Fat arrow leaks generated _this reference to other closures

[joshma]

Using a fat arrow generates a _this variable for use in the current closure, but due to how the JS engines work it gets leaked to all closures in the function scope:

@foo = 42

printFoo = => # This creates a "_this" variable
  console.log @foo

bar = ->
  debugger # When Chrome's debugger stops here, the "_this" variable is visible under the "Closure Scope" variables
  # console.log (eval '_this').foo # If this is executed, it'll frigging work!

bar()

(Thanks to @epidemian for the example, see end of #3052)

In the example above, printFoo uses a fat arrow to create a closure on _this, but bar accidentally gets a reference through the Closure Scope as well. Even if all references to printFoo are discarded, a reference to bar will indirectly hold a reference to _this and prevent it from being garbage collected.

Given that the use of a closure is an implementation detail and not part of the fat-arrow abstraction, this seems like an undesirable side-effect. As I suggested in #3052, an IIFE approach should provide the same abstraction without this issue:

bar = => console.log(@)
// ...translates to
bar = (function(_this) {
  return function() {
    console.log(_this);
  }
})(this)

(Both @epidemian and @michaelficarra seem to somewhat agree.)

[michaelficarra]

I forgot about arguments shadowing in #3052. The correct compilation would be

(function(){
  var $this = this;
  return function ...
}.apply(this, arguments))

as I've done in michaelficarra/CoffeeScriptRedux@2eca5c2

[joshma]

I'm relatively new to these neck of the woods, but for my own enlightenment why do arguments need to be passed in? It seems like it won't reach the inner-most function anyways, since that'll have its own arguments from when it's called.

[michaelficarra]

Ah crap, you're right. I'm so used to having to watch out for arguments shadowing when IIFE-wrapping things, I forgot that the bound function would shadow arguments anyway. This is why I didn't merge to master yet. Fixing.

[jashkenas]

It's really a shame that we have to add an extra closure wrapper (like we used to), for this ... but there's no other possible way around it, is there?

[jashkenas]

Let me know if that looks alright to y'all, or if it can be simplified...

[joshma]

👍 Thanks for getting to this! Looks good from my (very basic) understanding of the coffee source.