callback: callback, |
|
started: Date.now() |
|
}; |
|
context.requests.push(request); |
|
return request; |
|
}, |
|
|
|
// Set things up so that at least one pdfjsFontLoad event is |
|
// dispatched when all the @font-face |rules| for |fonts| have been |
|
// loaded in a subdocument. It's expected that the load of |rules| |
|
// has already started in this (outer) document, so that they should |
|
// be ordered before the load in the subdocument. |
|
prepareFontLoadEvent: function fontLoaderPrepareFontLoadEvent(rules, |
|
fonts, |
|
request) { |
|
/** Hack begin */ |
|
// There's no event when a font has finished downloading so the |
|
// following code is a dirty hack to 'guess' when a font is |
|
// ready. This code will be obsoleted by Mozilla bug 471915. |
|
// |
|
// The only reliable way to know if a font is loaded in Gecko |
|
// (at the moment) is document.onload in a document with |
|
// a @font-face rule defined in a "static" stylesheet. We use a |
|
// subdocument in an <iframe>, set up properly, to know when |
|
// our @font-face rule was loaded. However, the subdocument and |
|
// outer document can't share CSS rules, so the inner document |
|
// is only part of the puzzle. The second piece is an invisible |
|
// div created in order to force loading of the @font-face in |
|
// the *outer* document. (The font still needs to be loaded for |
|
// its metrics, for reflow). We create the div first for the |
|
// outer document, then create the iframe. Unless something |
|
// goes really wonkily, we expect the @font-face for the outer |
|
// document to be processed before the inner. That's still |
|
// fragile, but seems to work in practice. |
|
// |
|
// The postMessage() hackery was added to work around chrome bug |
|
// 82402. |
|
|
|
var requestId = request.id; |
|
// Validate the requestId parameter -- the value used to construct HTML. |
|
if (!/^[\w\-]+$/.test(requestId)) { |
|
error('Invalid request id: ' + requestId); |
|
|
|
// Normally the error-function throws. But if a malicious code |
|
// intercepts the function call then the return is needed. |
|
return; |
|
} |
|
|
|
var names = []; |
|
for (var i = 0, ii = fonts.length; i < ii; i++) |
|
names.push(fonts[i].loadedName); |
|
|
|
// Validate the names parameter -- the values can used to construct HTML. |
|
if (!/^\w+$/.test(names.join(''))) { |
|
error('Invalid font name(s): ' + names.join()); |
|
|
|
// Normally the error-function throws. But if a malicious code |
|
// intercepts the function call then the return is needed. |
|
return; |
|
} |
|
|
|
var div = document.createElement('div'); |
|
div.setAttribute('style', |
|
'visibility: hidden;' + |
|
'width: 10px; height: 10px;' + |
|
'position: absolute; top: 0px; left: 0px;'); |
|
var html = ''; |
|
for (var i = 0, ii = names.length; i < ii; ++i) { |
|
html += '<span style="font-family:' + names[i] + '">Hi</span>'; |
|
} |
|
div.innerHTML = html; |
|
document.body.appendChild(div); |
|
|
|
window.addEventListener( |
|
'message', |
|
function fontLoaderMessage(e) { |
|
if (e.data !== requestId) |
|
return; |
|
for (var i = 0, ii = fonts.length; i < ii; ++i) { |
|
var font = fonts[i]; |
|
font.loading = false; |
|
} |
|
request.complete(); |
|
// cleanup |
|
if (frame) { |
|
document.body.removeChild(frame); |
|
} |
|
window.removeEventListener('message', fontLoaderMessage, false); |
|
}, |
|
false); |
|
|
|
// XXX we should have a time-out here too, and maybe fire |
|
// pdfjsFontLoadFailed? |
|
var src = '<!DOCTYPE HTML><html><head><meta charset="utf-8">'; |
|
src += '<style type="text/css">'; |
|
for (var i = 0, ii = rules.length; i < ii; ++i) { |
|
src += rules[i]; |
|
} |
|
src += '</style>'; |
|
src += '<script type="application/javascript">'; |
|
src += ' window.onload = function fontLoaderOnload() {\n'; |
|
src += ' parent.postMessage("' + requestId + '", "*");\n'; |
|
// Chrome stuck on loading (see chrome issue 145227) - resetting url |
|
src += ' window.location = "about:blank";\n'; |
|
src += ' }'; |
|
// Hack so the end script tag isn't counted if this is inline JS. |
|
src += '</scr' + 'ipt></head><body>'; |
|
for (var i = 0, ii = names.length; i < ii; ++i) { |
|
src += '<p style="font-family:\'' + names[i] + '\'">Hi</p>'; |
|
} |
|
src += '</body></html>'; |
|
|
|
var MAX_IFRAME_SRC_LENGTH = 1000000, IFRAME_TIMEOUT = 2000; |
|
// Chrome fails for long src attributes (see issue 174023) |
|
if (src.length > MAX_IFRAME_SRC_LENGTH) { |
|
// ... waiting for some fixed period of time instead |
|
window.setTimeout(function() { |
|
window.postMessage(requestId, '*'); |
|
}, IFRAME_TIMEOUT); |
|
return; |
|
} |
|
|
|
var frame = document.createElement('iframe'); |
|
frame.src = 'data:text/html,' + src; |
|
frame.setAttribute('style', |
|
'visibility: hidden;' + |
|
'width: 10px; height: 10px;' + |
|
'position: absolute; top: 0px; left: 0px;'); |
|
document.body.appendChild(frame); |
|
/** Hack end */ |
|
} |
|
//#else |
|
//bind: function fontLoaderBind(fonts, callback) { |
|
// assert(!isWorker, 'bind() shall be called from main thread'); |
|
// |
|
// for (var i = 0, ii = fonts.length; i < ii; i++) { |
|
// var font = fonts[i]; |
|
// if (font.attached) |
|
// continue; |
|
// |
|
// font.attached = true; |
|
// font.bindDOM() |
|
// } |
|
// |
|
// setTimeout(callback); |
|
//} |
|
//#endif |
|
}; |
|
|
|
var UnicodeRanges = [ |
|
{ 'begin': 0x0000, 'end': 0x007F }, // Basic Latin |
|
{ 'begin': 0x0080, 'end': 0x00FF }, // Latin-1 Supplement |
|
{ 'begin': 0x0100, 'end': 0x017F }, // Latin Extended-A |
|
{ 'begin': 0x0180, 'end': 0x024F }, // Latin Extended-B |
|
{ 'begin': 0x0250, 'end': 0x02AF }, // IPA Extensions |
|
{ 'begin': 0x02B0, 'end': 0x02FF }, // Spacing Modifier Letters |
|
{ 'begin': 0x0300, 'end': 0x036F }, // Combining Diacritical Marks |
|
{ 'begin': 0x0370, 'end': 0x03FF }, // Greek and Coptic |
|
{ 'begin': 0x2C80, 'end': 0x2CFF }, // Coptic |
|
{ 'begin': 0x0400, 'end': 0x04FF }, // Cyrillic |
|
{ 'begin': 0x0530, 'end': 0x058F }, // Armenian |
|
{ 'begin': 0x0590, 'end': 0x05FF }, // Hebrew |
|
{ 'begin': 0xA500, 'end': 0xA63F }, // Vai |
|
{ 'begin': 0x0600, 'end': 0x06FF }, // Arabic |
|
{ 'begin': 0x07C0, 'end': 0x07FF }, // NKo |
|
{ 'begin': 0x0900, 'end': 0x097F }, // Devanagari |
|
{ 'begin': 0x0980, 'end': 0x09FF }, // Bengali |
|
{ 'begin': 0x0A00, 'end': 0x0A7F }, // Gurmukhi |
|
{ 'begin': 0x0A80, 'end': 0x0AFF }, // Gujarati |
|
{ 'begin': 0x0B00, 'end': 0x0B7F }, // Oriya |
|
{ 'begin': 0x0B80, 'end': 0x0BFF }, // Tamil |
|
{ 'begin': 0x0C00, 'end': 0x0C7F }, // Telugu |
|
{ 'begin': 0x0C80, 'end': 0x0CFF }, // Kannada |
|
{ 'begin': 0x0D00, 'end': 0x0D7F }, // Malayalam |
|
{ 'begin': 0x0E00, 'end': 0x0E7F }, // Thai |
|
{ 'begin': 0x0E80, 'end': 0x0EFF }, // Lao |
|
{ 'begin': 0x10A0, 'end': 0x10FF }, // Georgian |
|
{ 'begin': 0x1B00, 'end': 0x1B7F }, // Balinese |
|
{ 'begin': 0x1100, 'end': 0x11FF }, // Hangul Jamo |
|
{ 'begin': 0x1E00, 'end': 0x1EFF }, // Latin Extended Additional |
|
{ 'begin': 0x1F00, 'end': 0x1FFF }, // Greek Extended |
|
{ 'begin': 0x2000, 'end': 0x206F }, // General Punctuation |
|
{ 'begin': 0x2070, 'end': 0x209F }, // Superscripts And Subscripts |
|
{ 'begin': 0x20A0, 'end': 0x20CF }, // Currency Symbol |
|
{ 'begin': 0x20D0, 'end': 0x20FF }, // Combining Diacritical Marks For Symbols |
|
{ 'begin': 0x2100, 'end': 0x214F }, // Letterlike Symbols |
|
{ 'begin': 0x2150, 'end': 0x218F }, // Number Forms |
|
{ 'begin': 0x2190, 'end': 0x21FF }, // Arrows |
|
{ 'begin': 0x2200, 'end': 0x22FF }, // Mathematical Operators |
|
{ 'begin': 0x2300, 'end': 0x23FF }, // Miscellaneous Technical |
|
{ 'begin': 0x2400, 'end': 0x243F }, // Control Pictures |
|
{ 'begin': 0x2440, 'end': 0x245F }, // Optical Character Recognition |
|
{ 'begin': 0x2460, 'end': 0x24FF }, // Enclosed Alphanumerics |
|
{ 'begin': 0x2500, 'end': 0x257F }, // Box Drawing |
|
{ 'begin': 0x2580, 'end': 0x259F }, // Block Elements |
|
{ 'begin': 0x25A0, 'end': 0x25FF }, // Geometric Shapes |
|
{ 'begin': 0x2600, 'end': 0x26FF }, // Miscellaneous Symbols |
|
{ 'begin': 0x2700, 'end': 0x27BF }, // Dingbats |
|
{ 'begin': 0x3000, 'end': 0x303F }, // CJK Symbols And Punctuation |
|
{ 'begin': 0x3040, 'end': 0x309F }, // Hiragana |
|
{ 'begin': 0x30A0, 'end': 0x30FF }, // Katakana |
|
{ 'begin': 0x3100, 'end': 0x312F }, // Bopomofo |
|
{ 'begin': 0x3130, 'end': 0x318F }, // Hangul Compatibility Jamo |
|
{ 'begin': 0xA840, 'end': 0xA87F }, // Phags-pa |
|
{ 'begin': 0x3200, 'end': 0x32FF }, // Enclosed CJK Letters And Months |
|
{ 'begin': 0x3300, 'end': 0x33FF }, // CJK Compatibility |
|
{ 'begin': 0xAC00, 'end': 0xD7AF }, // Hangul Syllables |
|
{ 'begin': 0xD800, 'end': 0xDFFF }, // Non-Plane 0 * |
|
{ 'begin': 0x10900, 'end': 0x1091F }, // Phoenicia |
|
{ 'begin': 0x4E00, 'end': 0x9FFF }, // CJK Unified Ideographs |
|
{ 'begin': 0xE000, 'end': 0xF8FF }, // Private Use Area (plane 0) |
|
{ 'begin': 0x31C0, 'end': 0x31EF }, // CJK Strokes |
|
{ 'begin': 0xFB00, 'end': 0xFB4F }, // Alphabetic Presentation Forms |
|
{ 'begin': 0xFB50, 'end': 0xFDFF }, // Arabic Presentation Forms-A |
|
{ 'begin': 0xFE20, 'end': 0xFE2F }, // Combining Half Marks |
|
{ 'begin': 0xFE10, 'end': 0xFE1F }, // Vertical Forms |
|
{ 'begin': 0xFE50, 'end': 0xFE6F }, // Small Form Variants |
|
{ 'begin': 0xFE70, 'end': 0xFEFF }, // Arabic Presentation Forms-B |
|
{ 'begin': 0xFF00, 'end': 0xFFEF }, // Halfwidth And Fullwidth Forms |
|
{ 'begin': 0xFFF0, 'end': 0xFFFF }, // Specials |
|
{ 'begin': 0x0F00, 'end': 0x0FFF }, // Tibetan |
|
{ 'begin': 0x0700, 'end': 0x074F }, // Syriac |
|
{ 'begin': 0x0780, 'end': 0x07BF }, // Thaana |
|
{ 'begin': 0x0D80, 'end': 0x0DFF }, // Sinhala |
|
{ 'begin': 0x1000, 'end': 0x109F }, // Myanmar |
|
{ 'begin': 0x1200, 'end': 0x137F }, // Ethiopic |
|
{ 'begin': 0x13A0, 'end': 0x13FF }, // Cherokee |
|
{ 'begin': 0x1400, 'end': 0x167F }, // Unified Canadian Aboriginal Syllabics |
|
{ 'begin': 0x1680, 'end': 0x169F }, // Ogham |
|
{ 'begin': 0x16A0, 'end': 0x16FF }, // Runic |
|
{ 'begin': 0x1780, 'end': 0x17FF }, // Khmer |
|
{ 'begin': 0x1800, 'end': 0x18AF }, // Mongolian |
|
{ 'begin': 0x2800, 'end': 0x28FF }, // Braille Patterns |
|
{ 'begin': 0xA000, 'end': 0xA48F }, // Yi Syllables |
|
{ 'begin': 0x1700, 'end': 0x171F }, // Tagalog |
|
{ 'begin': 0x10300, 'end': 0x1032F }, // Old Italic |
|
{ 'begin': 0x10330, 'end': 0x1034F }, // Gothic |
|
{ 'begin': 0x10400, 'end': 0x1044F }, // Deseret |
|
{ 'begin': 0x1D000, 'end': 0x1D0FF }, // Byzantine Musical Symbols |
|
{ 'begin': 0x1D400, 'end': 0x1D7FF }, // Mathematical Alphanumeric Symbols |
|
{ 'begin': 0xFF000, 'end': 0xFFFFD }, // Private Use (plane 15) |
|
{ 'begin': 0xFE00, 'end': 0xFE0F }, // Variation Selectors |
|
{ 'begin': 0xE0000, 'end': 0xE007F }, // Tags |
|
{ 'begin': 0x1900, 'end': 0x194F }, // Limbu |
|
{ 'begin': 0x1950, 'end': 0x197F }, // Tai Le |
|
{ 'begin': 0x1980, 'end': 0x19DF }, // New Tai Lue |
|
{ 'begin': 0x1A00, 'end': 0x1A1F }, // Buginese |
|
{ 'begin': 0x2C00, 'end': 0x2C5F }, // Glagolitic |
|
{ 'begin': 0x2D30, 'end': 0x2D7F }, // Tifinagh |
|
{ 'begin': 0x4DC0, 'end': 0x4DFF }, // Yijing Hexagram Symbols |
|
{ 'begin': 0xA800, 'end': 0xA82F }, // Syloti Nagri |
|
{ 'begin': 0x10000, 'end': 0x1007F }, // Linear B Syllabary |
|
{ 'begin': 0x10140, 'end': 0x1018F }, // Ancient Greek Numbers |
|
{ 'begin': 0x10380, 'end': 0x1039F }, // Ugaritic |
|
{ 'begin': 0x103A0, 'end': 0x103DF }, // Old Persian |
|
{ 'begin': 0x10450, 'end': 0x1047F }, // Shavian |
|
{ 'begin': 0x10480, 'end': 0x104AF }, // Osmanya |
|
{ 'begin': 0x10800, 'end': 0x1083F }, // Cypriot Syllabary |
|
{ 'begin': 0x10A00, 'end': 0x10A5F }, // Kharoshthi |
|
{ 'begin': 0x1D300, 'end': 0x1D35F }, // Tai Xuan Jing Symbols |
|
{ 'begin': 0x12000, 'end': 0x123FF }, // Cuneiform |
|
{ 'begin': 0x1D360, 'end': 0x1D37F }, // Counting Rod Numerals |
|
{ 'begin': 0x1B80, 'end': 0x1BBF }, // Sundanese |
|
{ 'begin': 0x1C00, 'end': 0x1C4F }, // Lepcha |
|
{ 'begin': 0x1C50, 'end': 0x1C7F }, // Ol Chiki |
|
{ 'begin': 0xA880, 'end': 0xA8DF }, // Saurashtra |
|
{ 'begin': 0xA900, 'end': 0xA92F }, // Kayah Li |
|
{ 'begin': 0xA930, 'end': 0xA95F }, // Rejang |
|
{ 'begin': 0xAA00, 'end': 0xAA5F }, // Cham |
|
{ 'begin': 0x10190, 'end': 0x101CF }, // Ancient Symbols |
|
{ 'begin': 0x101D0, 'end': 0x101FF }, // Phaistos Disc |
|
{ 'begin': 0x102A0, 'end': 0x102DF }, // Carian |
|
{ 'begin': 0x1F030, 'end': 0x1F09F } // Domino Tiles |
|
]; |
|
|
|
var MacStandardGlyphOrdering = [ |
|
'.notdef', '.null', 'nonmarkingreturn', 'space', 'exclam', 'quotedbl', |
|
'numbersign', 'dollar', 'percent', 'ampersand', 'quotesingle', 'parenleft', |
|
'parenright', 'asterisk', 'plus', 'comma', 'hyphen', 'period', 'slash', |
|
'zero', 'one', 'two', 'three', 'four', 'five', 'six', 'seven', 'eight', |
|
'nine', 'colon', 'semicolon', 'less', 'equal', 'greater', 'question', 'at', |
|
'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I', 'J', 'K', 'L', 'M', 'N', 'O', |
|
'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', 'X', 'Y', 'Z', 'bracketleft', |
|
'backslash', 'bracketright', 'asciicircum', 'underscore', 'grave', 'a', 'b', |
|
'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', 'k', 'l', 'm', 'n', 'o', 'p', 'q', |
|
'r', 's', 't', 'u', 'v', 'w', 'x', 'y', 'z', 'braceleft', 'bar', 'braceright', |
|
'asciitilde', 'Adieresis', 'Aring', 'Ccedilla', 'Eacute', 'Ntilde', |
|
'Odieresis', 'Udieresis', 'aacute', 'agrave', 'acircumflex', 'adieresis', |
|
'atilde', 'aring', 'ccedilla', 'eacute', 'egrave', 'ecircumflex', 'edieresis', |
|
'iacute', 'igrave', 'icircumflex', 'idieresis', 'ntilde', 'oacute', 'ograve', |
|
'ocircumflex', 'odieresis', 'otilde', 'uacute', 'ugrave', 'ucircumflex', |
|
'udieresis', 'dagger', 'degree', 'cent', 'sterling', 'section', 'bullet', |
|
'paragraph', 'germandbls', 'registered', 'copyright', 'trademark', 'acute', |
|
'dieresis', 'notequal', 'AE', 'Oslash', 'infinity', 'plusminus', 'lessequal', |
|
'greaterequal', 'yen', 'mu', 'partialdiff', 'summation', 'product', 'pi', |
|
'integral', 'ordfeminine', 'ordmasculine', 'Omega', 'ae', 'oslash', |
|
'questiondown', 'exclamdown', 'logicalnot', 'radical', 'florin', |
|
'approxequal', 'Delta', 'guillemotleft', 'guillemotright', 'ellipsis', |
|
'nonbreakingspace', 'Agrave', 'Atilde', 'Otilde', 'OE', 'oe', 'endash', |
|
'emdash', 'quotedblleft', 'quotedblright', 'quoteleft', 'quoteright', |
|
'divide', 'lozenge', 'ydieresis', 'Ydieresis', 'fraction', 'currency', |
|
'guilsinglleft', 'guilsinglright', 'fi', 'fl', 'daggerdbl', 'periodcentered', |
|
'quotesinglbase', 'quotedblbase', 'perthousand', 'Acircumflex', |
|
'Ecircumflex', 'Aacute', 'Edieresis', 'Egrave', 'Iacute', 'Icircumflex', |
|
'Idieresis', 'Igrave', 'Oacute', 'Ocircumflex', 'apple', 'Ograve', 'Uacute', |
|
'Ucircumflex', 'Ugrave', 'dotlessi', 'circumflex', 'tilde', 'macron', |
|
'breve', 'dotaccent', 'ring', 'cedilla', 'hungarumlaut', 'ogonek', 'caron', |
|
'Lslash', 'lslash', 'Scaron', 'scaron', 'Zcaron', 'zcaron', 'brokenbar', |
|
'Eth', 'eth', 'Yacute', 'yacute', 'Thorn', 'thorn', 'minus', 'multiply', |
|
'onesuperior', 'twosuperior', 'threesuperior', 'onehalf', 'onequarter', |
|
'threequarters', 'franc', 'Gbreve', 'gbreve', 'Idotaccent', 'Scedilla', |
|
'scedilla', 'Cacute', 'cacute', 'Ccaron', 'ccaron', 'dcroat']; |
|
|
|
function getUnicodeRangeFor(value) { |
|
for (var i = 0, ii = UnicodeRanges.length; i < ii; i++) { |
|
var range = UnicodeRanges[i]; |
|
if (value >= range.begin && value < range.end) |
|
return i; |
|
} |
|
return -1; |
|
} |
|
|
|
function isRTLRangeFor(value) { |
|
var range = UnicodeRanges[13]; |
|
if (value >= range.begin && value < range.end) |
|
return true; |
|
range = UnicodeRanges[11]; |
|
if (value >= range.begin && value < range.end) |
|
return true; |
|
return false; |
|
} |
|
|
|
function isSpecialUnicode(unicode) { |
|
return (unicode <= 0x1F || (unicode >= 127 && unicode < GLYPH_AREA_SIZE)) || |
|
(unicode >= CMAP_GLYPH_OFFSET && |
|
unicode < CMAP_GLYPH_OFFSET + GLYPH_AREA_SIZE); |
|
} |
|
|
|
// The normalization table is obtained by filtering the Unicode characters |
|
// database with <compat> entries. |
|
var NormalizedUnicodes = { |
|
'\u00A8': '\u0020\u0308', |
|
'\u00AF': '\u0020\u0304', |
|
'\u00B4': '\u0020\u0301', |
|
'\u00B5': '\u03BC', |
|
'\u00B8': '\u0020\u0327', |
|
'\u0132': '\u0049\u004A', |
|
'\u0133': '\u0069\u006A', |
|
'\u013F': '\u004C\u00B7', |
|
'\u0140': '\u006C\u00B7', |
|
'\u0149': '\u02BC\u006E', |
|
'\u017F': '\u0073', |
|
'\u01C4': '\u0044\u017D', |
|
'\u01C5': '\u0044\u017E', |
|
'\u01C6': '\u0064\u017E', |
|
'\u01C7': '\u004C\u004A', |
|
'\u01C8': '\u004C\u006A', |
|
'\u01C9': '\u006C\u006A', |
|
'\u01CA': '\u004E\u004A', |
|
'\u01CB': '\u004E\u006A', |
|
'\u01CC': '\u006E\u006A', |
|
'\u01F1': '\u0044\u005A', |
|
'\u01F2': '\u0044\u007A', |
|
'\u01F3': '\u0064\u007A', |
|
'\u02D8': '\u0020\u0306', |
|
'\u02D9': '\u0020\u0307', |
|
'\u02DA': '\u0020\u030A', |
|
'\u02DB': '\u0020\u0328', |
|
'\u02DC': '\u0020\u0303', |
|
'\u02DD': '\u0020\u030B', |
|
'\u037A': '\u0020\u0345', |
|
'\u0384': '\u0020\u0301', |
|
'\u03D0': '\u03B2', |
|
'\u03D1': '\u03B8', |
|
'\u03D2': '\u03A5', |
|
'\u03D5': '\u03C6', |
|
'\u03D6': '\u03C0', |
|
'\u03F0': '\u03BA', |
|
'\u03F1': '\u03C1', |
|
'\u03F2': '\u03C2', |
|
'\u03F4': '\u0398', |
|
'\u03F5': '\u03B5', |
|
'\u03F9': '\u03A3', |
|
'\u0587': '\u0565\u0582', |
|
'\u0675': '\u0627\u0674', |
|
'\u0676': '\u0648\u0674', |
|
'\u0677': '\u06C7\u0674', |
|
'\u0678': '\u064A\u0674', |
|
'\u0E33': '\u0E4D\u0E32', |
|
'\u0EB3': '\u0ECD\u0EB2', |
|
'\u0EDC': '\u0EAB\u0E99', |
|
'\u0EDD': '\u0EAB\u0EA1', |
|
'\u0F77': '\u0FB2\u0F81', |
|
'\u0F79': '\u0FB3\u0F81', |
|
'\u1E9A': '\u0061\u02BE', |
|
'\u1FBD': '\u0020\u0313', |
|
'\u1FBF': '\u0020\u0313', |
|
'\u1FC0': '\u0020\u0342', |
|
'\u1FFE': '\u0020\u0314', |
|
'\u2002': '\u0020', |
|
'\u2003': '\u0020', |
|
'\u2004': '\u0020', |
|
'\u2005': '\u0020', |
|
'\u2006': '\u0020', |
|
'\u2008': '\u0020', |
|
'\u2009': '\u0020', |
|
'\u200A': '\u0020', |
|
'\u2017': '\u0020\u0333', |
|
'\u2024': '\u002E', |
|
'\u2025': '\u002E\u002E', |
|
'\u2026': '\u002E\u002E\u002E', |
|
'\u2033': '\u2032\u2032', |
|
'\u2034': '\u2032\u2032\u2032', |
|
'\u2036': '\u2035\u2035', |
|
'\u2037': '\u2035\u2035\u2035', |
|
'\u203C': '\u0021\u0021', |
|
'\u203E': '\u0020\u0305', |
|
'\u2047': '\u003F\u003F', |
|
'\u2048': '\u003F\u0021', |
|
'\u2049': '\u0021\u003F', |
|
'\u2057': '\u2032\u2032\u2032\u2032', |
|
'\u205F': '\u0020', |
|
'\u20A8': '\u0052\u0073', |
|
'\u2100': '\u0061\u002F\u0063', |
|
'\u2101': '\u0061\u002F\u0073', |
|
'\u2103': '\u00B0\u0043', |
|
'\u2105': '\u0063\u002F\u006F', |
|
'\u2106': '\u0063\u002F\u0075', |
|
'\u2107': '\u0190', |
|
'\u2109': '\u00B0\u0046', |
|
'\u2116': '\u004E\u006F', |
|
'\u2121': '\u0054\u0045\u004C', |
|
'\u2135': '\u05D0', |
|
'\u2136': '\u05D1', |
|
'\u2137': '\u05D2', |
|
'\u2138': '\u05D3', |
|
'\u213B': '\u0046\u0041\u0058', |
|
'\u2160': '\u0049', |
|
'\u2161': '\u0049\u0049', |
|
'\u2162': '\u0049\u0049\u0049', |
|
'\u2163': '\u0049\u0056', |
|
'\u2164': '\u0056', |
|
'\u2165': '\u0056\u0049', |
|
'\u2166': '\u0056\u0049\u0049', |
|
'\u2167': '\u0056\u0049\u0049\u0049', |
|
'\u2168': '\u0049\u0058', |
|
'\u2169': '\u0058', |
|
'\u216A': '\u0058\u0049', |
|
'\u216B': '\u0058\u0049\u0049', |
|
'\u216C': '\u004C', |
|
'\u216D': '\u0043', |
|
'\u216E': '\u0044', |
|
'\u216F': '\u004D', |
|
'\u2170': '\u0069', |
|
'\u2171': '\u0069\u0069', |
|
'\u2172': '\u0069\u0069\u0069', |
|
'\u2173': '\u0069\u0076', |
|
'\u2174': '\u0076', |
|
'\u2175': '\u0076\u0069', |
|
'\u2176': '\u0076\u0069\u0069', |
|
'\u2177': '\u0076\u0069\u0069\u0069', |
|
'\u2178': '\u0069\u0078', |
|
'\u2179': '\u0078', |
|
'\u217A': '\u0078\u0069', |
|
'\u217B': '\u0078\u0069\u0069', |
|
'\u217C': '\u006C', |
|
'\u217D': '\u0063', |
|
'\u217E': '\u0064', |
|
'\u217F': '\u006D', |
|
'\u222C': '\u222B\u222B', |
|
'\u222D': '\u222B\u222B\u222B', |
|
'\u222F': '\u222E\u222E', |
|
'\u2230': '\u222E\u222E\u222E', |
|
'\u2474': '\u0028\u0031\u0029', |
|
'\u2475': '\u0028\u0032\u0029', |
|
'\u2476': '\u0028\u0033\u0029', |
|
'\u2477': '\u0028\u0034\u0029', |
|
'\u2478': '\u0028\u0035\u0029', |
|
'\u2479': '\u0028\u0036\u0029', |
|
'\u247A': '\u0028\u0037\u0029', |
|
'\u247B': '\u0028\u0038\u0029', |
|
'\u247C': '\u0028\u0039\u0029', |
|
'\u247D': '\u0028\u0031\u0030\u0029', |
|
'\u247E': '\u0028\u0031\u0031\u0029', |
|
'\u247F': '\u0028\u0031\u0032\u0029', |
|
'\u2480': '\u0028\u0031\u0033\u0029', |
|
'\u2481': '\u0028\u0031\u0034\u0029', |
|
'\u2482': '\u0028\u0031\u0035\u0029', |
|
'\u2483': '\u0028\u0031\u0036\u0029', |
|
'\u2484': '\u0028\u0031\u0037\u0029', |
|
'\u2485': '\u0028\u0031\u0038\u0029', |
|
'\u2486': '\u0028\u0031\u0039\u0029', |
|
'\u2487': '\u0028\u0032\u0030\u0029', |
|
'\u2488': '\u0031\u002E', |
|
'\u2489': '\u0032\u002E', |
|
'\u248A': '\u0033\u002E', |
|
'\u248B': '\u0034\u002E', |
|
'\u248C': '\u0035\u002E', |
|
'\u248D': '\u0036\u002E', |
|
'\u248E': '\u0037\u002E', |
|
'\u248F': '\u0038\u002E', |
|
'\u2490': '\u0039\u002E', |
|
'\u2491': '\u0031\u0030\u002E', |
|
'\u2492': '\u0031\u0031\u002E', |
|
'\u2493': '\u0031\u0032\u002E', |
|
'\u2494': '\u0031\u0033\u002E', |
|
'\u2495': '\u0031\u0034\u002E', |
|
'\u2496': '\u0031\u0035\u002E', |
|
'\u2497': '\u0031\u0036\u002E', |
|
'\u2498': '\u0031\u0037\u002E', |
|
'\u2499': '\u0031\u0038\u002E', |
|
'\u249A': '\u0031\u0039\u002E', |
|
'\u249B': '\u0032\u0030\u002E', |
|
'\u249C': '\u0028\u0061\u0029', |
|
'\u249D': '\u0028\u0062\u0029', |
|
'\u249E': '\u0028\u0063\u0029', |
|
'\u249F': '\u0028\u0064\u0029', |
|
'\u24A0': '\u0028\u0065\u0029', |
|
'\u24A1': '\u0028\u0066\u0029', |
|
'\u24A2': '\u0028\u0067\u0029', |
|
'\u24A3': '\u0028\u0068\u0029', |
|
'\u24A4': '\u0028\u0069\u0029', |
|
'\u24A5': '\u0028\u006A\u0029', |
|
'\u24A6': '\u0028\u006B\u0029', |
|
'\u24A7': '\u0028\u006C\u0029', |
|
'\u24A8': '\u0028\u006D\u0029', |
|
'\u24A9': '\u0028\u006E\u0029', |
|
'\u24AA': '\u0028\u006F\u0029', |
|
'\u24AB': '\u0028\u0070\u0029', |
|
'\u24AC': '\u0028\u0071\u0029', |
|
'\u24AD': '\u0028\u0072\u0029', |
|
'\u24AE': '\u0028\u0073\u0029', |
|
'\u24AF': '\u0028\u0074\u0029', |
|
'\u24B0': '\u0028\u0075\u0029', |
|
'\u24B1': '\u0028\u0076\u0029', |
|
'\u24B2': '\u0028\u0077\u0029', |
|
'\u24B3': '\u0028\u0078\u0029', |
|
'\u24B4': '\u0028\u0079\u0029', |
|
'\u24B5': '\u0028\u007A\u0029', |
|
'\u2A0C': '\u222B\u222B\u222B\u222B', |
|
'\u2A74': '\u003A\u003A\u003D', |
|
'\u2A75': '\u003D\u003D', |
|
'\u2A76': '\u003D\u003D\u003D', |
|
'\u2E9F': '\u6BCD', |
|
'\u2EF3': '\u9F9F', |
|
'\u2F00': '\u4E00', |
|
'\u2F01': '\u4E28', |
|
'\u2F02': '\u4E36', |
|
'\u2F03': '\u4E3F', |
|
'\u2F04': '\u4E59', |
|
'\u2F05': '\u4E85', |
|
'\u2F06': '\u4E8C', |
|
'\u2F07': '\u4EA0', |
|
'\u2F08': '\u4EBA', |
|
'\u2F09': '\u513F', |
|
'\u2F0A': '\u5165', |
|
'\u2F0B': '\u516B', |
|
'\u2F0C': '\u5182', |
|
'\u2F0D': '\u5196', |
|
'\u2F0E': '\u51AB', |
|
'\u2F0F': '\u51E0', |
|
'\u2F10': '\u51F5', |
|
'\u2F11': '\u5200', |
|
'\u2F12': '\u529B', |
|
'\u2F13': '\u52F9', |
|
'\u2F14': '\u5315', |
|
'\u2F15': '\u531A', |
|
'\u2F16': '\u5338', |
|
'\u2F17': '\u5341', |
|
'\u2F18': '\u535C', |
|
'\u2F19': '\u5369', |
|
'\u2F1A': '\u5382', |
|
'\u2F1B': '\u53B6', |
|
'\u2F1C': '\u53C8', |
|
'\u2F1D': '\u53E3', |
|
'\u2F1E': '\u56D7', |
|
'\u2F1F': '\u571F', |
|
'\u2F20': '\u58EB', |
|
'\u2F21': '\u5902', |
|
'\u2F22': '\u590A', |
|
'\u2F23': '\u5915', |
|
'\u2F24': '\u5927', |
|
'\u2F25': '\u5973', |
|
'\u2F26': '\u5B50', |
|
'\u2F27': '\u5B80', |
|
'\u2F28': '\u5BF8', |
|
'\u2F29': '\u5C0F', |
|
'\u2F2A': '\u5C22', |
|
'\u2F2B': '\u5C38', |
|
'\u2F2C': '\u5C6E', |
|
'\u2F2D': '\u5C71', |
|
'\u2F2E': '\u5DDB', |
|
'\u2F2F': '\u5DE5', |
|
'\u2F30': '\u5DF1', |
|
'\u2F31': '\u5DFE', |
|
'\u2F32': '\u5E72', |
|
'\u2F33': '\u5E7A', |
|
'\u2F34': '\u5E7F', |
|
'\u2F35': '\u5EF4', |
|
'\u2F36': '\u5EFE', |
|
'\u2F37': '\u5F0B', |
|
'\u2F38': '\u5F13', |
|
'\u2F39': '\u5F50', |
|
'\u2F3A': '\u5F61', |
|
'\u2F3B': '\u5F73', |
|
'\u2F3C': '\u5FC3', |
|
'\u2F3D': '\u6208', |
|
'\u2F3E': '\u6236', |
|
'\u2F3F': '\u624B', |
|
'\u2F40': '\u652F', |
|
'\u2F41': '\u6534', |
|
'\u2F42': '\u6587', |
|
'\u2F43': '\u6597', |
|
'\u2F44': '\u65A4', |
|
'\u2F45': '\u65B9', |
|
'\u2F46': '\u65E0', |
|
'\u2F47': '\u65E5', |
|
'\u2F48': '\u66F0', |
|
'\u2F49': '\u6708', |
|
'\u2F4A': '\u6728', |
|
'\u2F4B': '\u6B20', |
|
'\u2F4C': '\u6B62', |
|
'\u2F4D': '\u6B79', |
|
'\u2F4E': '\u6BB3', |
|
'\u2F4F': '\u6BCB', |
|
'\u2F50': '\u6BD4', |
|
'\u2F51': '\u6BDB', |
|
'\u2F52': '\u6C0F', |
|
'\u2F53': '\u6C14', |
|
'\u2F54': '\u6C34', |
|
'\u2F55': '\u706B', |
|
'\u2F56': '\u722A', |
|
'\u2F57': '\u7236', |
|
'\u2F58': '\u723B', |
|
'\u2F59': '\u723F', |
|
'\u2F5A': '\u7247', |
|
'\u2F5B': '\u7259', |
|
'\u2F5C': '\u725B', |
|
'\u2F5D': '\u72AC', |
|
'\u2F5E': '\u7384', |
|
'\u2F5F': '\u7389', |
|
'\u2F60': '\u74DC', |
|
'\u2F61': '\u74E6', |
|
'\u2F62': '\u7518', |
|
'\u2F63': '\u751F', |
|
'\u2F64': '\u7528', |
|
'\u2F65': '\u7530', |
|
'\u2F66': '\u758B', |
|
'\u2F67': '\u7592', |
|
'\u2F68': '\u7676', |
|
'\u2F69': '\u767D', |
|
'\u2F6A': '\u76AE', |
|
'\u2F6B': '\u76BF', |
|
'\u2F6C': '\u76EE', |
|
'\u2F6D': '\u77DB', |
|
'\u2F6E': '\u77E2', |
|
'\u2F6F': '\u77F3', |
|
'\u2F70': '\u793A', |
|
'\u2F71': '\u79B8', |
|
'\u2F72': '\u79BE', |
|
'\u2F73': '\u7A74', |
|
'\u2F74': '\u7ACB', |
|
'\u2F75': '\u7AF9', |
|
'\u2F76': '\u7C73', |
|
'\u2F77': '\u7CF8', |
|
'\u2F78': '\u7F36', |
|
'\u2F79': '\u7F51', |
|
'\u2F7A': '\u7F8A', |
|
'\u2F7B': '\u7FBD', |
|
'\u2F7C': '\u8001', |
|
'\u2F7D': '\u800C', |
|
'\u2F7E': '\u8012', |
|
'\u2F7F': '\u8033', |
|
