Should mixins be able to add overloads?

[bzbarsky]

Consider this IDL:

interface A {
  void foo();
};

[NoInterfaceObject]
interface B {
  void foo(long arg);
};

A implements B;

Should this be allowed? I think the current spec more or less allows it, not least because it's pretty vague about when overload sets are actually determined. But it's not entirely obvious to me whether this is desirable, because this makes the mixin change the behavior of existing methods on the interface, which is something that should really go in the base spec, not a mixin, I would think.

[domenic]

Do we know about existing specs doing this? It seems very likely they might, but if nobody does, then yeah, disallowing would be nice...

[bzbarsky]

There aren't any that I'm aware of (in that the Gecko IDL parser explicitly errors out in this situation, and that's not being a problem for us so far), but WebGL is trying to add this sort of thing.

[annevk]

Hmm, did WebGL not get the memo that overloads in general are a poor API?

[bzbarsky]

Not really, no. They're basically trying to expose the GL APIs, and those are full of overloads or something.

[bzbarsky]

However, some of the webgl2 APIs are moving away from overloads more toward unions and optional args. They just kinda coexist with webgl1 apis that use the old setup.