wrangler-action/node_modules/chardet/encoding/utf8.js

85 lines
2.1 KiB
JavaScript
Raw Normal View History

2023-08-07 22:11:15 +02:00
var Match = require ('../match');
/**
* Charset recognizer for UTF-8
*/
module.exports = function() {
this.name = function() {
return 'UTF-8';
};
this.match = function(det) {
var hasBOM = false,
numValid = 0,
numInvalid = 0,
input = det.fRawInput,
trailBytes = 0,
confidence;
if (det.fRawLength >= 3 &&
(input[0] & 0xff) == 0xef && (input[1] & 0xff) == 0xbb && (input[2] & 0xff) == 0xbf) {
hasBOM = true;
}
// Scan for multi-byte sequences
for (var i = 0; i < det.fRawLength; i++) {
var b = input[i];
if ((b & 0x80) == 0)
continue; // ASCII
// Hi bit on char found. Figure out how long the sequence should be
if ((b & 0x0e0) == 0x0c0) {
trailBytes = 1;
} else if ((b & 0x0f0) == 0x0e0) {
trailBytes = 2;
} else if ((b & 0x0f8) == 0xf0) {
trailBytes = 3;
} else {
numInvalid++;
if (numInvalid > 5)
break;
trailBytes = 0;
}
// Verify that we've got the right number of trail bytes in the sequence
for (;;) {
i++;
if (i >= det.fRawLength)
break;
if ((input[i] & 0xc0) != 0x080) {
numInvalid++;
break;
}
if (--trailBytes == 0) {
numValid++;
break;
}
}
}
// Cook up some sort of confidence score, based on presense of a BOM
// and the existence of valid and/or invalid multi-byte sequences.
confidence = 0;
if (hasBOM && numInvalid == 0)
confidence = 100;
else if (hasBOM && numValid > numInvalid * 10)
confidence = 80;
else if (numValid > 3 && numInvalid == 0)
confidence = 100;
else if (numValid > 0 && numInvalid == 0)
confidence = 80;
else if (numValid == 0 && numInvalid == 0)
// Plain ASCII.
confidence = 10;
else if (numValid > numInvalid * 10)
// Probably corruput utf-8 data. Valid sequences aren't likely by chance.
confidence = 25;
else
return null
return new Match(det, this, confidence);
};
};