Stop using extended attributes for constructors

[Ms2ger]

There has been some discussion about this in #485; I'm filing this to consider changing [Constructor] in particular, since nobody seemed to particularly object to it.

I'd like to propose the following syntax:

interface Foo {
  constructor(...);
};

Without further changes, I believe the only WebIDL-defined extended attributes that would potentially start applying are [Exposed], [SecureContext] and [Unforgeable]. It seems fine to allow [Exposed] and [SecureContext], but I would disallow [Unforgeable].

We should probably not allow constructors to be specified within mixins.

Please comment asap if you have thoughts on this.

CC @yuki3 @littledan @bzbarsky @tabatkins @domenic @annevk

[TimothyGu]

It sounds a bit weird to me to see an interface whose constructors are exposed in certain situations but not others, even if the interface itself is exposed. If we don’t have a concrete use case for them yet then I’d vote for just disallowing any extended attributes right now, not that it’s too difficult to change them in the future.

[littledan]

We discussed multiple browser-specific use cases for extended attributes on specific constructor overloads in #485. Would it be reasonable to define what it means to include extended attributes on constructor overloads in WebIDL, even if we don't have a web spec that would use that capability at the moment?

cc @kmiller68

[annevk]

It seems okay to me. There might even be a use case today in clarifying the exact semantics of https://xhr.spec.whatwg.org/#interface-formdata. Namely that it doesn't have an optional argument in worker environments. (I think this is what the current prose means as well, but perhaps we should move to making such things more explicit.)

[bzbarsky]

[Exposed] and [SecureContext] are both meant to affect visibility/existence, and do so consistently at the moment. What are the proposed semantics for them for constructors? And if they're different from the normal semantics, do we actually want that?

[annevk]

I meant that I could see rewriting that example to have two (overloaded in a sense) constructors, one exposed in Window and one in Worker. So equivalent semantics to what they mean today.

[bzbarsky]

What would it mean to have them "not be exposed"? The constructor is not exposed per se; it just affects the behavior of the [[Construct]] of the interface object.

[domenic]

I think for now disallowing all of the spec-defined extended attributes is the best approach. But, allow extended attributes in the grammar, and that way UAs can use their internal extended attributes as desired.

[yuki3]

+1 for Timothy and Domenic's approach.

[tabatkins]

I'd be very happy to see constructors become normal methods. While I lean slightly towards calling them by the class's name, as that's how you actually call them, I'm sympathetic to the "just do it like Javascript" naming method (tho that was, I presume, at least partially motivated by the existence of anonymous classes, which WebIDL doesn't have).

[yuki3]

@Ms2ger just as a friendly ping from curiosity. I'm just wondering if there is any update. Is something blocking? Can we help something?

[Ms2ger]

I got a private suggestion that it might be better to hold off on this change in favour of doing #485 all at once; I haven't fully made up my mind yet if I agree with that.

[yuki3]

Thanks for the update. I understand the situation. :)

[domenic]

Hmm. I tend to think establishing a workflow for multiple incremental changes will be better.

[littledan]

I was the source of this suggestion. I just was thinking about how we could get everything updated and thought several incremental changes could be costly, and that it would make sense to batch them. I haven't posted about this on #485 to make this suggestion just because I have not taken the time dedicated to make a more clear and detailed proposal.

If others want to pursue a workflow for multiple independent changes, please go ahead.

[yuki3]

I'd personally like to have this change first compared to other changes because this proposal changes not only style but also functionality (what we can represent in Web IDL), and also because people already made agreement at some level.

I'm afraid that other purely-style changes would involve more discussions (it could be hard to make agreement on a style) and would take a long time.