'\u2F80': '\u807F', |
|
'\u2F81': '\u8089', |
|
'\u2F82': '\u81E3', |
|
'\u2F83': '\u81EA', |
|
'\u2F84': '\u81F3', |
|
'\u2F85': '\u81FC', |
|
'\u2F86': '\u820C', |
|
'\u2F87': '\u821B', |
|
'\u2F88': '\u821F', |
|
'\u2F89': '\u826E', |
|
'\u2F8A': '\u8272', |
|
'\u2F8B': '\u8278', |
|
'\u2F8C': '\u864D', |
|
'\u2F8D': '\u866B', |
|
'\u2F8E': '\u8840', |
|
'\u2F8F': '\u884C', |
|
'\u2F90': '\u8863', |
|
'\u2F91': '\u897E', |
|
'\u2F92': '\u898B', |
|
'\u2F93': '\u89D2', |
|
'\u2F94': '\u8A00', |
|
'\u2F95': '\u8C37', |
|
'\u2F96': '\u8C46', |
|
'\u2F97': '\u8C55', |
|
'\u2F98': '\u8C78', |
|
'\u2F99': '\u8C9D', |
|
'\u2F9A': '\u8D64', |
|
'\u2F9B': '\u8D70', |
|
'\u2F9C': '\u8DB3', |
|
'\u2F9D': '\u8EAB', |
|
'\u2F9E': '\u8ECA', |
|
'\u2F9F': '\u8F9B', |
|
'\u2FA0': '\u8FB0', |
|
'\u2FA1': '\u8FB5', |
|
'\u2FA2': '\u9091', |
|
'\u2FA3': '\u9149', |
|
'\u2FA4': '\u91C6', |
|
'\u2FA5': '\u91CC', |
|
'\u2FA6': '\u91D1', |
|
'\u2FA7': '\u9577', |
|
'\u2FA8': '\u9580', |
|
'\u2FA9': '\u961C', |
|
'\u2FAA': '\u96B6', |
|
'\u2FAB': '\u96B9', |
|
'\u2FAC': '\u96E8', |
|
'\u2FAD': '\u9751', |
|
'\u2FAE': '\u975E', |
|
'\u2FAF': '\u9762', |
|
'\u2FB0': '\u9769', |
|
'\u2FB1': '\u97CB', |
|
'\u2FB2': '\u97ED', |
|
'\u2FB3': '\u97F3', |
|
'\u2FB4': '\u9801', |
|
'\u2FB5': '\u98A8', |
|
'\u2FB6': '\u98DB', |
|
'\u2FB7': '\u98DF', |
|
'\u2FB8': '\u9996', |
|
'\u2FB9': '\u9999', |
|
'\u2FBA': '\u99AC', |
|
'\u2FBB': '\u9AA8', |
|
'\u2FBC': '\u9AD8', |
|
'\u2FBD': '\u9ADF', |
|
'\u2FBE': '\u9B25', |
|
'\u2FBF': '\u9B2F', |
|
'\u2FC0': '\u9B32', |
|
'\u2FC1': '\u9B3C', |
|
'\u2FC2': '\u9B5A', |
|
'\u2FC3': '\u9CE5', |
|
'\u2FC4': '\u9E75', |
|
'\u2FC5': '\u9E7F', |
|
'\u2FC6': '\u9EA5', |
|
'\u2FC7': '\u9EBB', |
|
'\u2FC8': '\u9EC3', |
|
'\u2FC9': '\u9ECD', |
|
'\u2FCA': '\u9ED1', |
|
'\u2FCB': '\u9EF9', |
|
'\u2FCC': '\u9EFD', |
|
'\u2FCD': '\u9F0E', |
|
'\u2FCE': '\u9F13', |
|
'\u2FCF': '\u9F20', |
|
'\u2FD0': '\u9F3B', |
|
'\u2FD1': '\u9F4A', |
|
'\u2FD2': '\u9F52', |
|
'\u2FD3': '\u9F8D', |
|
'\u2FD4': '\u9F9C', |
|
'\u2FD5': '\u9FA0', |
|
'\u3036': '\u3012', |
|
'\u3038': '\u5341', |
|
'\u3039': '\u5344', |
|
'\u303A': '\u5345', |
|
'\u309B': '\u0020\u3099', |
|
'\u309C': '\u0020\u309A', |
|
'\u3131': '\u1100', |
|
'\u3132': '\u1101', |
|
'\u3133': '\u11AA', |
|
'\u3134': '\u1102', |
|
'\u3135': '\u11AC', |
|
'\u3136': '\u11AD', |
|
'\u3137': '\u1103', |
|
'\u3138': '\u1104', |
|
'\u3139': '\u1105', |
|
'\u313A': '\u11B0', |
|
'\u313B': '\u11B1', |
|
'\u313C': '\u11B2', |
|
'\u313D': '\u11B3', |
|
'\u313E': '\u11B4', |
|
'\u313F': '\u11B5', |
|
'\u3140': '\u111A', |
|
'\u3141': '\u1106', |
|
'\u3142': '\u1107', |
|
'\u3143': '\u1108', |
|
'\u3144': '\u1121', |
|
'\u3145': '\u1109', |
|
'\u3146': '\u110A', |
|
'\u3147': '\u110B', |
|
'\u3148': '\u110C', |
|
'\u3149': '\u110D', |
|
'\u314A': '\u110E', |
|
'\u314B': '\u110F', |
|
'\u314C': '\u1110', |
|
'\u314D': '\u1111', |
|
'\u314E': '\u1112', |
|
'\u314F': '\u1161', |
|
'\u3150': '\u1162', |
|
'\u3151': '\u1163', |
|
'\u3152': '\u1164', |
|
'\u3153': '\u1165', |
|
'\u3154': '\u1166', |
|
'\u3155': '\u1167', |
|
'\u3156': '\u1168', |
|
'\u3157': '\u1169', |
|
'\u3158': '\u116A', |
|
'\u3159': '\u116B', |
|
'\u315A': '\u116C', |
|
'\u315B': '\u116D', |
|
'\u315C': '\u116E', |
|
'\u315D': '\u116F', |
|
'\u315E': '\u1170', |
|
'\u315F': '\u1171', |
|
'\u3160': '\u1172', |
|
'\u3161': '\u1173', |
|
'\u3162': '\u1174', |
|
'\u3163': '\u1175', |
|
'\u3164': '\u1160', |
|
'\u3165': '\u1114', |
|
'\u3166': '\u1115', |
|
'\u3167': '\u11C7', |
|
'\u3168': '\u11C8', |
|
'\u3169': '\u11CC', |
|
'\u316A': '\u11CE', |
|
'\u316B': '\u11D3', |
|
'\u316C': '\u11D7', |
|
'\u316D': '\u11D9', |
|
'\u316E': '\u111C', |
|
'\u316F': '\u11DD', |
|
'\u3170': '\u11DF', |
|
'\u3171': '\u111D', |
|
'\u3172': '\u111E', |
|
'\u3173': '\u1120', |
|
'\u3174': '\u1122', |
|
'\u3175': '\u1123', |
|
'\u3176': '\u1127', |
|
'\u3177': '\u1129', |
|
'\u3178': '\u112B', |
|
'\u3179': '\u112C', |
|
'\u317A': '\u112D', |
|
'\u317B': '\u112E', |
|
'\u317C': '\u112F', |
|
'\u317D': '\u1132', |
|
'\u317E': '\u1136', |
|
'\u317F': '\u1140', |
|
'\u3180': '\u1147', |
|
'\u3181': '\u114C', |
|
'\u3182': '\u11F1', |
|
'\u3183': '\u11F2', |
|
'\u3184': '\u1157', |
|
'\u3185': '\u1158', |
|
'\u3186': '\u1159', |
|
'\u3187': '\u1184', |
|
'\u3188': '\u1185', |
|
'\u3189': '\u1188', |
|
'\u318A': '\u1191', |
|
'\u318B': '\u1192', |
|
'\u318C': '\u1194', |
|
'\u318D': '\u119E', |
|
'\u318E': '\u11A1', |
|
'\u3200': '\u0028\u1100\u0029', |
|
'\u3201': '\u0028\u1102\u0029', |
|
'\u3202': '\u0028\u1103\u0029', |
|
'\u3203': '\u0028\u1105\u0029', |
|
'\u3204': '\u0028\u1106\u0029', |
|
'\u3205': '\u0028\u1107\u0029', |
|
'\u3206': '\u0028\u1109\u0029', |
|
'\u3207': '\u0028\u110B\u0029', |
|
'\u3208': '\u0028\u110C\u0029', |
|
'\u3209': '\u0028\u110E\u0029', |
|
'\u320A': '\u0028\u110F\u0029', |
|
'\u320B': '\u0028\u1110\u0029', |
|
'\u320C': '\u0028\u1111\u0029', |
|
'\u320D': '\u0028\u1112\u0029', |
|
'\u320E': '\u0028\u1100\u1161\u0029', |
|
'\u320F': '\u0028\u1102\u1161\u0029', |
|
'\u3210': '\u0028\u1103\u1161\u0029', |
|
'\u3211': '\u0028\u1105\u1161\u0029', |
|
'\u3212': '\u0028\u1106\u1161\u0029', |
|
'\u3213': '\u0028\u1107\u1161\u0029', |
|
'\u3214': '\u0028\u1109\u1161\u0029', |
|
'\u3215': '\u0028\u110B\u1161\u0029', |
|
'\u3216': '\u0028\u110C\u1161\u0029', |
|
'\u3217': '\u0028\u110E\u1161\u0029', |
|
'\u3218': '\u0028\u110F\u1161\u0029', |
|
'\u3219': '\u0028\u1110\u1161\u0029', |
|
'\u321A': '\u0028\u1111\u1161\u0029', |
|
'\u321B': '\u0028\u1112\u1161\u0029', |
|
'\u321C': '\u0028\u110C\u116E\u0029', |
|
'\u321D': '\u0028\u110B\u1169\u110C\u1165\u11AB\u0029', |
|
'\u321E': '\u0028\u110B\u1169\u1112\u116E\u0029', |
|
'\u3220': '\u0028\u4E00\u0029', |
|
'\u3221': '\u0028\u4E8C\u0029', |
|
'\u3222': '\u0028\u4E09\u0029', |
|
'\u3223': '\u0028\u56DB\u0029', |
|
'\u3224': '\u0028\u4E94\u0029', |
|
'\u3225': '\u0028\u516D\u0029', |
|
'\u3226': '\u0028\u4E03\u0029', |
|
'\u3227': '\u0028\u516B\u0029', |
|
'\u3228': '\u0028\u4E5D\u0029', |
|
'\u3229': '\u0028\u5341\u0029', |
|
'\u322A': '\u0028\u6708\u0029', |
|
'\u322B': '\u0028\u706B\u0029', |
|
'\u322C': '\u0028\u6C34\u0029', |
|
'\u322D': '\u0028\u6728\u0029', |
|
'\u322E': '\u0028\u91D1\u0029', |
|
'\u322F': '\u0028\u571F\u0029', |
|
'\u3230': '\u0028\u65E5\u0029', |
|
'\u3231': '\u0028\u682A\u0029', |
|
'\u3232': '\u0028\u6709\u0029', |
|
'\u3233': '\u0028\u793E\u0029', |
|
'\u3234': '\u0028\u540D\u0029', |
|
'\u3235': '\u0028\u7279\u0029', |
|
'\u3236': '\u0028\u8CA1\u0029', |
|
'\u3237': '\u0028\u795D\u0029', |
|
'\u3238': '\u0028\u52B4\u0029', |
|
'\u3239': '\u0028\u4EE3\u0029', |
|
'\u323A': '\u0028\u547C\u0029', |
|
'\u323B': '\u0028\u5B66\u0029', |
|
'\u323C': '\u0028\u76E3\u0029', |
|
'\u323D': '\u0028\u4F01\u0029', |
|
'\u323E': '\u0028\u8CC7\u0029', |
|
'\u323F': '\u0028\u5354\u0029', |
|
'\u3240': '\u0028\u796D\u0029', |
|
'\u3241': '\u0028\u4F11\u0029', |
|
'\u3242': '\u0028\u81EA\u0029', |
|
'\u3243': '\u0028\u81F3\u0029', |
|
'\u32C0': '\u0031\u6708', |
|
'\u32C1': '\u0032\u6708', |
|
'\u32C2': '\u0033\u6708', |
|
'\u32C3': '\u0034\u6708', |
|
'\u32C4': '\u0035\u6708', |
|
'\u32C5': '\u0036\u6708', |
|
'\u32C6': '\u0037\u6708', |
|
'\u32C7': '\u0038\u6708', |
|
'\u32C8': '\u0039\u6708', |
|
'\u32C9': '\u0031\u0030\u6708', |
|
'\u32CA': '\u0031\u0031\u6708', |
|
'\u32CB': '\u0031\u0032\u6708', |
|
'\u3358': '\u0030\u70B9', |
|
'\u3359': '\u0031\u70B9', |
|
'\u335A': '\u0032\u70B9', |
|
'\u335B': '\u0033\u70B9', |
|
'\u335C': '\u0034\u70B9', |
|
'\u335D': '\u0035\u70B9', |
|
'\u335E': '\u0036\u70B9', |
|
'\u335F': '\u0037\u70B9', |
|
'\u3360': '\u0038\u70B9', |
|
'\u3361': '\u0039\u70B9', |
|
'\u3362': '\u0031\u0030\u70B9', |
|
'\u3363': '\u0031\u0031\u70B9', |
|
'\u3364': '\u0031\u0032\u70B9', |
|
'\u3365': '\u0031\u0033\u70B9', |
|
'\u3366': '\u0031\u0034\u70B9', |
|
'\u3367': '\u0031\u0035\u70B9', |
|
'\u3368': '\u0031\u0036\u70B9', |
|
'\u3369': '\u0031\u0037\u70B9', |
|
'\u336A': '\u0031\u0038\u70B9', |
|
'\u336B': '\u0031\u0039\u70B9', |
|
'\u336C': '\u0032\u0030\u70B9', |
|
'\u336D': '\u0032\u0031\u70B9', |
|
'\u336E': '\u0032\u0032\u70B9', |
|
'\u336F': '\u0032\u0033\u70B9', |
|
'\u3370': '\u0032\u0034\u70B9', |
|
'\u33E0': '\u0031\u65E5', |
|
'\u33E1': '\u0032\u65E5', |
|
'\u33E2': '\u0033\u65E5', |
|
'\u33E3': '\u0034\u65E5', |
|
'\u33E4': '\u0035\u65E5', |
|
'\u33E5': '\u0036\u65E5', |
|
'\u33E6': '\u0037\u65E5', |
|
'\u33E7': '\u0038\u65E5', |
|
'\u33E8': '\u0039\u65E5', |
|
'\u33E9': '\u0031\u0030\u65E5', |
|
'\u33EA': '\u0031\u0031\u65E5', |
|
'\u33EB': '\u0031\u0032\u65E5', |
|
'\u33EC': '\u0031\u0033\u65E5', |
|
'\u33ED': '\u0031\u0034\u65E5', |
|
'\u33EE': '\u0031\u0035\u65E5', |
|
'\u33EF': '\u0031\u0036\u65E5', |
|
'\u33F0': '\u0031\u0037\u65E5', |
|
'\u33F1': '\u0031\u0038\u65E5', |
|
'\u33F2': '\u0031\u0039\u65E5', |
|
'\u33F3': '\u0032\u0030\u65E5', |
|
'\u33F4': '\u0032\u0031\u65E5', |
|
'\u33F5': '\u0032\u0032\u65E5', |
|
'\u33F6': '\u0032\u0033\u65E5', |
|
'\u33F7': '\u0032\u0034\u65E5', |
|
'\u33F8': '\u0032\u0035\u65E5', |
|
'\u33F9': '\u0032\u0036\u65E5', |
|
'\u33FA': '\u0032\u0037\u65E5', |
|
'\u33FB': '\u0032\u0038\u65E5', |
|
'\u33FC': '\u0032\u0039\u65E5', |
|
'\u33FD': '\u0033\u0030\u65E5', |
|
'\u33FE': '\u0033\u0031\u65E5', |
|
'\uFB00': '\u0066\u0066', |
|
'\uFB01': '\u0066\u0069', |
|
'\uFB02': '\u0066\u006C', |
|
'\uFB03': '\u0066\u0066\u0069', |
|
'\uFB04': '\u0066\u0066\u006C', |
|
'\uFB05': '\u017F\u0074', |
|
'\uFB06': '\u0073\u0074', |
|
'\uFB13': '\u0574\u0576', |
|
'\uFB14': '\u0574\u0565', |
|
'\uFB15': '\u0574\u056B', |
|
'\uFB16': '\u057E\u0576', |
|
'\uFB17': '\u0574\u056D', |
|
'\uFB4F': '\u05D0\u05DC', |
|
'\uFB50': '\u0671', |
|
'\uFB51': '\u0671', |
|
'\uFB52': '\u067B', |
|
'\uFB53': '\u067B', |
|
'\uFB54': '\u067B', |
|
'\uFB55': '\u067B', |
|
'\uFB56': '\u067E', |
|
'\uFB57': '\u067E', |
|
'\uFB58': '\u067E', |
|
'\uFB59': '\u067E', |
|
'\uFB5A': '\u0680', |
|
'\uFB5B': '\u0680', |
|
'\uFB5C': '\u0680', |
|
'\uFB5D': '\u0680', |
|
'\uFB5E': '\u067A', |
|
'\uFB5F': '\u067A', |
|
'\uFB60': '\u067A', |
|
'\uFB61': '\u067A', |
|
'\uFB62': '\u067F', |
|
'\uFB63': '\u067F', |
|
'\uFB64': '\u067F', |
|
'\uFB65': '\u067F', |
|
'\uFB66': '\u0679', |
|
'\uFB67': '\u0679', |
|
'\uFB68': '\u0679', |
|
'\uFB69': '\u0679', |
|
'\uFB6A': '\u06A4', |
|
'\uFB6B': '\u06A4', |
|
'\uFB6C': '\u06A4', |
|
'\uFB6D': '\u06A4', |
|
'\uFB6E': '\u06A6', |
|
'\uFB6F': '\u06A6', |
|
'\uFB70': '\u06A6', |
|
'\uFB71': '\u06A6', |
|
'\uFB72': '\u0684', |
|
'\uFB73': '\u0684', |
|
'\uFB74': '\u0684', |
|
'\uFB75': '\u0684', |
|
'\uFB76': '\u0683', |
|
'\uFB77': '\u0683', |
|
'\uFB78': '\u0683', |
|
'\uFB79': '\u0683', |
|
'\uFB7A': '\u0686', |
|
'\uFB7B': '\u0686', |
|
'\uFB7C': '\u0686', |
|
'\uFB7D': '\u0686', |
|
'\uFB7E': '\u0687', |
|
'\uFB7F': '\u0687', |
|
'\uFB80': '\u0687', |
|
'\uFB81': '\u0687', |
|
'\uFB82': '\u068D', |
|
'\uFB83': '\u068D', |
|
'\uFB84': '\u068C', |
|
'\uFB85': '\u068C', |
|
'\uFB86': '\u068E', |
|
'\uFB87': '\u068E', |
|
'\uFB88': '\u0688', |
|
'\uFB89': '\u0688', |
|
'\uFB8A': '\u0698', |
|
'\uFB8B': '\u0698', |
|
'\uFB8C': '\u0691', |
|
'\uFB8D': '\u0691', |
|
'\uFB8E': '\u06A9', |
|
'\uFB8F': '\u06A9', |
|
'\uFB90': '\u06A9', |
|
'\uFB91': '\u06A9', |
|
'\uFB92': '\u06AF', |
|
'\uFB93': '\u06AF', |
|
'\uFB94': '\u06AF', |
|
'\uFB95': '\u06AF', |
|
'\uFB96': '\u06B3', |
|
'\uFB97': '\u06B3', |
|
'\uFB98': '\u06B3', |
|
'\uFB99': '\u06B3', |
|
'\uFB9A': '\u06B1', |
|
'\uFB9B': '\u06B1', |
|
'\uFB9C': '\u06B1', |
|
'\uFB9D': '\u06B1', |
|
'\uFB9E': '\u06BA', |
|
'\uFB9F': '\u06BA', |
|
'\uFBA0': '\u06BB', |
|
'\uFBA1': '\u06BB', |
|
'\uFBA2': '\u06BB', |
|
'\uFBA3': '\u06BB', |
|
'\uFBA4': '\u06C0', |
|
'\uFBA5': '\u06C0', |
|
'\uFBA6': '\u06C1', |
|
'\uFBA7': '\u06C1', |
|
'\uFBA8': '\u06C1', |
|
'\uFBA9': '\u06C1', |
|
'\uFBAA': '\u06BE', |
|
'\uFBAB': '\u06BE', |
|
'\uFBAC': '\u06BE', |
|
'\uFBAD': '\u06BE', |
|
'\uFBAE': '\u06D2', |
|
'\uFBAF': '\u06D2', |
|
'\uFBB0': '\u06D3', |
|
'\uFBB1': '\u06D3', |
|
'\uFBD3': '\u06AD', |
|
'\uFBD4': '\u06AD', |
|
'\uFBD5': '\u06AD', |
|
'\uFBD6': '\u06AD', |
|
'\uFBD7': '\u06C7', |
|
'\uFBD8': '\u06C7', |
|
'\uFBD9': '\u06C6', |
|
'\uFBDA': '\u06C6', |
|
'\uFBDB': '\u06C8', |
|
'\uFBDC': '\u06C8', |
|
'\uFBDD': '\u0677', |
|
'\uFBDE': '\u06CB', |
|
'\uFBDF': '\u06CB', |
|
'\uFBE0': '\u06C5', |
|
'\uFBE1': '\u06C5', |
|
'\uFBE2': '\u06C9', |
|
'\uFBE3': '\u06C9', |
|
'\uFBE4': '\u06D0', |
|
'\uFBE5': '\u06D0', |
|
'\uFBE6': '\u06D0', |
|
'\uFBE7': '\u06D0', |
|
'\uFBE8': '\u0649', |
|
'\uFBE9': '\u0649', |
|
'\uFBEA': '\u0626\u0627', |
|
'\uFBEB': '\u0626\u0627', |
|
'\uFBEC': '\u0626\u06D5', |
|
'\uFBED': '\u0626\u06D5', |
|
'\uFBEE': '\u0626\u0648', |
|
'\uFBEF': '\u0626\u0648', |
|
'\uFBF0': '\u0626\u06C7', |
|
'\uFBF1': '\u0626\u06C7', |
|
'\uFBF2': '\u0626\u06C6', |
|
'\uFBF3': '\u0626\u06C6', |
|
'\uFBF4': '\u0626\u06C8', |
|
'\uFBF5': '\u0626\u06C8', |
|
'\uFBF6': '\u0626\u06D0', |
|
'\uFBF7': '\u0626\u06D0', |
|
'\uFBF8': '\u0626\u06D0', |
|
'\uFBF9': '\u0626\u0649', |
|
'\uFBFA': '\u0626\u0649', |
|
'\uFBFB': '\u0626\u0649', |
|
'\uFBFC': '\u06CC', |
|
'\uFBFD': '\u06CC', |
|
'\uFBFE': '\u06CC', |
|
'\uFBFF': '\u06CC', |
|
'\uFC00': '\u0626\u062C', |
|
'\uFC01': '\u0626\u062D', |
|
'\uFC02': '\u0626\u0645', |
|
'\uFC03': '\u0626\u0649', |
|
'\uFC04': '\u0626\u064A', |
|
'\uFC05': '\u0628\u062C', |
|
'\uFC06': '\u0628\u062D', |
|
'\uFC07': '\u0628\u062E', |
|
'\uFC08': '\u0628\u0645', |
|
'\uFC09': '\u0628\u0649', |
|
'\uFC0A': '\u0628\u064A', |
|
'\uFC0B': '\u062A\u062C', |
|
'\uFC0C': '\u062A\u062D', |
|
'\uFC0D': '\u062A\u062E', |
|
'\uFC0E': '\u062A\u0645', |
|
'\uFC0F': '\u062A\u0649', |
|
'\uFC10': '\u062A\u064A', |
|
'\uFC11': '\u062B\u062C', |
|
'\uFC12': '\u062B\u0645', |
|
'\uFC13': '\u062B\u0649', |
|
'\uFC14': '\u062B\u064A', |
|
'\uFC15': '\u062C\u062D', |
|
'\uFC16': '\u062C\u0645', |
|
'\uFC17': '\u062D\u062C', |
|
'\uFC18': '\u062D\u0645', |
|
'\uFC19': '\u062E\u062C', |
|
'\uFC1A': '\u062E\u062D', |
|
'\uFC1B': '\u062E\u0645', |
|
'\uFC1C': '\u0633\u062C', |
|
'\uFC1D': '\u0633\u062D', |
|
'\uFC1E': '\u0633\u062E', |
|
'\uFC1F': '\u0633\u0645', |
|
'\uFC20': '\u0635\u062D', |
|
'\uFC21': '\u0635\u0645', |
|
'\uFC22': '\u0636\u062C', |
|
'\uFC23': '\u0636\u062D', |
|
'\uFC24': '\u0636\u062E', |
|
'\uFC25': '\u0636\u0645', |
|
'\uFC26': '\u0637\u062D', |
|
'\uFC27': '\u0637\u0645', |
|
'\uFC28': '\u0638\u0645', |
|
'\uFC29': '\u0639\u062C', |
|
'\uFC2A': '\u0639\u0645', |
|
'\uFC2B': '\u063A\u062C', |
|
'\uFC2C': '\u063A\u0645', |
|
'\uFC2D': '\u0641\u062C', |
|
'\uFC2E': '\u0641\u062D', |
|
'\uFC2F': '\u0641\u062E', |
|
'\uFC30': '\u0641\u0645', |
|
'\uFC31': '\u0641\u0649', |
|
'\uFC32': '\u0641\u064A', |
|
'\uFC33': '\u0642\u062D', |
|
'\uFC34': '\u0642\u0645', |
|
'\uFC35': '\u0642\u0649', |
|
'\uFC36': '\u0642\u064A', |
|
'\uFC37': '\u0643\u0627', |
|
'\uFC38': '\u0643\u062C', |
|
'\uFC39': '\u0643\u062D', |
|
'\uFC3A': '\u0643\u062E', |
|
'\uFC3B': '\u0643\u0644', |
|
'\uFC3C': '\u0643\u0645', |
|
'\uFC3D': '\u0643\u0649', |
|
'\uFC3E': '\u0643\u064A', |
|
'\uFC3F': '\u0644\u062C', |
|
'\uFC40': '\u0644\u062D', |
|
'\uFC41': '\u0644\u062E', |
|
'\uFC42': '\u0644\u0645', |
|
'\uFC43': '\u0644\u0649', |
|
'\uFC44': '\u0644\u064A', |
|
'\uFC45': '\u0645\u062C', |
|
'\uFC46': '\u0645\u062D', |
|
'\uFC47': '\u0645\u062E', |
|
'\uFC48': '\u0645\u0645', |
|
'\uFC49': '\u0645\u0649', |
|
'\uFC4A': '\u0645\u064A', |
|
'\uFC4B': '\u0646\u062C', |
|
'\uFC4C': '\u0646\u062D', |
|
'\uFC4D': '\u0646\u062E', |
|
'\uFC4E': '\u0646\u0645', |
|
'\uFC4F': '\u0646\u0649', |
|
'\uFC50': '\u0646\u064A', |
|
'\uFC51': '\u0647\u062C', |
|
'\uFC52': '\u0647\u0645', |
|
'\uFC53': '\u0647\u0649', |
|
'\uFC54': '\u0647\u064A', |
|
'\uFC55': '\u064A\u062C', |
|
'\uFC56': '\u064A\u062D', |
|
'\uFC57': '\u064A\u062E', |
|
'\uFC58': '\u064A\u0645', |
|
'\uFC59': '\u064A\u0649', |
|
'\uFC5A': '\u064A\u064A', |
|
'\uFC5B': '\u0630\u0670', |
|
'\uFC5C': '\u0631\u0670', |
|
'\uFC5D': '\u0649\u0670', |
|
'\uFC5E': '\u0020\u064C\u0651', |
|
'\uFC5F': '\u0020\u064D\u0651', |
|
'\uFC60': '\u0020\u064E\u0651', |
|
'\uFC61': '\u0020\u064F\u0651', |
|
'\uFC62': '\u0020\u0650\u0651', |
|
'\uFC63': '\u0020\u0651\u0670', |
|
'\uFC64': '\u0626\u0631', |
|
'\uFC65': '\u0626\u0632', |
|
'\uFC66': '\u0626\u0645', |
|
'\uFC67': '\u0626\u0646', |
|
'\uFC68': '\u0626\u0649', |
|
'\uFC69': '\u0626\u064A', |
|
'\uFC6A': '\u0628\u0631', |
|
'\uFC6B': '\u0628\u0632', |
|
'\uFC6C': '\u0628\u0645', |
|
'\uFC6D': '\u0628\u0646', |
|
'\uFC6E': '\u0628\u0649', |
|
'\uFC6F': '\u0628\u064A', |
|
'\uFC70': '\u062A\u0631', |
|
'\uFC71': '\u062A\u0632', |
|
'\uFC72': '\u062A\u0645', |
|
'\uFC73': '\u062A\u0646', |
|
'\uFC74': '\u062A\u0649', |
|
'\uFC75': '\u062A\u064A', |
|
'\uFC76': '\u062B\u0631', |
|
'\uFC77': '\u062B\u0632', |
|
'\uFC78': '\u062B\u0645', |
|
'\uFC79': '\u062B\u0646', |
|
'\uFC7A': '\u062B\u0649', |
|
'\uFC7B': '\u062B\u064A', |
|
'\uFC7C': '\u0641\u0649', |
|
'\uFC7D': '\u0641\u064A', |
|
'\uFC7E': '\u0642\u0649', |
|
'\uFC7F': '\u0642\u064A', |
|
'\uFC80': '\u0643\u0627', |
|
'\uFC81': '\u0643\u0644', |
|
'\uFC82': '\u0643\u0645', |
|
'\uFC83': '\u0643\u0649', |
|
'\uFC84': '\u0643\u064A', |
|
'\uFC85': '\u0644\u0645', |
|
'\uFC86': '\u0644\u0649', |
|
'\uFC87': '\u0644\u064A', |
|
'\uFC88': '\u0645\u0627', |
|
'\uFC89': '\u0645\u0645', |
|
'\uFC8A': '\u0646\u0631', |
|
'\uFC8B': '\u0646\u0632', |
|
'\uFC8C': '\u0646\u0645', |
|
'\uFC8D': '\u0646\u0646', |
|
'\uFC8E': '\u0646\u0649', |
|
'\uFC8F': '\u0646\u064A', |
|
'\uFC90': '\u0649\u0670', |
|
'\uFC91': '\u064A\u0631', |
|
'\uFC92': '\u064A\u0632', |
|
'\uFC93': '\u064A\u0645', |
|
'\uFC94': '\u064A\u0646', |
|
'\uFC95': '\u064A\u0649', |
|
'\uFC96': '\u064A\u064A', |
|
'\uFC97': '\u0626\u062C', |
|
'\uFC98': '\u0626\u062D', |
|
'\uFC99': '\u0626\u062E', |
|
'\uFC9A': '\u0626\u0645', |
|
'\uFC9B': '\u0626\u0647', |
|
'\uFC9C': '\u0628\u062C', |
|
'\uFC9D': '\u0628\u062D', |
|
'\uFC9E': '\u0628\u062E', |
|
'\uFC9F': '\u0628\u0645', |
|
'\uFCA0': '\u0628\u0647', |
|
'\uFCA1': '\u062A\u062C', |
|
'\uFCA2': '\u062A\u062D', |
|
'\uFCA3': '\u062A\u062E', |
|
'\uFCA4': '\u062A\u0645', |
|
'\uFCA5': '\u062A\u0647', |
|
'\uFCA6': '\u062B\u0645', |
|
'\uFCA7': '\u062C\u062D', |
|
'\uFCA8': '\u062C\u0645', |
|
'\uFCA9': '\u062D\u062C', |
|
'\uFCAA': '\u062D\u0645', |
|
'\uFCAB': '\u062E\u062C', |
|
'\uFCAC': '\u062E\u0645', |
|
'\uFCAD': '\u0633\u062C', |
|
'\uFCAE': '\u0633\u062D', |
|
'\uFCAF': '\u0633\u062E', |
|
'\uFCB0': '\u0633\u0645', |
|
'\uFCB1': '\u0635\u062D', |
|
'\uFCB2': '\u0635\u062E', |
|
'\uFCB3': '\u0635\u0645', |
|
'\uFCB4': '\u0636\u062C', |
|
'\uFCB5': '\u0636\u062D', |
|
'\uFCB6': '\u0636\u062E', |
|
'\uFCB7': '\u0636\u0645', |
|
'\uFCB8': '\u0637\u062D', |
|
'\uFCB9': '\u0638\u0645', |
|
'\uFCBA': '\u0639\u062C', |
|
'\uFCBB': '\u0639\u0645', |
|
'\uFCBC': '\u063A\u062C', |
|
'\uFCBD': '\u063A\u0645', |
|
'\uFCBE': '\u0641\u062C', |
|
'\uFCBF': '\u0641\u062D', |
|
'\uFCC0': '\u0641\u062E', |
|
'\uFCC1': '\u0641\u0645', |
|
'\uFCC2': '\u0642\u062D', |
|
'\uFCC3': '\u0642\u0645', |
|
'\uFCC4': '\u0643\u062C', |
|
'\uFCC5': '\u0643\u062D', |
|
'\uFCC6': '\u0643\u062E', |
|
'\uFCC7': '\u0643\u0644', |
|
'\uFCC8': '\u0643\u0645', |
|
'\uFCC9': '\u0644\u062C', |
|
'\uFCCA': '\u0644\u062D', |
|
'\uFCCB': '\u0644\u062E', |
|
'\uFCCC': '\u0644\u0645', |
|
'\uFCCD': '\u0644\u0647', |
|
'\uFCCE': '\u0645\u062C', |
|
'\uFCCF': '\u0645\u062D', |
|
'\uFCD0': '\u0645\u062E', |
|
'\uFCD1': '\u0645\u0645', |
|
'\uFCD2': '\u0646\u062C', |
|
'\uFCD3': '\u0646\u062D', |
|
'\uFCD4': '\u0646\u062E', |
|
'\uFCD5': '\u0646\u0645', |
|
'\uFCD6': '\u0646\u0647', |
|
'\uFCD7': '\u0647\u062C', |
|
'\uFCD8': '\u0647\u0645', |
|
'\uFCD9': '\u0647\u0670', |
|
'\uFCDA': '\u064A\u062C', |
|
'\uFCDB': '\u064A\u062D', |
|
'\uFCDC': '\u064A\u062E', |
|
'\uFCDD': '\u064A\u0645', |
|
'\uFCDE': '\u064A\u0647', |
|
'\uFCDF': '\u0626\u0645', |
|
'\uFCE0': '\u0626\u0647', |
|
'\uFCE1': '\u0628\u0645', |
|
'\uFCE2': '\u0628\u0647', |
|
'\uFCE3': '\u062A\u0645', |
|
'\uFCE4': '\u062A\u0647', |
|
'\uFCE5': '\u062B\u0645', |
|
'\uFCE6': '\u062B\u0647', |
|
'\uFCE7': '\u0633\u0645', |
|
'\uFCE8': '\u0633\u0647', |
|
'\uFCE9': '\u0634\u0645', |
|
'\uFCEA': '\u0634\u0647', |
|
'\uFCEB': '\u0643\u0644', |
|
'\uFCEC': '\u0643\u0645', |
|
'\uFCED': '\u0644\u0645', |
|
'\uFCEE': '\u0646\u0645', |
|
'\uFCEF': '\u0646\u0647', |
|
'\uFCF0': '\u064A\u0645', |
|
'\uFCF1': '\u064A\u0647', |
|
'\uFCF2': '\u0640\u064E\u0651', |
|
'\uFCF3': '\u0640\u064F\u0651', |
|
'\uFCF4': '\u0640\u0650\u0651', |
|
'\uFCF5': '\u0637\u0649', |
|
'\uFCF6': '\u0637\u064A', |
|
'\uFCF7': '\u0639\u0649', |
|
'\uFCF8': '\u0639\u064A', |
|
'\uFCF9': '\u063A\u0649', |
|
'\uFCFA': '\u063A\u064A', |
|
'\uFCFB': '\u0633\u0649', |
|
'\uFCFC': '\u0633\u064A', |
|
'\uFCFD': '\u0634\u0649', |
|
'\uFCFE': '\u0634\u064A', |
|
'\uFCFF': '\u062D\u0649', |
|
'\uFD00': '\u062D\u064A', |
|
'\uFD01': '\u062C\u0649', |
|
'\uFD02': '\u062C\u064A', |
|
'\uFD03': '\u062E\u0649', |
|
'\uFD04': '\u062E\u064A', |
|
'\uFD05': '\u0635\u0649', |
|
'\uFD06': '\u0635\u064A', |
|
'\uFD07': '\u0636\u0649', |
|
'\uFD08': '\u0636\u064A', |
|
'\uFD09': '\u0634\u062C', |
|
'\uFD0A': '\u0634\u062D', |
|
'\uFD0B': '\u0634\u062E', |
|
'\uFD0C': '\u0634\u0645', |
|
'\uFD0D': '\u0634\u0631', |
|
'\uFD0E': '\u0633\u0631', |
|
'\uFD0F': '\u0635\u0631', |
|
'\uFD10': '\u0636\u0631', |
|
'\uFD11': '\u0637\u0649', |
|
'\uFD12': '\u0637\u064A', |
|
'\uFD13': '\u0639\u0649', |
|
'\uFD14': '\u0639\u064A', |
|
'\uFD15': '\u063A\u0649', |
|
'\uFD16': '\u063A\u064A', |
|
'\uFD17': '\u0633\u0649', |
|
'\uFD18': '\u0633\u064A', |
|
'\uFD19': '\u0634\u0649', |
|
'\uFD1A': '\u0634\u064A', |
|
'\uFD1B': '\u062D\u0649', |
|
'\uFD1C': '\u062D\u064A', |
|
'\uFD1D': '\u062C\u0649', |
|
'\uFD1E': '\u062C\u064A', |
|
'\uFD1F': '\u062E\u0649', |
|
'\uFD20': '\u062E\u064A', |
|
'\uFD21': '\u0635\u0649', |
|
'\uFD22': '\u0635\u064A', |
|
'\uFD23': '\u0636\u0649', |
|
'\uFD24': '\u0636\u064A', |
|
'\uFD25': '\u0634\u062C', |
|
'\uFD26': '\u0634\u062D', |
|
'\uFD27': '\u0634\u062E', |
|
'\uFD28': '\u0634\u0645', |
|
'\uFD29': '\u0634\u0631', |
|
'\uFD2A': '\u0633\u0631', |
|
'\uFD2B': '\u0635\u0631', |
|
'\uFD2C': '\u0636\u0631', |
|
'\uFD2D': '\u0634\u062C', |
|
'\uFD2E': '\u0634\u062D', |
|
'\uFD2F': '\u0634\u062E', |
|
'\uFD30': '\u0634\u0645', |
|
'\uFD31': '\u0633\u0647', |
|
'\uFD32': '\u0634\u0647', |
|
'\uFD33': '\u0637\u0645', |
|
'\uFD34': '\u0633\u062C', |
|
'\uFD35': '\u0633\u062D', |
|
'\uFD36': '\u0633\u062E', |
|
'\uFD37': '\u0634\u062C', |
|
'\uFD38': '\u0634\u062D', |
|
'\uFD39': '\u0634\u062E', |
|
'\uFD3A': '\u0637\u0645', |
|
'\uFD3B': '\u0638\u0645', |
|
'\uFD3C': '\u0627\u064B', |
|
'\uFD3D': '\u0627\u064B', |
|
'\uFD50': '\u062A\u062C\u0645', |
|
'\uFD51': '\u062A\u062D\u062C', |
|
'\uFD52': '\u062A\u062D\u062C', |
|
'\uFD53': '\u062A\u062D\u0645', |
|
'\uFD54': '\u062A\u062E\u0645', |
|
'\uFD55': '\u062A\u0645\u062C', |
|
'\uFD56': '\u062A\u0645\u062D', |
|
'\uFD57': '\u062A\u0645\u062E', |
|
'\uFD58': '\u062C\u0645\u062D', |
|
'\uFD59': '\u062C\u0645\u062D', |
|
'\uFD5A': '\u062D\u0645\u064A', |
|
'\uFD5B': '\u062D\u0645\u0649', |
|
'\uFD5C': '\u0633\u062D\u062C', |
|
'\uFD5D': '\u0633\u062C\u062D', |
|
'\uFD5E': '\u0633\u062C\u0649', |
|
'\uFD5F': '\u0633\u0645\u062D', |
|
'\uFD60': '\u0633\u0645\u062D', |
|
'\uFD61': '\u0633\u0645\u062C', |
|
'\uFD62': '\u0633\u0645\u0645', |
|
'\uFD63': '\u0633\u0645\u0645', |
|
'\uFD64': '\u0635\u062D\u062D', |
|
'\uFD65': '\u0635\u062D\u062D', |
|
'\uFD66': '\u0635\u0645\u0645', |
|
'\uFD67': '\u0634\u062D\u0645', |
|
'\uFD68': '\u0634\u062D\u0645', |
|
'\uFD69': '\u0634\u062C\u064A', |
|
'\uFD6A': '\u0634\u0645\u062E', |
|
'\uFD6B': '\u0634\u0645\u062E', |
|
'\uFD6C': '\u0634\u0645\u0645', |
|
'\uFD6D': '\u0634\u0645\u0645', |
|
'\uFD6E': '\u0636\u062D\u0649', |
|
'\uFD6F': '\u0636\u062E\u0645', |
|
'\uFD70': '\u0636\u062E\u0645', |
|
'\uFD71': '\u0637\u0645\u062D', |
|
'\uFD72': '\u0637\u0645\u062D', |
|
'\uFD73': '\u0637\u0645\u0645', |
|
'\uFD74': '\u0637\u0645\u064A', |
|
'\uFD75': '\u0639\u062C\u0645', |
|
'\uFD76': '\u0639\u0645\u0645', |
|
'\uFD77': '\u0639\u0645\u0645', |
|
'\uFD78': '\u0639\u0645\u0649', |
|
'\uFD79': '\u063A\u0645\u0645', |
|
'\uFD7A': '\u063A\u0645\u064A', |
|
'\uFD7B': '\u063A\u0645\u0649', |
|
'\uFD7C': '\u0641\u062E\u0645', |
|
'\uFD7D': '\u0641\u062E\u0645', |
|
'\uFD7E': '\u0642\u0645\u062D', |
|
'\uFD7F': '\u0642\u0645\u0645', |
|
'\uFD80': '\u0644\u062D\u0645', |
|
'\uFD81': '\u0644\u062D\u064A', |
|
'\uFD82': '\u0644\u062D\u0649', |
|
'\uFD83': '\u0644\u062C\u062C', |
|
'\uFD84': '\u0644\u062C\u062C', |
|
'\uFD85': '\u0644\u062E\u0645', |
|
'\uFD86': '\u0644\u062E\u0645', |
|
'\uFD87': '\u0644\u0645\u062D', |
|
'\uFD88': '\u0644\u0645\u062D', |
|
'\uFD89': '\u0645\u062D\u062C', |
|
'\uFD8A': '\u0645\u062D\u0645', |
|
'\uFD8B': '\u0645\u062D\u064A', |
|
'\uFD8C': '\u0645\u062C\u062D', |
|
'\uFD8D': '\u0645\u062C\u0645', |
|
'\uFD8E': '\u0645\u062E\u062C', |
|
'\uFD8F': '\u0645\u062E\u0645', |
|
'\uFD92': '\u0645\u062C\u062E', |
|
'\uFD93': '\u0647\u0645\u062C', |
|
'\uFD94': '\u0647\u0645\u0645', |
|
'\uFD95': '\u0646\u062D\u0645', |
|
'\uFD96': '\u0646\u062D\u0649', |
|
'\uFD97': '\u0646\u062C\u0645', |
|
'\uFD98': '\u0646\u062C\u0645', |
|
'\uFD99': '\u0646\u062C\u0649', |
|
'\uFD9A': '\u0646\u0645\u064A', |
|
'\uFD9B': '\u0646\u0645\u0649', |
|
'\uFD9C': '\u064A\u0645\u0645', |
|
'\uFD9D': '\u064A\u0645\u0645', |
|
'\uFD9E': '\u0628\u062E\u064A', |
|
'\uFD9F': '\u062A\u062C\u064A', |
|
'\uFDA0': '\u062A\u062C\u0649', |
|
'\uFDA1': '\u062A\u062E\u064A', |
