mirror of
https://github.com/idanoo/GoScrobble.git
synced 2024-11-25 09:55:15 +00:00
556 lines
21 KiB
JavaScript
556 lines
21 KiB
JavaScript
"use strict";
|
|
var Buffer = require("safer-buffer").Buffer;
|
|
|
|
// Multibyte codec. In this scheme, a character is represented by 1 or more bytes.
|
|
// Our codec supports UTF-16 surrogates, extensions for GB18030 and unicode sequences.
|
|
// To save memory and loading time, we read table files only when requested.
|
|
|
|
exports._dbcs = DBCSCodec;
|
|
|
|
var UNASSIGNED = -1,
|
|
GB18030_CODE = -2,
|
|
SEQ_START = -10,
|
|
NODE_START = -1000,
|
|
UNASSIGNED_NODE = new Array(0x100),
|
|
DEF_CHAR = -1;
|
|
|
|
for (var i = 0; i < 0x100; i++)
|
|
UNASSIGNED_NODE[i] = UNASSIGNED;
|
|
|
|
|
|
// Class DBCSCodec reads and initializes mapping tables.
|
|
function DBCSCodec(codecOptions, iconv) {
|
|
this.encodingName = codecOptions.encodingName;
|
|
if (!codecOptions)
|
|
throw new Error("DBCS codec is called without the data.")
|
|
if (!codecOptions.table)
|
|
throw new Error("Encoding '" + this.encodingName + "' has no data.");
|
|
|
|
// Load tables.
|
|
var mappingTable = codecOptions.table();
|
|
|
|
|
|
// Decode tables: MBCS -> Unicode.
|
|
|
|
// decodeTables is a trie, encoded as an array of arrays of integers. Internal arrays are trie nodes and all have len = 256.
|
|
// Trie root is decodeTables[0].
|
|
// Values: >= 0 -> unicode character code. can be > 0xFFFF
|
|
// == UNASSIGNED -> unknown/unassigned sequence.
|
|
// == GB18030_CODE -> this is the end of a GB18030 4-byte sequence.
|
|
// <= NODE_START -> index of the next node in our trie to process next byte.
|
|
// <= SEQ_START -> index of the start of a character code sequence, in decodeTableSeq.
|
|
this.decodeTables = [];
|
|
this.decodeTables[0] = UNASSIGNED_NODE.slice(0); // Create root node.
|
|
|
|
// Sometimes a MBCS char corresponds to a sequence of unicode chars. We store them as arrays of integers here.
|
|
this.decodeTableSeq = [];
|
|
|
|
// Actual mapping tables consist of chunks. Use them to fill up decode tables.
|
|
for (var i = 0; i < mappingTable.length; i++)
|
|
this._addDecodeChunk(mappingTable[i]);
|
|
|
|
this.defaultCharUnicode = iconv.defaultCharUnicode;
|
|
|
|
|
|
// Encode tables: Unicode -> DBCS.
|
|
|
|
// `encodeTable` is array mapping from unicode char to encoded char. All its values are integers for performance.
|
|
// Because it can be sparse, it is represented as array of buckets by 256 chars each. Bucket can be null.
|
|
// Values: >= 0 -> it is a normal char. Write the value (if <=256 then 1 byte, if <=65536 then 2 bytes, etc.).
|
|
// == UNASSIGNED -> no conversion found. Output a default char.
|
|
// <= SEQ_START -> it's an index in encodeTableSeq, see below. The character starts a sequence.
|
|
this.encodeTable = [];
|
|
|
|
// `encodeTableSeq` is used when a sequence of unicode characters is encoded as a single code. We use a tree of
|
|
// objects where keys correspond to characters in sequence and leafs are the encoded dbcs values. A special DEF_CHAR key
|
|
// means end of sequence (needed when one sequence is a strict subsequence of another).
|
|
// Objects are kept separately from encodeTable to increase performance.
