1
0
mirror of https://github.com/Jermolene/TiddlyWiki5 synced 2024-11-24 02:27:19 +00:00
TiddlyWiki5/core/modules/utils/base64-utf8/base64-utf8.module.js
Robin Munn c23eedd069
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.
2020-08-17 17:33:46 +01:00

142 lines
6.0 KiB
JavaScript

// From https://gist.github.com/Nijikokun/5192472
//
// UTF8 Module
//
// Cleaner and modularized utf-8 encoding and decoding library for javascript.
//
// copyright: MIT
// author: Nijiko Yonskai, @nijikokun, nijikokun@gmail.com
(function (name, definition, context, dependencies) {
if (typeof context['module'] !== 'undefined' && context['module']['exports']) { if (dependencies && context['require']) { for (var i = 0; i < dependencies.length; i++) context[dependencies[i]] = context['require'](dependencies[i]); } context['module']['exports'] = definition.apply(context); }
else if (typeof context['define'] !== 'undefined' && context['define'] === 'function' && context['define']['amd']) { define(name, (dependencies || []), definition); }
else { context[name] = definition.apply(context); }
})('utf8', function () {
return {
encode: function (string) {
if (typeof string !== 'string') return string;
else string = string.replace(/\r\n/g, "\n");
var output = "", i = 0, charCode;
for (i; i < string.length; i++) {
charCode = string.charCodeAt(i);
if (charCode < 128) {
output += String.fromCharCode(charCode);
} else if ((charCode > 127) && (charCode < 2048)) {
output += String.fromCharCode((charCode >> 6) | 192);
output += String.fromCharCode((charCode & 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;
},
decode: function (string) {
if (typeof string !== 'string') return string;
var output = "", i = 0, charCode = 0;
while (i < string.length) {
charCode = string.charCodeAt(i);
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));
i += 2;
} 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;
}
};
}, this);
// Base64 Module
//
// Cleaner, modularized and properly scoped base64 encoding and decoding module for strings.
//
// copyright: MIT
// author: Nijiko Yonskai, @nijikokun, nijikokun@gmail.com
(function (name, definition, context, dependencies) {
if (typeof context['module'] !== 'undefined' && context['module']['exports']) { if (dependencies && context['require']) { for (var i = 0; i < dependencies.length; i++) context[dependencies[i]] = context['require'](dependencies[i]); } context['module']['exports'] = definition.apply(context); }
else if (typeof context['define'] !== 'undefined' && context['define'] === 'function' && context['define']['amd']) { define(name, (dependencies || []), definition); }
else { context[name] = definition.apply(context); }
})('base64', function (utf8) {
var $this = this;
var $utf8 = utf8 || this.utf8;
var map = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/=";
return {
encode: function (input) {
if (typeof $utf8 === 'undefined') throw { error: "MissingMethod", message: "UTF8 Module is missing." };
if (typeof input !== 'string') return input;
else input = $utf8.encode(input);
var output = "", a, b, c, d, e, f, g, i = 0;
while (i < input.length) {
a = input.charCodeAt(i++);
b = input.charCodeAt(i++);
c = input.charCodeAt(i++);
d = a >> 2;
e = ((a & 3) << 4) | (b >> 4);
f = ((b & 15) << 2) | (c >> 6);
g = c & 63;
if (isNaN(b)) f = g = 64;
else if (isNaN(c)) g = 64;
output += map.charAt(d) + map.charAt(e) + map.charAt(f) + map.charAt(g);
}
return output;
},
decode: function (input) {
if (typeof $utf8 === 'undefined') throw { error: "MissingMethod", message: "UTF8 Module is missing." };
if (typeof input !== 'string') return input;
else input = input.replace(/[^A-Za-z0-9\+\/\=]/g, "");
var output = "", a, b, c, d, e, f, g, i = 0;
while (i < input.length) {
d = map.indexOf(input.charAt(i++));
e = map.indexOf(input.charAt(i++));
f = map.indexOf(input.charAt(i++));
g = map.indexOf(input.charAt(i++));
a = (d << 2) | (e >> 4);
b = ((e & 15) << 4) | (f >> 2);
c = ((f & 3) << 6) | g;
output += String.fromCharCode(a);
if (f != 64) output += String.fromCharCode(b);
if (g != 64) output += String.fromCharCode(c);
}
return $utf8.decode(output);
}
}
}, this, [ "utf8" ]);