The timeless repository

JSON: The JavaScript subset that isn't

Written by Magnus Holm.

From Wikipedia’s article on JSON

JSON was based on a subset of the JavaScript scripting language.

All JSON-formatted text is also syntactically legal JavaScript code.

JSON is a subset of JavaScript.

All these years we’ve heard it over and over again: “JSON is a JavaScript subset”. Guess what? They’re wrong. Wrong, wrong, wrong. You see, the devil’s in the details, and there’s no way to avoid it: Not all JSON-formatted text is legal JavaScript code:

{"JSON":"ro
cks!"}

Copy the exact code above and paste it into Firebug or the Web Inspector within a pair of parentheses (required to avoid an ambiguity in JavaScript’s syntax):

JSON sucks

Wait, what? SyntaxError: Unexpected token ILLEGAL? That doesn’t make any sense! It’s just a regular object literal, how can that be a SyntaxError?

Try the same code in a proper JSON parser. No problems at all.

Of course it’s not just a regular object literal. There’s a sneaky little Unicode character in there too: Right between “ro” and “cks!” there’s a tiny U+2028. Your browser probably doesn’t display it because it’s whitespace: LINE SEPARATOR, but it’s still there. If you replace the character with a U+2029 (PARAGRAPH SEPARATOR) you would have the exact same issue.

JSON + U+2028 = ☺

According to the JSON specification, you can safely use this character in any string. It’s not a quote, not a backslash, and not a control character. It’s just a weird Unicode whitespace character:

JSON string specification

JavaScript + U+2028 = ☹

ECMA-262 (the standard that JavaScript is based on) on the other hand defines strings a little differently: According to 7.8.4 String Literals, a string can contain anything as long as it’s not a quote, a backslash or a line terminator:

DoubleStringCharacter ::
    SourceCharacter but not double-quote " or backslash \ or LineTerminator 
    \ EscapeSequence

SingleStringCharacter ::
    SourceCharacter but not single-quote ' or backslash \ or LineTerminator 
    \ EscapeSequence

And what is a line terminator? Let’s have a look at 7.3 Line Terminators:

The following characters are consider to be line terminators:

Ouch.

No string in JavaScript can contain a literal U+2028 or a U+2029.

So what?

Because of these two invisible Unicode characters, JSON is not a subset of JavaScript. Close, but no cigar.

In most applications, you won’t notice this issue. First of all, the line separator and the paragraph separator aren’t exactly widely used. Secondly, any proper JSON parser will have no problems with parsing it.

However, when you’re dealing with JSONP there’s no way around: You’re forced to use the JavaScript parser in the browser. And if you’re sending data that other have entered, a tiny U+2028 or U+2029 might sneak in and break your pretty cross-domain API.

The solution

Luckily, the solution is simple: If we look at the JSON specification we see that the only place where a U+2028 or U+2029 can occur is in a string. Therefore we can simply replace every U+2028 with \u2028 (the escape sequence) and U+2029 with \u2029 whenever we need to send out some JSONP.

It’s already been fixed in Rack::JSONP and I encourage all frameworks or libraries that send out JSONP to do the same. It’s a one-line patch in most languages and the result is still 100% valid JSON.