|
'\uFDA2': '\u062A\u062E\u0649', |
|
'\uFDA3': '\u062A\u0645\u064A', |
|
'\uFDA4': '\u062A\u0645\u0649', |
|
'\uFDA5': '\u062C\u0645\u064A', |
|
'\uFDA6': '\u062C\u062D\u0649', |
|
'\uFDA7': '\u062C\u0645\u0649', |
|
'\uFDA8': '\u0633\u062E\u0649', |
|
'\uFDA9': '\u0635\u062D\u064A', |
|
'\uFDAA': '\u0634\u062D\u064A', |
|
'\uFDAB': '\u0636\u062D\u064A', |
|
'\uFDAC': '\u0644\u062C\u064A', |
|
'\uFDAD': '\u0644\u0645\u064A', |
|
'\uFDAE': '\u064A\u062D\u064A', |
|
'\uFDAF': '\u064A\u062C\u064A', |
|
'\uFDB0': '\u064A\u0645\u064A', |
|
'\uFDB1': '\u0645\u0645\u064A', |
|
'\uFDB2': '\u0642\u0645\u064A', |
|
'\uFDB3': '\u0646\u062D\u064A', |
|
'\uFDB4': '\u0642\u0645\u062D', |
|
'\uFDB5': '\u0644\u062D\u0645', |
|
'\uFDB6': '\u0639\u0645\u064A', |
|
'\uFDB7': '\u0643\u0645\u064A', |
|
'\uFDB8': '\u0646\u062C\u062D', |
|
'\uFDB9': '\u0645\u062E\u064A', |
|
'\uFDBA': '\u0644\u062C\u0645', |
|
'\uFDBB': '\u0643\u0645\u0645', |
|
'\uFDBC': '\u0644\u062C\u0645', |
|
'\uFDBD': '\u0646\u062C\u062D', |
|
'\uFDBE': '\u062C\u062D\u064A', |
|
'\uFDBF': '\u062D\u062C\u064A', |
|
'\uFDC0': '\u0645\u062C\u064A', |
|
'\uFDC1': '\u0641\u0645\u064A', |
|
'\uFDC2': '\u0628\u062D\u064A', |
|
'\uFDC3': '\u0643\u0645\u0645', |
|
'\uFDC4': '\u0639\u062C\u0645', |
|
'\uFDC5': '\u0635\u0645\u0645', |
|
'\uFDC6': '\u0633\u062E\u064A', |
|
'\uFDC7': '\u0646\u062C\u064A', |
|
'\uFE49': '\u203E', |
|
'\uFE4A': '\u203E', |
|
'\uFE4B': '\u203E', |
|
'\uFE4C': '\u203E', |
|
'\uFE4D': '\u005F', |
|
'\uFE4E': '\u005F', |
|
'\uFE4F': '\u005F', |
|
'\uFE80': '\u0621', |
|
'\uFE81': '\u0622', |
|
'\uFE82': '\u0622', |
|
'\uFE83': '\u0623', |
|
'\uFE84': '\u0623', |
|
'\uFE85': '\u0624', |
|
'\uFE86': '\u0624', |
|
'\uFE87': '\u0625', |
|
'\uFE88': '\u0625', |
|
'\uFE89': '\u0626', |
|
'\uFE8A': '\u0626', |
|
'\uFE8B': '\u0626', |
|
'\uFE8C': '\u0626', |
|
'\uFE8D': '\u0627', |
|
'\uFE8E': '\u0627', |
|
'\uFE8F': '\u0628', |
|
'\uFE90': '\u0628', |
|
'\uFE91': '\u0628', |
|
'\uFE92': '\u0628', |
|
'\uFE93': '\u0629', |
|
'\uFE94': '\u0629', |
|
'\uFE95': '\u062A', |
|
'\uFE96': '\u062A', |
|
'\uFE97': '\u062A', |
|
'\uFE98': '\u062A', |
|
'\uFE99': '\u062B', |
|
'\uFE9A': '\u062B', |
|
'\uFE9B': '\u062B', |
|
'\uFE9C': '\u062B', |
|
'\uFE9D': '\u062C', |
|
'\uFE9E': '\u062C', |
|
'\uFE9F': '\u062C', |
|
'\uFEA0': '\u062C', |
|
'\uFEA1': '\u062D', |
|
'\uFEA2': '\u062D', |
|
'\uFEA3': '\u062D', |
|
'\uFEA4': '\u062D', |
|
'\uFEA5': '\u062E', |
|
'\uFEA6': '\u062E', |
|
'\uFEA7': '\u062E', |
|
'\uFEA8': '\u062E', |
|
'\uFEA9': '\u062F', |
|
'\uFEAA': '\u062F', |
|
'\uFEAB': '\u0630', |
|
'\uFEAC': '\u0630', |
|
'\uFEAD': '\u0631', |
|
'\uFEAE': '\u0631', |
|
'\uFEAF': '\u0632', |
|
'\uFEB0': '\u0632', |
|
'\uFEB1': '\u0633', |
|
'\uFEB2': '\u0633', |
|
'\uFEB3': '\u0633', |
|
'\uFEB4': '\u0633', |
|
'\uFEB5': '\u0634', |
|
'\uFEB6': '\u0634', |
|
'\uFEB7': '\u0634', |
|
'\uFEB8': '\u0634', |
|
'\uFEB9': '\u0635', |
|
'\uFEBA': '\u0635', |
|
'\uFEBB': '\u0635', |
|
'\uFEBC': '\u0635', |
|
'\uFEBD': '\u0636', |
|
'\uFEBE': '\u0636', |
|
'\uFEBF': '\u0636', |
|
'\uFEC0': '\u0636', |
|
'\uFEC1': '\u0637', |
|
'\uFEC2': '\u0637', |
|
'\uFEC3': '\u0637', |
|
'\uFEC4': '\u0637', |
|
'\uFEC5': '\u0638', |
|
'\uFEC6': '\u0638', |
|
'\uFEC7': '\u0638', |
|
'\uFEC8': '\u0638', |
|
'\uFEC9': '\u0639', |
|
'\uFECA': '\u0639', |
|
'\uFECB': '\u0639', |
|
'\uFECC': '\u0639', |
|
'\uFECD': '\u063A', |
|
'\uFECE': '\u063A', |
|
'\uFECF': '\u063A', |
|
'\uFED0': '\u063A', |
|
'\uFED1': '\u0641', |
|
'\uFED2': '\u0641', |
|
'\uFED3': '\u0641', |
|
'\uFED4': '\u0641', |
|
'\uFED5': '\u0642', |
|
'\uFED6': '\u0642', |
|
'\uFED7': '\u0642', |
|
'\uFED8': '\u0642', |
|
'\uFED9': '\u0643', |
|
'\uFEDA': '\u0643', |
|
'\uFEDB': '\u0643', |
|
'\uFEDC': '\u0643', |
|
'\uFEDD': '\u0644', |
|
'\uFEDE': '\u0644', |
|
'\uFEDF': '\u0644', |
|
'\uFEE0': '\u0644', |
|
'\uFEE1': '\u0645', |
|
'\uFEE2': '\u0645', |
|
'\uFEE3': '\u0645', |
|
'\uFEE4': '\u0645', |
|
'\uFEE5': '\u0646', |
|
'\uFEE6': '\u0646', |
|
'\uFEE7': '\u0646', |
|
'\uFEE8': '\u0646', |
|
'\uFEE9': '\u0647', |
|
'\uFEEA': '\u0647', |
|
'\uFEEB': '\u0647', |
|
'\uFEEC': '\u0647', |
|
'\uFEED': '\u0648', |
|
'\uFEEE': '\u0648', |
|
'\uFEEF': '\u0649', |
|
'\uFEF0': '\u0649', |
|
'\uFEF1': '\u064A', |
|
'\uFEF2': '\u064A', |
|
'\uFEF3': '\u064A', |
|
'\uFEF4': '\u064A', |
|
'\uFEF5': '\u0644\u0622', |
|
'\uFEF6': '\u0644\u0622', |
|
'\uFEF7': '\u0644\u0623', |
|
'\uFEF8': '\u0644\u0623', |
|
'\uFEF9': '\u0644\u0625', |
|
'\uFEFA': '\u0644\u0625', |
|
'\uFEFB': '\u0644\u0627', |
|
'\uFEFC': '\u0644\u0627' |
|
}; |
|
|
|
function reverseIfRtl(chars) { |
|
var charsLength = chars.length; |
|
//reverse an arabic ligature |
|
if (charsLength <= 1 || !isRTLRangeFor(chars.charCodeAt(0))) |
|
return chars; |
|
|
|
var s = ''; |
|
for (var ii = charsLength - 1; ii >= 0; ii--) |
|
s += chars[ii]; |
|
return s; |
|
} |
|
|
|
function fontCharsToUnicode(charCodes, font) { |
|
var glyphs = font.charsToGlyphs(charCodes); |
|
var result = ''; |
|
for (var i = 0, ii = glyphs.length; i < ii; i++) { |
|
var glyph = glyphs[i]; |
|
if (!glyph) |
|
continue; |
|
|
|
var glyphUnicode = glyph.unicode; |
|
if (glyphUnicode in NormalizedUnicodes) |
|
glyphUnicode = NormalizedUnicodes[glyphUnicode]; |
|
result += reverseIfRtl(glyphUnicode); |
|
} |
|
return result; |
|
} |
|
|
|
function adjustWidths(properties) { |
|
if (properties.fontMatrix[0] === FONT_IDENTITY_MATRIX[0]) { |
|
return; |
|
} |
|
// adjusting width to fontMatrix scale |
|
var scale = 0.001 / properties.fontMatrix[0]; |
|
var glyphsWidths = properties.widths; |
|
for (var glyph in glyphsWidths) { |
|
glyphsWidths[glyph] *= scale; |
|
} |
|
properties.defaultWidth *= scale; |
|
} |
|
|
|
/** |
|
* 'Font' is the class the outside world should use, it encapsulate all the font |
|
* decoding logics whatever type it is (assuming the font type is supported). |
|
* |
|
* For example to read a Type1 font and to attach it to the document: |
|
* var type1Font = new Font("MyFontName", binaryFile, propertiesObject); |
|
* type1Font.bind(); |
|
*/ |
|
var Font = (function FontClosure() { |
|
function Font(name, file, properties) { |
|
if (arguments.length === 1) { |
|
// importing translated data |
|
var data = arguments[0]; |
|
for (var i in data) { |
|
this[i] = data[i]; |
|
} |
|
return; |
|
} |
|
|
|
this.name = name; |
|
this.loadedName = properties.loadedName; |
|
this.coded = properties.coded; |
|
this.loadCharProcs = properties.coded; |
|
this.sizes = []; |
|
|
|
var names = name.split('+'); |
|
names = names.length > 1 ? names[1] : names[0]; |
|
names = names.split(/[-,_]/g)[0]; |
|
this.isSerifFont = !!(properties.flags & FontFlags.Serif); |
|
this.isSymbolicFont = !!(properties.flags & FontFlags.Symbolic); |
|
this.isMonospace = !!(properties.flags & FontFlags.FixedPitch); |
|
|
|
var type = properties.type; |
|
this.type = type; |
|
|
|
this.fallbackName = this.isMonospace ? 'monospace' : |
|
this.isSerifFont ? 'serif' : 'sans-serif'; |
|
|
|
this.differences = properties.differences; |
|
this.widths = properties.widths; |
|
this.defaultWidth = properties.defaultWidth; |
|
this.composite = properties.composite; |
|
this.wideChars = properties.wideChars; |
|
this.hasEncoding = properties.hasEncoding; |
|
|
|
this.fontMatrix = properties.fontMatrix; |
|
if (properties.type == 'Type3') { |
|
this.encoding = properties.baseEncoding; |
|
return; |
|
} |
|
|
|
// Trying to fix encoding using glyph CIDSystemInfo. |
|
this.loadCidToUnicode(properties); |
|
this.cidEncoding = properties.cidEncoding; |
|
this.vertical = properties.vertical; |
|
if (this.vertical) { |
|
this.vmetrics = properties.vmetrics; |
|
this.defaultVMetrics = properties.defaultVMetrics; |
|
} |
|
|
|
if (properties.toUnicode && properties.toUnicode.length > 0) |
|
this.toUnicode = properties.toUnicode; |
|
else |
|
this.rebuildToUnicode(properties); |
|
|
|
this.toFontChar = this.buildToFontChar(this.toUnicode); |
|
|
|
if (!file) { |
|
// The file data is not specified. Trying to fix the font name |
|
// to be used with the canvas.font. |
|
var fontName = name.replace(/[,_]/g, '-'); |
|
fontName = stdFontMap[fontName] || nonStdFontMap[fontName] || fontName; |
|
|
|
this.bold = (fontName.search(/bold/gi) != -1); |
|
this.italic = (fontName.search(/oblique/gi) != -1) || |
|
(fontName.search(/italic/gi) != -1); |
|
|
|
// Use 'name' instead of 'fontName' here because the original |
|
// name ArialBlack for example will be replaced by Helvetica. |
|
this.black = (name.search(/Black/g) != -1); |
|
|
|
this.encoding = properties.baseEncoding; |
|
this.noUnicodeAdaptation = true; |
|
this.loadedName = fontName.split('-')[0]; |
|
this.loading = false; |
|
return; |
|
} |
|
|
|
// Some fonts might use wrong font types for Type1C or CIDFontType0C |
|
var subtype = properties.subtype; |
|
if (subtype == 'Type1C' && (type != 'Type1' && type != 'MMType1')) |
|
type = 'Type1'; |
|
if (subtype == 'CIDFontType0C' && type != 'CIDFontType0') |
|
type = 'CIDFontType0'; |
|
// XXX: Temporarily change the type for open type so we trigger a warning. |
|
// This should be removed when we add support for open type. |
|
if (subtype === 'OpenType') { |
|
type = 'OpenType'; |
|
} |
|
|
|
var data; |
|
switch (type) { |
|
case 'Type1': |
|
case 'CIDFontType0': |
|
this.mimetype = 'font/opentype'; |
|
|
|
var cff = (subtype == 'Type1C' || subtype == 'CIDFontType0C') ? |
|
new CFFFont(file, properties) : new Type1Font(name, file, properties); |
|
|
|
adjustWidths(properties); |
|
|
|
// Wrap the CFF data inside an OTF font file |
|
data = this.convert(name, cff, properties); |
|
break; |
|
|
|
case 'TrueType': |
|
case 'CIDFontType2': |
|
this.mimetype = 'font/opentype'; |
|
|
|
// Repair the TrueType file. It is can be damaged in the point of |
|
// view of the sanitizer |
|
data = this.checkAndRepair(name, file, properties); |
|
break; |
|
|
|
default: |
|
warn('Font ' + type + ' is not supported'); |
|
break; |
|
} |
|
|
|
this.data = data; |
|
|
|
// Transfer some properties again that could change during font conversion |
|
this.fontMatrix = properties.fontMatrix; |
|
this.widths = properties.widths; |
|
this.defaultWidth = properties.defaultWidth; |
|
this.encoding = properties.baseEncoding; |
|
this.seacMap = properties.seacMap; |
|
|
|
this.loading = true; |
|
} |
|
|
|
var numFonts = 0; |
|
function getUniqueName() { |
|
return 'pdfFont' + numFonts++; |
|
} |
|
|
|
function stringToArray(str) { |
|
var array = []; |
|
for (var i = 0, ii = str.length; i < ii; ++i) |
|
array[i] = str.charCodeAt(i); |
|
|
|
return array; |
|
} |
|
|
|
function arrayToString(arr) { |
|
var str = ''; |
|
for (var i = 0, ii = arr.length; i < ii; ++i) |
|
str += String.fromCharCode(arr[i]); |
|
|
|
return str; |
|
} |
|
|
|
function int16(bytes) { |
|
return (bytes[0] << 8) + (bytes[1] & 0xff); |
|
} |
|
|
|
function int32(bytes) { |
|
return (bytes[0] << 24) + (bytes[1] << 16) + |
|
(bytes[2] << 8) + (bytes[3] & 0xff); |
|
} |
|
|
|
function getMaxPower2(number) { |
|
var maxPower = 0; |
|
var value = number; |
|
while (value >= 2) { |
|
value /= 2; |
|
maxPower++; |
|
} |
|
|
|
value = 2; |
|
for (var i = 1; i < maxPower; i++) |
|
value *= 2; |
|
return value; |
|
} |
|
|
|
function string16(value) { |
|
return String.fromCharCode((value >> 8) & 0xff) + |
|
String.fromCharCode(value & 0xff); |
|
} |
|
|
|
function safeString16(value) { |
|
// clamp value to the 16-bit int range |
|
value = value > 0x7FFF ? 0x7FFF : value < -0x8000 ? -0x8000 : value; |
|
return String.fromCharCode((value >> 8) & 0xff) + |
|
String.fromCharCode(value & 0xff); |
|
} |
|
|
|
function string32(value) { |
|
return String.fromCharCode((value >> 24) & 0xff) + |
|
String.fromCharCode((value >> 16) & 0xff) + |
|
String.fromCharCode((value >> 8) & 0xff) + |
|
String.fromCharCode(value & 0xff); |
|
} |
|
|
|
function createOpenTypeHeader(sfnt, file, numTables) { |
|
// Windows hates the Mac TrueType sfnt version number |
|
if (sfnt == 'true') |
|
sfnt = string32(0x00010000); |
|
|
|
// sfnt version (4 bytes) |
|
var header = sfnt; |
|
|
|
// numTables (2 bytes) |
|
header += string16(numTables); |
|
|
|
// searchRange (2 bytes) |
|
var tablesMaxPower2 = getMaxPower2(numTables); |
|
var searchRange = tablesMaxPower2 * 16; |
|
header += string16(searchRange); |
|
|
|
// entrySelector (2 bytes) |
|
header += string16(Math.log(tablesMaxPower2) / Math.log(2)); |
|
|
|
// rangeShift (2 bytes) |
|
header += string16(numTables * 16 - searchRange); |
|
|
|
file.file += header; |
|
file.virtualOffset += header.length; |
|
} |
|
|
|
function createTableEntry(file, tag, data) { |
|
// offset |
|
var offset = file.virtualOffset; |
|
|
|
// length |
|
var length = data.length; |
|
|
|
// Per spec tables must be 4-bytes align so add padding as needed |
|
while (data.length & 3) |
|
data.push(0x00); |
|
|
|
while (file.virtualOffset & 3) |
|
file.virtualOffset++; |
|
|
|
// checksum |
|
var checksum = 0, n = data.length; |
|
for (var i = 0; i < n; i += 4) |
|
checksum = (checksum + int32([data[i], data[i + 1], data[i + 2], |
|
data[i + 3]])) | 0; |
|
|
|
var tableEntry = (tag + string32(checksum) + |
|
string32(offset) + string32(length)); |
|
file.file += tableEntry; |
|
file.virtualOffset += data.length; |
|
} |
|
|
|
function getRanges(glyphs) { |
|
// Array.sort() sorts by characters, not numerically, so convert to an |
|
// array of characters. |
|
var codes = []; |
|
var length = glyphs.length; |
|
for (var n = 0; n < length; ++n) |
|
codes.push({ unicode: glyphs[n].unicode, code: n }); |
|
codes.sort(function fontGetRangesSort(a, b) { |
|
return a.unicode - b.unicode; |
|
}); |
|
|
|
// Split the sorted codes into ranges. |
|
var ranges = []; |
|
for (var n = 0; n < length; ) { |
|
var start = codes[n].unicode; |
|
var codeIndices = [codes[n].code]; |
|
++n; |
|
var end = start; |
|
while (n < length && end + 1 == codes[n].unicode) { |
|
codeIndices.push(codes[n].code); |
|
++end; |
|
++n; |
|
if (end === 0x10000) { break; } |
|
} |
|
ranges.push([start, end, codeIndices]); |
|
} |
|
|
|
return ranges; |
|
} |
|
|
|
function createCmapTable(glyphs, deltas) { |
|
var ranges = getRanges(glyphs); |
|
|
|
var numTables = ranges[ranges.length - 1][1] > 0xFFFF ? 2 : 1; |
|
var cmap = '\x00\x00' + // version |
|
string16(numTables) + // numTables |
|
'\x00\x03' + // platformID |
|
'\x00\x01' + // encodingID |
|
string32(4 + numTables * 8); // start of the table record |
|
|
|
for (var i = ranges.length - 1; i >= 0; --i) { |
|
if (ranges[i][0] <= 0xFFFF) { break; } |
|
} |
|
var bmpLength = i + 1; |
|
|
|
var trailingRangesCount = ranges[bmpLength - 1][1] < 0xFFFF ? 1 : 0; |
|
var segCount = bmpLength + trailingRangesCount; |
|
var segCount2 = segCount * 2; |
|
var searchRange = getMaxPower2(segCount) * 2; |
|
var searchEntry = Math.log(segCount) / Math.log(2); |
|
var rangeShift = 2 * segCount - searchRange; |
|
|
|
// Fill up the 4 parallel arrays describing the segments. |
|
var startCount = ''; |
|
var endCount = ''; |
|
var idDeltas = ''; |
|
var idRangeOffsets = ''; |
|
var glyphsIds = ''; |
|
var bias = 0; |
|
|
|
if (deltas) { |
|
for (var i = 0, ii = bmpLength; i < ii; i++) { |
|
var range = ranges[i]; |
|
var start = range[0]; |
|
var end = range[1]; |
|
var offset = (segCount - i) * 2 + bias * 2; |
|
bias += (end - start + 1); |
|
|
|
startCount += string16(start); |
|
endCount += string16(end); |
|
idDeltas += string16(0); |
|
idRangeOffsets += string16(offset); |
|
|
|
var codes = range[2]; |
|
for (var j = 0, jj = codes.length; j < jj; ++j) |
|
glyphsIds += string16(deltas[codes[j]]); |
|
} |
|
} else { |
|
for (var i = 0, ii = bmpLength; i < ii; i++) { |
|
var range = ranges[i]; |
|
var start = range[0]; |
|
var end = range[1]; |
|
var startCode = range[2][0]; |
|
|
|
startCount += string16(start); |
|
endCount += string16(end); |
|
idDeltas += string16((startCode - start + 1) & 0xFFFF); |
|
idRangeOffsets += string16(0); |
|
} |
|
} |
|
|
|
if (trailingRangesCount > 0) { |
|
endCount += '\xFF\xFF'; |
|
startCount += '\xFF\xFF'; |
|
idDeltas += '\x00\x01'; |
|
idRangeOffsets += '\x00\x00'; |
|
} |
|
|
|
var format314 = '\x00\x00' + // language |
|
string16(segCount2) + |
|
string16(searchRange) + |
|
string16(searchEntry) + |
|
string16(rangeShift) + |
|
endCount + '\x00\x00' + startCount + |
|
idDeltas + idRangeOffsets + glyphsIds; |
|
|
|
var format31012 = ''; |
|
var header31012 = ''; |
|
if (numTables > 1) { |
|
cmap += '\x00\x03' + // platformID |
|
'\x00\x0A' + // encodingID |
|
string32(4 + numTables * 8 + |
|
4 + format314.length); // start of the table record |
|
format31012 = ''; |
|
if (deltas) { |
|
for (var i = 0, ii = ranges.length; i < ii; i++) { |
|
var range = ranges[i]; |
|
var start = range[0]; |
|
var codes = range[2]; |
|
var code = deltas[codes[0]]; |
|
for (var j = 1, jj = codes.length; j < jj; ++j) { |
|
if (deltas[codes[j]] !== deltas[codes[j - 1]] + 1) { |
|
var end = range[0] + j - 1; |
|
format31012 += string32(start) + // startCharCode |
|
string32(end) + // endCharCode |
|
string32(code); // startGlyphID |
|
start = end + 1; |
|
code = deltas[codes[j]]; |
|
} |
|
} |
|
format31012 += string32(start) + // startCharCode |
|
string32(range[1]) + // endCharCode |
|
string32(code); // startGlyphID |
|
} |
|
} else { |
|
for (var i = 0, ii = ranges.length; i < ii; i++) { |
|
var range = ranges[i]; |
|
var start = range[0]; |
|
var codes = range[2]; |
|
var code = codes[0]; |
|
for (var j = 1, jj = codes.length; j < jj; ++j) { |
|
if (codes[j] !== codes[j - 1] + 1) { |
|
var end = range[0] + j - 1; |
|
format31012 += string32(start) + // startCharCode |
|
string32(end) + // endCharCode |
|
string32(code); // startGlyphID |
|
start = end + 1; |
|
code = codes[j]; |
|
} |
|
} |
|
format31012 += string32(start) + // startCharCode |
|
string32(range[1]) + // endCharCode |
|
string32(code); // startGlyphID |
|
} |
|
} |
|
header31012 = '\x00\x0C' + // format |
|
'\x00\x00' + // reserved |
|
string32(format31012.length + 16) + // length |
|
'\x00\x00\x00\x00' + // language |
|
string32(format31012.length / 12); // nGroups |
|
} |
|
|
|
return stringToArray(cmap + |
|
'\x00\x04' + // format |
|
string16(format314.length + 4) + // length |
|
format314 + header31012 + format31012); |
|
} |
|
|
|
function validateOS2Table(os2) { |
|
var stream = new Stream(os2.data); |
|
var version = int16(stream.getBytes(2)); |
|
// TODO verify all OS/2 tables fields, but currently we validate only those |
|
// that give us issues |
|
stream.getBytes(60); // skipping type, misc sizes, panose, unicode ranges |
|
var selection = int16(stream.getBytes(2)); |
|
if (version < 4 && (selection & 0x0300)) { |
|
return false; |
|
} |
|
var firstChar = int16(stream.getBytes(2)); |
|
var lastChar = int16(stream.getBytes(2)); |
|
if (firstChar > lastChar) { |
|
return false; |
|
} |
|
stream.getBytes(6); // skipping sTypoAscender/Descender/LineGap |
|
var usWinAscent = int16(stream.getBytes(2)); |
|
if (usWinAscent === 0) { // makes font unreadable by windows |
|
return false; |
|
} |
|
|
|
// OS/2 appears to be valid, resetting some fields |
|
os2.data[8] = os2.data[9] = 0; // IE rejects fonts if fsType != 0 |
|
return true; |
|
} |
|
|
|
function createOS2Table(properties, charstrings, override) { |
|
override = override || { |
|
unitsPerEm: 0, |
|
yMax: 0, |
|
yMin: 0, |
|
ascent: 0, |
|
descent: 0 |
|
}; |
|
|
|
var ulUnicodeRange1 = 0; |
|
var ulUnicodeRange2 = 0; |
|
var ulUnicodeRange3 = 0; |
|
var ulUnicodeRange4 = 0; |
|
|
|
var firstCharIndex = null; |
|
var lastCharIndex = 0; |
|
|
|
if (charstrings) { |
|
for (var i = 0; i < charstrings.length; ++i) { |
|
var code = charstrings[i].unicode; |
|
if (firstCharIndex > code || !firstCharIndex) |
|
firstCharIndex = code; |
|
if (lastCharIndex < code) |
|
lastCharIndex = code; |
|
|
|
var position = getUnicodeRangeFor(code); |
|
if (position < 32) { |
|
ulUnicodeRange1 |= 1 << position; |
|
} else if (position < 64) { |
|
ulUnicodeRange2 |= 1 << position - 32; |
|
} else if (position < 96) { |
|
ulUnicodeRange3 |= 1 << position - 64; |
|
} else if (position < 123) { |
|
ulUnicodeRange4 |= 1 << position - 96; |
|
} else { |
|
error('Unicode ranges Bits > 123 are reserved for internal usage'); |
|
} |
|
} |
|
} else { |
|
// TODO |
|
firstCharIndex = 0; |
|
lastCharIndex = 255; |
|
} |
|
|
|
var bbox = properties.bbox || [0, 0, 0, 0]; |
|
var unitsPerEm = override.unitsPerEm || |
|
1 / (properties.fontMatrix || FONT_IDENTITY_MATRIX)[0]; |
|
|
|
// if the font units differ to the PDF glyph space units |
|
// then scale up the values |
|
var scale = properties.ascentScaled ? 1.0 : |
|
unitsPerEm / PDF_GLYPH_SPACE_UNITS; |
|
|
|
var typoAscent = override.ascent || Math.round(scale * |
|
(properties.ascent || bbox[3])); |
|
var typoDescent = override.descent || Math.round(scale * |
|
(properties.descent || bbox[1])); |
|
if (typoDescent > 0 && properties.descent > 0 && bbox[1] < 0) { |
|
typoDescent = -typoDescent; // fixing incorrect descent |
|
} |
|
var winAscent = override.yMax || typoAscent; |
|
var winDescent = -override.yMin || -typoDescent; |
|
|
|
return '\x00\x03' + // version |
|
'\x02\x24' + // xAvgCharWidth |
|
'\x01\xF4' + // usWeightClass |
|
'\x00\x05' + // usWidthClass |
|
'\x00\x00' + // fstype (0 to let the font loads via font-face on IE) |
|
'\x02\x8A' + // ySubscriptXSize |
|
'\x02\xBB' + // ySubscriptYSize |
|
'\x00\x00' + // ySubscriptXOffset |
|
'\x00\x8C' + // ySubscriptYOffset |
|
'\x02\x8A' + // ySuperScriptXSize |
|
'\x02\xBB' + // ySuperScriptYSize |
|
'\x00\x00' + // ySuperScriptXOffset |
|
'\x01\xDF' + // ySuperScriptYOffset |
|
'\x00\x31' + // yStrikeOutSize |
|
'\x01\x02' + // yStrikeOutPosition |
|
'\x00\x00' + // sFamilyClass |
|
'\x00\x00\x06' + |
|
String.fromCharCode(properties.fixedPitch ? 0x09 : 0x00) + |
|
'\x00\x00\x00\x00\x00\x00' + // Panose |
|
string32(ulUnicodeRange1) + // ulUnicodeRange1 (Bits 0-31) |
|
string32(ulUnicodeRange2) + // ulUnicodeRange2 (Bits 32-63) |
|
string32(ulUnicodeRange3) + // ulUnicodeRange3 (Bits 64-95) |
|
string32(ulUnicodeRange4) + // ulUnicodeRange4 (Bits 96-127) |
|
'\x2A\x32\x31\x2A' + // achVendID |
|
string16(properties.italicAngle ? 1 : 0) + // fsSelection |
|
string16(firstCharIndex || |
|
properties.firstChar) + // usFirstCharIndex |
|
string16(lastCharIndex || properties.lastChar) + // usLastCharIndex |
|
string16(typoAscent) + // sTypoAscender |
|
string16(typoDescent) + // sTypoDescender |
|
'\x00\x64' + // sTypoLineGap (7%-10% of the unitsPerEM value) |
|
string16(winAscent) + // usWinAscent |
|
string16(winDescent) + // usWinDescent |
|
'\x00\x00\x00\x00' + // ulCodePageRange1 (Bits 0-31) |
|
'\x00\x00\x00\x00' + // ulCodePageRange2 (Bits 32-63) |
|
string16(properties.xHeight) + // sxHeight |
|
string16(properties.capHeight) + // sCapHeight |
|
string16(0) + // usDefaultChar |
|
string16(firstCharIndex || properties.firstChar) + // usBreakChar |
|
'\x00\x03'; // usMaxContext |
|
} |
|
|
|
function createPostTable(properties) { |
|
var angle = Math.floor(properties.italicAngle * (Math.pow(2, 16))); |
|
return '\x00\x03\x00\x00' + // Version number |
|
string32(angle) + // italicAngle |
|
'\x00\x00' + // underlinePosition |
|
'\x00\x00' + // underlineThickness |
|
string32(properties.fixedPitch) + // isFixedPitch |
|
'\x00\x00\x00\x00' + // minMemType42 |
|
'\x00\x00\x00\x00' + // maxMemType42 |
|
'\x00\x00\x00\x00' + // minMemType1 |
|
'\x00\x00\x00\x00'; // maxMemType1 |
|
} |
|
|
|
function createNameTable(name, proto) { |
|
if (!proto) { |
|
proto = [[], []]; // no strings and unicode strings |
|
} |
|
|
|
var strings = [ |
|
proto[0][0] || 'Original licence', // 0.Copyright |
|
proto[0][1] || name, // 1.Font family |
|
proto[0][2] || 'Unknown', // 2.Font subfamily (font weight) |
|
proto[0][3] || 'uniqueID', // 3.Unique ID |
|
proto[0][4] || name, // 4.Full font name |
|
proto[0][5] || 'Version 0.11', // 5.Version |
|
proto[0][6] || '', // 6.Postscript name |
|
proto[0][7] || 'Unknown', // 7.Trademark |
|
proto[0][8] || 'Unknown', // 8.Manufacturer |
|
proto[0][9] || 'Unknown' // 9.Designer |
|
]; |
|
|
|
// Mac want 1-byte per character strings while Windows want |
|
// 2-bytes per character, so duplicate the names table |
|
var stringsUnicode = []; |
|
for (var i = 0, ii = strings.length; i < ii; i++) { |
|
var str = proto[1][i] || strings[i]; |
|
|
|
var strUnicode = ''; |
|
for (var j = 0, jj = str.length; j < jj; j++) |
|
strUnicode += string16(str.charCodeAt(j)); |
|
stringsUnicode.push(strUnicode); |
|
} |
|
|
|
var names = [strings, stringsUnicode]; |
|
var platforms = ['\x00\x01', '\x00\x03']; |
|
var encodings = ['\x00\x00', '\x00\x01']; |
|
var languages = ['\x00\x00', '\x04\x09']; |
|
|
|
var namesRecordCount = strings.length * platforms.length; |
|
var nameTable = |
|
'\x00\x00' + // format |
|
string16(namesRecordCount) + // Number of names Record |
|
string16(namesRecordCount * 12 + 6); // Storage |
|
|
|
// Build the name records field |
|
var strOffset = 0; |
|
for (var i = 0, ii = platforms.length; i < ii; i++) { |
|
var strs = names[i]; |
|
for (var j = 0, jj = strs.length; j < jj; j++) { |
|
var str = strs[j]; |
|
var nameRecord = |
|
platforms[i] + // platform ID |
|
encodings[i] + // encoding ID |
|
languages[i] + // language ID |
|
string16(j) + // name ID |
|
string16(str.length) + |
|
string16(strOffset); |
|
nameTable += nameRecord; |
|
strOffset += str.length; |
|
} |
|
} |
|
|
|
nameTable += strings.join('') + stringsUnicode.join(''); |
|
return nameTable; |
|
} |
|
|
|
// Normalize the charcodes in the cmap table into unicode values |
|
// that will work with the (3, 1) cmap table we will write out. |
|
