[CI] Enable audio tests on Chrome and Firefox #23665
Conversation
cwoffenden
force-pushed
the
cw-audio-fix-tests
branch
from
7c2c85f to
065759d
Compare
Yes, running with the browser flags documented earlier (or having interaction). I'll test all of the parameterised tests tomorrow, I didn't check all variants. |
Now checked, and yes, all the parameterised I'll work my way through the other browser tests next week now. It would be useful to have #23659 merged (which doesn't affect anything other than my code) so I can move all the browser audio tests to use the same pattern.
emscripten/.circleci/config.yml Line 309 in 6fd5a1f I removed EDIT: answering myself, it looks like there's no virtualised audio, given the assert: I'll add code to enumerate the hardware (if JS allows this). More notes:
|
|
The latest flags for Chrome I tried appear to work, with the AW tests for regular Chrome passing: For Firefox it's failing as expected since I've not added its fake audio flags ( |
| command: | | ||
| export EMTEST_BROWSER="/usr/bin/google-chrome $CHROME_FLAGS_BASE $CHROME_FLAGS_HEADLESS $CHROME_FLAGS_WASM $CHROME_FLAGS_NOCACHE" | ||
| export EMTEST_BROWSER="/usr/bin/google-chrome $CHROME_FLAGS_BASE $CHROME_FLAGS_HEADLESS $CHROME_FLAGS_WASM $CHROME_FLAGS_NOCACHE $CHROME_FLAGS_AUDIO" |
There was a problem hiding this comment.
If we make this change should we also remove the EMTEST_LACKS_SOUND_HARDWARE setting above?
I'm kind of inclined to make this change as its one separate PR, e.g. [CI] Enable audio testing on chrome?
There was a problem hiding this comment.
I didn't have much time today so it was mostly experimenting and coming back to it, but:
- I'll remove the
EMTEST_LACKS_SOUND_HARDWARE✅ - I'll rename to match its new purpose (I wasn't sure what I'd achieve when starting) ✅
- I'll disable all the wasm64 and 2GB tests with audio for now
Is it fine to always enable the dummy sound hardware and bypass the media interaction? I can't think what this would break (and where real sound hardware is present its used as a preference, at least from my manual testing).
There was a problem hiding this comment.
I imagine enabling dummy sounds hardware and bypass the interaction check should be fine for all out tests..
cwoffenden
force-pushed
the
cw-audio-fix-tests
branch
from
5483e24 to
f9a781b
Compare
cwoffenden
changed the title
|
Hmm, FF still broken. To do:
|
| #ifdef TEST_AND_EXIT | ||
| // We're in the test harness and exiting is via main_thread_tls_access() | ||
| emscripten_exit_with_live_runtime(); | ||
| #endif |
There was a problem hiding this comment.
I was thinking or a separate PR to enable audio testing CI.
PR1: enable audio testing in CI
PR2: fix audio worklet test so they fail correctly
There was a problem hiding this comment.
Sure, NP. I’ll see how I get on with FF on Monday but I might split it for Chrome first too (I set up a VM with the same Debian and FF and so far, even with no audio device, it doesn’t fail; left to try is headless).
Work-in-progress for #23131.
Confirmed that they're failing, now addingWorking as expected after retesting, I'll propagate the changes to the other AW tests (and move to@requires_sound_hardwareand retesting.btest_exitin the process). The current failures are unrelated (interestingly, if I test early morning CET I don't get this level of flakiness, it's very much an afternoon thing).