I've removed the stdint.h generation from libcubeb upstream since it was only needed for old MSVC (pre-2012) and updated the Gecko build to suit. https://treeherder.mozilla.org/#/jobs?repo=try&revision=00adfdb4b66a This also pulls in Paul's resampler rework. It might be good to get that landed sooner rather than later to expose it to more testing (if Paul's okay with that).

Note for when I get back to this on Monday: Breakage on Win32: update.sh needs to include cubeb_resampler_internal.h Breakage on Android: OpenSL backend needs updating for resampler API changes

The use of unique_ptr that I suggested breaks on Android... https://groups.google.com/forum/#!topic/mozilla.dev.platform/ph-63xS3zwA ...because using 5+ year old C++ features is too much to ask. With polyfill: https://treeherder.mozilla.org/#/jobs?repo=try&revision=cb761062f8cb

Forgot to define CUBEB_GECKO_BUILD in the src/moz.build file: https://treeherder.mozilla.org/#/jobs?repo=try&revision=67be7ba3afe4

These warnings in the resampler need investigating: 01:32:56 INFO - /home/worker/workspace/build/src/media/libcubeb/src/cubeb_resampler.cpp:237:11: warning: comparison between signed and unsigned integer expressions [-Wsign-compare] 01:32:57 INFO - if (got != output_frames_before_processing) { 01:32:57 INFO - /home/worker/workspace/build/src/media/libcubeb/src/cubeb_resampler.cpp:174:8: warning: variable 'got' set but not used [-Wunused-but-set-variable] 01:32:57 INFO - long got = 0; 1:32:57 INFO - /home/worker/workspace/build/src/media/libcubeb/src/cubeb_resampler.cpp:150:11: warning: comparison between signed and unsigned integer expressions [-Wsign-compare] 01:32:57 INFO - if (got != output_frames_before_processing) { 01:32:57 INFO - ^ 01:32:57 INFO - /home/worker/workspace/build/src/media/libcubeb/src/cubeb_resampler.cpp:135:7: warning: variable 'out_unprocessed' set but not used [-Wunused-but-set-variable] 01:32:57 INFO - T * out_unprocessed = nullptr;

Also: 17:59:18 INFO - Thread 89 (crashed) 17:59:18 INFO - 0 xul.dll!noop_resampler::fill(void *,long *,void *,long) [cubeb_resampler.cpp:67be7ba3afe4 : 65 + 0x6] noop_resampler::fill can't handle a null input_frames_count.

Last try run has a bunch of other crashes and test failures that need investigating too. Not planning to invest any more time on this right now.

Thanks, I'll finish this.

Includes all the full duplex work, including the resampler.

It is uploaded here to make it available to padenot. It contains a fix to select the devid instead of device_id during the device selection in webrtc engine.

Changes to media/libcubeb/include/moz.build

updated

This imports the wasapi backend, the resampler code, and various other commits. The OSX code is split to not completely destroy the blame info.

This is the OSX part.

Looks to me like the remaining issues (assuming they're basically just a cubeb import in the main wasapi patch) are all rubber-stamps/build-nits.

Comment on attachment 8734807 [details] [diff] [review] Update cubeb's udpate.sh script to account for new files. r= Review of attachment 8734807 [details] [diff] [review]: ----------------------------------------------------------------- ::: media/libcubeb/update.sh @@ +33,5 @@ > +cp $1/test/test_resampler.cpp tests/test_resampler.cpp > +cp $1/test/test_duplex.cpp tests/test_duplex.cpp > +cp $1/test/test_record.cpp tests/test_record.cpp > +cp $1/test/test_utils.cpp tests/test_utils.cpp > +cp $1/test/test_devices.c tests/test_devices.cpp Rename it upstream, otherwise it's a pain whenever a reverse-sync is necessary.

(In reply to Matthew Gregan [:kinetik] from comment #40) > Comment on attachment 8734807 [details] [diff] [review] > Update cubeb's udpate.sh script to account for new files. r= > > Review of attachment 8734807 [details] [diff] [review]: > ----------------------------------------------------------------- > > ::: media/libcubeb/update.sh > @@ +33,5 @@ > > +cp $1/test/test_resampler.cpp tests/test_resampler.cpp > > +cp $1/test/test_duplex.cpp tests/test_duplex.cpp > > +cp $1/test/test_record.cpp tests/test_record.cpp > > +cp $1/test/test_utils.cpp tests/test_utils.cpp > > +cp $1/test/test_devices.c tests/test_devices.cpp > > Rename it upstream, otherwise it's a pain whenever a reverse-sync is > necessary. test_devices is not included in gecko unit test. We could stop coping it.

https://hg.mozilla.org/integration/mozilla-inbound/rev/1b22933d97f2 https://hg.mozilla.org/integration/mozilla-inbound/rev/3deb965ce2e4 https://hg.mozilla.org/integration/mozilla-inbound/rev/0e4e9eba11ee https://hg.mozilla.org/integration/mozilla-inbound/rev/030e5bd797cd https://hg.mozilla.org/integration/mozilla-inbound/rev/ca8f6e82829b https://hg.mozilla.org/integration/mozilla-inbound/rev/54534d5c9382 https://hg.mozilla.org/integration/mozilla-inbound/rev/5e1216b00e35

https://hg.mozilla.org/mozilla-central/rev/1b22933d97f2 https://hg.mozilla.org/mozilla-central/rev/3deb965ce2e4 https://hg.mozilla.org/mozilla-central/rev/0e4e9eba11ee https://hg.mozilla.org/mozilla-central/rev/030e5bd797cd https://hg.mozilla.org/mozilla-central/rev/ca8f6e82829b https://hg.mozilla.org/mozilla-central/rev/54534d5c9382 https://hg.mozilla.org/mozilla-central/rev/5e1216b00e35 https://hg.mozilla.org/mozilla-central/rev/b0e1f8344f3e https://hg.mozilla.org/mozilla-central/rev/db4111c66699