1182678 - cubeb initialization on the main thread can take a long time on the main thread

Status: RESOLVED FIXED

(Core :: Audio/Video: MediaStreamGraph, defect)

Description

[(no longer active)]

This can apparently take up to 8 minutes!

anonymous namespace'::SafeRelease<IAudioClient>(IAudioClient *) (in xul.pdb)
wasapi_init (in xul.pdb)
cubeb_init (in xul.pdb)
mozilla::CubebUtils::GetCubebContextUnlocked() (in xul.pdb)
mozilla::CubebUtils::GetCubebContext() (in xul.pdb)
mozilla::MediaDecoder::Init(mozilla::MediaDecoderOwner *) (in xul.pdb)
mozilla::dom::HTMLMediaElement::InitializeDecoderForChannel(nsIChannel *,nsIStreamListener * *) (in xul.pdb)
mozilla::dom::HTMLMediaElement::MediaLoadListener::OnStartRequest(nsIRequest *,nsISupports *) (in xul.pdb)
nsJARChannel::OnStartRequest(nsIRequest *,nsISupports *) (in xul.pdb)
nsInputStreamPump::OnStateStart() (in xul.pdb)
nsInputStreamPump::OnInputStreamReady(nsIAsyncInputStream *) (in xul.pdb)
nsInputStreamReadyEvent::Run() (in xul.pdb)
nsThread::ProcessNextEvent(bool,bool *) (in xul.pdb)
NS_ProcessNextEvent(nsIThread *,bool) (in xul.pdb)
nsThread::Shutdown() (in xul.pdb)

This information is coming from http://telemetry.mozilla.org/chromehangs/.

Comment 1

[Paul Adenot (:padenot)]

Well we should not init cubeb on the main thread, this is the real bug.

Also, this is probably showing a stream that is failing to open, for whatever reason (SafeRelease at the bottom of the stack).

Matthew, do you know why we would try to open a cubeb stream on the main thread here, apparently for media playback ? Something fishy is going on.

Comment 2

[Paul Adenot (:padenot)]

Also, we have normal telemetry for the time it takes to open the first stream (setting up IPC and everything else between firefox and the audio server), and the subsequent stream. I was aware of streams that take 8 seconds on OSX, but not 8 minutes on windows, heh.

Comment 3

[Matthew Gregan [:kinetik]]

Bug 1103824 was initializing cubeb on the main thread to meet the requirement of initializing the first IAudioClient in an STA thread, but the patch has been backed out because of exactly these initialization hangs.

Comment 4

[Matthew Gregan [:kinetik]]

This was fixed by bug 1103824 being reverted.