400 lines
		
	
	
		
			16 KiB
		
	
	
	
		
			JavaScript
		
	
	
	
	
	
			
		
		
	
	
			400 lines
		
	
	
		
			16 KiB
		
	
	
	
		
			JavaScript
		
	
	
	
	
	
/* -*- Mode: Java; tab-width: 2; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
 | 
						|
/* vim: set shiftwidth=2 tabstop=2 autoindent cindent expandtab: */
 | 
						|
/* globals expect, it, describe, CFFCompiler, CFFParser, CFFIndex, CFFStrings,
 | 
						|
           SEAC_ANALYSIS_ENABLED:true, Type1Parser, StringStream */
 | 
						|
 | 
						|
'use strict';
 | 
						|
 | 
						|
describe('font', function() {
 | 
						|
  function hexDump(bytes) {
 | 
						|
    var line = '';
 | 
						|
    for (var i = 0, ii = bytes.length; i < ii; ++i) {
 | 
						|
      var b = bytes[i].toString(16);
 | 
						|
      if (b.length < 2) {
 | 
						|
        b = '0' + b;
 | 
						|
      }
 | 
						|
      line += b.toString(16);
 | 
						|
    }
 | 
						|
    return line;
 | 
						|
  }
 | 
						|
  // This example font comes from the CFF spec:
 | 
						|
  // http://www.adobe.com/content/dam/Adobe/en/devnet/font/pdfs/5176.CFF.pdf
 | 
						|
  var exampleFont = '0100040100010101134142434445462b' +
 | 
						|
                    '54696d65732d526f6d616e000101011f' +
 | 
						|
                    'f81b00f81c02f81d03f819041c6f000d' +
 | 
						|
                    'fb3cfb6efa7cfa1605e911b8f1120003' +
 | 
						|
                    '01010813183030312e30303754696d65' +
 | 
						|
                    '7320526f6d616e54696d657300000002' +
 | 
						|
                    '010102030e0e7d99f92a99fb7695f773' +
 | 
						|
                    '8b06f79a93fc7c8c077d99f85695f75e' +
 | 
						|
                    '9908fb6e8cf87393f7108b09a70adf0b' +
 | 
						|
                    'f78e14';
 | 
						|
  var fontData = [];
 | 
						|
  for (var i = 0; i < exampleFont.length; i += 2) {
 | 
						|
    var hex = exampleFont.substr(i, 2);
 | 
						|
    fontData.push(parseInt(hex, 16));
 | 
						|
  }
 | 
						|
  var bytes = new Uint8Array(fontData);
 | 
						|
  fontData = {
 | 
						|
    getBytes: function() {
 | 
						|
      return bytes;
 | 
						|
    }
 | 
						|
  };
 | 
						|
 | 
						|
  function bytesToString(bytesArray) {
 | 
						|
    var str = '';
 | 
						|
    for (var i = 0, ii = bytesArray.length; i < ii; i++) {
 | 
						|
      str += String.fromCharCode(bytesArray[i]);
 | 
						|
    }
 | 
						|
    return str;
 | 
						|
  }
 | 
						|
 | 
						|
  describe('CFFParser', function() {
 | 
						|
    var parser = new CFFParser(fontData, {});
 | 
						|
    var cff = parser.parse();
 | 
						|
 | 
						|
    it('parses header', function() {
 | 
						|
      var header = cff.header;
 | 
						|
      expect(header.major).toEqual(1);
 | 
						|
      expect(header.minor).toEqual(0);
 | 
						|
      expect(header.hdrSize).toEqual(4);
 | 
						|
      expect(header.offSize).toEqual(1);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses name index', function() {
 | 
						|
      var names = cff.names;
 | 
						|
      expect(names.length).toEqual(1);
 | 
						|
      expect(names[0]).toEqual('ABCDEF+Times-Roman');
 | 
						|
    });
 | 
						|
 | 
						|
    it('sanitizes name index', function() {
 | 
						|
      var index = new CFFIndex();
 | 
						|
      index.add(['['.charCodeAt(0), 'a'.charCodeAt(0)]);
 | 
						|
 | 
						|
      var names = parser.parseNameIndex(index);
 | 
						|
      expect(names).toEqual(['_a']);
 | 
						|
 | 
						|
      index = new CFFIndex();
 | 
						|
      var longName = [];
 | 
						|
      for (var i = 0; i < 129; i++) {
 | 
						|
        longName.push(0);
 | 
						|
      }
 | 
						|
      index.add(longName);
 | 
						|
      names = parser.parseNameIndex(index);
 | 
						|
      expect(names[0].length).toEqual(127);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses string index', function() {
 | 
						|
      var strings = cff.strings;
 | 
						|
      expect(strings.count).toEqual(3);
 | 
						|