|
|
this.encodeTableSeq = [];
|
|
|
|
// Some chars can be decoded, but need not be encoded.
|
|
var skipEncodeChars = {};
|
|
if (codecOptions.encodeSkipVals)
|
|
for (var i = 0; i < codecOptions.encodeSkipVals.length; i++) {
|
|
var val = codecOptions.encodeSkipVals[i];
|
|
if (typeof val === 'number')
|
|
skipEncodeChars[val] = true;
|
|
else
|
|
for (var j = val.from; j <= val.to; j++)
|
|
skipEncodeChars[j] = true;
|
|
}
|
|
|
|
// Use decode trie to recursively fill out encode tables.
|
|
this._fillEncodeTable(0, 0, skipEncodeChars);
|
|
|
|
// Add more encoding pairs when needed.
|
|
if (codecOptions.encodeAdd) {
|
|
for (var uChar in codecOptions.encodeAdd)
|
|
if (Object.prototype.hasOwnProperty.call(codecOptions.encodeAdd, uChar))
|
|
this._setEncodeChar(uChar.charCodeAt(0), codecOptions.encodeAdd[uChar]);
|
|
}
|
|
|
|
this.defCharSB = this.encodeTable[0][iconv.defaultCharSingleByte.charCodeAt(0)];
|
|
if (this.defCharSB === UNASSIGNED) this.defCharSB = this.encodeTable[0]['?'];
|
|
if (this.defCharSB === UNASSIGNED) this.defCharSB = "?".charCodeAt(0);
|
|
|
|
|
|
// Load & create GB18030 tables when needed.
|
|
if (typeof codecOptions.gb18030 === 'function') {
|
|
this.gb18030 = codecOptions.gb18030(); // Load GB18030 ranges.
|
|
|
|
// Add GB18030 decode tables.
|
|
var thirdByteNodeIdx = this.decodeTables.length;
|
|
var thirdByteNode = this.decodeTables[thirdByteNodeIdx] = UNASSIGNED_NODE.slice(0);
|
|
|
|
var fourthByteNodeIdx = this.decodeTables.length;
|
|
var fourthByteNode = this.decodeTables[fourthByteNodeIdx] = UNASSIGNED_NODE.slice(0);
|
|
|
|
for (var i = 0x81; i <= 0xFE; i++) {
|
|
var secondByteNodeIdx = NODE_START - this.decodeTables[0][i];
|
|
var secondByteNode = this.decodeTables[secondByteNodeIdx];
|
|
for (var j = 0x30; j <= 0x39; j++)
|
|
secondByteNode[j] = NODE_START - thirdByteNodeIdx;
|
|
}
|
|
for (var i = 0x81; i <= 0xFE; i++)
|
|
thirdByteNode[i] = NODE_START - fourthByteNodeIdx;
|
|
for (var i = 0x30; i <= 0x39; i++)
|
|
fourthByteNode[i] = GB18030_CODE
|
|
}
|
|
}
|
|
|
|
DBCSCodec.prototype.encoder = DBCSEncoder;
|
|
DBCSCodec.prototype.decoder = DBCSDecoder;
|
|
|
|
// Decoder helpers
|
|
DBCSCodec.prototype._getDecodeTrieNode = function(addr) {
|
|
var bytes = [];
|
|
for (; addr > 0; addr >>= 8)
|
|
bytes.push(addr & 0xFF);
|
|
if (bytes.length == 0)
|
|
bytes.push(0);
|
|
|
|
var node = this.decodeTables[0];
|
|
for (var i = bytes.length-1; i > 0; i--) { // Traverse nodes deeper into the trie.
|
|
var val = node[bytes[i]];
|
|
|
|
if (val == UNASSIGNED) { // Create new node.
|
|
node[bytes[i]] = NODE_START - this.decodeTables.length;
|
|
this.decodeTables.push(node = UNASSIGNED_NODE.slice(0));
|
|
}
|
|
else if (val <= NODE_START) { // Existing node.
