Compression of CoffeeScript output

[mbarkhau]

I'm wondering if UglifyJS2 might not be able to do a better job at compressing common output of CoffeeScript. Here an example:

test = (
  (s) -> w.toLowerCase() for w in s.trim().split(" ")
)("test")

Which compiles to this:

var test;

test = (function(s) {
  var w, _i, _len, _ref, _results;
  _ref = s.trim().split(" ");
  _results = [];
  for (_i = 0, _len = _ref.length; _i < _len; _i++) {
    w = _ref[_i];
    _results.push(w.toLowerCase());
  }
  return _results;
})("test");

Which is uglified to this:

var test;

test = function(t) {
    var e, r, s, n, o;
    for (n = t.trim().split(" "), o = [], r = 0, s = n.length; s > r; r++) e = n[r], 
    o.push(e.toLowerCase());
    return o;
}("test");

Shouldn't it also be possible to produce the following?

var test = function(t) {
    var n = t.trim().split(" "), o = [], r = 0, s = n.length;
    for (; s > r; r++) o.push(n[r].toLowerCase());
    return o;
}("test");

I removed the unnecessary temporary variable e from the last example, but mostly I'm referring to moving assignments up into the corresponding var declaration. I'm new to UglifyJS, so forgive me if I'm just missing a compiler flag.

[mbarkhau]

Is this out of scope for UglifyJS?

[mishoo]

Nope, not out of scope but very hard to get right in the general case. A pull request is welcome. ;-)

[mbarkhau]

Ok cool. I'll see what I can do.
Do you see any edge cases where this transformation might break code?

[mishoo]

Nope, absolutely not if the transformation is done properly. ;-) But it's hard. Good luck!

[ghost]

I think that this seems relevant to #320.

[dead-claudia]

I would like to note that this is generally the case with the output of many compilers targeting JS. LiveScript, Coco, and to an extent, Emscripten all do this. Such a pull request will be very useful. (Minified JS and minified compiled CoffeeScript should be at least close in size, particularly in these situations.)