Should probably spec a hasInstance for DOM interface objects

[bzbarsky]

Right now we have a [[HasInstance]] thing at http://heycam.github.io/webidl/#es-interface-hasinstance but that's gone in ES6.

Proposed prose with similar semantics is something like:

1. Let F be the this value.
2. Let V be the first argument we were passed.
3. If F is not an object, return false.
4. If F is not an interface object, return false (or delegate to OrdinaryHasInstance?)
5. If V is not an object, return false.
6. If V is an object that implements the interface for which F is the interface object (defined in terms of brands, whatever), return true.
7. Return OrdinaryHasInstance(F, V).

@domenic, thoughts?

An open question is whether Node[Symbol.hasInstance] == Element[Symbol.hasInstance] should test true or false. We could go either way, given the above prose, since they would implement the same algorithm...

[bzbarsky]

If F is not an interface object, return false (or delegate to OrdinaryHasInstance?)

Fwiw, the behavior I'm going to do in Firefox for now is to delegate.

[domenic]

Can you clarify the intent of the existing [[HasInstance]]? I'm unable to understand in what ways it is different from OrdinaryHasInstance.

Is the test "is a platform object that implements the interface for which O is the interface prototype object" meant to work cross-realm? It seems like youre revised version might be or might not be, depending on whether brands are cross-realm? If that's the desired semantics, how many browsers implement that?

Web IDL's current algorithm performs an observable Get of the "prototype" property, as does OrdinaryHasInstance. Both also do observable proto-walks. Your algorithm omits these. Is changing that a good idea?

[bzbarsky]

Is the test "is a platform object that implements the interface for which O is the interface prototype object" meant to work cross-realm?

Yes. This is in fact the entire intent of the existing [[HasInstance]].

It seems like youre revised version might be or might not be, depending on whether brands are cross-realm?

They better be, because DOM functions can be .call-ed cross-realm, yes?

If that's the desired semantics, how many browsers implement that?

I believe Firefox is the only one that does. That said, this is the behavior people agreed on when this was discussed on es-discuss several years ago... I can't help it if other browsers didn't make implementing Web IDL a very high priority. ;)

I agree that this does cause problems in terms of potential compat issues for whoever switches behavior.

Web IDL's current algorithm performs an observable Get of the "prototype" property

It's not observable: the object the Get is done on in the current algorithm is guaranteed to be a Web IDL interface object, which has such a property defined as a readonly non-configurable value property. At least modulo whatever happens with namespaces.

Both also do observable proto-walks. Your algorithm omits these.

My algorithm does the proto walk. It's right there in step 7.

[annevk]

Are https://lists.w3.org/Archives/Public/public-script-coord/2013JanMar/thread.html#msg1 and https://esdiscuss.org/topic/solving-the-how-do-i-tell-whether-i-have-an-html-element-or-image-element-or-whatever-problem those discussions? I guess nobody really objected.

[annevk]

(I'm still wary of this as it complicates moving objects between IDL and ECMAScript a bit. As happened with ArrayBuffer and friends, Promise in Gecko, and might happen for other objects in the future.)

[bzbarsky]

I don't recall where all this was discussed offhand. It's been a long time.

I agree that the insistence of ES on having no way to tell whether an object is branded in particular ways, which is partly a legacy of it pretending there is only one global and partly due to people philosophically objecting to branding afaict is a pain for everyone involved. It's a pain that was solved for Array via isArray and explicitly punted on for all sorts of other objects for years now....

[bzbarsky]

Oh, and what it means in practice is that people start relying on brittle Object.prototype.toString.call hacks, but ES6 is breaking those too, by making that sort of thing hookable...

[travisleithead]

Just happened to be reviewing some WebIDL tests and found a strange inconsistency. Not sure what behavior is desired (in case 3)...

In Firefox 46:

// "w" is a cross-realm windowproxy
test(function() {
  assert_true(document.implementation instanceof w.DOMImplementation);
}, "a DOMImplementation object is considered to be instanceof anotherWindow.DOMImplementation");
// This passes

This also passes (because of OrdinaryHasInstance?):

test(function() {
  var object = Object.create(document.implementation);
  assert_true(object instanceof DOMImplementation);
}, "an object whose [[Prototype]] is DOMImplementation.prototype is considered to be instanceof window.DOMImplementation");

However--and this is the oddity--the following fails:

test(function() {
  var object = Object.create(document.implementation);
  assert_true(object instanceof w.DOMImplementation);
}, "an object whose [[Prototype]] is DOMImplementation.prototype is considered to be instanceof anotherWindow.DOMImplementation");

This seems like it should also pass, because case 2 passed...We'll need an algorithm tweak to handle this.

[bzbarsky]

@travisleithead the Firefox behavior is basically the one described in #129 (comment). Your case 1 passes because of step 6, and case 2 passes because of step 7. But case 3 doesn't match in step 6 and as far as OrdinaryHasInstance is concerned that object has nothing to do with w.DOMImplementation.

I don't see any need for case 3 to pass in practice, honestly.... I think case 2 passing may be needed for web compat. Or at least I would need some convincing that it's not needed.

[esprehn]

I'd really rather we didn't do this, if we want branding checks we can add some kind of separate system to do the slow thing (Platform.isInstance or something), but I don't want to mess with instanceof at the platform level.

See also my reply here:
https://groups.google.com/a/chromium.org/d/msg/blink-reviews-bindings/6qzaSWen21c/OU792ZXFAwAJ