It would be nice to not have to think about explicitly adding usage telemetry.

qdot, can you point me at where this would happen? I'll take a stab at a patch.

Whoops, didn't mean to un-ni? myself. I'll take a look at this in a bit and let you know. While I'm here though: Are we wanting to just default to use-counting ALL web-exposed UIs, or do we want another decorator for webidl interfaces that explicitly marks them as use-counted?

(In reply to Kyle Machulis [:qdot] [:kmachulis] (if a patch has no decent commit message, automatic r-) from comment #2) > > While I'm here though: Are we wanting to just default to use-counting ALL > web-exposed UIs, or do we want another decorator for webidl interfaces that > explicitly marks them as use-counted? I think it depends on what sort of overhead we'd see here. I honestly don't know and just think it'd be cool to be able to have this data without thinking about it (default on). (Definitely P3 material here)

+1 to not knowing enough yet. I think we want to get to a place where we're not shipping things to the web that we're don't know whether anyone is actually using it. WebIDL auto-generation makes it zerowork for platform developers for those APIs. Doing this on Nightly seems like a low risk way to figure out how feasible this approach is.

One thing that would be cool is knowing how many unique origins use a feature. Peter pointed out this will of course be much more expensive ... maybe we could do that opt-in?

Alternatively we could have an explicit blacklist of things we need to be fast (Peter gives the example of Node).

I think I can clear my ni on this until more specific info is needed?