376 lines
11 KiB
JavaScript
376 lines
11 KiB
JavaScript
/**
|
|
* The Type2 reader code below is only used for debugging purpose since Type2
|
|
* is only a CharString format and is never used directly as a Font file.
|
|
*
|
|
* So the code here is useful for dumping the data content of a .cff file in
|
|
* order to investigate the similarity between a Type1 CharString and a Type2
|
|
* CharString.
|
|
*/
|
|
|
|
|
|
/**
|
|
* Build a charset by assigning the glyph name and the human readable form
|
|
* of the glyph data.
|
|
*/
|
|
function readCharset(aStream, aCharstrings) {
|
|
var charset = {};
|
|
|
|
var format = aStream.getByte();
|
|
if (format == 0) {
|
|
charset[".notdef"] = readCharstringEncoding(aCharstrings[0]);
|
|
|
|
var count = aCharstrings.length - 1;
|
|
for (var i = 1; i < count + 1; i++) {
|
|
var sid = aStream.getByte() << 8 | aStream.getByte();
|
|
charset[CFFStrings[sid]] = readCharstringEncoding(aCharstrings[i]);
|
|
//log(CFFStrings[sid] + "::" + charset[CFFStrings[sid]]);
|
|
}
|
|
} else if (format == 1) {
|
|
error("Charset Range are not supported");
|
|
} else {
|
|
error("Invalid charset format");
|
|
}
|
|
|
|
return charset;
|
|
};
|
|
|
|
/**
|
|
* Take a Type2 binary charstring as input and transform it to a human
|
|
* readable representation as specified by the 'The Type 2 Charstring Format',
|
|
* chapter 3.1.
|
|
*/
|
|
function readCharstringEncoding(aString) {
|
|
var charstringTokens = [];
|
|
|
|
var count = aString.length;
|
|
for (var i = 0; i < count; ) {
|
|
var value = aString[i++];
|
|
var token = null;
|
|
|
|
if (value < 0) {
|
|
continue;
|
|
} else if (value <= 11) {
|
|
token = CFFEncodingMap[value];
|
|
} else if (value == 12) {
|
|
token = CFFEncodingMap[value][aString[i++]];
|
|
} else if (value <= 18) {
|
|
token = CFFEncodingMap[value];
|
|
} else if (value <= 20) {
|
|
var mask = aString[i++];
|
|
token = CFFEncodingMap[value];
|
|
} else if (value <= 27) {
|
|
token = CFFEncodingMap[value];
|
|
} else if (value == 28) {
|
|
token = aString[i++] << 8 | aString[i++];
|
|
} else if (value <= 31) {
|
|
token = CFFEncodingMap[value];
|
|
} else if (value < 247) {
|
|
token = parseInt(value) - 139;
|
|
} else if (value < 251) {
|
|
token = ((value - 247) * 256) + aString[i++] + 108;
|
|
} else if (value < 255) {
|
|
token = -((value - 251) * 256) - aString[i++] - 108;
|
|
} else {// value == 255
|
|
token = aString[i++] << 24 | aString[i++] << 16 |
|
|
aString[i++] << 8 | aString[i];
|
|
}
|
|
|
|
charstringTokens.push(token);
|
|
}
|
|
|
|
return charstringTokens;
|
|
};
|
|
|
|
|
|
/**
|
|
* Take a binary DICT Data as input and transform it into a human readable
|
|
* form as specified by 'The Compact Font Format Specification', chapter 5.
|
|
*/
|
|
function readFontDictData(aString, aMap) {
|
|
var fontDictDataTokens = [];
|
|
|
|
var count = aString.length;
|
|
for (var i = 0; i < count; i) {
|
|
var value = aString[i++];
|
|
var token = null;
|
|
|
|
if (value == 12) {
|
|
token = aMap[value][aString[i++]];
|
|
} else if (value == 28) {
|
|
token = aString[i++] << 8 | aString[i++];
|
|
} else if (value == 29) {
|
|
token = aString[i++] << 24 |
|
|
aString[i++] << 16 |
|
|
aString[i++] << 8 |
|
|
aString[i++];
|
|
} else if (value == 30) {
|
|
token = "";
|
|
var parsed = false;
|
|
while (!parsed) {
|
|
var byte = aString[i++];
|
|
|
|
var nibbles = [parseInt(byte / 16), parseInt(byte % 16)];
|
|
for (var j = 0; j < nibbles.length; j++) {
|
|
var nibble = nibbles[j];
|
|
switch (nibble) {
|
|
case 0xA:
|
|
token += ".";
|
|
break;
|
|
case 0xB:
|
|
token += "E";
|
|
break;
|
|
case 0xC:
|
|
token += "E-";
|
|
break;
|
|
case 0xD:
|
|
break;
|
|
case 0xE:
|
|
token += "-";
|
|
break;
|
|
case 0xF:
|
|
parsed = true;
|
|
break;
|
|
default:
|
|
token += nibble;
|
|
break;
|
|
}
|
|
}
|
|
};
|
|
token = parseFloat(token);
|
|
} else if (value <= 31) {
|
|
token = aMap[value];
|
|
} else if (value <= 246) {
|
|
token = parseInt(value) - 139;
|
|
} else if (value <= 250) {
|
|
token = ((value - 247) * 256) + aString[i++] + 108;
|
|
} else if (value <= 254) {
|
|
token = -((value - 251) * 256) - aString[i++] - 108;
|
|
} else if (value == 255) {
|
|
error("255 is not a valid DICT command");
|
|
}
|
|
|
|
fontDictDataTokens.push(token);
|
|
}
|
|
|
|
return fontDictDataTokens;
|
|
};
|
|
|
|
|
|
/**
|
|
* Take a stream as input and return an array of objects.