|
|
node = this.decodeTables[NODE_START - val];
|
|
}
|
|
else
|
|
throw new Error("Overwrite byte in " + this.encodingName + ", addr: " + addr.toString(16));
|
|
}
|
|
return node;
|
|
}
|
|
|
|
|
|
DBCSCodec.prototype._addDecodeChunk = function(chunk) {
|
|
// First element of chunk is the hex mbcs code where we start.
|
|
var curAddr = parseInt(chunk[0], 16);
|
|
|
|
// Choose the decoding node where we'll write our chars.
|
|
var writeTable = this._getDecodeTrieNode(curAddr);
|
|
curAddr = curAddr & 0xFF;
|
|
|
|
// Write all other elements of the chunk to the table.
|
|
for (var k = 1; k < chunk.length; k++) {
|
|
var part = chunk[k];
|
|
if (typeof part === "string") { // String, write as-is.
|
|
for (var l = 0; l < part.length;) {
|
|
var code = part.charCodeAt(l++);
|
|
if (0xD800 <= code && code < 0xDC00) { // Decode surrogate
|
|
var codeTrail = part.charCodeAt(l++);
|
|
if (0xDC00 <= codeTrail && codeTrail < 0xE000)
|
|
writeTable[curAddr++] = 0x10000 + (code - 0xD800) * 0x400 + (codeTrail - 0xDC00);
|
|
else
|
|
throw new Error("Incorrect surrogate pair in " + this.encodingName + " at chunk " + chunk[0]);
|
|
}
|
|
else if (0x0FF0 < code && code <= 0x0FFF) { // Character sequence (our own encoding used)
|
|
var len = 0xFFF - code + 2;
|
|
var seq = [];
|
|
for (var m = 0; m < len; m++)
|
|
seq.push(part.charCodeAt(l++)); // Simple variation: don't support surrogates or subsequences in seq.
|
|
|
|
writeTable[curAddr++] = SEQ_START - this.decodeTableSeq.length;
|
|
this.decodeTableSeq.push(seq);
|
|
}
|
|
else
|
|
writeTable[curAddr++] = code; // Basic char
|
|
}
|
|
}
|
|
else if (typeof part === "number") { // Integer, meaning increasing sequence starting with prev character.
|
|
var charCode = writeTable[curAddr - 1] + 1;
|
|
for (var l = 0; l < part; l++)
|
|
writeTable[curAddr++] = charCode++;
|
|
}
|
|
else
|
|
throw new Error("Incorrect type '" + typeof part + "' given in " + this.encodingName + " at chunk " + chunk[0]);
|
|
}
|
|
if (curAddr > 0xFF)
|
|
throw new Error("Incorrect chunk in " + this.encodingName + " at addr " + chunk[0] + ": too long" + curAddr);
|
|
}
|
|
|
|
// Encoder helpers
|
|
DBCSCodec.prototype._getEncodeBucket = function(uCode) {
|
|
var high = uCode >> 8; // This could be > 0xFF because of astral characters.
|
|
if (this.encodeTable[high] === undefined)
|
|
this.encodeTable[high] = UNASSIGNED_NODE.slice(0); // Create bucket on demand.
|
|
return this.encodeTable[high];
|
|
}
|
|
|
|
DBCSCodec.prototype._setEncodeChar = function(uCode, dbcsCode) {
|
|
var bucket = this._getEncodeBucket(uCode);
|
|
var low = uCode & 0xFF;
|
|
if (bucket[low] <= SEQ_START)
|
|
this.encodeTableSeq[SEQ_START-bucket[low]][DEF_CHAR] = dbcsCode; // There's already a sequence, set a single-char subsequence of it.
|
|
else if (bucket[low] == UNASSIGNED)
|
|
bucket[low] = dbcsCode;
|
|
}
|
|
|
|
DBCSCodec.prototype._setEncodeSequence = function(seq, dbcsCode) {
|
|
|
|
// Get the root of character tree according to first character of the sequence.