      expect(strings.get(0)).toEqual('.notdef');
 | 
						|
      expect(strings.get(391)).toEqual('001.007');
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses top dict', function() {
 | 
						|
      var topDict = cff.topDict;
 | 
						|
      // 391 version 392 FullName 393 FamilyName 389 Weight 28416 UniqueID
 | 
						|
      // -168 -218 1000 898 FontBBox 94 CharStrings 45 102 Private
 | 
						|
      expect(topDict.getByName('version')).toEqual(391);
 | 
						|
      expect(topDict.getByName('FullName')).toEqual(392);
 | 
						|
      expect(topDict.getByName('FamilyName')).toEqual(393);
 | 
						|
      expect(topDict.getByName('Weight')).toEqual(389);
 | 
						|
      expect(topDict.getByName('UniqueID')).toEqual(28416);
 | 
						|
      expect(topDict.getByName('FontBBox')).toEqual([-168, -218, 1000, 898]);
 | 
						|
      expect(topDict.getByName('CharStrings')).toEqual(94);
 | 
						|
      expect(topDict.getByName('Private')).toEqual([45, 102]);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses a CharString having cntrmask', function() {
 | 
						|
      var bytes = new Uint8Array([0, 1, // count
 | 
						|
                                  1,  // offsetSize
 | 
						|
                                  0,  // offset[0]
 | 
						|
                                  38, // end
 | 
						|
                                  149, 149, 149, 149, 149, 149, 149, 149,
 | 
						|
                                  149, 149, 149, 149, 149, 149, 149, 149,
 | 
						|
                                  1,  // hstem
 | 
						|
                                  149, 149, 149, 149, 149, 149, 149, 149,
 | 
						|
                                  149, 149, 149, 149, 149, 149, 149, 149,
 | 
						|
                                  3,  // vstem
 | 
						|
                                  20, // cntrmask
 | 
						|
                                  22, 22, // fail if misparsed as hmoveto
 | 
						|
                                  14  // endchar
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var charStrings = parser.parseCharStrings(0).charStrings;
 | 
						|
      expect(charStrings.count).toEqual(1);
 | 
						|
      // shoudn't be sanitized
 | 
						|
      expect(charStrings.get(0).length).toEqual(38);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses a CharString endchar with 4 args w/seac enabled', function() {
 | 
						|
      var seacAnalysisState = SEAC_ANALYSIS_ENABLED;
 | 
						|
      try {
 | 
						|
        SEAC_ANALYSIS_ENABLED = true;
 | 
						|
        var bytes = new Uint8Array([0, 1, // count
 | 
						|
                                    1,  // offsetSize
 | 
						|
                                    0,  // offset[0]
 | 
						|
                                    237, 247, 22, 247, 72, 204, 247, 86, 14]);
 | 
						|
        parser.bytes = bytes;
 | 
						|
        var result = parser.parseCharStrings(0);
 | 
						|
        expect(result.charStrings.count).toEqual(1);
 | 
						|
        expect(result.charStrings.get(0).length).toEqual(1);
 | 
						|
        expect(result.seacs.length).toEqual(1);
 | 
						|
        expect(result.seacs[0].length).toEqual(4);
 | 
						|
        expect(result.seacs[0][0]).toEqual(130);
 | 
						|
        expect(result.seacs[0][1]).toEqual(180);
 | 
						|
        expect(result.seacs[0][2]).toEqual(65);
 | 
						|
        expect(result.seacs[0][3]).toEqual(194);
 | 
						|
      } finally {
 | 
						|
        SEAC_ANALYSIS_ENABLED = seacAnalysisState;
 | 
						|
      }
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses a CharString endchar with 4 args w/seac disabled', function() {
 | 
						|
      var seacAnalysisState = SEAC_ANALYSIS_ENABLED;
 | 
						|
      try {
 | 
						|
        SEAC_ANALYSIS_ENABLED = false;
 | 
						|
        var bytes = new Uint8Array([0, 1, // count
 | 
						|
                                    1,  // offsetSize
 | 
						|
                                    0,  // offset[0]
 | 
						|
                                    237, 247, 22, 247, 72, 204, 247, 86, 14]);