|
|
* In CFF an INDEX is a structure with the following format:
|
|
* {
|
|
* count: 2 bytes (Number of objects stored in INDEX),
|
|
* offsize: 1 byte (Offset array element size),
|
|
* offset: [count + 1] bytes (Offsets array),
|
|
* data: - (Objects data)
|
|
* }
|
|
*
|
|
* More explanation are given in the 'CFF Font Format Specification',
|
|
* chapter 5.
|
|
*/
|
|
function readFontIndexData(aStream, aIsByte) {
|
|
var count = aStream.getByte() << 8 | aStream.getByte();
|
|
var offsize = aStream.getByte();
|
|
|
|
function getNextOffset() {
|
|
switch (offsize) {
|
|
case 0:
|
|
return 0;
|
|
case 1:
|
|
return aStream.getByte();
|
|
case 2:
|
|
return aStream.getByte() << 8 | aStream.getByte();
|
|
case 3:
|
|
return aStream.getByte() << 16 | aStream.getByte() << 8 |
|
|
aStream.getByte();
|
|
case 4:
|
|
return aStream.getByte() << 24 | aStream.getByte() << 16 |
|
|
aStream.getByte() << 8 | aStream.getByte();
|
|
}
|
|
};
|
|
|
|
var offsets = [];
|
|
for (var i = 0; i < count + 1; i++)
|
|
offsets.push(getNextOffset());
|
|
|
|
log("Found " + count + " objects at offsets :" + offsets + " (offsize: " + offsize + ")");
|
|
|
|
// Now extract the objects
|
|
var relativeOffset = aStream.pos;
|
|
var objects = [];
|
|
for (var i = 0; i < count; i++) {
|
|
var offset = offsets[i];
|
|
aStream.pos = relativeOffset + offset - 1;
|
|
|
|
var data = [];
|
|
var length = offsets[i + 1] - 1;
|
|
for (var j = offset - 1; j < length; j++)
|
|
data.push(aIsByte ? aStream.getByte() : aStream.getChar());
|
|
objects.push(data);
|
|
}
|
|
|
|
return objects;
|
|
};
|
|
|
|
var Type2Parser = function(aFilePath) {
|
|
var font = new Dict();
|
|
|
|
// Turn on this flag for additional debugging logs
|
|
var debug = false;
|
|
|
|
function dump(aStr) {
|
|
if (debug)
|
|
log(aStr);
|
|
};
|
|
|
|
function parseAsToken(aString, aMap) {
|
|
var decoded = readFontDictData(aString, aMap);
|
|
|
|
var stack = [];
|
|
var count = decoded.length;
|
|
for (var i = 0; i < count; i++) {
|
|
var token = decoded[i];
|
|
if (IsNum(token)) {
|
|
stack.push(token);
|
|
} else {
|
|
switch (token.operand) {
|
|
case "SID":
|
|
font.set(token.name, CFFStrings[stack.pop()]);
|
|
break;
|
|
case "number number":
|
|
font.set(token.name, {
|
|
offset: stack.pop(),
|
|
size: stack.pop()
|
|
});
|
|
break;
|
|
case "boolean":
|
|
font.set(token.name, stack.pop());
|
|
break;
|
|
case "delta":
|
|
font.set(token.name, stack.pop());
|
|
break;
|
|
default:
|
|
if (token.operand && token.operand.length) {
|
|
var array = [];
|
|
for (var j = 0; j < token.operand.length; j++)
|
|
array.push(stack.pop());
|
|
font.set(token.name, array);
|
|
} else {
|
|
font.set(token.name, stack.pop());
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
};
|
|
|
|
this.parse = function(aStream) {
|
|
font.set("major", aStream.getByte());
|
|
font.set("minor", aStream.getByte());
|
|
font.set("hdrSize", aStream.getByte());
|
|
font.set("offsize", aStream.getByte());
|
|
|
|
// Move the cursor after the header
|
|
aStream.skip(font.get("hdrSize") - aStream.pos);
|
|
|
|
// Read the NAME Index
|
|
dump("Reading Index: Names");
|
|
font.set("Names", readFontIndexData(aStream));
|
|
log("Names: " + font.get("Names"));
|
|
|
|
// Read the Top Dict Index