|
|
var uCode = seq[0];
|
|
var bucket = this._getEncodeBucket(uCode);
|
|
var low = uCode & 0xFF;
|
|
|
|
var node;
|
|
if (bucket[low] <= SEQ_START) {
|
|
// There's already a sequence with - use it.
|
|
node = this.encodeTableSeq[SEQ_START-bucket[low]];
|
|
}
|
|
else {
|
|
// There was no sequence object - allocate a new one.
|
|
node = {};
|
|
if (bucket[low] !== UNASSIGNED) node[DEF_CHAR] = bucket[low]; // If a char was set before - make it a single-char subsequence.
|
|
bucket[low] = SEQ_START - this.encodeTableSeq.length;
|
|
this.encodeTableSeq.push(node);
|
|
}
|
|
|
|
// Traverse the character tree, allocating new nodes as needed.
|
|
for (var j = 1; j < seq.length-1; j++) {
|
|
var oldVal = node[uCode];
|
|
if (typeof oldVal === 'object')
|
|
node = oldVal;
|
|
else {
|
|
node = node[uCode] = {}
|
|
if (oldVal !== undefined)
|
|
node[DEF_CHAR] = oldVal
|
|
}
|
|
}
|
|
|
|
// Set the leaf to given dbcsCode.
|
|
uCode = seq[seq.length-1];
|
|
node[uCode] = dbcsCode;
|
|
}
|
|
|
|
DBCSCodec.prototype._fillEncodeTable = function(nodeIdx, prefix, skipEncodeChars) {
|
|
var node = this.decodeTables[nodeIdx];
|
|
for (var i = 0; i < 0x100; i++) {
|
|
var uCode = node[i];
|
|
var mbCode = prefix + i;
|
|
if (skipEncodeChars[mbCode])
|
|
continue;
|
|
|
|
if (uCode >= 0)
|
|
this._setEncodeChar(uCode, mbCode);
|
|
else if (uCode <= NODE_START)
|
|
this._fillEncodeTable(NODE_START - uCode, mbCode << 8, skipEncodeChars);
|
|
else if (uCode <= SEQ_START)
|
|
this._setEncodeSequence(this.decodeTableSeq[SEQ_START - uCode], mbCode);
|
|
}
|
|
}
|
|
|
|
|
|
|
|
// == Encoder ==================================================================
|
|
|
|
function DBCSEncoder(options, codec) {
|
|
// Encoder state
|
|
this.leadSurrogate = -1;
|
|
this.seqObj = undefined;
|
|
|
|
// Static data
|
|
this.encodeTable = codec.encodeTable;
|
|
this.encodeTableSeq = codec.encodeTableSeq;
|
|
this.defaultCharSingleByte = codec.defCharSB;
|
|
this.gb18030 = codec.gb18030;
|
|
}
|
|
|
|
DBCSEncoder.prototype.write = function(str) {
|
|
var newBuf = Buffer.alloc(str.length * (this.gb18030 ? 4 : 3)),
|
|
leadSurrogate = this.leadSurrogate,
|
|
seqObj = this.seqObj, nextChar = -1,
|
|
i = 0, j = 0;
|
|
|
|
while (true) {
|
|
// 0. Get next character.
|
|
if (nextChar === -1) {
|
|
if (i == str.length) break;
|
|
var uCode = str.charCodeAt(i++);
|
|
}
|
|
else {
|
|
var uCode = nextChar;
|
|
nextChar = -1;
|
|
}
|
|
|
|
// 1. Handle surrogates.
|
|
if (0xD800 <= uCode && uCode < 0xE000) { // Char is one of surrogates.
|
|
if (uCode < 0xDC00) { // We've got lead surrogate.
