CubebDeviceEnumerator [0] is the new way to enumerate input devices. An important improvement is that it does not do unnecessary enumerations, it saves a local device list until it has been notified that the list has changed. It would be great to expand CubebDeviceEnumerator to enumerate output devices too. We already hold an instance of the enumerator in MediaEngineWebRTC [1] so we could use it to get output devices too. However, it has some limitation currently. The enumerator makes use of the static cubeb context [2], which means we can have only one instance of the enumerator per process. This is because every new instantiation will replace the device collection callback pointer, inside cubeb, and the logic of reusing the local device list will work only for the last one. A second issue, with the output devices in place, is that the saved list will be invalidated for either input or output device collection changes. This will not be very accurate, it's not a big problem though. With this bug I would like to create a logic that the CubebDeviceEnumerator will own its own cubeb context instance instead of reusing the static one from CubebUtils. We could even have two cubeb context instances, one for input and one for output, to avoid the problem above. Cubeb context is cheap enough and I do not see any problem using extra instances. However, it would be important to respect the "media.cubeb.force_null_context" pref for every new instance, in order to have accurate results on testing. [0] https://searchfox.org/mozilla-central/rev/0f4d50ff5211e8dd85e119ef683d04b63062ed90/dom/media/webrtc/MediaEngineWebRTC.h#127 [1] https://searchfox.org/mozilla-central/rev/0f4d50ff5211e8dd85e119ef683d04b63062ed90/dom/media/webrtc/MediaEngineWebRTC.h#482 [2] https://searchfox.org/mozilla-central/rev/0f4d50ff5211e8dd85e119ef683d04b63062ed90/dom/media/CubebUtils.cpp#125 [3] https://searchfox.org/mozilla-central/rev/0f4d50ff5211e8dd85e119ef683d04b63062ed90/dom/media/CubebUtils.cpp#45

It would probably be better to have a more central CubebDeviceEnumerator: - Putting it in its own file, making it a singleton. - Allow multiple consumer to subscribe to device list notification callback in Gecko. - Using it from MediaEngineWebRTC and also other locations. - Possibly adding a feature where we invalidate either the input or output device list or both.

If we make it a singleton we need to make it thread safe. We can avoid that if we keep it as a "simple" object which creates internally it's own cubeb instances. Do you prefer that?

It's already thread safe isn't it? There is a lock for all methods and member access.

Yes it is, my point is that we could avoid that.

WIP patch

WIP patch

WIP patch. Affected by https://github.com/djg/audioipc-2/issues/50

WIP patch

In Linux _and_ when cubeb remote is on, we are affected by https://github.com/djg/audioipc-2/issues/50. Bug 1500468 creates a temporary solution and allow enumerator to enumerate every time. When audioipc issue is solved enumerator will be optimized to enumerate only after a device collection change.

Quering the stored devices list for an ID was not always correct. The stored devices list is expected to be empty before an enumeration takes place or after a collection change notification. In those cases, the method will fail to find an existing ID. The method has been updated to accept the device list as an argument. The device list can be created from the input and/or output device enumeration. Currently, the method is not being used.

Pushed by achronopoulos@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/ade887b7ba31 Move CubebDeviceEnumerator to its own source and header files. r=padenot https://hg.mozilla.org/integration/autoland/rev/18925888f5d7 Make CubebDeviceEnumerator singleton. r=padenot https://hg.mozilla.org/integration/autoland/rev/20a4fa9f6214 Implement enumeration of output devices. r=padenot https://hg.mozilla.org/integration/autoland/rev/662bbd0961e5 Create a unit test to verify all corner cases of the helper method. r=padenot

Backed out for causing windows build bustages on dom/media/CubebUtils.cpp Push with failures: https://treeherder.mozilla.org/#/jobs?repo=autoland&resultStatus=busted&fromchange=662bbd0961e58ad88e840c079f0a01fedb1ee035&tochange=a8b6a0a000c049313ed103512bf975a3a81693fa&selectedJob=210375013 Failure log: https://treeherder.mozilla.org/logviewer.html#?job_id=210375039&repo=autoland&lineNumber=41807 Backout link: https://hg.mozilla.org/integration/autoland/rev/a8b6a0a000c049313ed103512bf975a3a81693fa

The problem was that Windows MinGW 32/64 builds have WebRTc disabled due to Bug 1393901. A green try: https://treeherder.mozilla.org/#/jobs?repo=try&revision=3980b0b16e0d81e61beef708dae7e618335ff91c

Pushed by achronopoulos@mozilla.com: https://hg.mozilla.org/integration/autoland/rev/22d08a6ea7cc Move CubebDeviceEnumerator to its own source and header files. r=padenot https://hg.mozilla.org/integration/autoland/rev/8a075dabb0c5 Make CubebDeviceEnumerator singleton. r=padenot https://hg.mozilla.org/integration/autoland/rev/79ffde86ea1f Implement enumeration of output devices. r=padenot https://hg.mozilla.org/integration/autoland/rev/15bffbe82ec6 Create a unit test to verify all corner cases of the helper method. r=padenot

https://hg.mozilla.org/mozilla-central/rev/22d08a6ea7cc https://hg.mozilla.org/mozilla-central/rev/8a075dabb0c5 https://hg.mozilla.org/mozilla-central/rev/79ffde86ea1f https://hg.mozilla.org/mozilla-central/rev/15bffbe82ec6

Clearing NI