Force creation of a function in a loop? (unwanted optimiziation)

[determin1st]

Hi there, I was assembling constructor functions in a map with similar code:

# create constructors map
map = new !->
  for a of type
    # create constructor
    @[a] = b = !->
      @prop1 = false
      @prop2 = false
    # set its prototype (methods)
    b.prototype = type[a]

the output:

map = new function(){
  var a, b;
  for (a in type) {
    this[a] = b = fn$;
    b.prototype = type[a];
  }
  function fn$(){
    this.prop1 = false;
    this.prop2 = false;
  }
};

Is there a way to force creation (not definition) logic?

[rhendric]

Huh. That looks like a bug to me, or at the very least a completely undocumented gotcha.

I can think of two workarounds, at least: change for a to for let a, or change b = !-> to b = do -> !->. Both add some overhead at run time, which is unfortunate, but it might be mostly optimized out by a JS engine.

[determin1st]

ye, okay, i've created some factory function outside of the loop for now.. the construction overhead doesn't matter much in my case (it's a singleton object), it just doesn't look "good enough".

[determin1st]

well, i've got some answer from the JIT-man:

"Different constructor functions will have different hidden classes"

So my intent to make some set of "look-alike" constructors with the same hidden class won't work with this pattern. Maybe, it's not a bug, i see no reason to write such code constructs now..

[ceremcem]

Looks like very same issue: https://stackoverflow.com/a/41587540/1952991

[determin1st]

@ceremcem yes yes, it's the same.. almost.

[ceremcem]

Any update on this issue?

[rhendric]

Nobody's working on LiveScript in secret. What you see is what you get.

[ceremcem]

Would you accept a patch that entirely removes that optimization?

[rhendric]

Short answer, I don't know if that's a good idea. It's been a few years since I've dug into LiveScript. If you've done the research and can make a logical case that the optimization doesn't help, I'd look at it. Given LiveScript's status as a legacy language I don't want to remove anything that some project might be relying on.