|
|
dump("Reading Index: TopDict");
|
|
var topDict = readFontIndexData(aStream, true);
|
|
log("TopDict: " + topDict);
|
|
|
|
// Read the String Index
|
|
dump("Reading Index: Strings");
|
|
var strings = readFontIndexData(aStream);
|
|
log("strings: " + strings);
|
|
|
|
// Fill up the Strings dictionary with the new unique strings
|
|
for (var i = 0; i < strings.length; i++)
|
|
CFFStrings.push(strings[i].join(""));
|
|
|
|
// Parse the TopDict operator
|
|
var objects = [];
|
|
var count = topDict.length;
|
|
for (var i = 0; i < count; i++)
|
|
parseAsToken(topDict[i], CFFDictDataMap);
|
|
|
|
// Read the Global Subr Index that comes just after the Strings Index
|
|
// (cf. "The Compact Font Format Specification" Chapter 16)
|
|
dump("Reading Global Subr Index");
|
|
var subrs = readFontIndexData(aStream);
|
|
|
|
// Reading Private Dict
|
|
var private = font.get("Private");
|
|
log("Reading Private Dict (offset: " + private.offset + " size: " + private.size + ")");
|
|
aStream.pos = private.offset;
|
|
|
|
var privateDict = [];
|
|
for (var i = 0; i < private.size; i++)
|
|
privateDict.push(aStream.getByte());
|
|
dump("private:" + privateDict);
|
|
parseAsToken(privateDict, CFFDictPrivateDataMap);
|
|
|
|
for (var p in font.map)
|
|
dump(p + "::" + font.get(p));
|
|
|
|
// Read CharStrings Index
|
|
var charStringsOffset = font.get("CharStrings");
|
|
dump("Read CharStrings Index (offset: " + charStringsOffset + ")");
|
|
aStream.pos = charStringsOffset;
|
|
var charStrings = readFontIndexData(aStream, true);
|
|
|
|
// Read Charset
|
|
dump("Read Charset for " + charStrings.length + " glyphs");
|
|
var charsetEntry = font.get("charset");
|
|
if (charsetEntry == 0) {
|
|
error("Need to support CFFISOAdobeCharset");
|
|
} else if (charsetEntry == 1) {
|
|
error("Need to support CFFExpert");
|
|
} else if (charsetEntry == 2) {
|
|
error("Need to support CFFExpertSubsetCharset");
|
|
} else {
|
|
aStream.pos = charsetEntry;
|
|
var charset = readCharset(aStream, charStrings);
|
|
}
|
|
}
|
|
};
|
|
|
|
|
|
var xhr = new XMLHttpRequest();
|
|
xhr.open("GET", "titi.cff", false);
|
|
xhr.mozResponseType = xhr.responseType = "arraybuffer";
|
|
xhr.expected = (document.URL.indexOf("file:") == 0) ? 0 : 200;
|
|
xhr.send(null);
|
|
var cffData = xhr.mozResponseArrayBuffer || xhr.mozResponse ||
|
|
xhr.responseArrayBuffer || xhr.response;
|
|
var cff = new Type2Parser("titi.cff");
|
|
//cff.parse(new Stream(cffData));
|
|
|
|
|
|
/**
|
|
* Write to a file (works only on Firefox in privilege mode");
|
|
*/
|
|
function writeToFile(aBytes, aFilePath) {
|
|
netscape.security.PrivilegeManager.enablePrivilege("UniversalXPConnect");
|
|
var Cc = Components.classes,
|
|
Ci = Components.interfaces;
|
|
var file = Cc['@mozilla.org/file/local;1'].createInstance(Ci.nsILocalFile);
|
|
file.initWithPath(aFilePath);
|
|
|
|
var stream = Cc["@mozilla.org/network/file-output-stream;1"]
|
|
.createInstance(Ci.nsIFileOutputStream);
|
|
stream.init(file, 0x04 | 0x08 | 0x20, 0600, 0);
|
|
|
|
var bos = Cc["@mozilla.org/binaryoutputstream;1"]
|
|
.createInstance(Ci.nsIBinaryOutputStream);
|
|
bos.setOutputStream(stream);
|
|
bos.writeByteArray(aBytes, aBytes.length);
|
|
stream.close();
|
|
};
|
|
|