Using Javascript's atob to decode base64 doesn't properly decode utf-8 strings

Question

I\'m using the Javascript window.atob() function to decode a base64-encoded string (specifically the base64-encoded content from the GitHub API). Problem is I\'

Answer 1

I would assume that one might want a solution that produces a widely useable base64 URI. Please visit data:text/plain;charset=utf-8;base64,4pi44pi54pi64pi74pi84pi+4pi/ to see a demonstration (copy the data uri, open a new tab, paste the data URI into the address bar, then press enter to go to the page). Despite the fact that this URI is base64-encoded, the browser is still able to recognize the high code points and decode them properly. The minified encoder+decoder is 1058 bytes (+Gzip→589 bytes)

!function(e){"use strict";function h(b){var a=b.charCodeAt(0);if(55296<=a&&56319>=a)if(b=b.charCodeAt(1),b===b&&56320<=b&&57343>=b){if(a=1024*(a-55296)+b-56320+65536,65535>>18,128|a>>>12&63,128|a>>>6&63,128|a&63)}else return d(239,191,189);return 127>=a?inputString:2047>=a?d(192|a>>>6,128|a&63):d(224|a>>>12,128|a>>>6&63,128|a&63)}function k(b){var a=b.charCodeAt(0)<<24,f=l(~a),c=0,e=b.length,g="";if(5>f&&e>=f){a=a<>>24+f;for(c=1;c=a?g+=d(a):1114111>=a?(a-=65536,g+=d((a>>10)+55296,(a&1023)+56320)):c=0}for(;c>>0)/n|0},d=String.fromCharCode,p=atob,q=btoa;e.btoaUTF8=function(b,a){return q((a?"\u00ef\u00bb\u00bf":"")+b.replace(/[\x80-\uD7ff\uDC00-\uFFFF]|[\uD800-\uDBFF][\uDC00-\uDFFF]?/g,h))};e.atobUTF8=function(b,a){a||"\u00ef\u00bb\u00bf"!==b.substring(0,3)||(b=b.substring(3));return p(b).replace(/[\xc0-\xff][\x80-\xbf]*/g,k)}}(""+void 0==typeof global?""+void 0==typeof self?this:self:global)

Below is the source code used to generate it.

var fromCharCode = String.fromCharCode;
var btoaUTF8 = (function(btoa, replacer){"use strict";
    return function(inputString, BOMit){
        return btoa((BOMit ? "\xEF\xBB\xBF" : "") + inputString.replace(
            /[\x80-\uD7ff\uDC00-\uFFFF]|[\uD800-\uDBFF][\uDC00-\uDFFF]?/g, replacer
        ));
    }
})(btoa, function(nonAsciiChars){"use strict";
    // make the UTF string into a binary UTF-8 encoded string
    var point = nonAsciiChars.charCodeAt(0);
    if (point >= 0xD800 && point <= 0xDBFF) {
        var nextcode = nonAsciiChars.charCodeAt(1);
        if (nextcode !== nextcode) // NaN because string is 1 code point long
            return fromCharCode(0xef/*11101111*/, 0xbf/*10111111*/, 0xbd/*10111101*/);
        // https://mathiasbynens.be/notes/javascript-encoding#surrogate-formulae
        if (nextcode >= 0xDC00 && nextcode <= 0xDFFF) {
            point = (point - 0xD800) * 0x400 + nextcode - 0xDC00 + 0x10000;
            if (point > 0xffff)
                return fromCharCode(
                    (0x1e/*0b11110*/<<3) | (point>>>18),
                    (0x2/*0b10*/<<6) | ((point>>>12)&0x3f/*0b00111111*/),
                    (0x2/*0b10*/<<6) | ((point>>>6)&0x3f/*0b00111111*/),
                    (0x2/*0b10*/<<6) | (point&0x3f/*0b00111111*/)
                );
        } else return fromCharCode(0xef, 0xbf, 0xbd);
    }
    if (point <= 0x007f) return nonAsciiChars;
    else if (point <= 0x07ff) {
        return fromCharCode((0x6<<5)|(point>>>6), (0x2<<6)|(point&0x3f));
    } else return fromCharCode(
        (0xe/*0b1110*/<<4) | (point>>>12),
        (0x2/*0b10*/<<6) | ((point>>>6)&0x3f/*0b00111111*/),
        (0x2/*0b10*/<<6) | (point&0x3f/*0b00111111*/)
    );
});

