838540 - Implement ES6 @@create semantics

Status: RESOLVED INVALID

(Core :: JavaScript Engine, defect)

Description

[David Bruant]

I haven't found a good description of @@create.
It originated on es-discuss (jorendorff) [1] and it's been integrated to the latest draft [2].
@@create allow to subclass exotic objects (Array, Date, Element, etc.)
It seems that WebComponents want it too [3]

[1] https://mail.mozilla.org/pipermail/es-discuss/2012-November/026746.html
[2] http://wiki.ecmascript.org/doku.php?id=harmony:specification_drafts
[3] http://lists.w3.org/Archives/Public/public-webapps/2013JanMar/0250.html

Comment 1

[Brandon Benvie [:benvie]]

Some notes on @@create from a spec perspective:

* It's a Symbol keyed method of functions that is called during construction
* It brings prototypal inheritance into object construction since it's a regular property of the constructor, rather than an internal property
* Most builtins override the default Function.prototype.@@create with their own (for example, Array.@@create) that is responsible for adding the [[Class]] to the object 
* It splits object creation into two parts: allocation (in @@create) and initialization (the constructor)
* It is important for classes because it allows `super()` in constructors inheriting from builtins to actually work and it allows instances of subclassed builtins to have all the necessary internal state to be a legit instance of the builtin type. For example:

    class MyArray extends Array {
      constructor(...args){
        this.isMyArray = true;
        super(...args);
      }
    }

    let arr = new MyArray(5, 10, 20);
    Array.isArray(arr); // true
    arr.length; // 3

Comment 2

[Domenic Denicola]

A few points:

- To some extent, this is blocking DOM work on subclassing arrays for `Elements` [1].

- The easiest way to summarize @@create is `new X(...args)` <-> `X.call(X[@@create](), ...args)`.

- One particular user-facing consequence for current code, without ES6 class syntax, is:

  function X() { }
  X.__proto__ = Array;

  var x = new X();
  Array.isArray(x); // should be true

  because `X[@@create]()` calls `Array[@@create]()` which allocates an exotic array object. (In addition to changing `Array.isArray`'s answer, exotic array objects have the special DefineOwnProperty behavior that messes with `length`.)


[1]: https://gist.github.com/domenic/5864658

Comment 4

[Till Schneidereit [:till]]

To make this fast, we'll need to make isConstructing fast.

Comment 5

[Florian Bender]

I don't know if this is too much work, but would be awesome to have this in Moz33 as well since Symbol landed in the same version!

Comment 6

[Till Schneidereit [:till]]

Not only is this a very different feature from symbols (it relies on them, but nothing more), it's also a pretty big task, so it won't happen in this release cycle.

Additionally, it's not even clear what the final spec will say about this; there are at least two competing proposals for roughly what @@create is intended to be used for:

https://mail.mozilla.org/pipermail/es-discuss/2014-June/037810.html
https://mail.mozilla.org/pipermail/es-discuss/2014-June/038120.html

Comment 7

[Florian Bender]

Yeah, I thought so, but did not know the spec hasn't even stabilized yet. Please ignore my comment.

Comment 8

[Eric Faust [:efaust]]

This spec design has been abandoned in favor of super() and new.target

Comment 9

[Patrick Cloke [:clokep]]

Eric, is there a bug about implementing that or any other information you can point us to?

Comment 10

[Eric Faust [:efaust]]

Sure! The current work on this front is all in the dependency tree of bug 1141863. SuperCall is a ton of work, in the new spec, so it's proceeding slowly along.