function cmapCharcodeToUnicode(charcode, symbolic, platformId, encodingId) { |
|
var unicode; |
|
if (symbolic) { |
|
// These codes will be shifted into the range |
|
// SYMBOLIC_FONT_GLYPH_OFFSET to (SYMBOLIC_FONT_GLYPH_OFFSET + 0xFF) |
|
// so that they are not in the control character range that could |
|
// be displayed as spaces by browsers. |
|
if (platformId === 3 && encodingId === 0 || |
|
platformId === 1 && encodingId === 0) { |
|
unicode = SYMBOLIC_FONT_GLYPH_OFFSET | (charcode & 0xFF); |
|
} |
|
} else { |
|
if (platformId === 3 && encodingId === 1) { |
|
// A (3, 1) table is alredy unicode (Microsoft Unicode format) |
|
unicode = charcode; |
|
} else if (platformId === 1 && encodingId === 0) { |
|
// TODO(mack): Should apply the changes to convert the |
|
// MacRomanEncoding to Mac OS Roman encoding in 9.6.6.4 |
|
// table 115 of the pdf spec |
|
var glyphName = Encodings.MacRomanEncoding[charcode]; |
|
if (glyphName) { |
|
unicode = GlyphsUnicode[glyphName]; |
|
} |
|
} |
|
} |
|
return unicode; |
|
} |
|
|
|
|
|
Font.prototype = { |
|
name: null, |
|
font: null, |
|
mimetype: null, |
|
encoding: null, |
|
|
|
exportData: function Font_exportData() { |
|
var data = {}; |
|
for (var i in this) { |
|
if (this.hasOwnProperty(i)) |
|
data[i] = this[i]; |
|
} |
|
return data; |
|
}, |
|
|
|
checkAndRepair: function Font_checkAndRepair(name, font, properties) { |
|
function readTableEntry(file) { |
|
var tag = file.getBytes(4); |
|
tag = String.fromCharCode(tag[0]) + |
|
String.fromCharCode(tag[1]) + |
|
String.fromCharCode(tag[2]) + |
|
String.fromCharCode(tag[3]); |
|
|
|
var checksum = int32(file.getBytes(4)); |
|
var offset = int32(file.getBytes(4)); |
|
var length = int32(file.getBytes(4)); |
|
|
|
// Read the table associated data |
|
var previousPosition = file.pos; |
|
file.pos = file.start ? file.start : 0; |
|
file.skip(offset); |
|
var data = file.getBytes(length); |
|
file.pos = previousPosition; |
|
|
|
if (tag == 'head') { |
|
// clearing checksum adjustment |
|
data[8] = data[9] = data[10] = data[11] = 0; |
|
data[17] |= 0x20; //Set font optimized for cleartype flag |
|
} |
|
|
|
return { |
|
tag: tag, |
|
checksum: checksum, |
|
length: length, |
|
offset: offset, |
|
data: data |
|
}; |
|
} |
|
|
|
function readOpenTypeHeader(ttf) { |
|
return { |
|
version: arrayToString(ttf.getBytes(4)), |
|
numTables: int16(ttf.getBytes(2)), |
|
searchRange: int16(ttf.getBytes(2)), |
|
entrySelector: int16(ttf.getBytes(2)), |
|
rangeShift: int16(ttf.getBytes(2)) |
|
}; |
|
} |
|
|
|
function createGlyphNameMap(glyphs, ids, properties) { |
|
var glyphNames = properties.glyphNames; |
|
if (!glyphNames) { |
|
properties.glyphNameMap = {}; |
|
return; |
|
} |
|
var glyphsLength = glyphs.length; |
|
var glyphNameMap = {}; |
|
var encoding = []; |
|
for (var i = 0; i < glyphsLength; ++i) { |
|
var glyphName = glyphNames[ids[i]]; |
|
if (!glyphName) |
|
continue; |
|
var unicode = glyphs[i].unicode; |
|
glyphNameMap[glyphName] = unicode; |
|
var code = glyphs[i].code; |
|
encoding[code] = glyphName; |
|
} |
|
properties.glyphNameMap = glyphNameMap; |
|
if (!properties.hasEncoding) |
|
properties.baseEncoding = encoding; |
|
} |
|
|
|
/** |
|
* Read the appropriate subtable from the cmap according to 9.6.6.4 from |
|
* PDF spec |
|
*/ |
|
function readCmapTable(cmap, font, hasEncoding, isSymbolicFont) { |
|
var start = (font.start ? font.start : 0) + cmap.offset; |
|
font.pos = start; |
|
|
|
var version = int16(font.getBytes(2)); |
|
var numTables = int16(font.getBytes(2)); |
|
|
|
var potentialTable; |
|
var foundPreferredTable; |
|
// There's an order of preference in terms of which cmap subtable we |
|
// want to use. So scan through them to find our preferred table. |
|
for (var i = 0; i < numTables; i++) { |
|
var platformId = int16(font.getBytes(2)); |
|
var encodingId = int16(font.getBytes(2)); |
|
var offset = int32(font.getBytes(4)); |
|
var useTable = false; |
|
var canBreak = false; |
|
|
|
// The following block implements the following from the spec: |
|
// |
|
// When the font has no Encoding entry, or the font descriptor’s |
|
// Symbolic flag is set (in which case the Encoding entry |
|
// is ignored), this shall occur: |
|
// - If the font contains a (3, 0) subtable, the range of |
|
// - Otherwise, the (1, 0) subtable will be used. |
|
// Otherwise, if the font does have an encoding: |
|
// - Use the (3, 1) cmap subtable |
|
// - Otherwise, use the (1, 0) subtable if present |
|
// |
|
// The following diverges slightly from the above spec in order |
|
// to handle the case that hasEncoding and isSymbolicFont are both |
|
// true. In this, based on the ordering of the rules in the spec, |
|
// my interpretation is that we should be acting as if the font is |
|
// symbolic. |
|
// |
|
// However, in this case, the test pdf 'preistabelle.pdf' |
|
// is interpreting this case as a non-symbolic font. In this case |
|
// though, 'presitabelle.pdf' does contain a (3, 1) table and does |
|
// not contain a (3, 0) table which indicates it is non-symbolic. |
|
// |
|
// Thus, I am using this heurisitic of looking at which table is |
|
// found to truly determine whether or not the font is symbolic. |
|
// That is, if the specific symbolic/non-symbolic font specific |
|
// tables (3, 0) or (3, 1) is found, that information is used for |
|
// deciding if the font is symbolic or not. |
|
// |
|
// TODO(mack): This section needs some more thought on whether the |
|
// heuristic is good enough. For now, it passes all the regression |
|
// tests. |
|
if (isSymbolicFont && platformId === 3 && encodingId === 0) { |
|
useTable = true; |
|
canBreak = true; |
|
foundPreferredTable = true; |
|
} else if (hasEncoding && platformId === 3 && encodingId === 1) { |
|
useTable = true; |
|
canBreak = true; |
|
foundPreferredTable = true; |
|
// Update the isSymbolicFont based on this heuristic |
|
isSymbolicFont = false; |
|
} else if (platformId === 1 && encodingId === 0 && |
|
!foundPreferredTable) { |
|
useTable = true; |
|
foundPreferredTable = true; |
|
} else if (!potentialTable) { |
|
// We will use an arbitrary table if we cannot find a preferred |
|
// table |
|
useTable = true; |
|
} |
|
|
|
if (useTable) { |
|
potentialTable = { |
|
platformId: platformId, |
|
encodingId: encodingId, |
|
offset: offset, |
|
isSymbolicFont: isSymbolicFont |
|
}; |
|
} |
|
if (canBreak) { |
|
break; |
|
} |
|
} |
|
|
|
if (!potentialTable) { |
|
error('Could not find a cmap table'); |
|
return; |
|
} |
|
|
|
if (!foundPreferredTable) { |
|
warn('Did not find a cmap of suitable format. Interpreting (' + |
|
potentialTable.platformId + ', ' + potentialTable.encodingId + |
|
') as (3, 1) table'); |
|
potentialTable.platformId = 3; |
|
potentialTable.encodingId = 1; |
|
} |
|
|
|
font.pos = start + potentialTable.offset; |
|
var format = int16(font.getBytes(2)); |
|
var length = int16(font.getBytes(2)); |
|
var language = int16(font.getBytes(2)); |
|
|
|
var hasShortCmap = false; |
|
var mappings = []; |
|
|
|
// TODO(mack): refactor this cmap subtable reading logic out |
|
if (format === 0) { |
|
for (var j = 0; j < 256; j++) { |
|
var index = font.getByte(); |
|
if (!index) { |
|
continue; |
|
} |
|
mappings.push({ |
|
charcode: j, |
|
glyphId: index |
|
}); |
|
} |
|
hasShortCmap = true; |
|
} else if (format === 4) { |
|
// re-creating the table in format 4 since the encoding |
|
// might be changed |
|
var segCount = (int16(font.getBytes(2)) >> 1); |
|
font.getBytes(6); // skipping range fields |
|
var segIndex, segments = []; |
|
for (segIndex = 0; segIndex < segCount; segIndex++) { |
|
segments.push({ end: int16(font.getBytes(2)) }); |
|
} |
|
font.getBytes(2); |
|
for (segIndex = 0; segIndex < segCount; segIndex++) { |
|
segments[segIndex].start = int16(font.getBytes(2)); |
|
} |
|
|
|
for (segIndex = 0; segIndex < segCount; segIndex++) { |
|
segments[segIndex].delta = int16(font.getBytes(2)); |
|
} |
|
|
|
var offsetsCount = 0; |
|
for (segIndex = 0; segIndex < segCount; segIndex++) { |
|
var segment = segments[segIndex]; |
|
var rangeOffset = int16(font.getBytes(2)); |
|
if (!rangeOffset) { |
|
segment.offsetIndex = -1; |
|
continue; |
|
} |
|
|
|
var offsetIndex = (rangeOffset >> 1) - (segCount - segIndex); |
|
segment.offsetIndex = offsetIndex; |
|
offsetsCount = Math.max(offsetsCount, offsetIndex + |
|
segment.end - segment.start + 1); |
|
} |
|
|
|
var offsets = []; |
|
for (var j = 0; j < offsetsCount; j++) { |
|
offsets.push(int16(font.getBytes(2))); |
|
} |
|
|
|
for (segIndex = 0; segIndex < segCount; segIndex++) { |
|
var segment = segments[segIndex]; |
|
var start = segment.start, end = segment.end; |
|
var delta = segment.delta, offsetIndex = segment.offsetIndex; |
|
|
|
for (var j = start; j <= end; j++) { |
|
if (j == 0xFFFF) { |
|
continue; |
|
} |
|
|
|
var glyphId = offsetIndex < 0 ? j : |
|
offsets[offsetIndex + j - start]; |
|
glyphId = (glyphId + delta) & 0xFFFF; |
|
if (glyphId === 0) { |
|
continue; |
|
} |
|
mappings.push({ |
|
charcode: j, |
|
glyphId: glyphId |
|
}); |
|
} |
|
} |
|
} else if (format == 6) { |
|
// Format 6 is a 2-bytes dense mapping, which means the font data |
|
// lives glue together even if they are pretty far in the unicode |
|
// table. (This looks weird, so I can have missed something), this |
|
// works on Linux but seems to fails on Mac so let's rewrite the |
|
// cmap table to a 3-1-4 style |
|
var firstCode = int16(font.getBytes(2)); |
|
var entryCount = int16(font.getBytes(2)); |
|
|
|
var glyphs = []; |
|
var ids = []; |
|
for (var j = 0; j < entryCount; j++) { |
|
var glyphId = int16(font.getBytes(2)); |
|
var charcode = firstCode + j; |
|
|
|
mappings.push({ |
|
charcode: charcode, |
|
glyphId: glyphId |
|
}); |
|
} |
|
} else { |
|
error('cmap table has unsupported format: ' + format); |
|
} |
|
|
|
return { |
|
platformId: potentialTable.platformId, |
|
encodingId: potentialTable.encodingId, |
|
isSymbolicFont: potentialTable.isSymbolicFont, |
|
mappings: mappings, |
|
hasShortCmap: hasShortCmap |
|
}; |
|
} |
|
|
|
function sanitizeMetrics(font, header, metrics, numGlyphs) { |
|
if (!header) { |
|
if (metrics) { |
|
metrics.data = null; |
|
} |
|
return; |
|
} |
|
|
|
font.pos = (font.start ? font.start : 0) + header.offset; |
|
font.pos += header.length - 2; |
|
var numOfMetrics = int16(font.getBytes(2)); |
|
|
|
if (numOfMetrics > numGlyphs) { |
|
info('The numOfMetrics (' + numOfMetrics + ') should not be ' + |
|
'greater than the numGlyphs (' + numGlyphs + ')'); |
|
// Reduce numOfMetrics if it is greater than numGlyphs |
|
numOfMetrics = numGlyphs; |
|
header.data[34] = (numOfMetrics & 0xff00) >> 8; |
|
header.data[35] = numOfMetrics & 0x00ff; |
|
} |
|
|
|
var numOfSidebearings = numGlyphs - numOfMetrics; |
|
var numMissing = numOfSidebearings - |
|
((metrics.length - numOfMetrics * 4) >> 1); |
|
|
|
if (numMissing > 0) { |
|
font.pos = (font.start ? font.start : 0) + metrics.offset; |
|
var entries = ''; |
|
for (var i = 0, ii = metrics.length; i < ii; i++) |
|
entries += String.fromCharCode(font.getByte()); |
|
for (var i = 0; i < numMissing; i++) |
|
entries += '\x00\x00'; |
|
metrics.data = stringToArray(entries); |
|
} |
|
} |
|
|
|
function sanitizeGlyph(source, sourceStart, sourceEnd, dest, destStart) { |
|
if (sourceEnd - sourceStart <= 12) { |
|
// glyph with data less than 12 is invalid one |
|
return 0; |
|
} |
|
var glyf = source.subarray(sourceStart, sourceEnd); |
|
var contoursCount = (glyf[0] << 8) | glyf[1]; |
|
if (contoursCount & 0x8000) { |
|
// complex glyph, writing as is |
|
dest.set(glyf, destStart); |
|
return glyf.length; |
|
} |
|
|
|
var j = 10, flagsCount = 0; |
|
for (var i = 0; i < contoursCount; i++) { |
|
var endPoint = (glyf[j] << 8) | glyf[j + 1]; |
|
flagsCount = endPoint + 1; |
|
j += 2; |
|
} |
|
// skipping instructions |
|
var instructionsLength = (glyf[j] << 8) | glyf[j + 1]; |
|
j += 2 + instructionsLength; |
|
// validating flags |
|
var coordinatesLength = 0; |
|
for (var i = 0; i < flagsCount; i++) { |
|
var flag = glyf[j++]; |
|
if (flag & 0xC0) { |
|
// reserved flags must be zero, rejecting |
|
return 0; |
|
} |
|
var xyLength = ((flag & 2) ? 1 : (flag & 16) ? 0 : 2) + |
|
((flag & 4) ? 1 : (flag & 32) ? 0 : 2); |
|
coordinatesLength += xyLength; |
|
if (flag & 8) { |
|
var repeat = glyf[j++]; |
|
i += repeat; |
|
coordinatesLength += repeat * xyLength; |
|
} |
|
} |
|
var glyphDataLength = j + coordinatesLength; |
|
if (glyphDataLength > glyf.length) { |
|
// not enough data for coordinates |
|
return 0; |
|
} |
|
if (glyf.length - glyphDataLength > 3) { |
|
// truncating and aligning to 4 bytes the long glyph data |
|
glyphDataLength = (glyphDataLength + 3) & ~3; |
|
dest.set(glyf.subarray(0, glyphDataLength), destStart); |
|
return glyphDataLength; |
|
} |
|
// glyph data is fine |
|
dest.set(glyf, destStart); |
|
return glyf.length; |
|
} |
|
|
|
function sanitizeHead(head, numGlyphs, locaLength) { |
|
var data = head.data; |
|
|
|
// Validate version: |
|
// Should always be 0x00010000 |
|
var version = int32([data[0], data[1], data[2], data[3]]); |
|
if (version >> 16 !== 1) { |
|
info('Attempting to fix invalid version in head table: ' + version); |
|
data[0] = 0; |
|
data[1] = 1; |
|
data[2] = 0; |
|
data[3] = 0; |
|
} |
|
|
|
var indexToLocFormat = int16([data[50], data[51]]); |
|
if (indexToLocFormat < 0 || indexToLocFormat > 1) { |
|
info('Attempting to fix invalid indexToLocFormat in head table: ' + |
|
indexToLocFormat); |
|
|
|
// The value of indexToLocFormat should be 0 if the loca table |
|
// consists of short offsets, and should be 1 if the loca table |
|
// consists of long offsets. |
|
// |
|
// The number of entries in the loca table should be numGlyphs + 1. |
|
// |
|
// Using this information, we can work backwards to deduce if the |
|
// size of each offset in the loca table, and thus figure out the |
|
// appropriate value for indexToLocFormat. |
|
|
|
var numGlyphsPlusOne = numGlyphs + 1; |
|
if (locaLength === numGlyphsPlusOne << 1) { |
|
// 0x0000 indicates the loca table consists of short offsets |
|
data[50] = 0; |
|
data[51] = 0; |
|
} else if (locaLength === numGlyphsPlusOne << 2) { |
|
// 0x0001 indicates the loca table consists of long offsets |
|
data[50] = 0; |
|
data[51] = 1; |
|
} else { |
|
warn('Could not fix indexToLocFormat: ' + indexToLocFormat); |
|
} |
|
} |
|
} |
|
|
|
function sanitizeGlyphLocations(loca, glyf, numGlyphs, |
|
isGlyphLocationsLong) { |
|
var itemSize, itemDecode, itemEncode; |
|
if (isGlyphLocationsLong) { |
|
itemSize = 4; |
|
itemDecode = function fontItemDecodeLong(data, offset) { |
|
return (data[offset] << 24) | (data[offset + 1] << 16) | |
|
(data[offset + 2] << 8) | data[offset + 3]; |
|
}; |
|
itemEncode = function fontItemEncodeLong(data, offset, value) { |
|
data[offset] = (value >>> 24) & 0xFF; |
|
data[offset + 1] = (value >> 16) & 0xFF; |
|
data[offset + 2] = (value >> 8) & 0xFF; |
|
data[offset + 3] = value & 0xFF; |
|
}; |
|
} else { |
|
itemSize = 2; |
|
itemDecode = function fontItemDecode(data, offset) { |
|
return (data[offset] << 9) | (data[offset + 1] << 1); |
|
}; |
|
itemEncode = function fontItemEncode(data, offset, value) { |
|
data[offset] = (value >> 9) & 0xFF; |
|
data[offset + 1] = (value >> 1) & 0xFF; |
|
}; |
|
} |
|
var locaData = loca.data; |
|
// removing the invalid glyphs |
|
var oldGlyfData = glyf.data; |
|
var oldGlyfDataLength = oldGlyfData.length; |
|
var newGlyfData = new Uint8Array(oldGlyfDataLength); |
|
var startOffset = itemDecode(locaData, 0); |
|
var writeOffset = 0; |
|
itemEncode(locaData, 0, writeOffset); |
|
for (var i = 0, j = itemSize; i < numGlyphs; i++, j += itemSize) { |
|
var endOffset = itemDecode(locaData, j); |
|
if (endOffset > oldGlyfDataLength) { |
|
// glyph end offset points outside glyf data, rejecting the glyph |
|
itemEncode(locaData, j, writeOffset); |
|
startOffset = endOffset; |
|
continue; |
|
} |
|
|
|
var newLength = sanitizeGlyph(oldGlyfData, startOffset, endOffset, |
|
newGlyfData, writeOffset); |
|
writeOffset += newLength; |
|
itemEncode(locaData, j, writeOffset); |
|
startOffset = endOffset; |
|
} |
|
|
|
if (writeOffset === 0) { |
|
// glyf table cannot be empty -- redoing the glyf and loca tables |
|
// to have single glyph with one point |
|
var simpleGlyph = new Uint8Array( |
|
[0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 49, 0]); |
|
for (var i = 0, j = itemSize; i < numGlyphs; i++, j += itemSize) |
|
itemEncode(locaData, j, simpleGlyph.length); |
|
glyf.data = simpleGlyph; |
|
return; |
|
} |
|
|
|
glyf.data = newGlyfData.subarray(0, writeOffset); |
|
} |
|
|
|
function findEmptyGlyphs(locaTable, isGlyphLocationsLong, emptyGlyphIds) { |
|
var itemSize, itemDecode; |
|
if (isGlyphLocationsLong) { |
|
itemSize = 4; |
|
itemDecode = function fontItemDecodeLong(data, offset) { |
|
return (data[offset] << 24) | (data[offset + 1] << 16) | |
|
(data[offset + 2] << 8) | data[offset + 3]; |
|
}; |
|
} else { |
|
itemSize = 2; |
|
itemDecode = function fontItemDecode(data, offset) { |
|
return (data[offset] << 9) | (data[offset + 1] << 1); |
|
}; |
|
} |
|
var data = locaTable.data, length = data.length; |
|
var lastOffset = itemDecode(data, 0); |
|
for (var i = itemSize, j = 0; i < length; i += itemSize, j++) { |
|
var offset = itemDecode(data, i); |
|
if (offset == lastOffset) |
|
emptyGlyphIds[j] = true; |
|
lastOffset = offset; |
|
} |
|
} |
|
|
|
function readPostScriptTable(post, properties, maxpNumGlyphs) { |
|
var start = (font.start ? font.start : 0) + post.offset; |
|
font.pos = start; |
|
|
|
var length = post.length, end = start + length; |
|
var version = int32(font.getBytes(4)); |
|
// skip rest to the tables |
|
font.getBytes(28); |
|
|
|
var glyphNames; |
|
var valid = true; |
|
switch (version) { |
|
case 0x00010000: |
|
glyphNames = MacStandardGlyphOrdering; |
|
break; |
|
case 0x00020000: |
|
var numGlyphs = int16(font.getBytes(2)); |
|
if (numGlyphs != maxpNumGlyphs) { |
|
valid = false; |
|
break; |
|
} |
|
var glyphNameIndexes = []; |
|
for (var i = 0; i < numGlyphs; ++i) { |
|
var index = int16(font.getBytes(2)); |
|
if (index >= 32768) { |
|
valid = false; |
|
break; |
|
} |
|
glyphNameIndexes.push(index); |
|
} |
|
if (!valid) { |
|
break; |
|
} |
|
var customNames = []; |
|
while (font.pos < end) { |
|
var stringLength = font.getByte(); |
|
var string = ''; |
|
for (var i = 0; i < stringLength; ++i) |
|
string += font.getChar(); |
|
customNames.push(string); |
|
} |
|
glyphNames = []; |
|
for (var i = 0; i < numGlyphs; ++i) { |
|
var j = glyphNameIndexes[i]; |
|
if (j < 258) { |
|
glyphNames.push(MacStandardGlyphOrdering[j]); |
|
continue; |
|
} |
|
glyphNames.push(customNames[j - 258]); |
|
} |
|
break; |
|
case 0x00030000: |
|
break; |
|
default: |
|
warn('Unknown/unsupported post table version ' + version); |
|
valid = false; |
|
break; |
|
} |
|
properties.glyphNames = glyphNames; |
|
return valid; |
|
} |
|
|
|
function readNameTable(nameTable) { |
|
var start = (font.start ? font.start : 0) + nameTable.offset; |
|
font.pos = start; |
|
|
|
var names = [[], []]; |
|
var length = nameTable.length, end = start + length; |
|
var format = int16(font.getBytes(2)); |
|
var FORMAT_0_HEADER_LENGTH = 6; |
|
if (format !== 0 || length < FORMAT_0_HEADER_LENGTH) { |
|
// unsupported name table format or table "too" small |
|
return names; |
|
} |
|
var numRecords = int16(font.getBytes(2)); |
|
var stringsStart = int16(font.getBytes(2)); |
|
var records = []; |
|
var NAME_RECORD_LENGTH = 12; |
|
for (var i = 0; i < numRecords && |
|
font.pos + NAME_RECORD_LENGTH <= end; i++) { |
|
var r = { |
|
platform: int16(font.getBytes(2)), |
|
encoding: int16(font.getBytes(2)), |
|
language: int16(font.getBytes(2)), |
|
name: int16(font.getBytes(2)), |
|
length: int16(font.getBytes(2)), |
|
offset: int16(font.getBytes(2)) |
|
}; |
|
// using only Macintosh and Windows platform/encoding names |
|
if ((r.platform == 1 && r.encoding === 0 && r.language === 0) || |
|
(r.platform == 3 && r.encoding == 1 && r.language == 0x409)) { |
|
records.push(r); |
|
} |
|
} |
|
for (var i = 0, ii = records.length; i < ii; i++) { |
|
var record = records[i]; |
|
var pos = start + stringsStart + record.offset; |
|
if (pos + record.length > end) { |
|
continue; // outside of name table, ignoring |
|
} |
|
font.pos = pos; |
|
var nameIndex = record.name; |
|
var encoding = record.encoding ? 1 : 0; |
|
if (record.encoding) { |
|
// unicode |
|
var str = ''; |
|
for (var j = 0, jj = record.length; j < jj; j += 2) { |
|
str += String.fromCharCode(int16(font.getBytes(2))); |
|
} |
|
names[1][nameIndex] = str; |
|
} else { |
|
names[0][nameIndex] = bytesToString(font.getBytes(record.length)); |
|
} |
|
} |
|
return names; |
|
} |
|
|
|
var TTOpsStackDeltas = [ |
|
0, 0, 0, 0, 0, 0, 0, 0, -2, -2, -2, -2, 0, 0, -2, -5, |
|
-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, -1, 0, -1, -1, -1, -1, |
|
1, -1, -999, 0, 1, 0, 0, -2, 0, -1, -2, -1, -999, -999, -1, -1, |
|
0, 0, -999, -999, -1, -1, -1, -1, -2, -999, -2, -2, -2, 0, -2, -2, |
|
0, 0, -2, 0, -2, 0, 0, 0, -2, -1, -1, 1, 1, 0, 0, -1, |
|
-1, -1, -1, -1, -1, -1, 0, 0, -1, 0, -1, -1, 0, -999, -1, -1, |
|
-1, -1, -1, -1, 0, 0, 0, 0, -1, -1, 0, 0, 0, 0, 0, 0, |
|
-2, -999, -999, -999, -999, -999, -1, -1, -2, -2, 0, 0, 0, 0, -1, -1, |
|
-999, -2, -2, 0, 0, -1, -2, -2, 0, -999, 0, 0, 0, -1, -2]; |
|
// 0xC0-DF == -1 and 0xE0-FF == -2 |
|
|
|
function sanitizeTTProgram(table, ttContext) { |
|
var data = table.data; |
|
var i = 0, n, lastEndf = 0, lastDeff = 0; |
|
var stack = []; |
|
var tooComplexToFollowFunctions = |
|
ttContext.tooComplexToFollowFunctions; |
|
for (var ii = data.length; i < ii;) { |
|
var op = data[i++]; |
|
// The TrueType instruction set docs can be found at |
|
// https://developer.apple.com/fonts/TTRefMan/RM05/Chap5.html |
|
if (op === 0x40) { // NPUSHB - pushes n bytes |
|
n = data[i++]; |
|
for (var j = 0; j < n; j++) { |
|
stack.push(data[i++]); |
|
} |
|
} else if (op === 0x41) { // NPUSHW - pushes n words |
|
n = data[i++]; |
|
for (var j = 0; j < n; j++) { |
|
var b = data[i++]; |
|
stack.push((b << 8) | data[i++]); |
|
} |
|
} else if ((op & 0xF8) === 0xB0) { // PUSHB - pushes bytes |
|
n = op - 0xB0 + 1; |
|
for (var j = 0; j < n; j++) { |
|
stack.push(data[i++]); |
|
} |
|
} else if ((op & 0xF8) === 0xB8) { // PUSHW - pushes words |
|
n = op - 0xB8 + 1; |
|
for (var j = 0; j < n; j++) { |
|
var b = data[i++]; |
|
stack.push((b << 8) | data[i++]); |
|
} |
|
} else if (op === 0x2B && !tooComplexToFollowFunctions) { // CALL |
|
// collecting inforamtion about which functions are used |
|
var funcId = stack[stack.length - 1]; |
|
ttContext.functionsUsed[funcId] = true; |
|
if (i >= 2 && data[i - 2] === 0x2B) { |
|
// all data in stack, calls are performed in sequence |
|
tooComplexToFollowFunctions = true; |
|
} |
|
} else if (op === 0x2C && !tooComplexToFollowFunctions) { // FDEF |
|
// collecting inforamtion about which functions are defined |
|
lastDeff = i; |
|
var funcId = stack[stack.length - 1]; |
|
ttContext.functionsDefined[funcId] = true; |
|
if (i >= 2 && data[i - 2] === 0x2D) { |
|
// all function ids in stack, FDEF/ENDF perfomed in sequence |
|
tooComplexToFollowFunctions = true; |
|
} |
|
} else if (op === 0x2D) { // ENDF - end of function |
|
lastEndf = i; |
|
} else if (op === 0x89) { // IDEF - instruction definition |
|
// recording it as a function to track ENDF |
|
lastDeff = i; |
|
} |
|
// Adjusting stack not extactly, but just enough to get function id |
|
var stackDelta = op <= 0x8E ? TTOpsStackDeltas[op] : |
|
op >= 0xC0 && op <= 0xDF ? -1 : op >= 0xE0 ? -2 : 0; |
|
while (stackDelta < 0 && stack.length > 0) { |
|
stack.pop(); |
|
stackDelta++; |
|
} |
|
while (stackDelta > 0) { |
|
stack.push(NaN); // pushing any number into stack |
|
stackDelta--; |
|
} |
|
} |
|
ttContext.tooComplexToFollowFunctions = tooComplexToFollowFunctions; |
|
var content = [data]; |
|
if (i > data.length) { |
|
content.push(new Uint8Array(i - data.length)); |
|
} |
|
if (lastDeff > lastEndf) { |
|
// new function definition started, but not finished |
|
// complete function by [CLEAR, ENDF] |
|
content.push(new Uint8Array([0x22, 0x2D])); |
|
} |
|
if (ttContext.defineMissingFunctions && !tooComplexToFollowFunctions) { |
|
for (var j = 0, jj = ttContext.functionsUsed.length; j < jj; j++) { |
|
if (!ttContext.functionsUsed[j] || ttContext.functionsDefined[j]) { |
|
continue; |
|
} |
|
// function is used, but not defined |
|
// creating empty one [PUSHB, function-id, FDEF, ENDF] |
|
content.push(new Uint8Array([0xB0, j, 0x2C, 0x2D])); |
|
} |
|
} |
|
if (content.length > 1) { |
|
// concatenating the content items |
|
var newLength = 0; |
|
for (var j = 0, jj = content.length; j < jj; j++) { |
|
newLength += content[j].length; |
|
} |
|
newLength = (newLength + 3) & ~3; |
|
var result = new Uint8Array(newLength); |
|
var pos = 0; |
|
for (var j = 0, jj = content.length; j < jj; j++) { |
|
result.set(content[j], pos); |
|
pos += content[j].length; |
|
} |
|
table.data = result; |
|
table.length = newLength; |
|
} |
|
} |
|
|
|
function sanitizeTTPrograms(fpgm, prep) { |
|
var ttContext = { |
|
functionsDefined: [], |
|
functionsUsed: [], |
|
tooComplexToFollowFunctions: false |
|
}; |
|
if (prep) { |
|
// collecting prep functions info first |
|
sanitizeTTProgram(prep, ttContext); |
|
} |
|
if (fpgm) { |
|
ttContext.defineMissingFunctions = true; |
|
sanitizeTTProgram(fpgm, ttContext); |
|
} |
|
} |
|
|
|
// Check that required tables are present |
|
var requiredTables = ['OS/2', 'cmap', 'head', 'hhea', |
|
'hmtx', 'maxp', 'name', 'post']; |
|
|
|
var header = readOpenTypeHeader(font); |
|
var numTables = header.numTables; |
