+1

Even Android?

@timmywil, please read my report again, I covered Android.

Oh, sorry. I don't even know which Android browsers we want to support.

I guess >= 2.3 currently, unfortunately. :( Anyway, I doubt anyone writes sites only for Android <4.1 so relying on the ability to redefine undefined wouldn't work.

IE seems to have mutable undefined, after all. I used kangax tables for desktop browsers:

http://kangax.github.io/es5-compat-table/

but they seem to be wrong about IE9.

Anyway, the spec must be more complicated than what I anticipaded. The following code provides the same results in Chrome, Firefox, Opera & IE10 (IE9 results in [0, 1, ..., 7]):

(function ( undefined ) {
	console.log( '0: ' + undefined );
})( 0 ); // => 0

(function ( undefined ) {
	'use strict';
	console.log( '1: ' + undefined );
})( 1 ); // => 1

(function () {
	undefined = 2;
	console.log( '2: ' + undefined );
})(); // => 2

(function () {
	'use strict';
	try {
		undefined = 3;
		console.log( '3: ' + undefined );
	} catch (e) {
		console.log( '3: ' + e );
	}
})(); // => TypeError

(function () {
	var undefined = 4;
	console.log( '4: ' + undefined );
})(); // => 4

(function () {
	'use strict';
	var undefined = 5;
	console.log( '5: ' + undefined );
})(); // => 5

var undefined = 6;
(function () {
	console.log( '6: ' + undefined );
})(); // => undefined

var undefined = 7;
(function () {
	'use strict';
	console.log( '7: ' + undefined );
})(); // => undefined

On jsFiddle every browser behaves like IE9, though...

http://jsfiddle.net/m_gol/3KrZW/2/

On IRC, dcherman suggested just telling uglify to not compress undefined, which he says saves 40-something bytes. Not quite as much saved, but no downside.

Yeah... I didn't suspect ES5's meaning of undefined is that you can't do undefined = 'a' but you can do var undefined = 'a'. Sooo helpful...

Pull request: https://github.com/jquery/jquery/pull/1239

Tell uglifyjs to not mangle undefined; saves 44 bytes. Fixes #13759. Close gh-1239.

Changeset: 8576e24bfe3fce886bcb63ebc24ee16b3fa1a6a1

BTW, orkel was right, this change actually increases the gzipped 1.x-master size so it makes sense only on master.