https://bugzilla.mozilla.org/show_bug.cgi?id=1070216#c32 <jesup> so why is it on MainThread? <jesup> MediaManager.cpp:1441 <jesup> So, that lambda then DispatchToMainThread()s to resolve the pledges... let's see if that's it <jesup> Yeah, it's in the Pledge stuff <jesup> Aha. In processing the Fitness distance code, it does size_t num = NumCapabilities(); - but it's now back on mainthread <jesup> It needs to cache the capabilities while still on the MediaManager thread I think, and just use them in the thing we send back to MainThread <jesup> Ok, we need a bug on this, assigned to jib. And we should add thread-asserts to a few more MediaEngine APIs (at least !Is_MainThread())

Bug 1210852 - do SelectSettings of device capabilities on media thread.

https://reviewboard.mozilla.org/r/21215/#review19073 Should work, but not done. ::: dom/media/MediaManager.cpp:1147 (Diff revision 1) > + MediaManager::PostTask(FROM_HERE, NewTaskFrom([id, aConstraints, > + aSources]() mutable { Note to self: need to add reference counting or something here, since the algorithm modifies the passed-in array. The caller keeps it alive normally, but if things are torn down at the wrong time...

(In reply to Jan-Ivar Bruaroey [:jib] from comment #2) > https://reviewboard.mozilla.org/r/21215/#review19073 > > Should work, but not done. Just tested, the deadlock is gone! Thanks :-) It seems like I get an audio track when I request video-only gUM() though, but I guess you're on top of that.

Yeah I get that too. I messed something up clearly. Will take a look later today.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Bug 1210852 - do SelectSettings of device capabilities on media thread.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Bug 1210852 - do SelectSettings of device capabilities on media thread.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Bug 1210852 - do SelectSettings of device capabilities on media thread.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. https://reviewboard.mozilla.org/r/21215/#review19259 ::: dom/media/systemservices/MediaUtils.h:325 (Diff revision 4) > +/* media::Refcountable - Add threadsafe ref-counting to something that hasn't. hasn't -> isn't Consider proposing this for mfbt somewhere

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Bug 1210852 - do SelectSettings of device capabilities on media thread.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Bug 1210852 - do SelectSettings of device capabilities on media thread.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Thanks! I've incorporated your feedback, and I also did some minor changes to the tests for whether MediaManager is still around in the callbacks. If you could take another look that would be great.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. Bug 1210852 - do SelectSettings of device capabilities on media thread.

Comment on attachment 8669486 [details] MozReview Request: Bug 1210852 - do SelectSettings of device capabilities on media thread. https://reviewboard.mozilla.org/r/21215/#review19289

[Tracking Requested - why for this release]: It deadlocks. Bug 1209033 comment 13 demonstrates this in the wild, so we should probably uplift this if we can. This is likely a regression from 39.

A patch rebased for beta is available in Bug 1209033 comment 20.

Beta uplift request made in Bug 1209033.

Not tracking for 42 as we released and I don't think it is worst shipping in a dot release.

Not tracking this for 43 either, since the fix for it is uplifting in bug 1209033.