This repository has been archived on 2018-10-12. You can view files and clone it, but cannot push or open issues or pull requests.
node-task/node_modules/body-parser/node_modules/iconv-lite/encodings/utf16.js
2014-09-18 15:35:58 -04:00

203 lines
5.1 KiB
JavaScript

// == UTF16-BE codec. ==========================================================
exports.utf16be = function(options) {
return {
encoder: utf16beEncoder,
decoder: utf16beDecoder,
bom: new Buffer([0xFE, 0xFF]),
};
};
// -- Encoding
function utf16beEncoder(options) {
return {
write: utf16beEncoderWrite,
end: function() {},
}
}
function utf16beEncoderWrite(str) {
var buf = new Buffer(str, 'ucs2');
for (var i = 0; i < buf.length; i += 2) {
var tmp = buf[i]; buf[i] = buf[i+1]; buf[i+1] = tmp;
}
return buf;
}
// -- Decoding
function utf16beDecoder(options) {
return {
write: utf16beDecoderWrite,
end: function() {},
overflowByte: -1,
};
}
function utf16beDecoderWrite(buf) {
if (buf.length == 0)
return '';
var buf2 = new Buffer(buf.length + 1),
i = 0, j = 0;
if (this.overflowByte !== -1) {
buf2[0] = buf[0];
buf2[1] = this.overflowByte;
i = 1; j = 2;
}
for (; i < buf.length-1; i += 2, j+= 2) {
buf2[j] = buf[i+1];
buf2[j+1] = buf[i];
}
this.overflowByte = (i == buf.length-1) ? buf[buf.length-1] : -1;
return buf2.slice(0, j).toString('ucs2');
}
// == UTF-16 codec =============================================================
// Decoder chooses automatically from UTF-16LE and UTF-16BE using BOM and space-based heuristic.
// Defaults to UTF-16BE, according to RFC 2781, although it is against some industry practices, see
// http://en.wikipedia.org/wiki/UTF-16 and http://encoding.spec.whatwg.org/#utf-16le
// Decoder default can be changed: iconv.decode(buf, 'utf16', {default: 'utf-16le'});
// Encoder prepends BOM and uses UTF-16BE.
// Endianness can also be changed: iconv.encode(str, 'utf16', {use: 'utf-16le'});
exports.utf16 = function(options) {
return {
encoder: utf16Encoder,
decoder: utf16Decoder,
getCodec: options.iconv.getCodec,
};
};
// -- Encoding
function utf16Encoder(options) {
options = options || {};
var codec = this.getCodec(options.use || 'utf-16be');
if (!codec.bom)
throw new Error("iconv-lite: in UTF-16 encoder, 'use' parameter should be either UTF-16BE or UTF16-LE.");
return {
write: utf16EncoderWrite,
end: utf16EncoderEnd,
bom: codec.bom,
internalEncoder: codec.encoder(options),
};
}
function utf16EncoderWrite(str) {
var buf = this.internalEncoder.write(str);
if (this.bom) {
buf = Buffer.concat([this.bom, buf]);
this.bom = null;
}
return buf;
}
function utf16EncoderEnd() {
return this.internalEncoder.end();
}
// -- Decoding
function utf16Decoder(options) {
return {
write: utf16DecoderWrite,
end: utf16DecoderEnd,
internalDecoder: null,
initialBytes: [],
initialBytesLen: 0,
options: options || {},
getCodec: this.getCodec,
};
}
function utf16DecoderWrite(buf) {
if (this.internalDecoder)
return this.internalDecoder.write(buf);
// Codec is not chosen yet. Accumulate initial bytes.
this.initialBytes.push(buf);
this.initialBytesLen += buf.length;
if (this.initialBytesLen < 16) // We need > 2 bytes to use space heuristic (see below)
return '';
// We have enough bytes -> decide endianness.
return utf16DecoderDecideEndianness.call(this);
}
function utf16DecoderEnd() {
if (this.internalDecoder)
return this.internalDecoder.end();
var res = utf16DecoderDecideEndianness.call(this);
var trail;
if (this.internalDecoder)
trail = this.internalDecoder.end();
return (trail && trail.length > 0) ? (res + trail) : res;
}
function utf16DecoderDecideEndianness() {
var buf = Buffer.concat(this.initialBytes);
this.initialBytes.length = this.initialBytesLen = 0;
if (buf.length < 2)
return ''; // Not a valid UTF-16 sequence anyway.
// Default encoding.
var enc = this.options.default || 'utf-16be';
// Check BOM.
if (buf[0] == 0xFE && buf[1] == 0xFF) { // UTF-16BE BOM
enc = 'utf-16be'; buf = buf.slice(2);
}
else if (buf[0] == 0xFF && buf[1] == 0xFE) { // UTF-16LE BOM
enc = 'utf-16le'; buf = buf.slice(2);
}
else {
// No BOM found. Try to deduce encoding from initial content.
// Most of the time, the content has spaces (U+0020), but the opposite (U+2000) is very uncommon.
// So, we count spaces as if it was LE or BE, and decide from that.
var spaces = [0, 0], // Counts of space chars in both positions
_len = Math.min(buf.length - (buf.length % 2), 64); // Len is always even.
for (var i = 0; i < _len; i += 2) {
if (buf[i] == 0x00 && buf[i+1] == 0x20) spaces[0]++;
if (buf[i] == 0x20 && buf[i+1] == 0x00) spaces[1]++;
}
if (spaces[0] > 0 && spaces[1] == 0)
enc = 'utf-16be';
else if (spaces[0] == 0 && spaces[1] > 0)
enc = 'utf-16le';
}
this.internalDecoder = this.getCodec(enc).decoder(this.options);
return this.internalDecoder.write(buf);
}