I agree on principle, but that ship has sailed.

How would we distinguish between an interface that shouldn't be exposed because [exposed] is missing from an interface that should be because it was spec'ed before we made this change?

We'd distinguish by going and editing every single spec accordingly...

(In reply to Boris Zbarsky from comment #2)
> We'd distinguish by going and editing every single spec accordingly...

Sure. Seems like a lot of work for something that appears to be essentially cosmetics, though. Or am I missing the broader picture?

Yeah, the problem is that specifications don't consider workers, or service workers, etc. Not having a default would require specifications to consider what they're doing, which I think is what we want. (There's also been some confusion from developers in the past as to how interfaces end up being exposed. There being an explicit syntax link would help with that too.)

OK, that makes sense. We need a strategy to make every spec fix their this though.

Making an interface without [Exposed] into a bikeshed and respec error would catch a good fraction of maintained specs.

And once those get updated, we can change browser IDL parsers and file bugs on any remaining specs that this catches....

Is there a good doc I can point to somewhere in Bikeshed's error message letting people know how to fix things correctly?  (I'm not willing to add errors where the solution is "ask someone in #whatwg what to do".)

(In reply to Tab Atkins Jr. from comment #8)
> Is there a good doc I can point to somewhere in Bikeshed's error message
> letting people know how to fix things correctly?  (I'm not willing to add
> errors where the solution is "ask someone in #whatwg what to do".)

I'll add documentation in WebIDL itself once we get to this and will coordinate with you.

I've filed https://github.com/w3ctag/design-principles/issues/35 to hold the design advice Tab's looking for.

See also https://github.com/heycam/webidl/issues/365.

See comment above.