|
|
|
var cmap, post, maxp, hhea, hmtx, head, os2; |
|
var glyf, fpgm, loca, prep, cvt; |
|
var tables = []; |
|
for (var i = 0; i < numTables; i++) { |
|
var table = readTableEntry(font); |
|
var index = requiredTables.indexOf(table.tag); |
|
if (index != -1) { |
|
if (table.tag == 'cmap') |
|
cmap = table; |
|
else if (table.tag == 'post') |
|
post = table; |
|
else if (table.tag == 'maxp') |
|
maxp = table; |
|
else if (table.tag == 'hhea') |
|
hhea = table; |
|
else if (table.tag == 'hmtx') |
|
hmtx = table; |
|
else if (table.tag == 'head') |
|
head = table; |
|
else if (table.tag == 'OS/2') |
|
os2 = table; |
|
|
|
requiredTables.splice(index, 1); |
|
} else { |
|
if (table.tag == 'loca') |
|
loca = table; |
|
else if (table.tag == 'glyf') |
|
glyf = table; |
|
else if (table.tag == 'fpgm') |
|
fpgm = table; |
|
else if (table.tag == 'prep') |
|
prep = table; |
|
else if (table.tag == 'cvt ') |
|
cvt = table; |
|
else // skipping table if it's not a required or optional table |
|
continue; |
|
} |
|
tables.push(table); |
|
} |
|
|
|
var numTables = tables.length + requiredTables.length; |
|
|
|
// header and new offsets. Table entry information is appended to the |
|
// end of file. The virtualOffset represents where to put the actual |
|
// data of a particular table; |
|
var ttf = { |
|
file: '', |
|
virtualOffset: numTables * (4 * 4) |
|
}; |
|
|
|
// The new numbers of tables will be the last one plus the num |
|
// of missing tables |
|
createOpenTypeHeader(header.version, ttf, numTables); |
|
|
|
// Ensure the hmtx table contains the advance width and |
|
// sidebearings information for numGlyphs in the maxp table |
|
font.pos = (font.start || 0) + maxp.offset; |
|
var version = int16(font.getBytes(4)); |
|
var numGlyphs = int16(font.getBytes(2)); |
|
|
|
sanitizeMetrics(font, hhea, hmtx, numGlyphs); |
|
|
|
sanitizeTTPrograms(fpgm, prep); |
|
|
|
if (head) { |
|
sanitizeHead(head, numGlyphs, loca.length); |
|
} |
|
|
|
var isGlyphLocationsLong = int16([head.data[50], head.data[51]]); |
|
if (head && loca && glyf) { |
|
sanitizeGlyphLocations(loca, glyf, numGlyphs, isGlyphLocationsLong); |
|
} |
|
|
|
var emptyGlyphIds = []; |
|
if (glyf) |
|
findEmptyGlyphs(loca, isGlyphLocationsLong, emptyGlyphIds); |
|
|
|
// Sanitizer reduces the glyph advanceWidth to the maxAdvanceWidth |
|
// Sometimes it's 0. That needs to be fixed |
|
if (hhea.data[10] === 0 && hhea.data[11] === 0) { |
|
hhea.data[10] = 0xFF; |
|
hhea.data[11] = 0xFF; |
|
} |
|
|
|
// The 'post' table has glyphs names. |
|
if (post) { |
|
var valid = readPostScriptTable(post, properties, numGlyphs); |
|
if (!valid) { |
|
tables.splice(tables.indexOf(post), 1); |
|
post = null; |
|
} |
|
} |
|
|
|
var glyphs, ids; |
|
if (properties.type == 'CIDFontType2') { |
|
// Replace the old CMAP table with a shiny new one |
|
// Type2 composite fonts map characters directly to glyphs so the cmap |
|
// table must be replaced. |
|
// canvas fillText will reencode some characters even if the font has a |
|
// glyph at that position - e.g. newline is converted to a space and |
|
// U+00AD (soft hyphen) is not drawn. |
|
// So, offset all the glyphs by 0xFF to avoid these cases and use |
|
// the encoding to map incoming characters to the new glyph positions |
|
if (!cmap) { |
|
cmap = { |
|
tag: 'cmap', |
|
data: null |
|
}; |
|
tables.push(cmap); |
|
} |
|
|
|
var cidToGidMap = properties.cidToGidMap || []; |
|
var gidToCidMap = [0]; |
|
if (cidToGidMap.length > 0) { |
|
for (var j = cidToGidMap.length - 1; j >= 0; j--) { |
|
var gid = cidToGidMap[j]; |
|
if (gid) |
|
gidToCidMap[gid] = j; |
|
} |
|
// filling the gaps using CID above the CIDs currently used in font |
|
var nextCid = cidToGidMap.length; |
|
for (var i = 1; i < numGlyphs; i++) { |
|
if (!gidToCidMap[i]) |
|
gidToCidMap[i] = nextCid++; |
|
} |
|
} |
|
|
|
glyphs = []; |
|
ids = []; |
|
|
|
var usedUnicodes = []; |
|
var unassignedUnicodeItems = []; |
|
var toFontChar = this.cidToFontChar || this.toFontChar; |
|
for (var i = 1; i < numGlyphs; i++) { |
|
var cid = gidToCidMap[i] || i; |
|
var unicode = toFontChar[cid]; |
|
if (!unicode || typeof unicode !== 'number' || |
|
isSpecialUnicode(unicode) || unicode in usedUnicodes) { |
|
unassignedUnicodeItems.push(i); |
|
continue; |
|
} |
|
usedUnicodes[unicode] = true; |
|
glyphs.push({ unicode: unicode, code: cid }); |
|
ids.push(i); |
|
} |
|
// unassigned codepoints will never be used for non-Identity CMap |
|
// because the input will be Unicode |
|
if (!this.cidToFontChar) { |
|
// trying to fit as many unassigned symbols as we can |
|
// in the range allocated for the user defined symbols |
|
var unusedUnicode = CMAP_GLYPH_OFFSET; |
|
for (var j = 0, jj = unassignedUnicodeItems.length; j < jj; j++) { |
|
var i = unassignedUnicodeItems[j]; |
|
var cid = gidToCidMap[i] || i; |
|
while (unusedUnicode in usedUnicodes) |
|
unusedUnicode++; |
|
if (unusedUnicode >= CMAP_GLYPH_OFFSET + GLYPH_AREA_SIZE) |
|
break; |
|
var unicode = unusedUnicode++; |
|
this.toFontChar[cid] = unicode; |
|
usedUnicodes[unicode] = true; |
|
glyphs.push({ unicode: unicode, code: cid }); |
|
ids.push(i); |
|
} |
|
} |
|
} else { |
|
this.useToFontChar = true; |
|
// Most of the following logic in this code branch is based on the |
|
// 9.6.6.4 of the PDF spec. |
|
|
|
// TODO(mack): |
|
// We are using this.hasEncoding to mean that the encoding is either |
|
// MacRomanEncoding or WinAnsiEncoding (following spec in 9.6.6.4), |
|
// but this.hasEncoding is currently true for any encodings on the |
|
// Encodings object (e.g. MacExpertEncoding). So should consider using |
|
// better check for this. |
|
var cmapTable = readCmapTable(cmap, font, this.hasEncoding, |
|
this.isSymbolicFont); |
|
|
|
// TODO(mack): If the (3, 0) cmap table used, then the font is |
|
// symbolic. The range of charcodes in the cmap table should be |
|
// one of the following: |
|
// -> 0x0000 - 0x00FF |
|
// -> 0xF000 - 0xF0FF |
|
// -> 0xF100 - 0xF1FF |
|
// -> 0xF200 - 0xF2FF |
|
// If it is not, we should change not consider this a symbolic font |
|
this.isSymbolicFont = cmapTable.isSymbolicFont; |
|
|
|
var cmapPlatformId = cmapTable.platformId; |
|
var cmapEncodingId = cmapTable.encodingId; |
|
var cmapMappings = cmapTable.mappings; |
|
var cmapMappingsLength = cmapMappings.length; |
|
var glyphs = []; |
|
var ids = []; |
|
for (var i = 0; i < cmapMappingsLength; ++i) { |
|
var cmapMapping = cmapMappings[i]; |
|
var charcode = cmapMapping.charcode; |
|
var unicode = cmapCharcodeToUnicode(charcode, this.isSymbolicFont, |
|
cmapPlatformId, cmapEncodingId); |
|
|
|
if (!unicode) { |
|
// TODO(mack): gotta check if skipping mappings where we cannot find |
|
// a unicode is the correct behaviour |
|
continue; |
|
} |
|
glyphs.push({ |
|
code: charcode, |
|
unicode: unicode |
|
}); |
|
ids.push(cmapMapping.glyphId); |
|
} |
|
|
|
var hasShortCmap = cmapTable.hasShortCmap; |
|
var toFontChar = this.toFontChar; |
|
|
|
if (hasShortCmap && ids.length == numGlyphs) { |
|
// Fixes the short cmap tables -- some generators use incorrect |
|
// glyph id. |
|
for (var i = 0, ii = ids.length; i < ii; i++) { |
|
ids[i] = i; |
|
} |
|
} |
|
|
|
// Rewrite the whole toFontChar dictionary with a new one using the |
|
// information from the mappings in the cmap table. |
|
var newToFontChar = []; |
|
if (this.isSymbolicFont) { |
|
for (var i = 0, ii = glyphs.length; i < ii; i++) { |
|
var glyph = glyphs[i]; |
|
// For (3, 0) cmap tables: |
|
// The charcode key being stored in toFontChar is the lower byte |
|
// of the two-byte charcodes of the cmap table since according to |
|
// the spec: 'each byte from the string shall be prepended with the |
|
// high byte of the range [of charcodes in the cmap table], to form |
|
// a two-byte character, which shall be used to select the |
|
// associated glyph description from the subtable'. |
|
// |
|
// For (1, 0) cmap tables: |
|
// 'single bytes from the string shall be used to look up the |
|
// associated glyph descriptions from the subtable'. This means |
|
// charcodes in the cmap will be single bytes, so no-op since |
|
// glyph.code & 0xFF === glyph.code |
|
newToFontChar[glyph.code & 0xFF] = glyph.unicode; |
|
} |
|
} else { |
|
|
|
var encoding = properties.baseEncoding; |
|
var differences = properties.differences; |
|
|
|
// TODO(mack): check if it is necessary to shift control characters |
|
// for non-symbolic fonts so that browsers dont't render them using |
|
// space characters |
|
|
|
var glyphCodeMapping = cmapTable.glyphCodeMapping; |
|
for (var charcode = 0; charcode < encoding.length; ++charcode) { |
|
if (!encoding.hasOwnProperty(charcode)) { |
|
continue; |
|
} |
|
|
|
// Since the cmap table that we will be writing out is a (3, 1) |
|
// unicode table, in this section we will rewrites the charcodes |
|
// in the pdf into unicodes |
|
|
|
var glyphName = encoding[charcode]; |
|
// A nonsymbolic font should not have a Differences array, but |
|
// if it does have one, we should still use it |
|
if (charcode in differences) { |
|
glyphName = differences[charcode]; |
|
} |
|
|
|
// Finally, any undefined entries in the table shall be filled |
|
// using StandardEncoding |
|
if (!glyphName) { |
|
glyphName = Encodings.StandardEncoding[charcode]; |
|
} |
|
|
|
// TODO(mack): Handle the case that the glyph name cannot be |
|
// mapped as specified, in which case the glyph name shall be |
|
// looked up in the font program's 'post' table (if one is |
|
// present) and the associated glyph id shall be used. |
|
// |
|
// For now, we're just using the '.notdef' glyph name in this |
|
// case. |
|
glyphName = glyphName || '.notdef'; |
|
|
|
var unicode = GlyphsUnicode[glyphName]; |
|
newToFontChar[charcode] = unicode; |
|
} |
|
} |
|
this.toFontChar = toFontChar = newToFontChar; |
|
|
|
createGlyphNameMap(glyphs, ids, properties); |
|
this.glyphNameMap = properties.glyphNameMap; |
|
} |
|
|
|
if (glyphs.length === 0) { |
|
// defines at least one glyph |
|
glyphs.push({ unicode: 0xF000, code: 0xF000, glyph: '.notdef' }); |
|
ids.push(0); |
|
} |
|
|
|
// Converting glyphs and ids into font's cmap table |
|
cmap.data = createCmapTable(glyphs, ids); |
|
var unicodeIsEnabled = []; |
|
for (var i = 0, ii = glyphs.length; i < ii; i++) { |
|
unicodeIsEnabled[glyphs[i].unicode] = true; |
|
} |
|
this.unicodeIsEnabled = unicodeIsEnabled; |
|
|
|
if (os2 && !validateOS2Table(os2)) { |
|
tables.splice(tables.indexOf(os2), 1); |
|
os2 = null; |
|
} |
|
|
|
if (!os2) { |
|
// extract some more font properties from the OpenType head and |
|
// hhea tables; yMin and descent value are always negative |
|
var override = { |
|
unitsPerEm: int16([head.data[18], head.data[19]]), |
|
yMax: int16([head.data[42], head.data[43]]), |
|
yMin: int16([head.data[38], head.data[39]]) - 0x10000, |
|
ascent: int16([hhea.data[4], hhea.data[5]]), |
|
descent: int16([hhea.data[6], hhea.data[7]]) - 0x10000 |
|
}; |
|
|
|
tables.push({ |
|
tag: 'OS/2', |
|
data: stringToArray(createOS2Table(properties, glyphs, override)) |
|
}); |
|
} |
|
|
|
// Rewrite the 'post' table if needed |
|
if (!post) { |
|
tables.push({ |
|
tag: 'post', |
|
data: stringToArray(createPostTable(properties)) |
|
}); |
|
} |
|
|
|
// Re-creating 'name' table |
|
if (requiredTables.indexOf('name') != -1) { |
|
tables.push({ |
|
tag: 'name', |
|
data: stringToArray(createNameTable(this.name)) |
|
}); |
|
} else { |
|
// ... using existing 'name' table as prototype |
|
for (var i = 0, ii = tables.length; i < ii; i++) { |
|
var table = tables[i]; |
|
if (table.tag === 'name') { |
|
var namePrototype = readNameTable(table); |
|
table.data = stringToArray(createNameTable(name, namePrototype)); |
|
break; |
|
} |
|
} |
|
} |
|
|
|
// Tables needs to be written by ascendant alphabetic order |
|
tables.sort(function tables_sort(a, b) { |
|
return (a.tag > b.tag) - (a.tag < b.tag); |
|
}); |
|
|
|
// rewrite the tables but tweak offsets |
|
for (var i = 0, ii = tables.length; i < ii; i++) { |
|
var table = tables[i]; |
|
var data = []; |
|
|
|
var tableData = table.data; |
|
for (var j = 0, jj = tableData.length; j < jj; j++) |
|
data.push(tableData[j]); |
|
createTableEntry(ttf, table.tag, data); |
|
} |
|
|
|
// Add the table datas |
|
for (var i = 0, ii = tables.length; i < ii; i++) { |
|
var table = tables[i]; |
|
var tableData = table.data; |
|
ttf.file += arrayToString(tableData); |
|
|
|
// 4-byte aligned data |
|
while (ttf.file.length & 3) |
|
ttf.file += String.fromCharCode(0); |
|
} |
|
|
|
return stringToArray(ttf.file); |
|
}, |
|
|
|
convert: function Font_convert(fontName, font, properties) { |
|
function isFixedPitch(glyphs) { |
|
for (var i = 0, ii = glyphs.length - 1; i < ii; i++) { |
|
if (glyphs[i] != glyphs[i + 1]) |
|
return false; |
|
} |
|
return true; |
|
} |
|
|
|
// The offsets object holds at the same time a representation of where |
|
// to write the table entry information about a table and another offset |
|
// representing the offset where to draw the actual data of a particular |
|
// table |
|
var REQ_TABLES_CNT = 9; |
|
|
|
var otf = { |
|
file: '', |
|
virtualOffset: 9 * (4 * 4) |
|
}; |
|
|
|
createOpenTypeHeader('\x4F\x54\x54\x4F', otf, 9); |
|
|
|
var charstrings = font.charstrings; |
|
properties.fixedPitch = isFixedPitch(charstrings); |
|
|
|
var glyphNameMap = {}; |
|
for (var i = 0; i < charstrings.length; ++i) { |
|
var charstring = charstrings[i]; |
|
glyphNameMap[charstring.glyph] = charstring.unicode; |
|
} |
|
this.glyphNameMap = glyphNameMap; |
|
|
|
var seacs = font.seacs; |
|
if (SEAC_ANALYSIS_ENABLED && seacs) { |
|
var seacMap = []; |
|
var matrix = properties.fontMatrix || FONT_IDENTITY_MATRIX; |
|
for (var i = 0; i < charstrings.length; ++i) { |
|
var charstring = charstrings[i]; |
|
var seac = seacs[charstring.gid]; |
|
if (!seac) { |
|
continue; |
|
} |
|
var baseGlyphName = Encodings.StandardEncoding[seac[2]]; |
|
var baseUnicode = glyphNameMap[baseGlyphName]; |
|
var accentGlyphName = Encodings.StandardEncoding[seac[3]]; |
|
var accentUnicode = glyphNameMap[accentGlyphName]; |
|
if (!baseUnicode || !accentUnicode) { |
|
continue; |
|
} |
|
var accentOffset = { |
|
x: seac[0] * matrix[0] + seac[1] * matrix[2] + matrix[4], |
|
y: seac[0] * matrix[1] + seac[1] * matrix[3] + matrix[5] |
|
}; |
|
seacMap[charstring.unicode] = { |
|
baseUnicode: baseUnicode, |
|
accentUnicode: accentUnicode, |
|
accentOffset: accentOffset |
|
}; |
|
} |
|
properties.seacMap = seacMap; |
|
} |
|
|
|
if (!properties.hasEncoding && (properties.subtype == 'Type1C' || |
|
properties.subtype == 'CIDFontType0C')) { |
|
var encoding = []; |
|
for (var i = 0; i < charstrings.length; ++i) { |
|
var charstring = charstrings[i]; |
|
encoding[charstring.code] = charstring.glyph; |
|
} |
|
properties.baseEncoding = encoding; |
|
} |
|
if (properties.subtype == 'CIDFontType0C') { |
|
var toFontChar = []; |
|
for (var i = 0; i < charstrings.length; ++i) { |
|
var charstring = charstrings[i]; |
|
toFontChar[charstring.code] = charstring.unicode; |
|
} |
|
this.toFontChar = toFontChar; |
|
} |
|
var unitsPerEm = 1 / (properties.fontMatrix || FONT_IDENTITY_MATRIX)[0]; |
|
|
|
var fields = { |
|
// PostScript Font Program |
|
'CFF ': font.data, |
|
|
|
// OS/2 and Windows Specific metrics |
|
'OS/2': stringToArray(createOS2Table(properties, charstrings)), |
|
|
|
// Character to glyphs mapping |
|
'cmap': createCmapTable(charstrings.slice(), |
|
('glyphIds' in font) ? font.glyphIds : null), |
|
|
|
// Font header |
|
'head': (function fontFieldsHead() { |
|
return stringToArray( |
|
'\x00\x01\x00\x00' + // Version number |
|
'\x00\x00\x10\x00' + // fontRevision |
|
'\x00\x00\x00\x00' + // checksumAdjustement |
|
'\x5F\x0F\x3C\xF5' + // magicNumber |
|
'\x00\x00' + // Flags |
|
safeString16(unitsPerEm) + // unitsPerEM |
|
'\x00\x00\x00\x00\x9e\x0b\x7e\x27' + // creation date |
|
'\x00\x00\x00\x00\x9e\x0b\x7e\x27' + // modifification date |
|
'\x00\x00' + // xMin |
|
safeString16(properties.descent) + // yMin |
|
'\x0F\xFF' + // xMax |
|
safeString16(properties.ascent) + // yMax |
|
string16(properties.italicAngle ? 2 : 0) + // macStyle |
|
'\x00\x11' + // lowestRecPPEM |
|
'\x00\x00' + // fontDirectionHint |
|
'\x00\x00' + // indexToLocFormat |
|
'\x00\x00'); // glyphDataFormat |
|
})(), |
|
|
|
// Horizontal header |
|
'hhea': (function fontFieldsHhea() { |
|
return stringToArray( |
|
'\x00\x01\x00\x00' + // Version number |
|
safeString16(properties.ascent) + // Typographic Ascent |
|
safeString16(properties.descent) + // Typographic Descent |
|
'\x00\x00' + // Line Gap |
|
'\xFF\xFF' + // advanceWidthMax |
|
'\x00\x00' + // minLeftSidebearing |
|
'\x00\x00' + // minRightSidebearing |
|
'\x00\x00' + // xMaxExtent |
|
safeString16(properties.capHeight) + // caretSlopeRise |
|
safeString16(Math.tan(properties.italicAngle) * |
|
properties.xHeight) + // caretSlopeRun |
|
'\x00\x00' + // caretOffset |
|
'\x00\x00' + // -reserved- |
|
'\x00\x00' + // -reserved- |
|
'\x00\x00' + // -reserved- |
|
'\x00\x00' + // -reserved- |
|
'\x00\x00' + // metricDataFormat |
|
string16(charstrings.length + 1)); // Number of HMetrics |
|
})(), |
|
|
|
// Horizontal metrics |
|
'hmtx': (function fontFieldsHmtx() { |
|
var hmtx = '\x00\x00\x00\x00'; // Fake .notdef |
|
for (var i = 0, ii = charstrings.length; i < ii; i++) { |
|
var charstring = charstrings[i]; |
|
var width = 'width' in charstring ? charstring.width : 0; |
|
hmtx += string16(width) + string16(0); |
|
} |
|
return stringToArray(hmtx); |
|
})(), |
|
|
|
// Maximum profile |
|
'maxp': (function fontFieldsMaxp() { |
|
return stringToArray( |
|
'\x00\x00\x50\x00' + // Version number |
|
string16(charstrings.length + 1)); // Num of glyphs |
|
})(), |
|
|
|
// Naming tables |
|
'name': stringToArray(createNameTable(fontName)), |
|
|
|
// PostScript informations |
|
'post': stringToArray(createPostTable(properties)) |
|
}; |
|
|
|
for (var field in fields) |
|
createTableEntry(otf, field, fields[field]); |
|
|
|
for (var field in fields) { |
|
var table = fields[field]; |
|
otf.file += arrayToString(table); |
|
} |
|
|
|
return stringToArray(otf.file); |
|
}, |
|
|
|
buildToFontChar: function Font_buildToFontChar(toUnicode) { |
|
var result = []; |
|
var unusedUnicode = CMAP_GLYPH_OFFSET; |
|
for (var i = 0, ii = toUnicode.length; i < ii; i++) { |
|
var unicode = toUnicode[i]; |
|
var fontCharCode = typeof unicode === 'object' ? unusedUnicode++ : |
|
unicode; |
|
if (typeof unicode !== 'undefined') |
|
result[i] = fontCharCode; |
|
} |
|
return result; |
|
}, |
|
|
|
rebuildToUnicode: function Font_rebuildToUnicode(properties) { |
|
var firstChar = properties.firstChar, lastChar = properties.lastChar; |
|
var map = []; |
|
var toUnicode = this.toUnicode || this.cidToUnicode; |
|
if (toUnicode) { |
|
var isIdentityMap = toUnicode.length === 0; |
|
for (var i = firstChar, ii = lastChar; i <= ii; i++) { |
|
// TODO missing map the character according font's CMap |
|
map[i] = isIdentityMap ? i : toUnicode[i]; |
|
} |
|
} else { |
|
for (var i = firstChar, ii = lastChar; i <= ii; i++) { |
|
var glyph = properties.differences[i]; |
|
if (!glyph) |
|
glyph = properties.baseEncoding[i]; |
|
if (!!glyph && (glyph in GlyphsUnicode)) |
|
map[i] = GlyphsUnicode[glyph]; |
|
} |
|
} |
|
this.toUnicode = map; |
|
}, |
|
|
|
loadCidToUnicode: function Font_loadCidToUnicode(properties) { |
|
if (!properties.cidSystemInfo) |
|
return; |
|
|
|
var cidToUnicodeMap = [], unicodeToCIDMap = []; |
|
this.cidToUnicode = cidToUnicodeMap; |
|
this.unicodeToCID = unicodeToCIDMap; |
|
|
|
var cidEncoding = properties.cidEncoding; |
|
if (properties.toUnicode) { |
|
if (cidEncoding && cidEncoding.indexOf('Identity-') !== 0) { |
|
TODO('Need to create a reverse mapping from \'ToUnicode\' CMap'); |
|
} |
|
return; // 'ToUnicode' CMap will be used |
|
} |
|
|
|
var cidSystemInfo = properties.cidSystemInfo; |
|
var cidToUnicode; |
|
if (cidSystemInfo) { |
|
cidToUnicode = CIDToUnicodeMaps[ |
|
cidSystemInfo.registry + '-' + cidSystemInfo.ordering]; |
|
} |
|
|
|
if (!cidToUnicode) |
|
return; // identity encoding |
|
|
|
var overwrite = HalfwidthCMaps[cidEncoding]; |
|
var cid = 1, i, j, k, ii; |
|
for (i = 0, ii = cidToUnicode.length; i < ii; ++i) { |
|
var unicode = cidToUnicode[i]; |
|
if (isArray(unicode)) { |
|
var length = unicode.length; |
|
for (j = 0; j < length; j++) { |
|
cidToUnicodeMap[cid] = k = unicode[j]; |
|
if (!unicodeToCIDMap[k] || overwrite) { |
|
unicodeToCIDMap[k] = cid; |
|
} |
|
} |
|
cid++; |
|
} else if (typeof unicode === 'object') { |
|
var fillLength = unicode.f; |
|
if (fillLength) { |
|
k = unicode.c; |
|
for (j = 0; j < fillLength; ++j) { |
|
cidToUnicodeMap[cid] = k; |
|
if (!unicodeToCIDMap[k] || overwrite) { |
|
unicodeToCIDMap[k] = cid; |
|
} |
|
cid++; |
|
k++; |
|
} |
|
} else |
|
cid += unicode.s; |
|
} else if (unicode) { |
|
cidToUnicodeMap[cid] = unicode; |
|
if (!unicodeToCIDMap[unicode] || overwrite) { |
|
unicodeToCIDMap[unicode] = cid; |
|
} |
|
cid++; |
|
} else |
|
cid++; |
|
} |
|
|
|
if (!cidEncoding) { |
|
return; |
|
} |
|
if (cidEncoding.indexOf('Identity-') !== 0) { |
|
// input is already Unicode for non-Identity CMap encodings. |
|
this.cidToUnicode = []; |
|
// For CIDFontType2, however, we need cid-to-Unicode conversion |
|
// to rebuild cmap. |
|
if (properties.type == 'CIDFontType2') { |
|
this.cidToFontChar = cidToUnicodeMap; |
|
} |
|
} else { |
|
// We don't have to do reverse conversions if the string is |
|
// already CID. |
|
this.unicodeToCID = []; |
|
} |
|
}, |
|
|
|
bindDOM: function Font_bindDOM() { |
|
if (!this.data) |
|
return null; |
|
|
|
var data = bytesToString(this.data); |
|
var fontName = this.loadedName; |
|
|
|
// Add the font-face rule to the document |
|
var url = ('url(data:' + this.mimetype + ';base64,' + |
|
window.btoa(data) + ');'); |
|
var rule = '@font-face { font-family:"' + fontName + '";src:' + url + '}'; |
|
|
|
var styleElement = document.getElementById('PDFJS_FONT_STYLE_TAG'); |
|
if (!styleElement) { |
|
styleElement = document.createElement('style'); |
|
styleElement.id = 'PDFJS_FONT_STYLE_TAG'; |
|
document.documentElement.getElementsByTagName('head')[0].appendChild( |
|
styleElement); |
|
} |
|
|
|
var styleSheet = styleElement.sheet; |
|
styleSheet.insertRule(rule, styleSheet.cssRules.length); |
|
|
|
if (PDFJS.pdfBug && 'FontInspector' in globalScope && |
|
globalScope['FontInspector'].enabled) |
|
globalScope['FontInspector'].fontAdded(this, url); |
|
|
|
return rule; |
|
}, |
|
|
|
get spaceWidth() { |
|
if ('_shadowWidth' in this) { |
|
return this._shadowWidth; |
|
} |
|
|
|
// trying to estimate space character width |
|
var possibleSpaceReplacements = ['space', 'minus', 'one', 'i']; |
|
var width; |
|
for (var i = 0, ii = possibleSpaceReplacements.length; i < ii; i++) { |
|
var glyphName = possibleSpaceReplacements[i]; |
|
// if possible, getting width by glyph name |
|
if (glyphName in this.widths) { |
|
width = this.widths[glyphName]; |
|
break; |
|
} |
|
var glyphUnicode = GlyphsUnicode[glyphName]; |
|
// finding the charcode via unicodeToCID map |
|
var charcode = 0; |
|
if (this.composite) |
|
charcode = this.unicodeToCID[glyphUnicode]; |
|
// ... via toUnicode map |
|
if (!charcode && 'toUnicode' in this) |
|
charcode = this.toUnicode.indexOf(glyphUnicode); |
|
// setting it to unicode if negative or undefined |
|
if (charcode <= 0) |
|
charcode = glyphUnicode; |
|
// trying to get width via charcode |
|
width = this.widths[charcode]; |
|
if (width) |
|
break; // the non-zero width found |
|
} |
|
width = width || this.defaultWidth; |
|
// Do not shadow the property here. See discussion: |
|
// https://github.com/mozilla/pdf.js/pull/2127#discussion_r1662280 |
|
this._shadowWidth = width; |
|
return width; |
|
}, |
|
|
|
charToGlyph: function Font_charToGlyph(charcode) { |
|
var fontCharCode, width, operatorList, disabled; |
|
|
|
var width = this.widths[charcode]; |
|
var vmetric = this.vmetrics && this.vmetrics[charcode]; |
|
|
|
switch (this.type) { |
|
case 'CIDFontType0': |
|
var cid = this.unicodeToCID[charcode] || charcode; |
|
if (this.unicodeToCID.length > 0) { |
|
width = this.widths[cid]; |
|
vmetric = this.vmetrics && this.vmetrics[cid]; |
|
} |
|
if (this.noUnicodeAdaptation) { |
|
fontCharCode = this.toFontChar[charcode] || charcode; |
|
break; |
|
} |
|
// CIDFontType0 is not encoded in Unicode. |
|
fontCharCode = this.toFontChar[cid] || cid; |
|
break; |
|
case 'CIDFontType2': |
|
if (this.unicodeToCID.length > 0) { |
|
var cid = this.unicodeToCID[charcode] || charcode; |
|
width = this.widths[cid]; |
|
vmetric = this.vmetrics && this.vmetrics[cid]; |
|
fontCharCode = charcode; |
|
break; |
|
} |
|
fontCharCode = this.toFontChar[charcode] || charcode; |
|
break; |
|
case 'Type1': |
|
var glyphName = this.differences[charcode] || this.encoding[charcode]; |
|
if (!isNum(width)) |
|
width = this.widths[glyphName]; |
|
if (this.noUnicodeAdaptation) { |
|
fontCharCode = mapPrivateUseChars(GlyphsUnicode[glyphName] || |
|
charcode); |
|
break; |
|
} |
|
fontCharCode = this.glyphNameMap[glyphName] || |
|
GlyphsUnicode[glyphName] || charcode; |
|
break; |
|
case 'Type3': |
|
var glyphName = this.differences[charcode] || this.encoding[charcode]; |
|
operatorList = this.charProcOperatorList[glyphName]; |
|
fontCharCode = charcode; |
|
break; |
|
case 'TrueType': |
|
if (this.useToFontChar) { |
|
fontCharCode = this.toFontChar[charcode] || charcode; |
|
break; |
|
} |
|
var glyphName = this.differences[charcode] || this.encoding[charcode]; |
|
if (!glyphName) |
|
glyphName = Encodings.StandardEncoding[charcode]; |
|
if (!isNum(width)) |
|
width = this.widths[glyphName]; |