Then, to decode the base64 data, either HTTP get the data as a data URI or use the function below.

var clz32 = Math.clz32 || (function(log, LN2){"use strict";
    return function(x) {return 31 - log(x >>> 0) / LN2 | 0};
})(Math.log, Math.LN2);
var fromCharCode = String.fromCharCode;
var atobUTF8 = (function(atob, replacer){"use strict";
    return function(inputString, keepBOM){
        inputString = atob(inputString);
        if (!keepBOM && inputString.substring(0,3) === "\xEF\xBB\xBF")
            inputString = inputString.substring(3); // eradicate UTF-8 BOM
        // 0xc0 => 0b11000000; 0xff => 0b11111111; 0xc0-0xff => 0b11xxxxxx
        // 0x80 => 0b10000000; 0xbf => 0b10111111; 0x80-0xbf => 0b10xxxxxx
        return inputString.replace(/[\xc0-\xff][\x80-\xbf]*/g, replacer);
    }
})(atob, function(encoded){"use strict";
    var codePoint = encoded.charCodeAt(0) << 24;
    var leadingOnes = clz32(~codePoint);
    var endPos = 0, stringLen = encoded.length;
    var result = "";
    if (leadingOnes < 5 && stringLen >= leadingOnes) {
        codePoint = (codePoint<>>(24+leadingOnes);
        for (endPos = 1; endPos < leadingOnes; ++endPos)
            codePoint = (codePoint<<6) | (encoded.charCodeAt(endPos)&0x3f/*0b00111111*/);
        if (codePoint <= 0xFFFF) { // BMP code point
          result += fromCharCode(codePoint);
        } else if (codePoint <= 0x10FFFF) {
          // https://mathiasbynens.be/notes/javascript-encoding#surrogate-formulae
          codePoint -= 0x10000;
          result += fromCharCode(
            (codePoint >> 10) + 0xD800,  // highSurrogate
            (codePoint & 0x3ff) + 0xDC00 // lowSurrogate
          );
        } else endPos = 0; // to fill it in with INVALIDs
    }
    for (; endPos < stringLen; ++endPos) result += "\ufffd"; // replacement character
    return result;
});

The advantage of being more standard is that this encoder and this decoder are more widely applicable because they can be used as a valid URL that displays correctly. Observe.

(function(window){
    "use strict";
    var sourceEle = document.getElementById("source");
    var urlBarEle = document.getElementById("urlBar");
    var mainFrameEle = document.getElementById("mainframe");
    var gotoButton = document.getElementById("gotoButton");
    var parseInt = window.parseInt;
    var fromCodePoint = String.fromCodePoint;
    var parse = JSON.parse;
    
    function unescape(str){
        return str.replace(/\\u[\da-f]{0,4}|\\x[\da-f]{0,2}|\\u{[^}]*}|\\[bfnrtv"'\\]|\\0[0-7]{1,3}|\\\d{1,3}/g, function(match){
          try{
            if (match.startsWith("\\u{"))
              return fromCodePoint(parseInt(match.slice(2,-1),16));
            if (match.startsWith("\\u") || match.startsWith("\\x"))
              return fromCodePoint(parseInt(match.substring(2),16));
            if (match.startsWith("\\0") && match.length > 2)
              return fromCodePoint(parseInt(match.substring(2),8));
            if (/^\\\d/.test(match)) return fromCodePoint(+match.slice(1));
          }catch(e){return "\ufffd".repeat(match.length)}
          return parse('"' + match + '"');
        });
    }
    
    function whenChange(){
      try{ urlBarEle.value = "data:text/plain;charset=UTF-8;base64," + btoaUTF8(unescape(sourceEle.value), true);
      } finally{ gotoURL(); }
    }
    sourceEle.addEventListener("change",whenChange,{passive:1});
    sourceEle.addEventListener("input",whenChange,{passive:1});
    