 | 
						|
        parser.bytes = bytes;
 | 
						|
        var result = parser.parseCharStrings(0);
 | 
						|
        expect(result.charStrings.count).toEqual(1);
 | 
						|
        expect(result.charStrings.get(0).length).toEqual(9);
 | 
						|
        expect(result.seacs.length).toEqual(0);
 | 
						|
      } finally {
 | 
						|
        SEAC_ANALYSIS_ENABLED = seacAnalysisState;
 | 
						|
      }
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses a CharString endchar no args', function() {
 | 
						|
      var bytes = new Uint8Array([0, 1, // count
 | 
						|
                                  1,  // offsetSize
 | 
						|
                                  0,  // offset[0]
 | 
						|
                                  14]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var result = parser.parseCharStrings(0);
 | 
						|
      expect(result.charStrings.count).toEqual(1);
 | 
						|
      expect(result.charStrings.get(0)[0]).toEqual(14);
 | 
						|
      expect(result.seacs.length).toEqual(0);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses predefined charsets', function() {
 | 
						|
      var charset = parser.parseCharsets(0, 0, null, true);
 | 
						|
      expect(charset.predefined).toEqual(true);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses charset format 0', function() {
 | 
						|
      // The first three bytes make the offset large enough to skip predefined.
 | 
						|
      var bytes = new Uint8Array([0x00, 0x00, 0x00,
 | 
						|
                                  0x00, // format
 | 
						|
                                  0x00, 0x02 // sid/cid
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var charset = parser.parseCharsets(3, 2, new CFFStrings(), false);
 | 
						|
      expect(charset.charset[1]).toEqual('exclam');
 | 
						|
 | 
						|
      // CID font
 | 
						|
      var charset = parser.parseCharsets(3, 2, new CFFStrings(), true);
 | 
						|
      expect(charset.charset[1]).toEqual(2);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses charset format 1', function() {
 | 
						|
      // The first three bytes make the offset large enough to skip predefined.
 | 
						|
      var bytes = new Uint8Array([0x00, 0x00, 0x00,
 | 
						|
                                  0x01, // format
 | 
						|
                                  0x00, 0x08, // sid/cid start
 | 
						|
                                  0x01 // sid/cid left
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var charset = parser.parseCharsets(3, 2, new CFFStrings(), false);
 | 
						|
      expect(charset.charset).toEqual(['.notdef', 'quoteright', 'parenleft']);
 | 
						|
 | 
						|
      // CID font
 | 
						|
      var charset = parser.parseCharsets(3, 2, new CFFStrings(), true);
 | 
						|
      expect(charset.charset).toEqual(['.notdef', 8, 9]);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses charset format 2', function() {
 | 
						|
      // format 2 is the same as format 1 but the left is card16
 | 
						|
      // The first three bytes make the offset large enough to skip predefined.
 | 
						|
      var bytes = new Uint8Array([0x00, 0x00, 0x00,
 | 
						|
                                  0x02, // format
 | 
						|
                                  0x00, 0x08, // sid/cid start
 | 
						|
                                  0x00, 0x01 // sid/cid left
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var charset = parser.parseCharsets(3, 2, new CFFStrings(), false);
 | 
						|
      expect(charset.charset).toEqual(['.notdef', 'quoteright', 'parenleft']);
 | 
						|
 | 
						|
      // CID font
 | 
						|
      var charset = parser.parseCharsets(3, 2, new CFFStrings(), true);
 | 
						|
      expect(charset.charset).toEqual(['.notdef', 8, 9]);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses encoding format 0', function() {
 | 
						|
      // The first two bytes make the offset large enough to skip predefined.
 | 
						|
      var bytes = new Uint8Array([0x00, 0x00,
 | 