|
if (this.noUnicodeAdaptation) { |
|
fontCharCode = GlyphsUnicode[glyphName] || charcode; |
|
break; |
|
} |
|
if (!this.hasEncoding || this.isSymbolicFont) { |
|
fontCharCode = this.useToFontChar ? this.toFontChar[charcode] : |
|
charcode; |
|
break; |
|
} |
|
|
|
// MacRoman encoding address by re-encoding the cmap table |
|
|
|
fontCharCode = glyphName in this.glyphNameMap ? |
|
this.glyphNameMap[glyphName] : GlyphsUnicode[glyphName]; |
|
break; |
|
default: |
|
warn('Unsupported font type: ' + this.type); |
|
break; |
|
} |
|
|
|
var unicodeChars = !('toUnicode' in this) ? charcode : |
|
this.toUnicode[charcode] || charcode; |
|
if (typeof unicodeChars === 'number') { |
|
unicodeChars = String.fromCharCode(unicodeChars); |
|
} |
|
|
|
width = isNum(width) ? width : this.defaultWidth; |
|
disabled = this.unicodeIsEnabled ? |
|
!this.unicodeIsEnabled[fontCharCode] : false; |
|
|
|
var accent = null; |
|
if (this.seacMap && this.seacMap[fontCharCode]) { |
|
var seac = this.seacMap[fontCharCode]; |
|
fontCharCode = seac.baseUnicode; |
|
accent = { |
|
fontChar: String.fromCharCode(seac.accentUnicode), |
|
offset: seac.accentOffset |
|
}; |
|
} |
|
|
|
return { |
|
fontChar: String.fromCharCode(fontCharCode), |
|
unicode: unicodeChars, |
|
accent: accent, |
|
width: width, |
|
vmetric: vmetric, |
|
disabled: disabled, |
|
operatorList: operatorList |
|
}; |
|
}, |
|
|
|
charsToGlyphs: function Font_charsToGlyphs(chars) { |
|
var charsCache = this.charsCache; |
|
var glyphs; |
|
|
|
// if we translated this string before, just grab it from the cache |
|
if (charsCache) { |
|
glyphs = charsCache[chars]; |
|
if (glyphs) |
|
return glyphs; |
|
} |
|
|
|
// lazily create the translation cache |
|
if (!charsCache) |
|
charsCache = this.charsCache = Object.create(null); |
|
|
|
glyphs = []; |
|
var charsCacheKey = chars; |
|
|
|
var converter; |
|
var cidEncoding = this.cidEncoding; |
|
if (cidEncoding) { |
|
converter = CMapConverterList[cidEncoding]; |
|
if (converter) { |
|
chars = converter(chars); |
|
} else if (cidEncoding.indexOf('Uni') !== 0 && |
|
cidEncoding.indexOf('Identity-') !== 0) { |
|
warn('Unsupported CMap: ' + cidEncoding); |
|
} |
|
} |
|
if (!converter && this.wideChars) { |
|
// composite fonts have multi-byte strings convert the string from |
|
// single-byte to multi-byte |
|
// XXX assuming CIDFonts are two-byte - later need to extract the |
|
// correct byte encoding according to the PDF spec |
|
var length = chars.length - 1; // looping over two bytes at a time so |
|
// loop should never end on the last byte |
|
for (var i = 0; i < length; i++) { |
|
var charcode = int16([chars.charCodeAt(i++), chars.charCodeAt(i)]); |
|
var glyph = this.charToGlyph(charcode); |
|
glyphs.push(glyph); |
|
// placing null after each word break charcode (ASCII SPACE) |
|
if (charcode == 0x20) |
|
glyphs.push(null); |
|
} |
|
} |
|
else { |
|
for (var i = 0, ii = chars.length; i < ii; ++i) { |
|
var charcode = chars.charCodeAt(i); |
|
var glyph = this.charToGlyph(charcode); |
|
glyphs.push(glyph); |
|
if (charcode == 0x20) |
|
glyphs.push(null); |
|
} |
|
} |
|
|
|
// Enter the translated string into the cache |
|
return (charsCache[charsCacheKey] = glyphs); |
|
} |
|
}; |
|
|
|
return Font; |
|
})(); |
|
|
|
var ErrorFont = (function ErrorFontClosure() { |
|
function ErrorFont(error) { |
|
this.error = error; |
|
} |
|
|
|
ErrorFont.prototype = { |
|
charsToGlyphs: function ErrorFont_charsToGlyphs() { |
|
return []; |
|
}, |
|
exportData: function ErrorFont_exportData() { |
|
return {error: this.error}; |
|
} |
|
}; |
|
|
|
return ErrorFont; |
|
})(); |
|
|
|
/* |
|
* CharStrings are encoded following the the CharString Encoding sequence |
|
* describe in Chapter 6 of the "Adobe Type1 Font Format" specification. |
|
* The value in a byte indicates a command, a number, or subsequent bytes |
|
* that are to be interpreted in a special way. |
|
* |
|
* CharString Number Encoding: |
|
* A CharString byte containing the values from 32 through 255 inclusive |
|
* indicate an integer. These values are decoded in four ranges. |
|
* |
|
* 1. A CharString byte containing a value, v, between 32 and 246 inclusive, |
|
* indicate the integer v - 139. Thus, the integer values from -107 through |
|
* 107 inclusive may be encoded in single byte. |
|
* |
|
* 2. A CharString byte containing a value, v, between 247 and 250 inclusive, |
|
* indicates an integer involving the next byte, w, according to the formula: |
|
* [(v - 247) x 256] + w + 108 |
|
* |
|
* 3. A CharString byte containing a value, v, between 251 and 254 inclusive, |
|
* indicates an integer involving the next byte, w, according to the formula: |
|
* -[(v - 251) * 256] - w - 108 |
|
* |
|
* 4. A CharString containing the value 255 indicates that the next 4 bytes |
|
* are a two complement signed integer. The first of these bytes contains the |
|
* highest order bits, the second byte contains the next higher order bits |
|
* and the fourth byte contain the lowest order bits. |
|
* |
|
* |
|
* CharString Command Encoding: |
|
* CharStrings commands are encoded in 1 or 2 bytes. |
|
* |
|
* Single byte commands are encoded in 1 byte that contains a value between |
|
* 0 and 31 inclusive. |
|
* If a command byte contains the value 12, then the value in the next byte |
|
* indicates a command. This "escape" mechanism allows many extra commands |
|
* to be encoded and this encoding technique helps to minimize the length of |
|
* the charStrings. |
|
*/ |
|
var Type1CharString = (function Type1CharStringClosure() { |
|
var COMMAND_MAP = { |
|
'hstem': [1], |
|
'vstem': [3], |
|
'vmoveto': [4], |
|
'rlineto': [5], |
|
'hlineto': [6], |
|
'vlineto': [7], |
|
'rrcurveto': [8], |
|
'callsubr': [10], |
|
'flex': [12, 35], |
|
'drop' : [12, 18], |
|
'endchar': [14], |
|
'rmoveto': [21], |
|
'hmoveto': [22], |
|
'vhcurveto': [30], |
|
'hvcurveto': [31] |
|
}; |
|
|
|
function Type1CharString() { |
|
this.width = 0; |
|
this.lsb = 0; |
|
this.flexing = false; |
|
this.output = []; |
|
this.stack = []; |
|
} |
|
|
|
Type1CharString.prototype = { |
|
convert: function Type1CharString_convert(encoded, subrs) { |
|
var count = encoded.length; |
|
var error = false; |
|
for (var i = 0; i < count; i++) { |
|
var value = encoded[i]; |
|
if (value < 32) { |
|
if (value === 12) { |
|
value = (value << 8) + encoded[++i]; |
|
} |
|
switch (value) { |
|
case 1: // hstem |
|
if (!HINTING_ENABLED) { |
|
this.stack = []; |
|
break; |
|
} |
|
error = this.executeCommand(2, COMMAND_MAP.hstem); |
|
break; |
|
case 3: // vstem |
|
if (!HINTING_ENABLED) { |
|
this.stack = []; |
|
break; |
|
} |
|
error = this.executeCommand(2, COMMAND_MAP.vstem); |
|
break; |
|
case 4: // vmoveto |
|
if (this.flexing) { |
|
if (this.stack.length < 1) { |
|
error = true; |
|
break; |
|
} |
|
// Add the dx for flex and but also swap the values so they are |
|
// the right order. |
|
var dy = this.stack.pop(); |
|
this.stack.push(0, dy); |
|
break; |
|
} |
|
error = this.executeCommand(1, COMMAND_MAP.vmoveto); |
|
break; |
|
case 5: // rlineto |
|
error = this.executeCommand(2, COMMAND_MAP.rlineto); |
|
break; |
|
case 6: // hlineto |
|
error = this.executeCommand(1, COMMAND_MAP.hlineto); |
|
break; |
|
case 7: // vlineto |
|
error = this.executeCommand(1, COMMAND_MAP.vlineto); |
|
break; |
|
case 8: // rrcurveto |
|
error = this.executeCommand(6, COMMAND_MAP.rrcurveto); |
|
break; |
|
case 9: // closepath |
|
// closepath is a Type1 command that does not take argument and is |
|
// useless in Type2 and it can simply be ignored. |
|
this.stack = []; |
|
break; |
|
case 10: // callsubr |
|
if (this.stack.length < 1) { |
|
error = true; |
|
break; |
|
} |
|
var subrNumber = this.stack.pop(); |
|
error = this.convert(subrs[subrNumber], subrs); |
|
break; |
|
case 11: // return |
|
return error; |
|
case 13: // hsbw |
|
if (this.stack.length < 2) { |
|
error = true; |
|
break; |
|
} |
|
// To convert to type2 we have to move the width value to the |
|
// first part of the charstring and then use hmoveto with lsb. |
|
var wx = this.stack.pop(); |
|
var sbx = this.stack.pop(); |
|
this.lsb = sbx; |
|
this.width = wx; |
|
this.stack.push(sbx); |
|
error = this.executeCommand(1, COMMAND_MAP.hmoveto); |
|
break; |
|
case 14: // endchar |
|
this.output.push(COMMAND_MAP.endchar[0]); |
|
break; |
|
case 21: // rmoveto |
|
if (this.flexing) { |
|
break; |
|
} |
|
error = this.executeCommand(2, COMMAND_MAP.rmoveto); |
|
break; |
|
case 22: // hmoveto |
|
if (this.flexing) { |
|
// Add the dy for flex. |
|
this.stack.push(0); |
|
break; |
|
} |
|
error = this.executeCommand(1, COMMAND_MAP.hmoveto); |
|
break; |
|
case 30: // vhcurveto |
|
error = this.executeCommand(4, COMMAND_MAP.vhcurveto); |
|
break; |
|
case 31: // hvcurveto |
|
error = this.executeCommand(4, COMMAND_MAP.hvcurveto); |
|
break; |
|
case (12 << 8) + 0: // dotsection |
|
// dotsection is a Type1 command to specify some hinting feature |
|
// for dots that do not take a parameter and it can safely be |
|
// ignored for Type2. |
|
this.stack = []; |
|
break; |
|
case (12 << 8) + 1: // vstem3 |
|
if (!HINTING_ENABLED) { |
|
this.stack = []; |
|
break; |
|
} |
|
// [vh]stem3 are Type1 only and Type2 supports [vh]stem with |
|
// multiple parameters, so instead of returning [vh]stem3 take a |
|
// shortcut and return [vhstem] instead. |
|
error = this.executeCommand(2, COMMAND_MAP.vstem); |
|
break; |
|
case (12 << 8) + 2: // hstem3 |
|
if (!HINTING_ENABLED) { |
|
this.stack = []; |
|
break; |
|
} |
|
// See vstem3. |
|
error = this.executeCommand(2, COMMAND_MAP.hstem); |
|
break; |
|
case (12 << 8) + 6: // seac |
|
// seac is like type 2's special endchar but it doesn't use the |
|
// first argument asb, so remove it. |
|
if (SEAC_ANALYSIS_ENABLED) { |
|
this.seac = this.stack.splice(-4, 4); |
|
error = this.executeCommand(0, COMMAND_MAP.endchar); |
|
} else { |
|
error = this.executeCommand(4, COMMAND_MAP.endchar); |
|
} |
|
break; |
|
case (12 << 8) + 7: // sbw |
|
if (this.stack.length < 4) { |
|
error = true; |
|
break; |
|
} |
|
// To convert to type2 we have to move the width value to the |
|
// first part of the charstring and then use rmoveto with |
|
// (dx, dy). The height argument will not be used for vmtx and |
|
// vhea tables reconstruction -- ignoring it. |
|
var wy = this.stack.pop(); |
|
var wx = this.stack.pop(); |
|
var sby = this.stack.pop(); |
|
var sbx = this.stack.pop(); |
|
this.lsb = sbx; |
|
this.width = wx; |
|
this.stack.push(sbx, sby); |
|
error = this.executeCommand(2, COMMAND_MAP.rmoveto); |
|
break; |
|
case (12 << 8) + 12: // div |
|
if (this.stack.length < 2) { |
|
error = true; |
|
break; |
|
} |
|
var num2 = this.stack.pop(); |
|
var num1 = this.stack.pop(); |
|
this.stack.push(num1 / num2); |
|
break; |
|
case (12 << 8) + 16: // callothersubr |
|
if (this.stack.length < 2) { |
|
error = true; |
|
break; |
|
} |
|
var subrNumber = this.stack.pop(); |
|
var numArgs = this.stack.pop(); |
|
if (subrNumber === 0 && numArgs === 3) { |
|
var flexArgs = this.stack.splice(this.stack.length - 17, 17); |
|
this.stack.push( |
|
flexArgs[2] + flexArgs[0], // bcp1x + rpx |
|
flexArgs[3] + flexArgs[1], // bcp1y + rpy |
|
flexArgs[4], // bcp2x |
|
flexArgs[5], // bcp2y |
|
flexArgs[6], // p2x |
|
flexArgs[7], // p2y |
|
flexArgs[8], // bcp3x |
|
flexArgs[9], // bcp3y |
|
flexArgs[10], // bcp4x |
|
flexArgs[11], // bcp4y |
|
flexArgs[12], // p3x |
|
flexArgs[13], // p3y |
|
flexArgs[14] // flexDepth |
|
// 15 = finalx unused by flex |
|
// 16 = finaly unused by flex |
|
); |
|
error = this.executeCommand(13, COMMAND_MAP.flex, true); |
|
this.flexing = false; |
|
this.stack.push(flexArgs[15], flexArgs[16]); |
|
} else if (subrNumber === 1 && numArgs === 0) { |
|
this.flexing = true; |
|
} |
|
break; |
|
case (12 << 8) + 17: // pop |
|
// Ignore this since it is only used with othersubr. |
|
break; |
|
case (12 << 8) + 33: // setcurrentpoint |
|
// Ignore for now. |
|
this.stack = []; |
|
break; |
|
default: |
|
warn('Unknown type 1 charstring command of "' + value + '"'); |
|
break; |
|
} |
|
if (error) { |
|
break; |
|
} |
|
continue; |
|
} else if (value <= 246) { |
|
value = value - 139; |
|
} else if (value <= 250) { |
|
value = ((value - 247) * 256) + encoded[++i] + 108; |
|
} else if (value <= 254) { |
|
value = -((value - 251) * 256) - encoded[++i] - 108; |
|
} else { |
|
value = (encoded[++i] & 0xff) << 24 | (encoded[++i] & 0xff) << 16 | |
|
(encoded[++i] & 0xff) << 8 | (encoded[++i] & 0xff) << 0; |
|
} |
|
this.stack.push(value); |
|
} |
|
return error; |
|
}, |
|
|
|
executeCommand: function(howManyArgs, command, keepStack) { |
|
var stackLength = this.stack.length; |
|
if (howManyArgs > stackLength) { |
|
return true; |
|
} |
|
var start = stackLength - howManyArgs; |
|
for (var i = start; i < stackLength; i++) { |
|
var value = this.stack[i]; |
|
if (value === (value | 0)) { // int |
|
this.output.push(28, (value >> 8) & 0xff, value & 0xff); |
|
} else { // fixed point |
|
value = (65536 * value) | 0; |
|
this.output.push(255, |
|
(value >> 24) & 0xFF, |
|
(value >> 16) & 0xFF, |
|
(value >> 8) & 0xFF, |
|
value & 0xFF); |
|
} |
|
} |
|
this.output.push.apply(this.output, command); |
|
if (keepStack) { |
|
this.stack.splice(start, howManyArgs); |
|
} else { |
|
this.stack = []; |
|
} |
|
return false; |
|
} |
|
}; |
|
|
|
return Type1CharString; |
|
})(); |
|
|
|
/* |
|
* Type1Parser encapsulate the needed code for parsing a Type1 font |
|
* program. Some of its logic depends on the Type2 charstrings |
|
* structure. |
|
*/ |
|
var Type1Parser = function type1Parser() { |
|
/* |
|
* Decrypt a Sequence of Ciphertext Bytes to Produce the Original Sequence |
|
* of Plaintext Bytes. The function took a key as a parameter which can be |
|
* for decrypting the eexec block of for decoding charStrings. |
|
*/ |
|
var EEXEC_ENCRYPT_KEY = 55665; |
|
var CHAR_STRS_ENCRYPT_KEY = 4330; |
|
|
|
function decrypt(stream, key, discardNumber) { |
|
var r = key, c1 = 52845, c2 = 22719; |
|
var decryptedString = []; |
|
|
|
var value = ''; |
|
var count = stream.length; |
|
for (var i = 0; i < count; i++) { |
|
value = stream[i]; |
|
decryptedString[i] = value ^ (r >> 8); |
|
r = ((value + r) * c1 + c2) & ((1 << 16) - 1); |
|
} |
|
return decryptedString.slice(discardNumber); |
|
} |
|
|
|
/* |
|
* Returns an object containing a Subrs array and a CharStrings |
|
* array extracted from and eexec encrypted block of data |
|
*/ |
|
function readNumberArray(str, index) { |
|
var start = index; |
|
while (str[index++] != '[') |
|
start++; |
|
start++; |
|
|
|
var count = 0; |
|
while (str[index++] != ']') |
|
count++; |
|
|
|
str = str.substr(start, count); |
|
|
|
str = str.trim(); |
|
// Remove adjacent spaces |
|
str = str.replace(/\s+/g, ' '); |
|
|
|
var array = str.split(' '); |
|
for (var i = 0, ii = array.length; i < ii; i++) |
|
array[i] = parseFloat(array[i] || 0); |
|
return array; |
|
} |
|
|
|
function readNumber(str, index) { |
|
while (str[index] == ' ') |
|
index++; |
|
|
|
var start = index; |
|
|
|
var count = 0; |
|
while (str[index++] != ' ') |
|
count++; |
|
|
|
return parseFloat(str.substr(start, count) || 0); |
|
} |
|
|
|
function readBoolean(str, index) { |
|
while (str[index] == ' ') |
|
index++; |
|
|
|
var start = index; |
|
|
|
var count = 0; |
|
var length = str.length; |
|
while (index < length && str[index++] != ' ') { |
|
count++; |
|
} |
|
|
|
// Use 1 and 0 since that's what type2 charstrings use. |
|
return str.substr(start, count) === 'true' ? 1 : 0; |
|
} |
|
|
|
|
|
function isSeparator(c) { |
|
return c == ' ' || c == '\n' || c == '\x0d'; |
|
} |
|
|
|
this.extractFontProgram = function Type1Parser_extractFontProgram(stream) { |
|
var eexec = decrypt(stream, EEXEC_ENCRYPT_KEY, 4); |
|
var eexecStr = ''; |
|
for (var i = 0, ii = eexec.length; i < ii; i++) |
|
eexecStr += String.fromCharCode(eexec[i]); |
|
|
|
var glyphsSection = false, subrsSection = false; |
|
var subrs = [], charstrings = []; |
|
var program = { |
|
subrs: [], |
|
charstrings: [], |
|
properties: { |
|
'privateData': { |
|
'lenIV': 4 |
|
} |
|
} |
|
}; |
|
|
|
var glyph = ''; |
|
var token = ''; |
|
var length = 0; |
|
|
|
var c = ''; |
|
var count = eexecStr.length; |
|
for (var i = 0; i < count; i++) { |
|
var getToken = function getToken() { |
|
while (i < count && isSeparator(eexecStr[i])) |
|
++i; |
|
|
|
var token = ''; |
|
while (i < count && !isSeparator(eexecStr[i])) |
|
token += eexecStr[i++]; |
|
|
|
return token; |
|
}; |
|
var c = eexecStr[i]; |
|
|
|
if ((glyphsSection || subrsSection) && |
|
(token == 'RD' || token == '-|')) { |
|
i++; |
|
var data = eexec.slice(i, i + length); |
|
var lenIV = program.properties.privateData['lenIV']; |
|
var encoded = decrypt(data, CHAR_STRS_ENCRYPT_KEY, lenIV); |
|
|
|
if (glyphsSection) { |
|
charstrings.push({ |
|
glyph: glyph, |
|
encoded: encoded |
|
}); |
|
} else { |
|
subrs.push(encoded); |
|
} |
|
i += length; |
|
token = ''; |
|
} else if (isSeparator(c)) { |
|
// Use '| 0' to prevent setting a double into length such as the double |
|
// does not flow into the loop variable. |
|
length = parseInt(token, 10) | 0; |
|
token = ''; |
|
} else { |
|
token += c; |
|
if (!glyphsSection) { |
|
switch (token) { |
|
case '/CharString': |
|
glyphsSection = true; |
|
break; |
|
case '/Subrs': |
|
++i; |
|
var num = parseInt(getToken(), 10); |
|
getToken(); // read in 'array' |
|
for (var j = 0; j < num; ++j) { |
|
var t = getToken(); // read in 'dup' |
|
if (t == 'ND' || t == '|-' || t == 'noaccess') |
|
break; |
|
var index = parseInt(getToken(), 10); |
|
if (index > j) |
|
j = index; |
|
var length = parseInt(getToken(), 10); |
|
getToken(); // read in 'RD' |
|
var data = eexec.slice(i + 1, i + 1 + length); |
|
var lenIV = program.properties.privateData['lenIV']; |
|
var encoded = decrypt(data, CHAR_STRS_ENCRYPT_KEY, lenIV); |
|
i = i + 1 + length; |
|
t = getToken(); // read in 'NP' |
|
if (t == 'noaccess') |
|
getToken(); // read in 'put' |
|
subrs[index] = encoded; |
|
} |
|
break; |
|
case '/BlueValues': |
|
case '/OtherBlues': |
|
case '/FamilyBlues': |
|
case '/FamilyOtherBlues': |
|
var blueArray = readNumberArray(eexecStr, i + 1); |
|
// *Blue* values may contain invalid data: disables reading of |
|
// those values when hinting is disabled. |
|
if (blueArray.length > 0 && (blueArray.length % 2) === 0 && |
|
HINTING_ENABLED) { |
|
program.properties.privateData[token.substring(1)] = blueArray; |
|
} |
|
break; |
|
case '/StemSnapH': |
|
case '/StemSnapV': |
|
program.properties.privateData[token.substring(1)] = |
|
readNumberArray(eexecStr, i + 1); |
|
break; |
|
case '/StdHW': |
|
case '/StdVW': |
|
program.properties.privateData[token.substring(1)] = |
|
readNumberArray(eexecStr, i + 1)[0]; |
|
break; |
|
case '/BlueShift': |
|
case '/lenIV': |
|
case '/BlueFuzz': |
|
case '/BlueScale': |
|
case '/LanguageGroup': |
|
case '/ExpansionFactor': |
|
program.properties.privateData[token.substring(1)] = |
|
readNumber(eexecStr, i + 1); |
|
break; |
|
case '/ForceBold': |
|
program.properties.privateData[token.substring(1)] = |
|
readBoolean(eexecStr, i + 1); |
|
break; |
|
} |
|
} else if (c == '/') { |
|
token = glyph = ''; |
|
while ((c = eexecStr[++i]) != ' ') |
|
glyph += c; |
|
} |
|
} |
|
} |
|
|
|
for (var i = 0; i < charstrings.length; i++) { |
|
var glyph = charstrings[i].glyph; |
|
var encoded = charstrings[i].encoded; |
|
var charString = new Type1CharString(); |
|
var error = charString.convert(encoded, subrs); |
|
var output = charString.output; |
|
if (error) { |
|
// It seems when FreeType encounters an error while evaluating a glyph |
|
// that it completely ignores the glyph so we'll mimic that behaviour |
|
// here and put an endchar to make the validator happy. |
|
output = [14]; |
|
} |
|
program.charstrings.push({ |
|
glyph: glyph, |
|
data: output, |
|
seac: charString.seac, |
|
lsb: charString.lsb, |
|
width: charString.width |
|
}); |
|
} |
|
|
|
return program; |
|
}; |
|
|
|
this.extractFontHeader = function Type1Parser_extractFontHeader(stream, |
|
properties) { |
|
var headerString = ''; |
|
for (var i = 0, ii = stream.length; i < ii; i++) |
|
headerString += String.fromCharCode(stream[i]); |
|
|
|
var token = ''; |
|
var count = headerString.length; |
|
for (var i = 0; i < count; i++) { |
|
var getToken = function getToken() { |
|
var character = headerString[i]; |
|
while (i < count && (isSeparator(character) || character == '/')) |
|
character = headerString[++i]; |
|
|
|
var token = ''; |
|
while (i < count && !(isSeparator(character) || character == '/')) { |
|
token += character; |
|
character = headerString[++i]; |
|
} |
|
|
|
return token; |
|
}; |
|
|
|
var c = headerString[i]; |
|
if (isSeparator(c)) { |
|
switch (token) { |
|
case '/FontMatrix': |
|
var matrix = readNumberArray(headerString, i + 1); |
|
properties.fontMatrix = matrix; |
|
break; |
|
case '/Encoding': |
|
var encodingArg = getToken(); |
|
var encoding; |
|
if (!/^\d+$/.test(encodingArg)) { |
|
// encoding name is specified |
|
encoding = Encodings[encodingArg]; |
|
} else { |
|
encoding = []; |
|
var size = parseInt(encodingArg, 10); |
|
getToken(); // read in 'array' |
|
|
|
for (var j = 0; j < size; j++) { |
|
var token = getToken(); |
|
if (token == 'dup') { |
|
var index = parseInt(getToken(), 10); |
|
var glyph = getToken(); |
|
encoding[index] = glyph; |
|
getToken(); // read the in 'put' |
|
} |
|
} |
|
} |
|
if (!properties.hasEncoding && encoding) { |
|
properties.baseEncoding = encoding; |
|
break; |
|
} |
|
break; |
|
} |
|
token = ''; |
|
} else { |
|
token += c; |
|
} |
|
} |
|
}; |
|
}; |
|
|
|
/** |
|
* The CFF class takes a Type1 file and wrap it into a |
|
* 'Compact Font Format' which itself embed Type2 charstrings. |
|
*/ |
|
var CFFStandardStrings = [ |
|
'.notdef', 'space', 'exclam', 'quotedbl', 'numbersign', 'dollar', 'percent', |
|
'ampersand', 'quoteright', 'parenleft', 'parenright', 'asterisk', 'plus', |
|
'comma', 'hyphen', 'period', 'slash', 'zero', 'one', 'two', 'three', 'four', |
|
'five', 'six', 'seven', 'eight', 'nine', 'colon', 'semicolon', 'less', |
|
'equal', 'greater', 'question', 'at', 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', |
|
'I', 'J', 'K', 'L', 'M', 'N', 'O', 'P', 'Q', 'R', 'S', 'T', 'U', 'V', 'W', |
|
'X', 'Y', 'Z', 'bracketleft', 'backslash', 'bracketright', 'asciicircum', |
|
'underscore', 'quoteleft', 'a', 'b', 'c', 'd', 'e', 'f', 'g', 'h', 'i', 'j', |
|
'k', 'l', 'm', 'n', 'o', 'p', 'q', 'r', 's', 't', 'u', 'v', 'w', 'x', 'y', |
|
'z', 'braceleft', 'bar', 'braceright', 'asciitilde', 'exclamdown', 'cent', |
|
'sterling', 'fraction', 'yen', 'florin', 'section', 'currency', |
|
'quotesingle', 'quotedblleft', 'guillemotleft', 'guilsinglleft', |
|
'guilsinglright', 'fi', 'fl', 'endash', 'dagger', 'daggerdbl', |
|
'periodcentered', 'paragraph', 'bullet', 'quotesinglbase', 'quotedblbase', |
|
'quotedblright', 'guillemotright', 'ellipsis', 'perthousand', 'questiondown', |
|
'grave', 'acute', 'circumflex', 'tilde', 'macron', 'breve', 'dotaccent', |
|
'dieresis', 'ring', 'cedilla', 'hungarumlaut', 'ogonek', 'caron', 'emdash', |
|
'AE', 'ordfeminine', 'Lslash', 'Oslash', 'OE', 'ordmasculine', 'ae', |
|
'dotlessi', 'lslash', 'oslash', 'oe', 'germandbls', 'onesuperior', |
|
'logicalnot', 'mu', 'trademark', 'Eth', 'onehalf', 'plusminus', 'Thorn', |
|
'onequarter', 'divide', 'brokenbar', 'degree', 'thorn', 'threequarters', |
|
'twosuperior', 'registered', 'minus', 'eth', 'multiply', 'threesuperior', |
|
'copyright', 'Aacute', 'Acircumflex', 'Adieresis', 'Agrave', 'Aring', |
|
'Atilde', 'Ccedilla', 'Eacute', 'Ecircumflex', 'Edieresis', 'Egrave', |
|
'Iacute', 'Icircumflex', 'Idieresis', 'Igrave', 'Ntilde', 'Oacute', |
|
'Ocircumflex', 'Odieresis', 'Ograve', 'Otilde', 'Scaron', 'Uacute', |
|
'Ucircumflex', 'Udieresis', 'Ugrave', 'Yacute', 'Ydieresis', 'Zcaron', |
|
'aacute', 'acircumflex', 'adieresis', 'agrave', 'aring', 'atilde', |
|
'ccedilla', 'eacute', 'ecircumflex', 'edieresis', 'egrave', 'iacute', |
|
'icircumflex', 'idieresis', 'igrave', 'ntilde', 'oacute', 'ocircumflex', |
|
'odieresis', 'ograve', 'otilde', 'scaron', 'uacute', 'ucircumflex', |
|
'udieresis', 'ugrave', 'yacute', 'ydieresis', 'zcaron', 'exclamsmall', |
|
'Hungarumlautsmall', 'dollaroldstyle', 'dollarsuperior', 'ampersandsmall', |
|
'Acutesmall', 'parenleftsuperior', 'parenrightsuperior', '266 ff', |
|
'onedotenleader', 'zerooldstyle', 'oneoldstyle', 'twooldstyle', |
|
'threeoldstyle', 'fouroldstyle', 'fiveoldstyle', 'sixoldstyle', |
|
'sevenoldstyle', 'eightoldstyle', 'nineoldstyle', 'commasuperior', |
|
'threequartersemdash', 'periodsuperior', 'questionsmall', 'asuperior', |
|
'bsuperior', 'centsuperior', 'dsuperior', 'esuperior', 'isuperior', |
|
'lsuperior', 'msuperior', 'nsuperior', 'osuperior', 'rsuperior', 'ssuperior', |
|
'tsuperior', 'ff', 'ffi', 'ffl', 'parenleftinferior', 'parenrightinferior', |
|
'Circumflexsmall', 'hyphensuperior', 'Gravesmall', 'Asmall', 'Bsmall', |
|
'Csmall', 'Dsmall', 'Esmall', 'Fsmall', 'Gsmall', 'Hsmall', 'Ismall', |
|
'Jsmall', 'Ksmall', 'Lsmall', 'Msmall', 'Nsmall', 'Osmall', 'Psmall', |
|
'Qsmall', 'Rsmall', 'Ssmall', 'Tsmall', 'Usmall', 'Vsmall', 'Wsmall', |
|
'Xsmall', 'Ysmall', 'Zsmall', 'colonmonetary', 'onefitted', 'rupiah', |
|
'Tildesmall', 'exclamdownsmall', 'centoldstyle', 'Lslashsmall', |
|
'Scaronsmall', 'Zcaronsmall', 'Dieresissmall', 'Brevesmall', 'Caronsmall', |
|
'Dotaccentsmall', 'Macronsmall', 'figuredash', 'hypheninferior', |
|
'Ogoneksmall', 'Ringsmall', 'Cedillasmall', 'questiondownsmall', 'oneeighth', |
|
'threeeighths', 'fiveeighths', 'seveneighths', 'onethird', 'twothirds', |
|
