@milpa/comal 0.0.0-stage → 0.1.0

package/node_modules/iconv-lite/encodings/dbcs-codec.js

@@ -0,0 +1,597 @@
+"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]);
+
+    // Load & create GB18030 tables when needed.
+    if (typeof codecOptions.gb18030 === 'function') {
+        this.gb18030 = codecOptions.gb18030(); // Load GB18030 ranges.
+
+        // Add GB18030 common decode nodes.
+        var commonThirdByteNodeIdx = this.decodeTables.length;
+        this.decodeTables.push(UNASSIGNED_NODE.slice(0));
+
+        var commonFourthByteNodeIdx = this.decodeTables.length;
+        this.decodeTables.push(UNASSIGNED_NODE.slice(0));
+
+        // Fill out the tree
+        var firstByteNode = this.decodeTables[0];
+        for (var i = 0x81; i <= 0xFE; i++) {
+            var secondByteNode = this.decodeTables[NODE_START - firstByteNode[i]];
+            for (var j = 0x30; j <= 0x39; j++) {
+                if (secondByteNode[j] === UNASSIGNED) {
+                    secondByteNode[j] = NODE_START - commonThirdByteNodeIdx;
+                } else if (secondByteNode[j] > NODE_START) {
+                    throw new Error("gb18030 decode tables conflict at byte 2");
+                }
+
+                var thirdByteNode = this.decodeTables[NODE_START - secondByteNode[j]];
+                for (var k = 0x81; k <= 0xFE; k++) {
+                    if (thirdByteNode[k] === UNASSIGNED) {
+                        thirdByteNode[k] = NODE_START - commonFourthByteNodeIdx;
+                    } else if (thirdByteNode[k] === NODE_START - commonFourthByteNodeIdx) {
+                        continue;
+                    } else if (thirdByteNode[k] > NODE_START) {
+                        throw new Error("gb18030 decode tables conflict at byte 3");
+                    }
+
+                    var fourthByteNode = this.decodeTables[NODE_START - thirdByteNode[k]];
+                    for (var l = 0x30; l <= 0x39; l++) {
+                        if (fourthByteNode[l] === UNASSIGNED)
+                            fourthByteNode[l] = GB18030_CODE;
+                    }
+                }
+            }
+        }
+    }
+
+    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);
+}
+
+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];
+    var hasValues = false;
+    var subNodeEmpty = {};
+    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);
+            hasValues = true;
+        } else if (uCode <= NODE_START) {
+            var subNodeIdx = NODE_START - uCode;
+            if (!subNodeEmpty[subNodeIdx]) {  // Skip empty subtrees (they are too large in gb18030).
+                var newPrefix = (mbCode << 8) >>> 0;  // NOTE: '>>> 0' keeps 32-bit num positive.
+                if (this._fillEncodeTable(subNodeIdx, newPrefix, skipEncodeChars))
+                    hasValues = true;
+                else
+                    subNodeEmpty[subNodeIdx] = true;
+            }
+        } else if (uCode <= SEQ_START) {
+            this._setEncodeSequence(this.decodeTableSeq[SEQ_START - uCode], mbCode);
+            hasValues = true;
+        }
+    }
+    return hasValues;
+}
+
+
+
+// == 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 if (dbcsCode < 0x1000000) {
+            newBuf[j++] = dbcsCode >> 16;
+            newBuf[j++] = (dbcsCode >> 8) & 0xFF;
+            newBuf[j++] = dbcsCode & 0xFF;
+        } else {
+            newBuf[j++] = dbcsCode >>> 24;
+            newBuf[j++] = (dbcsCode >>> 16) & 0xFF;
+            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.prevBytes = [];
+
+    // 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, 
+        prevBytes = this.prevBytes, prevOffset = this.prevBytes.length,
+        seqStart = -this.prevBytes.length, // idx of the start of current parsed sequence.
+        uCode;
+
+    for (var i = 0, j = 0; i < buf.length; i++) {
+        var curByte = (i >= 0) ? buf[i] : prevBytes[i + prevOffset];
+
+        // 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.
+            uCode = this.defaultCharUnicode.charCodeAt(0);
+            i = seqStart; // Skip one byte ('i' will be incremented by the for loop) and try to parse again.
+        }
+        else if (uCode === GB18030_CODE) {
+            if (i >= 3) {
+                var ptr = (buf[i-3]-0x81)*12600 + (buf[i-2]-0x30)*1260 + (buf[i-1]-0x81)*10 + (curByte-0x30);
+            } else {
+                var ptr = (prevBytes[i-3+prevOffset]-0x81)*12600 + 
+                          (((i-2 >= 0) ? buf[i-2] : prevBytes[i-2+prevOffset])-0x30)*1260 + 
+                          (((i-1 >= 0) ? buf[i-1] : prevBytes[i-1+prevOffset])-0x81)*10 + 
+                          (curByte-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 >= 0x10000) { 
+            uCode -= 0x10000;
+            var uCodeLead = 0xD800 | (uCode >> 10);
+            newBuf[j++] = uCodeLead & 0xFF;
+            newBuf[j++] = uCodeLead >> 8;
+
+            uCode = 0xDC00 | (uCode & 0x3FF);
+        }
+        newBuf[j++] = uCode & 0xFF;
+        newBuf[j++] = uCode >> 8;
+
+        // Reset trie node.
+        nodeIdx = 0; seqStart = i+1;
+    }
+
+    this.nodeIdx = nodeIdx;
+    this.prevBytes = (seqStart >= 0)
+        ? Array.prototype.slice.call(buf, seqStart)
+        : prevBytes.slice(seqStart + prevOffset).concat(Array.prototype.slice.call(buf));
+
+    return newBuf.slice(0, j).toString('ucs2');
+}
+
+DBCSDecoder.prototype.end = function() {
+    var ret = '';
+
+    // Try to parse all remaining chars.
+    while (this.prevBytes.length > 0) {
+        // Skip 1 character in the buffer.
+        ret += this.defaultCharUnicode;
+        var bytesArr = this.prevBytes.slice(1);
+
+        // Parse remaining as usual.
+        this.prevBytes = [];
+        this.nodeIdx = 0;
+        if (bytesArr.length > 0)
+            ret += this.write(bytesArr);
+    }
+
+    this.prevBytes = [];
+    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 + ((r-l+1) >> 1);
+        if (table[mid] <= val)
+            l = mid;
+        else
+            r = mid;
+    }
+    return l;
+}
+