						|
                                  0x00, // format
 | 
						|
                                  0x01, // count
 | 
						|
                                  0x08  // start
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var encoding = parser.parseEncoding(2, {}, new CFFStrings(), null);
 | 
						|
      expect(encoding.encoding).toEqual({0x8: 1});
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses encoding format 1', function() {
 | 
						|
      // The first two bytes make the offset large enough to skip predefined.
 | 
						|
      var bytes = new Uint8Array([0x00, 0x00,
 | 
						|
                                  0x01, // format
 | 
						|
                                  0x01, // num ranges
 | 
						|
                                  0x07, // range1 start
 | 
						|
                                  0x01 // range2 left
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var encoding = parser.parseEncoding(2, {}, new CFFStrings(), null);
 | 
						|
      expect(encoding.encoding).toEqual({0x7: 0x01, 0x08: 0x02});
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses fdselect format 0', function() {
 | 
						|
      var bytes = new Uint8Array([0x00, // format
 | 
						|
                                  0x00, // gid: 0 fd: 0
 | 
						|
                                  0x01 // gid: 1 fd: 1
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var fdSelect = parser.parseFDSelect(0, 2);
 | 
						|
      expect(fdSelect.fdSelect).toEqual([0, 1]);
 | 
						|
    });
 | 
						|
 | 
						|
    it('parses fdselect format 3', function() {
 | 
						|
      var bytes = new Uint8Array([0x03, // format
 | 
						|
                                  0x00, 0x02, // range count
 | 
						|
                                  0x00, 0x00, // first gid
 | 
						|
                                  0x09, // font dict 1 id
 | 
						|
                                  0x00, 0x02, // nex gid
 | 
						|
                                  0x0a, // font dict 2 gid
 | 
						|
                                  0x00, 0x04 // sentinel (last gid)
 | 
						|
                                ]);
 | 
						|
      parser.bytes = bytes;
 | 
						|
      var fdSelect = parser.parseFDSelect(0, 2);
 | 
						|
      expect(fdSelect.fdSelect).toEqual([9, 9, 0xa, 0xa]);
 | 
						|
    });
 | 
						|
    // TODO fdArray
 | 
						|
  });
 | 
						|
  describe('CFFCompiler', function() {
 | 
						|
    it('encodes integers', function() {
 | 
						|
      var c = new CFFCompiler();
 | 
						|
      // all the examples from the spec
 | 
						|
      expect(c.encodeInteger(0)).toEqual([0x8b]);
 | 
						|
      expect(c.encodeInteger(100)).toEqual([0xef]);
 | 
						|
      expect(c.encodeInteger(-100)).toEqual([0x27]);
 | 
						|
      expect(c.encodeInteger(1000)).toEqual([0xfa, 0x7c]);
 | 
						|
      expect(c.encodeInteger(-1000)).toEqual([0xfe, 0x7c]);
 | 
						|
      expect(c.encodeInteger(10000)).toEqual([0x1c, 0x27, 0x10]);
 | 
						|
      expect(c.encodeInteger(-10000)).toEqual([0x1c, 0xd8, 0xf0]);
 | 
						|
      expect(c.encodeInteger(100000)).toEqual([0x1d, 0x00, 0x01, 0x86, 0xa0]);
 | 
						|
      expect(c.encodeInteger(-100000)).toEqual([0x1d, 0xff, 0xfe, 0x79, 0x60]);
 | 
						|
    });
 | 
						|
    it('encodes floats', function() {
 | 
						|
      var c = new CFFCompiler();
 | 
						|
      expect(c.encodeFloat(-2.25)).toEqual([0x1e, 0xe2, 0xa2, 0x5f]);
 | 
						|
      expect(c.encodeFloat(5e-11)).toEqual([0x1e, 0x5c, 0x11, 0xff]);
 | 
						|
    });
 | 
						|
    // TODO a lot more compiler tests
 | 
						|
  });
 | 
						|
 | 
						|
  describe('Type1Parser', function() {
 | 
						|
 | 
						|
    it('splits tokens', function() {
 | 
						|
      var stream = new StringStream('/BlueValues[-17 0]noaccess def');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.getToken()).toEqual('/');
 | 
						|
      expect(parser.getToken()).toEqual('BlueValues');
 | 
						|
      expect(parser.getToken()).toEqual('[');
 | 
						|