'zerosuperior', 'foursuperior', 'fivesuperior', 'sixsuperior', |
|
'sevensuperior', 'eightsuperior', 'ninesuperior', 'zeroinferior', |
|
'oneinferior', 'twoinferior', 'threeinferior', 'fourinferior', |
|
'fiveinferior', 'sixinferior', 'seveninferior', 'eightinferior', |
|
'nineinferior', 'centinferior', 'dollarinferior', 'periodinferior', |
|
'commainferior', 'Agravesmall', 'Aacutesmall', 'Acircumflexsmall', |
|
'Atildesmall', 'Adieresissmall', 'Aringsmall', 'AEsmall', 'Ccedillasmall', |
|
'Egravesmall', 'Eacutesmall', 'Ecircumflexsmall', 'Edieresissmall', |
|
'Igravesmall', 'Iacutesmall', 'Icircumflexsmall', 'Idieresissmall', |
|
'Ethsmall', 'Ntildesmall', 'Ogravesmall', 'Oacutesmall', 'Ocircumflexsmall', |
|
'Otildesmall', 'Odieresissmall', 'OEsmall', 'Oslashsmall', 'Ugravesmall', |
|
'Uacutesmall', 'Ucircumflexsmall', 'Udieresissmall', 'Yacutesmall', |
|
'Thornsmall', 'Ydieresissmall', '001.000', '001.001', '001.002', '001.003', |
|
'Black', 'Bold', 'Book', 'Light', 'Medium', 'Regular', 'Roman', 'Semibold' |
|
]; |
|
|
|
var type1Parser = new Type1Parser(); |
|
|
|
// Type1Font is also a CIDFontType0. |
|
var Type1Font = function Type1Font(name, file, properties) { |
|
// Get the data block containing glyphs and subrs informations |
|
var headerBlock = file.getBytes(properties.length1); |
|
type1Parser.extractFontHeader(headerBlock, properties); |
|
|
|
// Decrypt the data blocks and retrieve it's content |
|
var eexecBlock = file.getBytes(properties.length2); |
|
var data = type1Parser.extractFontProgram(eexecBlock); |
|
for (var info in data.properties) |
|
properties[info] = data.properties[info]; |
|
|
|
var charstrings = this.getOrderedCharStrings(data.charstrings, properties); |
|
var type2Charstrings = this.getType2Charstrings(charstrings); |
|
var subrs = this.getType2Subrs(data.subrs); |
|
|
|
this.charstrings = charstrings; |
|
this.data = this.wrap(name, type2Charstrings, this.charstrings, |
|
subrs, properties); |
|
this.seacs = this.getSeacs(data.charstrings); |
|
}; |
|
|
|
Type1Font.prototype = { |
|
getOrderedCharStrings: function Type1Font_getOrderedCharStrings(glyphs, |
|
properties) { |
|
var charstrings = []; |
|
var i, length, glyphName; |
|
var unusedUnicode = CMAP_GLYPH_OFFSET; |
|
for (i = 0, length = glyphs.length; i < length; i++) { |
|
var item = glyphs[i]; |
|
var glyphName = item.glyph; |
|
var unicode = glyphName in GlyphsUnicode ? |
|
GlyphsUnicode[glyphName] : unusedUnicode++; |
|
charstrings.push({ |
|
glyph: glyphName, |
|
unicode: unicode, |
|
gid: i, |
|
charstring: item.data, |
|
width: item.width, |
|
lsb: item.lsb |
|
}); |
|
} |
|
|
|
charstrings.sort(function charstrings_sort(a, b) { |
|
return a.unicode - b.unicode; |
|
}); |
|
return charstrings; |
|
}, |
|
|
|
getSeacs: function Type1Font_getSeacs(charstrings) { |
|
var i, ii; |
|
var seacMap = []; |
|
for (i = 0, ii = charstrings.length; i < ii; i++) { |
|
var charstring = charstrings[i]; |
|
if (charstring.seac) { |
|
seacMap[i] = charstring.seac; |
|
} |
|
} |
|
return seacMap; |
|
}, |
|
|
|
getType2Charstrings: function Type1Font_getType2Charstrings( |
|
type1Charstrings) { |
|
var type2Charstrings = []; |
|
for (var i = 0, ii = type1Charstrings.length; i < ii; i++) { |
|
type2Charstrings.push(type1Charstrings[i].charstring); |
|
} |
|
return type2Charstrings; |
|
}, |
|
|
|
getType2Subrs: function Type1Font_getType2Subrs(type1Subrs) { |
|
var bias = 0; |
|
var count = type1Subrs.length; |
|
if (count < 1133) |
|
bias = 107; |
|
else if (count < 33769) |
|
bias = 1131; |
|
else |
|
bias = 32768; |
|
|
|
// Add a bunch of empty subrs to deal with the Type2 bias |
|
var type2Subrs = []; |
|
for (var i = 0; i < bias; i++) |
|
type2Subrs.push([0x0B]); |
|
|
|
for (var i = 0; i < count; i++) { |
|
type2Subrs.push(type1Subrs[i]); |
|
} |
|
|
|
return type2Subrs; |
|
}, |
|
|
|
wrap: function Type1Font_wrap(name, glyphs, charstrings, subrs, properties) { |
|
var cff = new CFF(); |
|
cff.header = new CFFHeader(1, 0, 4, 4); |
|
|
|
cff.names = [name]; |
|
|
|
var topDict = new CFFTopDict(); |
|
// CFF strings IDs 0...390 are predefined names, so refering |
|
// to entries in our own String INDEX starts at SID 391. |
|
topDict.setByName('version', 391); |
|
topDict.setByName('Notice', 392); |
|
topDict.setByName('FullName', 393); |
|
topDict.setByName('FamilyName', 394); |
|
topDict.setByName('Weight', 395); |
|
topDict.setByName('Encoding', null); // placeholder |
|
topDict.setByName('FontMatrix', properties.fontMatrix); |
|
topDict.setByName('FontBBox', properties.bbox); |
|
topDict.setByName('charset', null); // placeholder |
|
topDict.setByName('CharStrings', null); // placeholder |
|
topDict.setByName('Private', null); // placeholder |
|
cff.topDict = topDict; |
|
|
|
var strings = new CFFStrings(); |
|
strings.add('Version 0.11'); // Version |
|
strings.add('See original notice'); // Notice |
|
strings.add(name); // FullName |
|
strings.add(name); // FamilyName |
|
strings.add('Medium'); // Weight |
|
cff.strings = strings; |
|
|
|
cff.globalSubrIndex = new CFFIndex(); |
|
|
|
var count = glyphs.length; |
|
var charsetArray = [0]; |
|
for (var i = 0; i < count; i++) { |
|
var index = CFFStandardStrings.indexOf(charstrings[i].glyph); |
|
// Some characters like asterikmath && circlecopyrt are |
|
// missing from the original strings, for the moment let's |
|
// map them to .notdef and see later if it cause any |
|
// problems |
|
if (index == -1) |
|
index = 0; |
|
|
|
charsetArray.push((index >> 8) & 0xff, index & 0xff); |
|
} |
|
cff.charset = new CFFCharset(false, 0, [], charsetArray); |
|
|
|
var charStringsIndex = new CFFIndex(); |
|
charStringsIndex.add([0x8B, 0x0E]); // .notdef |
|
for (var i = 0; i < count; i++) { |
|
charStringsIndex.add(glyphs[i]); |
|
} |
|
cff.charStrings = charStringsIndex; |
|
|
|
var privateDict = new CFFPrivateDict(); |
|
privateDict.setByName('Subrs', null); // placeholder |
|
var fields = [ |
|
'BlueValues', |
|
'OtherBlues', |
|
'FamilyBlues', |
|
'FamilyOtherBlues', |
|
'StemSnapH', |
|
'StemSnapV', |
|
'BlueShift', |
|
'BlueFuzz', |
|
'BlueScale', |
|
'LanguageGroup', |
|
'ExpansionFactor', |
|
'ForceBold', |
|
'StdHW', |
|
'StdVW' |
|
]; |
|
for (var i = 0, ii = fields.length; i < ii; i++) { |
|
var field = fields[i]; |
|
if (!properties.privateData.hasOwnProperty(field)) |
|
continue; |
|
privateDict.setByName(field, properties.privateData[field]); |
|
} |
|
cff.topDict.privateDict = privateDict; |
|
|
|
var subrIndex = new CFFIndex(); |
|
for (var i = 0, ii = subrs.length; i < ii; i++) { |
|
subrIndex.add(subrs[i]); |
|
} |
|
privateDict.subrsIndex = subrIndex; |
|
|
|
var compiler = new CFFCompiler(cff); |
|
return compiler.compile(); |
|
} |
|
}; |
|
|
|
var CFFFont = (function CFFFontClosure() { |
|
function CFFFont(file, properties) { |
|
this.properties = properties; |
|
|
|
var parser = new CFFParser(file, properties); |
|
var cff = parser.parse(); |
|
var compiler = new CFFCompiler(cff); |
|
this.readExtra(cff); |
|
try { |
|
this.data = compiler.compile(); |
|
} catch (e) { |
|
warn('Failed to compile font ' + properties.loadedName); |
|
// There may have just been an issue with the compiler, set the data |
|
// anyway and hope the font loaded. |
|
this.data = file; |
|
} |
|
} |
|
|
|
CFFFont.prototype = { |
|
readExtra: function CFFFont_readExtra(cff) { |
|
// charstrings contains info about glyphs (one element per glyph |
|
// containing mappings for {unicode, width}) |
|
var charset = cff.charset.charset; |
|
var encoding = cff.encoding ? cff.encoding.encoding : null; |
|
var charstrings = this.getCharStrings(charset, encoding); |
|
|
|
// create the mapping between charstring and glyph id |
|
var glyphIds = []; |
|
for (var i = 0, ii = charstrings.length; i < ii; i++) |
|
glyphIds.push(charstrings[i].gid); |
|
|
|
this.charstrings = charstrings; |
|
this.glyphIds = glyphIds; |
|
this.seacs = cff.seacs; |
|
}, |
|
getCharStrings: function CFFFont_getCharStrings(charsets, encoding) { |
|
var charstrings = []; |
|
var unicodeUsed = []; |
|
var unassignedUnicodeItems = []; |
|
var inverseEncoding = []; |
|
// CID fonts don't have an encoding. |
|
if (encoding !== null) |
|
for (var charcode in encoding) |
|
inverseEncoding[encoding[charcode]] = charcode | 0; |
|
else |
|
inverseEncoding = charsets; |
|
var i = charsets[0] == '.notdef' ? 1 : 0; |
|
for (var ii = charsets.length; i < ii; i++) { |
|
var glyph = charsets[i]; |
|
|
|
var code = inverseEncoding[i]; |
|
if (!code || isSpecialUnicode(code)) { |
|
unassignedUnicodeItems.push(i); |
|
continue; |
|
} |
|
charstrings.push({ |
|
unicode: code, |
|
code: code, |
|
gid: i, |
|
glyph: glyph |
|
}); |
|
unicodeUsed[code] = true; |
|
} |
|
|
|
var nextUnusedUnicode = CMAP_GLYPH_OFFSET; |
|
for (var j = 0, jj = unassignedUnicodeItems.length; j < jj; ++j) { |
|
var i = unassignedUnicodeItems[j]; |
|
// giving unicode value anyway |
|
while (nextUnusedUnicode in unicodeUsed) |
|
nextUnusedUnicode++; |
|
var unicode = nextUnusedUnicode++; |
|
charstrings.push({ |
|
unicode: unicode, |
|
code: inverseEncoding[i] || 0, |
|
gid: i, |
|
glyph: charsets[i] |
|
}); |
|
} |
|
|
|
// sort the array by the unicode value (again) |
|
charstrings.sort(function getCharStringsSort(a, b) { |
|
return a.unicode - b.unicode; |
|
}); |
|
return charstrings; |
|
} |
|
}; |
|
|
|
return CFFFont; |
|
})(); |
|
|
|
var CFFParser = (function CFFParserClosure() { |
|
var CharstringValidationData = [ |
|
null, |
|
{ id: 'hstem', min: 2, resetStack: true, stem: true }, |
|
null, |
|
{ id: 'vstem', min: 2, resetStack: true, stem: true }, |
|
{ id: 'vmoveto', min: 1, resetStack: true }, |
|
{ id: 'rlineto', min: 2, resetStack: true }, |
|
{ id: 'hlineto', min: 1, resetStack: true }, |
|
{ id: 'vlineto', min: 1, resetStack: true }, |
|
{ id: 'rrcurveto', min: 6, resetStack: true }, |
|
null, |
|
{ id: 'callsubr', min: 1, undefStack: true }, |
|
{ id: 'return', min: 0, undefStack: true }, |
|
null, // 12 |
|
null, |
|
null, // endchar |
|
null, |
|
null, |
|
null, |
|
{ id: 'hstemhm', min: 2, resetStack: true, stem: true }, |
|
null, // hintmask |
|
null, // cntrmask |
|
{ id: 'rmoveto', min: 2, resetStack: true }, |
|
{ id: 'hmoveto', min: 1, resetStack: true }, |
|
{ id: 'vstemhm', min: 2, resetStack: true, stem: true }, |
|
{ id: 'rcurveline', min: 8, resetStack: true }, |
|
{ id: 'rlinecurve', min: 8, resetStack: true }, |
|
{ id: 'vvcurveto', min: 4, resetStack: true }, |
|
{ id: 'hhcurveto', min: 4, resetStack: true }, |
|
null, // shortint |
|
{ id: 'callgsubr', min: 1, undefStack: true }, |
|
{ id: 'vhcurveto', min: 4, resetStack: true }, |
|
{ id: 'hvcurveto', min: 4, resetStack: true } |
|
]; |
|
var CharstringValidationData12 = [ |
|
null, |
|
null, |
|
null, |
|
{ id: 'and', min: 2, stackDelta: -1 }, |
|
{ id: 'or', min: 2, stackDelta: -1 }, |
|
{ id: 'not', min: 1, stackDelta: 0 }, |
|
null, |
|
null, |
|
null, |
|
{ id: 'abs', min: 1, stackDelta: 0 }, |
|
{ id: 'add', min: 2, stackDelta: -1, |
|
stackFn: function stack_div(stack, index) { |
|
stack[index - 2] = stack[index - 2] + stack[index - 1]; |
|
} |
|
}, |
|
{ id: 'sub', min: 2, stackDelta: -1, |
|
stackFn: function stack_div(stack, index) { |
|
stack[index - 2] = stack[index - 2] - stack[index - 1]; |
|
} |
|
}, |
|
{ id: 'div', min: 2, stackDelta: -1, |
|
stackFn: function stack_div(stack, index) { |
|
stack[index - 2] = stack[index - 2] / stack[index - 1]; |
|
} |
|
}, |
|
null, |
|
{ id: 'neg', min: 1, stackDelta: 0, |
|
stackFn: function stack_div(stack, index) { |
|
stack[index - 1] = -stack[index - 1]; |
|
} |
|
}, |
|
{ id: 'eq', min: 2, stackDelta: -1 }, |
|
null, |
|
null, |
|
{ id: 'drop', min: 1, stackDelta: -1 }, |
|
null, |
|
{ id: 'put', min: 2, stackDelta: -2 }, |
|
{ id: 'get', min: 1, stackDelta: 0 }, |
|
{ id: 'ifelse', min: 4, stackDelta: -3 }, |
|
{ id: 'random', min: 0, stackDelta: 1 }, |
|
{ id: 'mul', min: 2, stackDelta: -1, |
|
stackFn: function stack_div(stack, index) { |
|
stack[index - 2] = stack[index - 2] * stack[index - 1]; |
|
} |
|
}, |
|
null, |
|
{ id: 'sqrt', min: 1, stackDelta: 0 }, |
|
{ id: 'dup', min: 1, stackDelta: 1 }, |
|
{ id: 'exch', min: 2, stackDelta: 0 }, |
|
{ id: 'index', min: 2, stackDelta: 0 }, |
|
{ id: 'roll', min: 3, stackDelta: -2 }, |
|
null, |
|
null, |
|
null, |
|
{ id: 'hflex', min: 7, resetStack: true }, |
|
{ id: 'flex', min: 13, resetStack: true }, |
|
{ id: 'hflex1', min: 9, resetStack: true }, |
|
{ id: 'flex1', min: 11, resetStack: true } |
|
]; |
|
|
|
function CFFParser(file, properties) { |
|
this.bytes = file.getBytes(); |
|
this.properties = properties; |
|
} |
|
CFFParser.prototype = { |
|
parse: function CFFParser_parse() { |
|
var properties = this.properties; |
|
var cff = new CFF(); |
|
this.cff = cff; |
|
|
|
// The first five sections must be in order, all the others are reached |
|
// via offsets contained in one of the below. |
|
var header = this.parseHeader(); |
|
var nameIndex = this.parseIndex(header.endPos); |
|
var topDictIndex = this.parseIndex(nameIndex.endPos); |
|
var stringIndex = this.parseIndex(topDictIndex.endPos); |
|
var globalSubrIndex = this.parseIndex(stringIndex.endPos); |
|
|
|
var topDictParsed = this.parseDict(topDictIndex.obj.get(0)); |
|
var topDict = this.createDict(CFFTopDict, topDictParsed, cff.strings); |
|
|
|
cff.header = header.obj; |
|
cff.names = this.parseNameIndex(nameIndex.obj); |
|
cff.strings = this.parseStringIndex(stringIndex.obj); |
|
cff.topDict = topDict; |
|
cff.globalSubrIndex = globalSubrIndex.obj; |
|
|
|
this.parsePrivateDict(cff.topDict); |
|
|
|
cff.isCIDFont = topDict.hasName('ROS'); |
|
|
|
var charStringOffset = topDict.getByName('CharStrings'); |
|
var charStringsAndSeacs = this.parseCharStrings(charStringOffset); |
|
cff.charStrings = charStringsAndSeacs.charStrings; |
|
cff.seacs = charStringsAndSeacs.seacs; |
|
|
|
var fontMatrix = topDict.getByName('FontMatrix'); |
|
if (fontMatrix) { |
|
properties.fontMatrix = fontMatrix; |
|
} |
|
|
|
var fontBBox = topDict.getByName('FontBBox'); |
|
if (fontBBox) { |
|
// adjusting ascent/descent |
|
properties.ascent = fontBBox[3]; |
|
properties.descent = fontBBox[1]; |
|
properties.ascentScaled = true; |
|
} |
|
|
|
var charset, encoding; |
|
if (cff.isCIDFont) { |
|
var fdArrayIndex = this.parseIndex(topDict.getByName('FDArray')).obj; |
|
for (var i = 0, ii = fdArrayIndex.count; i < ii; ++i) { |
|
var dictRaw = fdArrayIndex.get(i); |
|
var fontDict = this.createDict(CFFTopDict, this.parseDict(dictRaw), |
|
cff.strings); |
|
this.parsePrivateDict(fontDict); |
|
cff.fdArray.push(fontDict); |
|
} |
|
// cid fonts don't have an encoding |
|
encoding = null; |
|
charset = this.parseCharsets(topDict.getByName('charset'), |
|
cff.charStrings.count, cff.strings, true); |
|
cff.fdSelect = this.parseFDSelect(topDict.getByName('FDSelect'), |
|
cff.charStrings.count); |
|
} else { |
|
charset = this.parseCharsets(topDict.getByName('charset'), |
|
cff.charStrings.count, cff.strings, false); |
|
encoding = this.parseEncoding(topDict.getByName('Encoding'), |
|
properties, |
|
cff.strings, charset.charset); |
|
} |
|
cff.charset = charset; |
|
cff.encoding = encoding; |
|
|
|
return cff; |
|
}, |
|
parseHeader: function CFFParser_parseHeader() { |
|
var bytes = this.bytes; |
|
var offset = 0; |
|
|
|
while (bytes[offset] != 1) |
|
++offset; |
|
|
|
if (offset !== 0) { |
|
info('cff data is shifted'); |
|
bytes = bytes.subarray(offset); |
|
this.bytes = bytes; |
|
} |
|
var major = bytes[0]; |
|
var minor = bytes[1]; |
|
var hdrSize = bytes[2]; |
|
var offSize = bytes[3]; |
|
var header = new CFFHeader(major, minor, hdrSize, offSize); |
|
return {obj: header, endPos: hdrSize}; |
|
}, |
|
parseDict: function CFFParser_parseDict(dict) { |
|
var pos = 0; |
|
|
|
function parseOperand() { |
|
var value = dict[pos++]; |
|
if (value === 30) { |
|
return parseFloatOperand(pos); |
|
} else if (value === 28) { |
|
value = dict[pos++]; |
|
value = ((value << 24) | (dict[pos++] << 16)) >> 16; |
|
return value; |
|
} else if (value === 29) { |
|
value = dict[pos++]; |
|
value = (value << 8) | dict[pos++]; |
|
value = (value << 8) | dict[pos++]; |
|
value = (value << 8) | dict[pos++]; |
|
return value; |
|
} else if (value >= 32 && value <= 246) { |
|
return value - 139; |
|
} else if (value >= 247 && value <= 250) { |
|
return ((value - 247) * 256) + dict[pos++] + 108; |
|
} else if (value >= 251 && value <= 254) { |
|
return -((value - 251) * 256) - dict[pos++] - 108; |
|
} else { |
|
error('255 is not a valid DICT command'); |
|
} |
|
return -1; |
|
} |
|
|
|
function parseFloatOperand() { |
|
var str = ''; |
|
var eof = 15; |
|
var lookup = ['0', '1', '2', '3', '4', '5', '6', '7', '8', |
|
'9', '.', 'E', 'E-', null, '-']; |
|
var length = dict.length; |
|
while (pos < length) { |
|
var b = dict[pos++]; |
|
var b1 = b >> 4; |
|
var b2 = b & 15; |
|
|
|
if (b1 == eof) |
|
break; |
|
str += lookup[b1]; |
|
|
|
if (b2 == eof) |
|
break; |
|
str += lookup[b2]; |
|
} |
|
return parseFloat(str); |
|
} |
|
|
|
var operands = []; |
|
var entries = []; |
|
|
|
var pos = 0; |
|
var end = dict.length; |
|
while (pos < end) { |
|
var b = dict[pos]; |
|
if (b <= 21) { |
|
if (b === 12) |
|
b = (b << 8) | dict[++pos]; |
|
entries.push([b, operands]); |
|
operands = []; |
|
++pos; |
|
} else { |
|
operands.push(parseOperand()); |
|
} |
|
} |
|
return entries; |
|
}, |
|
parseIndex: function CFFParser_parseIndex(pos) { |
|
var cffIndex = new CFFIndex(); |
|
var bytes = this.bytes; |
|
var count = (bytes[pos++] << 8) | bytes[pos++]; |
|
var offsets = []; |
|
var start = pos; |
|
var end = pos; |
|
|
|
if (count !== 0) { |
|
var offsetSize = bytes[pos++]; |
|
// add 1 for offset to determine size of last object |
|
var startPos = pos + ((count + 1) * offsetSize) - 1; |
|
|
|
for (var i = 0, ii = count + 1; i < ii; ++i) { |
|
var offset = 0; |
|
for (var j = 0; j < offsetSize; ++j) { |
|
offset <<= 8; |
|
offset += bytes[pos++]; |
|
} |
|
offsets.push(startPos + offset); |
|
} |
|
end = offsets[count]; |
|
} |
|
for (var i = 0, ii = offsets.length - 1; i < ii; ++i) { |
|
var offsetStart = offsets[i]; |
|
var offsetEnd = offsets[i + 1]; |
|
cffIndex.add(bytes.subarray(offsetStart, offsetEnd)); |
|
} |
|
return {obj: cffIndex, endPos: end}; |
|
}, |
|
parseNameIndex: function CFFParser_parseNameIndex(index) { |
|
var names = []; |
|
for (var i = 0, ii = index.count; i < ii; ++i) { |
|
var name = index.get(i); |
|
// OTS doesn't allow names to be over 127 characters. |
|
var length = Math.min(name.length, 127); |
|
var data = []; |
|
// OTS also only permits certain characters in the name. |
|
for (var j = 0; j < length; ++j) { |
|
var c = name[j]; |
|
if (j === 0 && c === 0) { |
|
data[j] = c; |
|
continue; |
|
} |
|
if ((c < 33 || c > 126) || c === 91 /* [ */ || c === 93 /* ] */ || |
|
c === 40 /* ( */ || c === 41 /* ) */ || c === 123 /* { */ || |
|
c === 125 /* } */ || c === 60 /* < */ || c === 62 /* > */ || |
|
c === 47 /* / */ || c === 37 /* % */) { |
|
data[j] = 95; |
|
continue; |
|
} |
|
data[j] = c; |
|
} |
|
names.push(String.fromCharCode.apply(null, data)); |
|
} |
|
return names; |
|
}, |
|
parseStringIndex: function CFFParser_parseStringIndex(index) { |
|
var strings = new CFFStrings(); |
|
for (var i = 0, ii = index.count; i < ii; ++i) { |
|
var data = index.get(i); |
|
strings.add(String.fromCharCode.apply(null, data)); |
|
} |
|
return strings; |
|
}, |
|
createDict: function CFFParser_createDict(Type, dict, strings) { |
|
var cffDict = new Type(strings); |
|
var types = cffDict.types; |
|
|
|
for (var i = 0, ii = dict.length; i < ii; ++i) { |
|
var pair = dict[i]; |
|
var key = pair[0]; |
|
var value = pair[1]; |
|
cffDict.setByKey(key, value); |
|
} |
|
return cffDict; |
|
}, |
|
parseCharStrings: function CFFParser_parseCharStrings(charStringOffset) { |
|
var charStrings = this.parseIndex(charStringOffset).obj; |
|
var seacs = []; |
|
var count = charStrings.count; |
|
for (var i = 0; i < count; i++) { |
|
var charstring = charStrings.get(i); |
|
|
|
var stackSize = 0; |
|
var stack = []; |
|
var undefStack = true; |
|
var hints = 0; |
|
var valid = true; |
|
var data = charstring; |
|
var length = data.length; |
|
for (var j = 0; j < length;) { |
|
var value = data[j++]; |
|
var validationCommand = null; |
|
if (value == 12) { |
|
var q = data[j++]; |
|
if (q === 0) { |
|
// The CFF specification state that the 'dotsection' command |
|
// (12, 0) is deprecated and treated as a no-op, but all Type2 |
|
// charstrings processors should support them. Unfortunately |
|
// the font sanitizer don't. As a workaround the sequence (12, 0) |
|
// is replaced by a useless (0, hmoveto). |
|
data[j - 2] = 139; |
|
data[j - 1] = 22; |
|
stackSize = 0; |
|
} else { |
|
validationCommand = CharstringValidationData12[q]; |
|
} |
|
} else if (value === 28) { // number (16 bit) |
|
stack[stackSize] = ((data[j] << 24) | (data[j + 1] << 16)) >> 16; |
|
j += 2; |
|
stackSize++; |
|
} else if (value == 14) { |
|
if (stackSize >= 4) { |
|
stackSize -= 4; |
|
if (SEAC_ANALYSIS_ENABLED) { |
|
seacs[i] = stack.slice(stackSize, stackSize + 4); |
|
valid = false; |
|
} |
|
} |
|
} else if (value >= 32 && value <= 246) { // number |
|
stack[stackSize] = value - 139; |
|
stackSize++; |
|
} else if (value >= 247 && value <= 254) { // number (+1 bytes) |
|
stack[stackSize] = value < 251 ? |
|
((value - 247) << 8) + data[j] + 108 : |
|
-((value - 251) << 8) - data[j] - 108; |
|
j++; |
|
stackSize++; |
|
} else if (value == 255) { // number (32 bit) |
|
stack[stackSize] = ((data[j] << 24) | (data[j + 1] << 16) | |
|
(data[j + 2] << 8) | data[j + 3]) / 65536; |
|
j += 4; |
|
stackSize++; |
|
} else if (value == 19 || value == 20) { |
|
hints += stackSize >> 1; |
|
j += (hints + 7) >> 3; // skipping right amount of hints flag data |
|
stackSize = 0; |
|
} else { |
|
validationCommand = CharstringValidationData[value]; |
|
} |
|
if (validationCommand) { |
|
if (validationCommand.stem) { |
|
hints += stackSize >> 1; |
|
} |
|
if ('min' in validationCommand) { |
|
if (!undefStack && stackSize < validationCommand.min) { |
|
warn('Not enough parameters for ' + validationCommand.id + |
|
'; actual: ' + stackSize + |
|
', expected: ' + validationCommand.min); |
|
valid = false; |
|
break; |
|
} |
|
} |
|
if ('stackDelta' in validationCommand) { |
|
if ('stackFn' in validationCommand) { |
|
validationCommand.stackFn(stack, stackSize); |
|
} |
|
stackSize += validationCommand.stackDelta; |
|
} else if (validationCommand.resetStack) { |
|
stackSize = 0; |
|
undefStack = false; |
|
} else if (validationCommand.undefStack) { |
|
stackSize = 0; |
|
undefStack = true; |
|
} |
|
} |
|
} |
|
if (!valid) { |
|
// resetting invalid charstring to single 'endchar' |
|
charStrings.set(i, new Uint8Array([14])); |
|
} |
|
} |
|
return { charStrings: charStrings, seacs: seacs }; |
|
}, |
|
emptyPrivateDictionary: |
|
function CFFParser_emptyPrivateDictionary(parentDict) { |
|
var privateDict = this.createDict(CFFPrivateDict, [], |
|
parentDict.strings); |
|
parentDict.setByKey(18, [0, 0]); |
|
parentDict.privateDict = privateDict; |
|
}, |
|
parsePrivateDict: function CFFParser_parsePrivateDict(parentDict) { |
|
// no private dict, do nothing |
|
if (!parentDict.hasName('Private')) { |
|
this.emptyPrivateDictionary(parentDict); |
|
return; |
|
} |
|
var privateOffset = parentDict.getByName('Private'); |
|
// make sure the params are formatted correctly |
|
if (!isArray(privateOffset) || privateOffset.length !== 2) { |
|
parentDict.removeByName('Private'); |
|
return; |
|
} |
|
var size = privateOffset[0]; |
|
var offset = privateOffset[1]; |
|
// remove empty dicts or ones that refer to invalid location |
|
if (size === 0 || offset >= this.bytes.length) { |
|
this.emptyPrivateDictionary(parentDict); |
|
return; |
|
} |
|
|
|
var privateDictEnd = offset + size; |
|
var dictData = this.bytes.subarray(offset, privateDictEnd); |
|
var dict = this.parseDict(dictData); |
|
var privateDict = this.createDict(CFFPrivateDict, dict, |
|
parentDict.strings); |
|
parentDict.privateDict = privateDict; |
|
|
|
// Parse the Subrs index also since it's relative to the private dict. |
|
if (!privateDict.getByName('Subrs')) |
|
return; |
|
var subrsOffset = privateDict.getByName('Subrs'); |
|
var relativeOffset = offset + subrsOffset; |
|
// Validate the offset. |
|
if (subrsOffset === 0 || relativeOffset >= this.bytes.length) { |
|
this.emptyPrivateDictionary(parentDict); |
|
return; |
|
} |
|
var subrsIndex = this.parseIndex(relativeOffset); |
|
privateDict.subrsIndex = subrsIndex.obj; |
|
}, |
|
parseCharsets: function CFFParser_parseCharsets(pos, length, strings, cid) { |
|
if (pos === 0) { |
|
return new CFFCharset(true, CFFCharsetPredefinedTypes.ISO_ADOBE, |
|
ISOAdobeCharset); |
|
} else if (pos == 1) { |
|
return new CFFCharset(true, CFFCharsetPredefinedTypes.EXPERT, |
|
ExpertCharset); |
|
} else if (pos == 2) { |
|
return new CFFCharset(true, CFFCharsetPredefinedTypes.EXPERT_SUBSET, |
|
ExpertSubsetCharset); |
|
} |
|
|
|
var bytes = this.bytes; |
|
var start = pos; |
|
var format = bytes[pos++]; |
|
var charset = ['.notdef']; |
|
|
|
// subtract 1 for the .notdef glyph |
|
length -= 1; |
|
|
|
switch (format) { |
|
case 0: |
|
for (var i = 0; i < length; i++) { |
|
var id = (bytes[pos++] << 8) | bytes[pos++]; |
|
charset.push(cid ? id : strings.get(id)); |
|
} |
|
break; |
|
case 1: |
|
while (charset.length <= length) { |
|
var id = (bytes[pos++] << 8) | bytes[pos++]; |
|
var count = bytes[pos++]; |
|
for (var i = 0; i <= count; i++) |
|
charset.push(cid ? id++ : strings.get(id++)); |
|
} |
|
break; |
|
case 2: |
|
while (charset.length <= length) { |
|
var id = (bytes[pos++] << 8) | bytes[pos++]; |
|