|
|
if (leadSurrogate === -1) {
|
|
leadSurrogate = uCode;
|
|
continue;
|
|
} else {
|
|
leadSurrogate = uCode;
|
|
// Double lead surrogate found.
|
|
uCode = UNASSIGNED;
|
|
}
|
|
} else { // We've got trail surrogate.
|
|
if (leadSurrogate !== -1) {
|
|
uCode = 0x10000 + (leadSurrogate - 0xD800) * 0x400 + (uCode - 0xDC00);
|
|
leadSurrogate = -1;
|
|
} else {
|
|
// Incomplete surrogate pair - only trail surrogate found.
|
|
uCode = UNASSIGNED;
|
|
}
|
|
|
|
}
|
|
}
|
|
else if (leadSurrogate !== -1) {
|
|
// Incomplete surrogate pair - only lead surrogate found.
|
|
nextChar = uCode; uCode = UNASSIGNED; // Write an error, then current char.
|
|
leadSurrogate = -1;
|
|
}
|
|
|
|
// 2. Convert uCode character.
|
|
var dbcsCode = UNASSIGNED;
|
|
if (seqObj !== undefined && uCode != UNASSIGNED) { // We are in the middle of the sequence
|
|
var resCode = seqObj[uCode];
|
|
if (typeof resCode === 'object') { // Sequence continues.
|
|
seqObj = resCode;
|
|
continue;
|
|
|
|
} else if (typeof resCode == 'number') { // Sequence finished. Write it.
|
|
dbcsCode = resCode;
|
|
|
|
} else if (resCode == undefined) { // Current character is not part of the sequence.
|
|
|
|
// Try default character for this sequence
|
|
resCode = seqObj[DEF_CHAR];
|
|
if (resCode !== undefined) {
|
|
dbcsCode = resCode; // Found. Write it.
|
|
nextChar = uCode; // Current character will be written too in the next iteration.
|
|
|
|
} else {
|
|
// TODO: What if we have no default? (resCode == undefined)
|
|
// Then, we should write first char of the sequence as-is and try the rest recursively.
|
|
// Didn't do it for now because no encoding has this situation yet.
|
|
// Currently, just skip the sequence and write current char.
|
|
}
|
|
}
|
|
seqObj = undefined;
|
|
}
|
|
else if (uCode >= 0) { // Regular character
|
|
var subtable = this.encodeTable[uCode >> 8];
|
|
if (subtable !== undefined)
|
|
dbcsCode = subtable[uCode & 0xFF];
|
|
|
|
if (dbcsCode <= SEQ_START) { // Sequence start
|
|
seqObj = this.encodeTableSeq[SEQ_START-dbcsCode];
|
|
continue;
|
|
}
|
|
|
|
if (dbcsCode == UNASSIGNED && this.gb18030) {
|
|
// Use GB18030 algorithm to find character(s) to write.
|
|
var idx = findIdx(this.gb18030.uChars, uCode);
|
|
if (idx != -1) {
|
|
var dbcsCode = this.gb18030.gbChars[idx] + (uCode - this.gb18030.uChars[idx]);
|
|
newBuf[j++] = 0x81 + Math.floor(dbcsCode / 12600); dbcsCode = dbcsCode % 12600;
|
|
newBuf[j++] = 0x30 + Math.floor(dbcsCode / 1260); dbcsCode = dbcsCode % 1260;
|
|
newBuf[j++] = 0x81 + Math.floor(dbcsCode / 10); dbcsCode = dbcsCode % 10;
|
|
newBuf[j++] = 0x30 + dbcsCode;
|
|
continue;
|
|
}
|
|
}
|
|
}
|
|
|
|
// 3. Write dbcsCode character.