      expect(parser.getToken()).toEqual('-17');
 | 
						|
      expect(parser.getToken()).toEqual('0');
 | 
						|
      expect(parser.getToken()).toEqual(']');
 | 
						|
      expect(parser.getToken()).toEqual('noaccess');
 | 
						|
      expect(parser.getToken()).toEqual('def');
 | 
						|
      expect(parser.getToken()).toEqual(null);
 | 
						|
    });
 | 
						|
    it('handles glued tokens', function() {
 | 
						|
      var stream = new StringStream('dup/CharStrings');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.getToken()).toEqual('dup');
 | 
						|
      expect(parser.getToken()).toEqual('/');
 | 
						|
      expect(parser.getToken()).toEqual('CharStrings');
 | 
						|
    });
 | 
						|
    it('ignores whitespace', function() {
 | 
						|
      var stream = new StringStream('\nab   c\t');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.getToken()).toEqual('ab');
 | 
						|
      expect(parser.getToken()).toEqual('c');
 | 
						|
    });
 | 
						|
    it('parses numbers', function() {
 | 
						|
      var stream = new StringStream('123');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.readNumber()).toEqual(123);
 | 
						|
    });
 | 
						|
    it('parses booleans', function() {
 | 
						|
      var stream = new StringStream('true false');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.readBoolean()).toEqual(1);
 | 
						|
      expect(parser.readBoolean()).toEqual(0);
 | 
						|
    });
 | 
						|
    it('parses number arrays', function() {
 | 
						|
      var stream = new StringStream('[1 2]');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.readNumberArray()).toEqual([1, 2]);
 | 
						|
      // Variation on spacing.
 | 
						|
      var stream = new StringStream('[ 1 2 ]');
 | 
						|
      parser = new Type1Parser(stream);
 | 
						|
      expect(parser.readNumberArray()).toEqual([1, 2]);
 | 
						|
    });
 | 
						|
    it('skips comments', function() {
 | 
						|
      var stream = new StringStream(
 | 
						|
        '%!PS-AdobeFont-1.0: CMSY10 003.002\n' +
 | 
						|
        '%%Title: CMSY10\n' +
 | 
						|
        '%Version: 003.002\n' +
 | 
						|
        'FontDirectory');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      expect(parser.getToken()).toEqual('FontDirectory');
 | 
						|
    });
 | 
						|
    it('parses font program', function() {
 | 
						|
      var stream = new StringStream(
 | 
						|
        '/ExpansionFactor  99\n' +
 | 
						|
        '/Subrs 1 array\n' +
 | 
						|
        'dup 0 1 RD x noaccess put\n'+
 | 
						|
        'end\n' +
 | 
						|
        '/CharStrings 46 dict dup begin\n' +
 | 
						|
        '/.notdef 1 RD x ND' + '\n' +
 | 
						|
        'end');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      var program = parser.extractFontProgram();
 | 
						|
      expect(program.charstrings.length).toEqual(1);
 | 
						|
      expect(program.properties.privateData.ExpansionFactor).toEqual(99);
 | 
						|
    });
 | 
						|
    it('parses font header font matrix', function() {
 | 
						|
      var stream = new StringStream(
 | 
						|
        '/FontMatrix [0.001 0 0 0.001 0 0 ]readonly def\n');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      var props = {};
 | 
						|
      var program = parser.extractFontHeader(props);
 | 
						|
      expect(props.fontMatrix).toEqual([0.001, 0, 0, 0.001, 0, 0]);
 | 
						|
    });
 | 
						|
    it('parses font header encoding', function() {
 | 
						|
      var stream = new StringStream(
 | 
						|
        '/Encoding 256 array\n' +
 | 
						|
        '0 1 255 {1 index exch /.notdef put} for\n' +
 | 
						|
        'dup 33 /arrowright put\n' +
 | 
						|
        'readonly def\n');
 | 
						|
      var parser = new Type1Parser(stream);
 | 
						|
      var props = { overridableEncoding: true };
 | 
						|
      var program = parser.extractFontHeader(props);
 | 
						|
      expect(props.builtInEncoding[33]).toEqual('arrowright');
 | 
						|
    });
 | 
						|
  });
 | 
						|
});
 |