var count = (bytes[pos++] << 8) | bytes[pos++]; |
|
for (var i = 0; i <= count; i++) |
|
charset.push(cid ? id++ : strings.get(id++)); |
|
} |
|
break; |
|
default: |
|
error('Unknown charset format'); |
|
} |
|
// Raw won't be needed if we actually compile the charset. |
|
var end = pos; |
|
var raw = bytes.subarray(start, end); |
|
|
|
return new CFFCharset(false, format, charset, raw); |
|
}, |
|
parseEncoding: function CFFParser_parseEncoding(pos, |
|
properties, |
|
strings, |
|
charset) { |
|
var encoding = {}; |
|
var bytes = this.bytes; |
|
var predefined = false; |
|
var hasSupplement = false; |
|
var format; |
|
var raw = null; |
|
|
|
function readSupplement() { |
|
var supplementsCount = bytes[pos++]; |
|
for (var i = 0; i < supplementsCount; i++) { |
|
var code = bytes[pos++]; |
|
var sid = (bytes[pos++] << 8) + (bytes[pos++] & 0xff); |
|
encoding[code] = properties.differences.indexOf(strings.get(sid)); |
|
} |
|
} |
|
|
|
if (pos === 0 || pos == 1) { |
|
predefined = true; |
|
format = pos; |
|
var baseEncoding = pos ? Encodings.ExpertEncoding : |
|
Encodings.StandardEncoding; |
|
for (var i = 0, ii = charset.length; i < ii; i++) { |
|
var index = baseEncoding.indexOf(charset[i]); |
|
if (index != -1) { |
|
encoding[index] = i; |
|
} |
|
} |
|
} else { |
|
var dataStart = pos; |
|
var format = bytes[pos++]; |
|
switch (format & 0x7f) { |
|
case 0: |
|
var glyphsCount = bytes[pos++]; |
|
for (var i = 1; i <= glyphsCount; i++) |
|
encoding[bytes[pos++]] = i; |
|
break; |
|
|
|
case 1: |
|
var rangesCount = bytes[pos++]; |
|
var gid = 1; |
|
for (var i = 0; i < rangesCount; i++) { |
|
var start = bytes[pos++]; |
|
var left = bytes[pos++]; |
|
for (var j = start; j <= start + left; j++) |
|
encoding[j] = gid++; |
|
} |
|
break; |
|
|
|
default: |
|
error('Unknow encoding format: ' + format + ' in CFF'); |
|
break; |
|
} |
|
var dataEnd = pos; |
|
if (format & 0x80) { |
|
// The font sanitizer does not support CFF encoding with a |
|
// supplement, since the encoding is not really used to map |
|
// between gid to glyph, let's overwrite what is declared in |
|
// the top dictionary to let the sanitizer think the font use |
|
// StandardEncoding, that's a lie but that's ok. |
|
bytes[dataStart] &= 0x7f; |
|
readSupplement(); |
|
hasSupplement = true; |
|
} |
|
raw = bytes.subarray(dataStart, dataEnd); |
|
} |
|
format = format & 0x7f; |
|
return new CFFEncoding(predefined, format, encoding, raw); |
|
}, |
|
parseFDSelect: function CFFParser_parseFDSelect(pos, length) { |
|
var start = pos; |
|
var bytes = this.bytes; |
|
var format = bytes[pos++]; |
|
var fdSelect = []; |
|
switch (format) { |
|
case 0: |
|
for (var i = 0; i < length; ++i) { |
|
var id = bytes[pos++]; |
|
fdSelect.push(id); |
|
} |
|
break; |
|
case 3: |
|
var rangesCount = (bytes[pos++] << 8) | bytes[pos++]; |
|
for (var i = 0; i < rangesCount; ++i) { |
|
var first = (bytes[pos++] << 8) | bytes[pos++]; |
|
var fdIndex = bytes[pos++]; |
|
var next = (bytes[pos] << 8) | bytes[pos + 1]; |
|
for (var j = first; j < next; ++j) |
|
fdSelect.push(fdIndex); |
|
} |
|
// Advance past the sentinel(next). |
|
pos += 2; |
|
break; |
|
default: |
|
error('Unknown fdselect format ' + format); |
|
break; |
|
} |
|
var end = pos; |
|
return new CFFFDSelect(fdSelect, bytes.subarray(start, end)); |
|
} |
|
}; |
|
return CFFParser; |
|
})(); |
|
|
|
// Compact Font Format |
|
var CFF = (function CFFClosure() { |
|
function CFF() { |
|
this.header = null; |
|
this.names = []; |
|
this.topDict = null; |
|
this.strings = new CFFStrings(); |
|
this.globalSubrIndex = null; |
|
|
|
// The following could really be per font, but since we only have one font |
|
// store them here. |
|
this.encoding = null; |
|
this.charset = null; |
|
this.charStrings = null; |
|
this.fdArray = []; |
|
this.fdSelect = null; |
|
|
|
this.isCIDFont = false; |
|
} |
|
return CFF; |
|
})(); |
|
|
|
var CFFHeader = (function CFFHeaderClosure() { |
|
function CFFHeader(major, minor, hdrSize, offSize) { |
|
this.major = major; |
|
this.minor = minor; |
|
this.hdrSize = hdrSize; |
|
this.offSize = offSize; |
|
} |
|
return CFFHeader; |
|
})(); |
|
|
|
var CFFStrings = (function CFFStringsClosure() { |
|
function CFFStrings() { |
|
this.strings = []; |
|
} |
|
CFFStrings.prototype = { |
|
get: function CFFStrings_get(index) { |
|
if (index >= 0 && index <= 390) |
|
return CFFStandardStrings[index]; |
|
if (index - 391 <= this.strings.length) |
|
return this.strings[index - 391]; |
|
return CFFStandardStrings[0]; |
|
}, |
|
add: function CFFStrings_add(value) { |
|
this.strings.push(value); |
|
}, |
|
get count() { |
|
return this.strings.length; |
|
} |
|
}; |
|
return CFFStrings; |
|
})(); |
|
|
|
var CFFIndex = (function CFFIndexClosure() { |
|
function CFFIndex() { |
|
this.objects = []; |
|
this.length = 0; |
|
} |
|
CFFIndex.prototype = { |
|
add: function CFFIndex_add(data) { |
|
this.length += data.length; |
|
this.objects.push(data); |
|
}, |
|
set: function CFFIndex_set(index, data) { |
|
this.length += data.length - this.objects[index].length; |
|
this.objects[index] = data; |
|
}, |
|
get: function CFFIndex_get(index) { |
|
return this.objects[index]; |
|
}, |
|
get count() { |
|
return this.objects.length; |
|
} |
|
}; |
|
return CFFIndex; |
|
})(); |
|
|
|
var CFFDict = (function CFFDictClosure() { |
|
function CFFDict(tables, strings) { |
|
this.keyToNameMap = tables.keyToNameMap; |
|
this.nameToKeyMap = tables.nameToKeyMap; |
|
this.defaults = tables.defaults; |
|
this.types = tables.types; |
|
this.opcodes = tables.opcodes; |
|
this.order = tables.order; |
|
this.strings = strings; |
|
this.values = {}; |
|
} |
|
CFFDict.prototype = { |
|
// value should always be an array |
|
setByKey: function CFFDict_setByKey(key, value) { |
|
if (!(key in this.keyToNameMap)) |
|
return false; |
|
// ignore empty values |
|
if (value.length === 0) |
|
return true; |
|
var type = this.types[key]; |
|
// remove the array wrapping these types of values |
|
if (type === 'num' || type === 'sid' || type === 'offset') |
|
value = value[0]; |
|
this.values[key] = value; |
|
return true; |
|
}, |
|
setByName: function CFFDict_setByName(name, value) { |
|
if (!(name in this.nameToKeyMap)) { |
|
error('Invalid dictionary name "' + name + '"'); |
|
} |
|
this.values[this.nameToKeyMap[name]] = value; |
|
}, |
|
hasName: function CFFDict_hasName(name) { |
|
return this.nameToKeyMap[name] in this.values; |
|
}, |
|
getByName: function CFFDict_getByName(name) { |
|
if (!(name in this.nameToKeyMap)) |
|
error('Invalid dictionary name "' + name + '"'); |
|
var key = this.nameToKeyMap[name]; |
|
if (!(key in this.values)) |
|
return this.defaults[key]; |
|
return this.values[key]; |
|
}, |
|
removeByName: function CFFDict_removeByName(name) { |
|
delete this.values[this.nameToKeyMap[name]]; |
|
} |
|
}; |
|
CFFDict.createTables = function CFFDict_createTables(layout) { |
|
var tables = { |
|
keyToNameMap: {}, |
|
nameToKeyMap: {}, |
|
defaults: {}, |
|
types: {}, |
|
opcodes: {}, |
|
order: [] |
|
}; |
|
for (var i = 0, ii = layout.length; i < ii; ++i) { |
|
var entry = layout[i]; |
|
var key = isArray(entry[0]) ? (entry[0][0] << 8) + entry[0][1] : entry[0]; |
|
tables.keyToNameMap[key] = entry[1]; |
|
tables.nameToKeyMap[entry[1]] = key; |
|
tables.types[key] = entry[2]; |
|
tables.defaults[key] = entry[3]; |
|
tables.opcodes[key] = isArray(entry[0]) ? entry[0] : [entry[0]]; |
|
tables.order.push(key); |
|
} |
|
return tables; |
|
}; |
|
return CFFDict; |
|
})(); |
|
|
|
var CFFTopDict = (function CFFTopDictClosure() { |
|
var layout = [ |
|
[[12, 30], 'ROS', ['sid', 'sid', 'num'], null], |
|
[[12, 20], 'SyntheticBase', 'num', null], |
|
[0, 'version', 'sid', null], |
|
[1, 'Notice', 'sid', null], |
|
[[12, 0], 'Copyright', 'sid', null], |
|
[2, 'FullName', 'sid', null], |
|
[3, 'FamilyName', 'sid', null], |
|
[4, 'Weight', 'sid', null], |
|
[[12, 1], 'isFixedPitch', 'num', 0], |
|
[[12, 2], 'ItalicAngle', 'num', 0], |
|
[[12, 3], 'UnderlinePosition', 'num', -100], |
|
[[12, 4], 'UnderlineThickness', 'num', 50], |
|
[[12, 5], 'PaintType', 'num', 0], |
|
[[12, 6], 'CharstringType', 'num', 2], |
|
[[12, 7], 'FontMatrix', ['num', 'num', 'num', 'num', 'num', 'num'], |
|
[0.001, 0, 0, 0.001, 0, 0]], |
|
[13, 'UniqueID', 'num', null], |
|
[5, 'FontBBox', ['num', 'num', 'num', 'num'], [0, 0, 0, 0]], |
|
[[12, 8], 'StrokeWidth', 'num', 0], |
|
[14, 'XUID', 'array', null], |
|
[15, 'charset', 'offset', 0], |
|
[16, 'Encoding', 'offset', 0], |
|
[17, 'CharStrings', 'offset', 0], |
|
[18, 'Private', ['offset', 'offset'], null], |
|
[[12, 21], 'PostScript', 'sid', null], |
|
[[12, 22], 'BaseFontName', 'sid', null], |
|
[[12, 23], 'BaseFontBlend', 'delta', null], |
|
[[12, 31], 'CIDFontVersion', 'num', 0], |
|
[[12, 32], 'CIDFontRevision', 'num', 0], |
|
[[12, 33], 'CIDFontType', 'num', 0], |
|
[[12, 34], 'CIDCount', 'num', 8720], |
|
[[12, 35], 'UIDBase', 'num', null], |
|
// XXX: CID Fonts on DirectWrite 6.1 only seem to work if FDSelect comes |
|
// before FDArray. |
|
[[12, 37], 'FDSelect', 'offset', null], |
|
[[12, 36], 'FDArray', 'offset', null], |
|
[[12, 38], 'FontName', 'sid', null] |
|
]; |
|
var tables = null; |
|
function CFFTopDict(strings) { |
|
if (tables === null) |
|
tables = CFFDict.createTables(layout); |
|
CFFDict.call(this, tables, strings); |
|
this.privateDict = null; |
|
} |
|
CFFTopDict.prototype = Object.create(CFFDict.prototype); |
|
return CFFTopDict; |
|
})(); |
|
|
|
var CFFPrivateDict = (function CFFPrivateDictClosure() { |
|
var layout = [ |
|
[6, 'BlueValues', 'delta', null], |
|
[7, 'OtherBlues', 'delta', null], |
|
[8, 'FamilyBlues', 'delta', null], |
|
[9, 'FamilyOtherBlues', 'delta', null], |
|
[[12, 9], 'BlueScale', 'num', 0.039625], |
|
[[12, 10], 'BlueShift', 'num', 7], |
|
[[12, 11], 'BlueFuzz', 'num', 1], |
|
[10, 'StdHW', 'num', null], |
|
[11, 'StdVW', 'num', null], |
|
[[12, 12], 'StemSnapH', 'delta', null], |
|
[[12, 13], 'StemSnapV', 'delta', null], |
|
[[12, 14], 'ForceBold', 'num', 0], |
|
[[12, 17], 'LanguageGroup', 'num', 0], |
|
[[12, 18], 'ExpansionFactor', 'num', 0.06], |
|
[[12, 19], 'initialRandomSeed', 'num', 0], |
|
[20, 'defaultWidthX', 'num', 0], |
|
[21, 'nominalWidthX', 'num', 0], |
|
[19, 'Subrs', 'offset', null] |
|
]; |
|
var tables = null; |
|
function CFFPrivateDict(strings) { |
|
if (tables === null) |
|
tables = CFFDict.createTables(layout); |
|
CFFDict.call(this, tables, strings); |
|
this.subrsIndex = null; |
|
} |
|
CFFPrivateDict.prototype = Object.create(CFFDict.prototype); |
|
return CFFPrivateDict; |
|
})(); |
|
|
|
var CFFCharsetPredefinedTypes = { |
|
ISO_ADOBE: 0, |
|
EXPERT: 1, |
|
EXPERT_SUBSET: 2 |
|
}; |
|
var CFFCharsetEmbeddedTypes = { |
|
FORMAT0: 0, |
|
FORMAT1: 1, |
|
FORMAT2: 2 |
|
}; |
|
var CFFCharset = (function CFFCharsetClosure() { |
|
function CFFCharset(predefined, format, charset, raw) { |
|
this.predefined = predefined; |
|
this.format = format; |
|
this.charset = charset; |
|
this.raw = raw; |
|
} |
|
return CFFCharset; |
|
})(); |
|
|
|
var CFFEncodingPredefinedTypes = { |
|
STANDARD: 0, |
|
EXPERT: 1 |
|
}; |
|
var CFFCharsetEmbeddedTypes = { |
|
FORMAT0: 0, |
|
FORMAT1: 1 |
|
}; |
|
var CFFEncoding = (function CFFEncodingClosure() { |
|
function CFFEncoding(predefined, format, encoding, raw) { |
|
this.predefined = predefined; |
|
this.format = format; |
|
this.encoding = encoding; |
|
this.raw = raw; |
|
} |
|
return CFFEncoding; |
|
})(); |
|
|
|
var CFFFDSelect = (function CFFFDSelectClosure() { |
|
function CFFFDSelect(fdSelect, raw) { |
|
this.fdSelect = fdSelect; |
|
this.raw = raw; |
|
} |
|
return CFFFDSelect; |
|
})(); |
|
|
|
// Helper class to keep track of where an offset is within the data and helps |
|
// filling in that offset once it's known. |
|
var CFFOffsetTracker = (function CFFOffsetTrackerClosure() { |
|
function CFFOffsetTracker() { |
|
this.offsets = {}; |
|
} |
|
CFFOffsetTracker.prototype = { |
|
isTracking: function CFFOffsetTracker_isTracking(key) { |
|
return key in this.offsets; |
|
}, |
|
track: function CFFOffsetTracker_track(key, location) { |
|
if (key in this.offsets) |
|
error('Already tracking location of ' + key); |
|
this.offsets[key] = location; |
|
}, |
|
offset: function CFFOffsetTracker_offset(value) { |
|
for (var key in this.offsets) { |
|
this.offsets[key] += value; |
|
} |
|
}, |
|
setEntryLocation: function CFFOffsetTracker_setEntryLocation(key, |
|
values, |
|
output) { |
|
if (!(key in this.offsets)) |
|
error('Not tracking location of ' + key); |
|
var data = output.data; |
|
var dataOffset = this.offsets[key]; |
|
var size = 5; |
|
for (var i = 0, ii = values.length; i < ii; ++i) { |
|
var offset0 = i * size + dataOffset; |
|
var offset1 = offset0 + 1; |
|
var offset2 = offset0 + 2; |
|
var offset3 = offset0 + 3; |
|
var offset4 = offset0 + 4; |
|
// It's easy to screw up offsets so perform this sanity check. |
|
if (data[offset0] !== 0x1d || data[offset1] !== 0 || |
|
data[offset2] !== 0 || data[offset3] !== 0 || data[offset4] !== 0) |
|
error('writing to an offset that is not empty'); |
|
var value = values[i]; |
|
data[offset0] = 0x1d; |
|
data[offset1] = (value >> 24) & 0xFF; |
|
data[offset2] = (value >> 16) & 0xFF; |
|
data[offset3] = (value >> 8) & 0xFF; |
|
data[offset4] = value & 0xFF; |
|
} |
|
} |
|
}; |
|
return CFFOffsetTracker; |
|
})(); |
|
|
|
// Takes a CFF and converts it to the binary representation. |
|
var CFFCompiler = (function CFFCompilerClosure() { |
|
function stringToArray(str) { |
|
var array = []; |
|
for (var i = 0, ii = str.length; i < ii; ++i) |
|
array[i] = str.charCodeAt(i); |
|
|
|
return array; |
|
} |
|
function CFFCompiler(cff) { |
|
this.cff = cff; |
|
} |
|
CFFCompiler.prototype = { |
|
compile: function CFFCompiler_compile() { |
|
var cff = this.cff; |
|
var output = { |
|
data: [], |
|
length: 0, |
|
add: function CFFCompiler_add(data) { |
|
this.data = this.data.concat(data); |
|
this.length = this.data.length; |
|
} |
|
}; |
|
|
|
// Compile the five entries that must be in order. |
|
var header = this.compileHeader(cff.header); |
|
output.add(header); |
|
|
|
var nameIndex = this.compileNameIndex(cff.names); |
|
output.add(nameIndex); |
|
|
|
var compiled = this.compileTopDicts([cff.topDict], |
|
output.length, |
|
cff.isCIDFont); |
|
output.add(compiled.output); |
|
var topDictTracker = compiled.trackers[0]; |
|
|
|
var stringIndex = this.compileStringIndex(cff.strings.strings); |
|
output.add(stringIndex); |
|
|
|
var globalSubrIndex = this.compileIndex(cff.globalSubrIndex); |
|
output.add(globalSubrIndex); |
|
|
|
// Now start on the other entries that have no specfic order. |
|
if (cff.encoding && cff.topDict.hasName('Encoding')) { |
|
if (cff.encoding.predefined) { |
|
topDictTracker.setEntryLocation('Encoding', [cff.encoding.format], |
|
output); |
|
} else { |
|
var encoding = this.compileEncoding(cff.encoding); |
|
topDictTracker.setEntryLocation('Encoding', [output.length], output); |
|
output.add(encoding); |
|
} |
|
} |
|
|
|
if (cff.charset && cff.topDict.hasName('charset')) { |
|
if (cff.charset.predefined) { |
|
topDictTracker.setEntryLocation('charset', [cff.charset.format], |
|
output); |
|
} else { |
|
var charset = this.compileCharset(cff.charset); |
|
topDictTracker.setEntryLocation('charset', [output.length], output); |
|
output.add(charset); |
|
} |
|
} |
|
|
|
var charStrings = this.compileCharStrings(cff.charStrings); |
|
topDictTracker.setEntryLocation('CharStrings', [output.length], output); |
|
output.add(charStrings); |
|
|
|
if (cff.isCIDFont) { |
|
// For some reason FDSelect must be in front of FDArray on windows. OSX |
|
// and linux don't seem to care. |
|
topDictTracker.setEntryLocation('FDSelect', [output.length], output); |
|
var fdSelect = this.compileFDSelect(cff.fdSelect.raw); |
|
output.add(fdSelect); |
|
// It is unclear if the sub font dictionary can have CID related |
|
// dictionary keys, but the sanitizer doesn't like them so remove them. |
|
var compiled = this.compileTopDicts(cff.fdArray, output.length, true); |
|
topDictTracker.setEntryLocation('FDArray', [output.length], output); |
|
output.add(compiled.output); |
|
var fontDictTrackers = compiled.trackers; |
|
|
|
this.compilePrivateDicts(cff.fdArray, fontDictTrackers, output); |
|
} |
|
|
|
this.compilePrivateDicts([cff.topDict], [topDictTracker], output); |
|
|
|
// If the font data ends with INDEX whose object data is zero-length, |
|
// the sanitizer will bail out. Add a dummy byte to avoid that. |
|
output.add([0]); |
|
|
|
return output.data; |
|
}, |
|
encodeNumber: function CFFCompiler_encodeNumber(value) { |
|
if (parseFloat(value) == parseInt(value, 10) && !isNaN(value)) // isInt |
|
return this.encodeInteger(value); |
|
else |
|
return this.encodeFloat(value); |
|
}, |
|
encodeFloat: function CFFCompiler_encodeFloat(num) { |
|
var value = num.toString(); |
|
var nibbles = ''; |
|
for (var i = 0, ii = value.length; i < ii; ++i) { |
|
var a = value[i]; |
|
if (a === 'e') { |
|
nibbles += value[++i] === '-' ? 'c' : 'b'; |
|
} else if (a === '.') { |
|
nibbles += 'a'; |
|
} else if (a === '-') { |
|
nibbles += 'e'; |
|
} else { |
|
nibbles += a; |
|
} |
|
} |
|
nibbles += (nibbles.length & 1) ? 'f' : 'ff'; |
|
var out = [30]; |
|
for (var i = 0, ii = nibbles.length; i < ii; i += 2) { |
|
out.push(parseInt(nibbles.substr(i, 2), 16)); |
|
} |
|
return out; |
|
}, |
|
encodeInteger: function CFFCompiler_encodeInteger(value) { |
|
var code; |
|
if (value >= -107 && value <= 107) { |
|
code = [value + 139]; |
|
} else if (value >= 108 && value <= 1131) { |
|
value = [value - 108]; |
|
code = [(value >> 8) + 247, value & 0xFF]; |
|
} else if (value >= -1131 && value <= -108) { |
|
value = -value - 108; |
|
code = [(value >> 8) + 251, value & 0xFF]; |
|
} else if (value >= -32768 && value <= 32767) { |
|
code = [0x1c, (value >> 8) & 0xFF, value & 0xFF]; |
|
} else { |
|
code = [0x1d, |
|
(value >> 24) & 0xFF, |
|
(value >> 16) & 0xFF, |
|
(value >> 8) & 0xFF, |
|
value & 0xFF]; |
|
} |
|
return code; |
|
}, |
|
compileHeader: function CFFCompiler_compileHeader(header) { |
|
return [ |
|
header.major, |
|
header.minor, |
|
header.hdrSize, |
|
header.offSize |
|
]; |
|
}, |
|
compileNameIndex: function CFFCompiler_compileNameIndex(names) { |
|
var nameIndex = new CFFIndex(); |
|
for (var i = 0, ii = names.length; i < ii; ++i) |
|
nameIndex.add(stringToArray(names[i])); |
|
return this.compileIndex(nameIndex); |
|
}, |
|
compileTopDicts: function CFFCompiler_compileTopDicts(dicts, |
|
length, |
|
removeCidKeys) { |
|
var fontDictTrackers = []; |
|
var fdArrayIndex = new CFFIndex(); |
|
for (var i = 0, ii = dicts.length; i < ii; ++i) { |
|
var fontDict = dicts[i]; |
|
if (removeCidKeys) { |
|
fontDict.removeByName('CIDFontVersion'); |
|
fontDict.removeByName('CIDFontRevision'); |
|
fontDict.removeByName('CIDFontType'); |
|
fontDict.removeByName('CIDCount'); |
|
fontDict.removeByName('UIDBase'); |
|
} |
|
var fontDictTracker = new CFFOffsetTracker(); |
|
var fontDictData = this.compileDict(fontDict, fontDictTracker); |
|
fontDictTrackers.push(fontDictTracker); |
|
fdArrayIndex.add(fontDictData); |
|
fontDictTracker.offset(length); |
|
} |
|
fdArrayIndex = this.compileIndex(fdArrayIndex, fontDictTrackers); |
|
return { |
|
trackers: fontDictTrackers, |
|
output: fdArrayIndex |
|
}; |
|
}, |
|
compilePrivateDicts: function CFFCompiler_compilePrivateDicts(dicts, |
|
trackers, |
|
output) { |
|
for (var i = 0, ii = dicts.length; i < ii; ++i) { |
|
var fontDict = dicts[i]; |
|
assert(fontDict.privateDict && fontDict.hasName('Private'), |
|
'There must be an private dictionary.'); |
|
var privateDict = fontDict.privateDict; |
|
var privateDictTracker = new CFFOffsetTracker(); |
|
var privateDictData = this.compileDict(privateDict, privateDictTracker); |
|
|
|
var outputLength = output.length; |
|
privateDictTracker.offset(outputLength); |
|
if (!privateDictData.length) { |
|
// The private dictionary was empty, set the output length to zero to |
|
// ensure the offset length isn't out of bounds in the eyes of the |
|
// sanitizer. |
|
outputLength = 0; |
|
} |
|
|
|
trackers[i].setEntryLocation('Private', |
|
[privateDictData.length, outputLength], |
|
output); |
|
output.add(privateDictData); |
|
|
|
if (privateDict.subrsIndex && privateDict.hasName('Subrs')) { |
|
var subrs = this.compileIndex(privateDict.subrsIndex); |
|
privateDictTracker.setEntryLocation('Subrs', [privateDictData.length], |
|
output); |
|
output.add(subrs); |
|
} |
|
} |
|
}, |
|
compileDict: function CFFCompiler_compileDict(dict, offsetTracker) { |
|
var out = []; |
|
// The dictionary keys must be in a certain order. |
|
var order = dict.order; |
|
for (var i = 0; i < order.length; ++i) { |
|
var key = order[i]; |
|
if (!(key in dict.values)) |
|
continue; |
|
var values = dict.values[key]; |
|
var types = dict.types[key]; |
|
if (!isArray(types)) types = [types]; |
|
if (!isArray(values)) values = [values]; |
|
|
|
// Remove any empty dict values. |
|
if (values.length === 0) |
|
continue; |
|
|
|
for (var j = 0, jj = types.length; j < jj; ++j) { |
|
var type = types[j]; |
|
var value = values[j]; |
|
switch (type) { |
|
case 'num': |
|
case 'sid': |
|
out = out.concat(this.encodeNumber(value)); |
|
break; |
|
case 'offset': |
|
// For offsets we just insert a 32bit integer so we don't have to |
|
// deal with figuring out the length of the offset when it gets |
|
// replaced later on by the compiler. |
|
var name = dict.keyToNameMap[key]; |
|
// Some offsets have the offset and the length, so just record the |
|
// position of the first one. |
|
if (!offsetTracker.isTracking(name)) |
|
offsetTracker.track(name, out.length); |
|
out = out.concat([0x1d, 0, 0, 0, 0]); |
|
break; |
|
case 'array': |
|
case 'delta': |
|
out = out.concat(this.encodeNumber(value)); |
|
for (var k = 1, kk = values.length; k < kk; ++k) |
|
out = out.concat(this.encodeNumber(values[k])); |
|
break; |
|
default: |
|
error('Unknown data type of ' + type); |
|
break; |
|
} |
|
} |
|
out = out.concat(dict.opcodes[key]); |
|
} |
|
return out; |
|
}, |
|
compileStringIndex: function CFFCompiler_compileStringIndex(strings) { |
|
var stringIndex = new CFFIndex(); |
|
for (var i = 0, ii = strings.length; i < ii; ++i) |
|
stringIndex.add(stringToArray(strings[i])); |
|
return this.compileIndex(stringIndex); |
|
}, |
|
compileGlobalSubrIndex: function CFFCompiler_compileGlobalSubrIndex() { |
|
var globalSubrIndex = this.cff.globalSubrIndex; |
|
this.out.writeByteArray(this.compileIndex(globalSubrIndex)); |
|
}, |
|
compileCharStrings: function CFFCompiler_compileCharStrings(charStrings) { |
|
return this.compileIndex(charStrings); |
|
}, |
|
compileCharset: function CFFCompiler_compileCharset(charset) { |
|
return this.compileTypedArray(charset.raw); |
|
}, |
|
compileEncoding: function CFFCompiler_compileEncoding(encoding) { |
|
return this.compileTypedArray(encoding.raw); |
|
}, |
|
compileFDSelect: function CFFCompiler_compileFDSelect(fdSelect) { |
|
return this.compileTypedArray(fdSelect); |
|
}, |
|
compileTypedArray: function CFFCompiler_compileTypedArray(data) { |
|
var out = []; |
|
for (var i = 0, ii = data.length; i < ii; ++i) |
|
out[i] = data[i]; |
|
return out; |
|
}, |
|
compileIndex: function CFFCompiler_compileIndex(index, trackers) { |
|
trackers = trackers || []; |
|
var objects = index.objects; |
|
// First 2 bytes contains the number of objects contained into this index |
|
var count = objects.length; |
|
|
|
// If there is no object, just create an index. This technically |
|
// should just be [0, 0] but OTS has an issue with that. |
|
if (count === 0) |
|
return [0, 0, 0]; |
|
|
|
var data = [(count >> 8) & 0xFF, count & 0xff]; |
|
|
|
var lastOffset = 1; |
|
for (var i = 0; i < count; ++i) |
|
lastOffset += objects[i].length; |
|
|
|
var offsetSize; |
|
if (lastOffset < 0x100) |
|
offsetSize = 1; |
|
else if (lastOffset < 0x10000) |
|
offsetSize = 2; |
|
else if (lastOffset < 0x1000000) |
|
offsetSize = 3; |
|
else |
|
offsetSize = 4; |
|
|
|
// Next byte contains the offset size use to reference object in the file |
|
data.push(offsetSize); |
|
|
|
// Add another offset after this one because we need a new offset |
|
var relativeOffset = 1; |
|
for (var i = 0; i < count + 1; i++) { |
|
if (offsetSize === 1) { |
|
data.push(relativeOffset & 0xFF); |
|
} else if (offsetSize === 2) { |
|
data.push((relativeOffset >> 8) & 0xFF, |
|
relativeOffset & 0xFF); |
|
} else if (offsetSize === 3) { |
|
data.push((relativeOffset >> 16) & 0xFF, |
|
(relativeOffset >> 8) & 0xFF, |
|
relativeOffset & 0xFF); |
|
} else { |
|
data.push((relativeOffset >>> 24) & 0xFF, |
|
(relativeOffset >> 16) & 0xFF, |
|
(relativeOffset >> 8) & 0xFF, |
|
relativeOffset & 0xFF); |
|
} |
|
|
|
if (objects[i]) |
|
relativeOffset += objects[i].length; |
|
} |
|
var offset = data.length; |
|
|
|
for (var i = 0; i < count; i++) { |
|
// Notify the tracker where the object will be offset in the data. |
|
if (trackers[i]) |
|
trackers[i].offset(data.length); |
|
for (var j = 0, jj = objects[i].length; j < jj; j++) |
|
data.push(objects[i][j]); |
|
} |
|
return data; |
|
} |
|
}; |
|
return CFFCompiler; |
|
})(); |
|
|
|
// Workaround for seac on Windows. |
|
(function checkSeacSupport() { |
|
if (/Windows/.test(navigator.userAgent)) { |
|
SEAC_ANALYSIS_ENABLED = true; |
|
} |
|
})(); |
|
|
|
// Workaround for Private Use Area characters in Chrome on Windows |
|
// http://code.google.com/p/chromium/issues/detail?id=122465 |
|
// https://github.com/mozilla/pdf.js/issues/1689 |
|
(function checkChromeWindows() { |
|
if (/Windows.*Chrome/.test(navigator.userAgent)) { |
|
SYMBOLIC_FONT_GLYPH_OFFSET = 0xF100; |
|
} |
|
})();
|
|
|