|
|
if (dbcsCode === UNASSIGNED)
|
|
dbcsCode = this.defaultCharSingleByte;
|
|
|
|
if (dbcsCode < 0x100) {
|
|
newBuf[j++] = dbcsCode;
|
|
}
|
|
else if (dbcsCode < 0x10000) {
|
|
newBuf[j++] = dbcsCode >> 8; // high byte
|
|
newBuf[j++] = dbcsCode & 0xFF; // low byte
|
|
}
|
|
else {
|
|
newBuf[j++] = dbcsCode >> 16;
|
|
newBuf[j++] = (dbcsCode >> 8) & 0xFF;
|
|
newBuf[j++] = dbcsCode & 0xFF;
|
|
}
|
|
}
|
|
|
|
this.seqObj = seqObj;
|
|
this.leadSurrogate = leadSurrogate;
|
|
return newBuf.slice(0, j);
|
|
}
|
|
|
|
DBCSEncoder.prototype.end = function() {
|
|
if (this.leadSurrogate === -1 && this.seqObj === undefined)
|
|
return; // All clean. Most often case.
|
|
|
|
var newBuf = Buffer.alloc(10), j = 0;
|
|
|
|
if (this.seqObj) { // We're in the sequence.
|
|
var dbcsCode = this.seqObj[DEF_CHAR];
|
|
if (dbcsCode !== undefined) { // Write beginning of the sequence.
|
|
if (dbcsCode < 0x100) {
|
|
newBuf[j++] = dbcsCode;
|
|
}
|
|
else {
|
|
newBuf[j++] = dbcsCode >> 8; // high byte
|
|
newBuf[j++] = dbcsCode & 0xFF; // low byte
|
|
}
|
|
} else {
|
|
// See todo above.
|
|
}
|
|
this.seqObj = undefined;
|
|
}
|
|
|
|
if (this.leadSurrogate !== -1) {
|
|
// Incomplete surrogate pair - only lead surrogate found.
|
|
newBuf[j++] = this.defaultCharSingleByte;
|
|
this.leadSurrogate = -1;
|
|
}
|
|
|
|
return newBuf.slice(0, j);
|
|
}
|
|
|
|
// Export for testing
|
|
DBCSEncoder.prototype.findIdx = findIdx;
|
|
|
|
|
|
// == Decoder ==================================================================
|
|
|
|
function DBCSDecoder(options, codec) {
|
|
// Decoder state
|
|
this.nodeIdx = 0;
|
|
this.prevBuf = Buffer.alloc(0);
|
|
|
|
// Static data
|
|
this.decodeTables = codec.decodeTables;
|
|
this.decodeTableSeq = codec.decodeTableSeq;
|
|
this.defaultCharUnicode = codec.defaultCharUnicode;
|
|
this.gb18030 = codec.gb18030;
|
|
}
|
|
|
|
DBCSDecoder.prototype.write = function(buf) {
|
|
var newBuf = Buffer.alloc(buf.length*2),
|
|
nodeIdx = this.nodeIdx,
|
|
prevBuf = this.prevBuf, prevBufOffset = this.prevBuf.length,
|
|
seqStart = -this.prevBuf.length, // idx of the start of current parsed sequence.
|
|
uCode;
|
|
|
|
if (prevBufOffset > 0) // Make prev buf overlap a little to make it easier to slice later.
|
|
prevBuf = Buffer.concat([prevBuf, buf.slice(0, 10)]);
|
|
|
|
for (var i = 0, j = 0; i < buf.length; i++) {
|
|
var curByte = (i >= 0) ? buf[i] : prevBuf[i + prevBufOffset];
|
|
|
|
// Lookup in current trie node.
|
|
var uCode = this.decodeTables[nodeIdx][curByte];
|
|
|
|
if (uCode >= 0) {
|
|
// Normal character, just use it.
|
|
}
|
|
else if (uCode === UNASSIGNED) { // Unknown char.
|
|
// TODO: Callback with seq.
|
|
//var curSeq = (seqStart >= 0) ? buf.slice(seqStart, i+1) : prevBuf.slice(seqStart + prevBufOffset, i+1 + prevBufOffset);
|
|
i = seqStart; // Try to parse again, after skipping first byte of the sequence ('i' will be incremented by 'for' cycle).
