media-source/waiting-for-audio.html sometimes times out on Chrome

[nsatragno]

Hello! media-source/waiting-for-audio.html is flaky on Chrome, it sometimes times out.

Check out https://crbug.com/427585592 which has more information.

Karl, is this test flaky on firefox as well? Could there be a problem with the test itself, or do you think it's more likely we got some interop bug on Chrome? I'm afraid I'm not very familiar with media source.

[karlt]

The test consistently passes in Firefox.

Can you point me at the results of a failing run in Chromium please?
LUCI is showing Error [object Object] when trying to view results.
What proportion of runs is the test failing?

Does the recent date on https://issues.chromium.org/issues/427585592 indicate that the test has been generating variable results in Chromium only recently?

The test passes for me with Chrome 136.0.7103.48 on https://wpt.live/media-source/waiting-for-audio.html

https://wpt.fyi/results/media-source/waiting-for-audio.html?label=experimental&label=master&aligned shows that test has been passing consistently with Chrome since Feb 2.
It also shows that the test has always been failing in Edge with PROMISE_REJECT(object "QuotaExceededError: Failed to execute 'appendBuffer' on 'SourceBuffer': The SourceBuffer is full, and cannot free space to append additional buffers."). If that is occurring while readyState < HAVE_FUTURE_DATA, then that sounds like a user-agent problem.

As indicated in the test, some user-agent implementations do not transition to HAVE_FUTURE_DATA on the first frame beyond currentTime. That is contrary to the spec. Delaying HAVE_FUTURE_DATA might be the cause of QuotaExceededError.

[nsatragno]

The tests are timing out with no output. The proportion seems to vary a lot depending on the platform, but it fails frequently enough to be on the top flakes.

As indicated in the test, some user-agent implementations do not transition to HAVE_FUTURE_DATA on the first frame beyond currentTime. That is contrary to the spec. Delaying HAVE_FUTURE_DATA might be the cause of QuotaExceededError.

Could this cause a timeout as well? I sent it to some of our media people to take a look.

[karlt]

Perhaps. It would make the test take longer to run, and the test does take longer in Chrome.
Many asserts should be running well before the timeout is reached.