    // IFrame Setup:
    function gotoURL(){mainFrameEle.src = urlBarEle.value}
    gotoButton.addEventListener("click", gotoURL, {passive: 1});
    function urlChanged(){urlBarEle.value = mainFrameEle.src}
    mainFrameEle.addEventListener("load", urlChanged, {passive: 1});
    urlBarEle.addEventListener("keypress", function(evt){
      if (evt.key === "enter") evt.preventDefault(), urlChanged();
    }, {passive: 1});
    
        
    var fromCharCode = String.fromCharCode;
    var btoaUTF8 = (function(btoa, replacer){
		    "use strict";
        return function(inputString, BOMit){
        	return btoa((BOMit?"\xEF\xBB\xBF":"") + inputString.replace(
        		/[\x80-\uD7ff\uDC00-\uFFFF]|[\uD800-\uDBFF][\uDC00-\uDFFF]?/g, replacer
    		));
    	}
    })(btoa, function(nonAsciiChars){
		"use strict";
    	// make the UTF string into a binary UTF-8 encoded string
    	var point = nonAsciiChars.charCodeAt(0);
    	if (point >= 0xD800 && point <= 0xDBFF) {
    		var nextcode = nonAsciiChars.charCodeAt(1);
    		if (nextcode !== nextcode) { // NaN because string is 1code point long
    			return fromCharCode(0xef/*11101111*/, 0xbf/*10111111*/, 0xbd/*10111101*/);
    		}
    		// https://mathiasbynens.be/notes/javascript-encoding#surrogate-formulae
    		if (nextcode >= 0xDC00 && nextcode <= 0xDFFF) {
    			point = (point - 0xD800) * 0x400 + nextcode - 0xDC00 + 0x10000;
    			if (point > 0xffff) {
    				return fromCharCode(
    					(0x1e/*0b11110*/<<3) | (point>>>18),
    					(0x2/*0b10*/<<6) | ((point>>>12)&0x3f/*0b00111111*/),
    					(0x2/*0b10*/<<6) | ((point>>>6)&0x3f/*0b00111111*/),
    					(0x2/*0b10*/<<6) | (point&0x3f/*0b00111111*/)
    				);
    			}
    		} else {
    			return fromCharCode(0xef, 0xbf, 0xbd);
    		}
    	}
    	if (point <= 0x007f) { return inputString; }
    	else if (point <= 0x07ff) {
    		return fromCharCode((0x6<<5)|(point>>>6), (0x2<<6)|(point&0x3f/*00111111*/));
    	} else {
    		return fromCharCode(
    			(0xe/*0b1110*/<<4) | (point>>>12),
    			(0x2/*0b10*/<<6) | ((point>>>6)&0x3f/*0b00111111*/),
    			(0x2/*0b10*/<<6) | (point&0x3f/*0b00111111*/)
    		);
    	}
    });
    setTimeout(whenChange, 0);
})(window);

img:active{opacity:0.8}

Hello \u1234 W\186\0256ld!
Enter text into the top box. Then the URL will update automatically.

In addition to being very standardized, the above code snippets are also very fast. Instead of an indirect chain of succession where the data has to be converted several times between various forms (such as in Riccardo Galli's response), the above code snippet is as direct as performantly possible. It uses only one simple fast String.prototype.replace call to process the data when encoding, and only one to decode the data when decoding. Another plus is that (especially for big strings), String.prototype.replace allows the browser to automatically handle the underlying memory management of resizing the string, leading a significant performance boost especially in evergreen browsers like Chrome and Firefox that heavily optimize String.prototype.replace. Finally, the icing on the cake is that for you latin script exclūsīvō users, strings which don't contain any code points above 0x7f are extra fast to process because the string remains unmodified by the replacement algorithm.

I have created a github repository for this solution at https://github.com/anonyco/BestBase64EncoderDecoder/