|
|
uCode = this.defaultCharUnicode.charCodeAt(0);
|
|
}
|
|
else if (uCode === GB18030_CODE) {
|
|
var curSeq = (seqStart >= 0) ? buf.slice(seqStart, i+1) : prevBuf.slice(seqStart + prevBufOffset, i+1 + prevBufOffset);
|
|
var ptr = (curSeq[0]-0x81)*12600 + (curSeq[1]-0x30)*1260 + (curSeq[2]-0x81)*10 + (curSeq[3]-0x30);
|
|
var idx = findIdx(this.gb18030.gbChars, ptr);
|
|
uCode = this.gb18030.uChars[idx] + ptr - this.gb18030.gbChars[idx];
|
|
}
|
|
else if (uCode <= NODE_START) { // Go to next trie node.
|
|
nodeIdx = NODE_START - uCode;
|
|
continue;
|
|
}
|
|
else if (uCode <= SEQ_START) { // Output a sequence of chars.
|
|
var seq = this.decodeTableSeq[SEQ_START - uCode];
|
|
for (var k = 0; k < seq.length - 1; k++) {
|
|
uCode = seq[k];
|
|
newBuf[j++] = uCode & 0xFF;
|
|
newBuf[j++] = uCode >> 8;
|
|
}
|
|
uCode = seq[seq.length-1];
|
|
}
|
|
else
|
|
throw new Error("iconv-lite internal error: invalid decoding table value " + uCode + " at " + nodeIdx + "/" + curByte);
|
|
|
|
// Write the character to buffer, handling higher planes using surrogate pair.
|
|
if (uCode > 0xFFFF) {
|
|
uCode -= 0x10000;
|
|
var uCodeLead = 0xD800 + Math.floor(uCode / 0x400);
|
|
newBuf[j++] = uCodeLead & 0xFF;
|
|
newBuf[j++] = uCodeLead >> 8;
|
|
|
|
uCode = 0xDC00 + uCode % 0x400;
|
|
}
|
|
newBuf[j++] = uCode & 0xFF;
|
|
newBuf[j++] = uCode >> 8;
|
|
|
|
// Reset trie node.
|
|
nodeIdx = 0; seqStart = i+1;
|
|
}
|
|
|
|
this.nodeIdx = nodeIdx;
|
|
this.prevBuf = (seqStart >= 0) ? buf.slice(seqStart) : prevBuf.slice(seqStart + prevBufOffset);
|
|
return newBuf.slice(0, j).toString('ucs2');
|
|
}
|
|
|
|
DBCSDecoder.prototype.end = function() {
|
|
var ret = '';
|
|
|
|
// Try to parse all remaining chars.
|
|
while (this.prevBuf.length > 0) {
|
|
// Skip 1 character in the buffer.
|
|
ret += this.defaultCharUnicode;
|
|
var buf = this.prevBuf.slice(1);
|
|
|
|
// Parse remaining as usual.
|
|
this.prevBuf = Buffer.alloc(0);
|
|
this.nodeIdx = 0;
|
|
if (buf.length > 0)
|
|
ret += this.write(buf);
|
|
}
|
|
|
|
this.nodeIdx = 0;
|
|
return ret;
|
|
}
|
|
|
|
// Binary search for GB18030. Returns largest i such that table[i] <= val.
|
|
function findIdx(table, val) {
|
|
if (table[0] > val)
|
|
return -1;
|
|
|
|
var l = 0, r = table.length;
|
|
while (l < r-1) { // always table[l] <= val < table[r]
|
|
var mid = l + Math.floor((r-l+1)/2);
|
|
if (table[mid] <= val)
|
|
l = mid;
|
|
else
|
|
r = mid;
|
|
}
|
|
return l;
|
|
}
|
|
|