Fix incorrect base64 encoding of astral-plane text (#4813)

Most astral-plane text is emojis like U+1F4DA BOOKS (📚), but some
languages like Osage have their alphabet entirely in the supplementary
multilingual plane as well. For proper support of languages like Osage,
and newer emojis, the UTF-8 decode and encode functions need to properly
handle codepoints above U+FFFF, which are represented by a surrogate
pair in Javascript strings.
This commit is contained in:
Robin Munn
2020-08-17 23:33:46 +07:00
committed by GitHub
parent f74c49f393
commit c23eedd069
3 changed files with 43 additions and 12 deletions

View File

@@ -20,15 +20,27 @@
for (i; i < string.length; i++) {
charCode = string.charCodeAt(i);
if (charCode < 128)
if (charCode < 128) {
output += String.fromCharCode(charCode);
else if ((charCode > 127) && (charCode < 2048))
output += String.fromCharCode((charCode >> 6) | 192),
} else if ((charCode > 127) && (charCode < 2048)) {
output += String.fromCharCode((charCode >> 6) | 192);
output += String.fromCharCode((charCode & 63) | 128);
else
output += String.fromCharCode((charCode >> 12) | 224),
output += String.fromCharCode(((charCode >> 6) & 63) | 128),
} else if ((charCode > 55295) && (charCode < 57344) && string.length > i+1) {
// Surrogate pair
var hiSurrogate = charCode;
var loSurrogate = string.charCodeAt(i+1);
i++; // Skip the low surrogate on the next loop pass
var codePoint = (((hiSurrogate - 55296) << 10) | (loSurrogate - 56320)) + 65536;
output += String.fromCharCode((codePoint >> 18) | 240);
output += String.fromCharCode(((codePoint >> 12) & 63) | 128);
output += String.fromCharCode(((codePoint >> 6) & 63) | 128);
output += String.fromCharCode((codePoint & 63) | 128);
} else {
// Not a surrogate pair, or a dangling surrogate without its partner that we'll just encode as-is
output += String.fromCharCode((charCode >> 12) | 224);
output += String.fromCharCode(((charCode >> 6) & 63) | 128);
output += String.fromCharCode((charCode & 63) | 128);
}
}
return output;
@@ -41,15 +53,21 @@
while (i < string.length) {
charCode = string.charCodeAt(i);
if (charCode < 128)
if (charCode < 128) {
output += String.fromCharCode(charCode),
i++;
else if ((charCode > 191) && (charCode < 224))
output += String.fromCharCode(((charCode & 31) << 6) | (string.charCodeAt(i + 1) & 63)),
} else if ((charCode > 191) && (charCode < 224)) {
output += String.fromCharCode(((charCode & 31) << 6) | (string.charCodeAt(i + 1) & 63));
i += 2;
else
output += String.fromCharCode(((charCode & 15) << 12) | ((string.charCodeAt(i + 1) & 63) << 6) | (string.charCodeAt(i + 2) & 63)),
} else if ((charCode > 223) && (charCode < 240)) {
output += String.fromCharCode(((charCode & 15) << 12) | ((string.charCodeAt(i + 1) & 63) << 6) | (string.charCodeAt(i + 2) & 63));
i += 3;
} else {
var codePoint = ((charCode & 7) << 18) | ((string.charCodeAt(i + 1) & 63) << 12) | ((string.charCodeAt(i + 2) & 63) << 6) | (string.charCodeAt(i + 3) & 63);
// output += String.fromCodePoint(codePoint); // Can't do this because Internet Explorer doesn't have String.fromCodePoint
output += String.fromCharCode(((codePoint - 65536) >> 10) + 55296) + String.fromCharCode(((codePoint - 65536) & 1023) + 56320); // So we do this instead
i += 4;
}
}
return output;