Bugzilla

I don't see anything obvious with the tests. They appear to take constant time on Windows. MacOS seems to be fairing the worst by a large margin. Linux oddly seems to be somewhere in the middle, but oddly recovers?

In contrast, when I test locally on macOS the tests appear to run in constant time:
``
mach mochitest --run-until-failure browser/base/content/test/webrtc/browser_devices_select_audio_output.js 
```

Could it be related to the new voice engine stuff in macOS Andreas was working on? — But AFAICT the tests only touch the front-end part of speaker selection, not playback. But the prompt code probably tickles cubeb. Is that right, Karl?

I don't see anything obvious with the tests. They appear to take constant time on Windows. MacOS seems to be fairing the worst by a large margin. Linux oddly seems to be somewhere in the middle, but oddly recovers?

In contrast, when I test locally on macOS the tests appear to run in constant time:
```
mach mochitest --run-until-failure browser/base/content/test/webrtc/browser_devices_select_audio_output.js 
```
Could it be related to the new voice engine stuff in macOS Andreas was working on? — But AFAICT the tests only touch the front-end part of speaker selection, not playback. But the prompt code probably tickles cubeb. Is that right, Karl?

I don't see anything obvious with the tests. They appear to take constant time on Windows. MacOS seems to be fairing the worst by a large margin. Linux oddly seems to be somewhere in the middle, but oddly recovers? (from GC maybe?)

In contrast, when I test locally on macOS the tests appear to run in constant time:
```
mach mochitest --run-until-failure browser/base/content/test/webrtc/browser_devices_select_audio_output.js 
```
Could it be related to the new voice engine stuff in macOS Andreas was working on? — But AFAICT the tests only touch the front-end part of speaker selection, not playback. But the prompt code probably tickles cubeb. Is that right, Karl?