Bug 174241 – [GTK] Layout test media/video-inactive-playback.html is timing out

• Status: RESOLVED
• Resolution: FIXED
• Component: WebKitGTK

[Michael Catanzaro]

Layout test media/video-inactive-playback.html is timing out on the GTK port since it was added in r218417 "[iOS] Do not pause playing video when application resigns active state." Updating expectations accordingly.

[Ms2ger (he/him; ⌚ UTC+1/+2)]

I looked into this and the problem is that the test sets restrictions on the "videoaudio" session:

        run('internals.setMediaSessionRestrictions("videoaudio", "")');
        run('internals.setMediaSessionRestrictions("videoaudio", "inactiveprocessplaybackrestricted")');

while forEachSession only finds a "video" session. This means nothing special happens and neither the expected "pause" or the expected "playing" event are fired.

(However, a "playing" event is fired as the result of the video finishing naturally.)

It's not clear to me why this test works at all on other platforms.

Jer, thoughts?

[Jer Noble]

(In reply to Ms2ger from comment #1)
> I looked into this and the problem is that the test sets restrictions on the
> "videoaudio" session:
> 
>         run('internals.setMediaSessionRestrictions("videoaudio", "")');
>         run('internals.setMediaSessionRestrictions("videoaudio",
> "inactiveprocessplaybackrestricted")');
> 
> while forEachSession only finds a "video" session. This means nothing
> special happens and neither the expected "pause" or the expected "playing"
> event are fired.
> 
> (However, a "playing" event is fired as the result of the video finishing
> naturally.)
> 
> It's not clear to me why this test works at all on other platforms.
> 
> Jer, thoughts?

The file LayoutTests/media/content/test.{mp4,ogv} should have both an audio and a video track, so the resulting session should be of MediaType "VideoAudio", and not just "Video".  It sounds like that's the source of your problem.

If you look at HTMLMediaElement::mediaType(), it will return VideoAudio if the readyState >= HAVE_METADATA, the MediaPlayerPrivate return true for hasVideo() & hasAudio(), and the element is not muted.  So either this test is being run before the "loadedmetadata" event is fired, or the GTK media player doesn't think the file has audio.

[Ms2ger (he/him; ⌚ UTC+1/+2)]

The problem appears to be that Page::m_mutedState is AudioIsMuted. I have not yet been able to figure out how it got that way.

[Ms2ger (he/him; ⌚ UTC+1/+2)]

Looks like this and a bunch of other tests were broken by bug 162582.

[Jer Noble]

(In reply to Ms2ger from comment #4)
> Looks like this and a bunch of other tests were broken by bug 162582.

Sounds like the GTK media player is reporting !hasAudio() when the page state is muted. That seems wrong.

[Michael Catanzaro]

(In reply to Jer Noble from comment #5)
> (In reply to Ms2ger from comment #4)
> > Looks like this and a bunch of other tests were broken by bug 162582.
> 
> Sounds like the GTK media player is reporting !hasAudio() when the page
> state is muted. That seems wrong.

Charlie, this would be a good one for you or Calvaris to look into.

[Ms2ger (he/him; ⌚ UTC+1/+2)]

The code isn't specific to GTK, though:

bool HTMLMediaElement::effectiveMuted() const
{
    return muted() || (document().page() && document().page()->isAudioMuted());
}

[Philippe Normand]

Ms2ger was right in comment 3, the page is muted by default by WKTR so the test should un-mute it in its go function with internals.setPageMuted("");

I'm also a bit confused as to why this passes on other platforms, I'll try to dig this down further...

[Philippe Normand]

The current explanation I have is that in the player the volume element gets muted based on MediaPlayer::muted() value, which is true because the page is muted. I think this behavior is different on other platforms, so that's why they don't need to explicitely un-mute the page.

[Philippe Normand]

Another potential issue is that hasAudio() might return false before the gst pipeline is fully built and the audio track was discovered. This effectively short-cuts the page state too...

[Philippe Normand]

On mac the test passes because the AVF player doesn't support muting, so the volume is set to 0 from the player POV in HTMLMediaElement::updatePlayState() but the HTMLMediaElement m_muted attribute remains false. This is an inconsistency :(

Moreover the AVF player ::setMuted() method is implemented for both mac and iOS but supportsMuting() returns true only for iOS, yet another inconsistency.

If I replace the m_player->setMuted() call with a setMuted() call in HTMLMediaElement::updatePlayState() I get the test timeout on mac, same as in GTK.

[Philippe Normand]

Fixed in https://trac.webkit.org/r226948

But maybe the Apple folks should consider checking the little inconsistencies I found in the AVF player?