1137632 - Assertion failure after tab/browser sharing and using replaceTrack ( outputTrack->GetEnd() == GraphTimeToStreamTime(interval.mStart) )

Status: REOPENED

(Core :: WebRTC, defect, P4)

Description

[Mark Banner (:standard8)]

STR:

1) Use in-development Firefox Hello tab sharing and the patches on bug 1136871.
2) Set up Hello call in-room
3) Share Tabs
4) Stop sharing tabs (without switching)
5) Leave conversation
6) Close conversation window

=> No crash/assertion

7) Re-enter the room
8) Share Tabs
9) Switch the tab that is shown in the window
10) Stop sharing tabs
11) Leave conversation
12) Close conversation window

=> Crashes with a debug assertion - see the stack below:

Assertion failure: outputTrack->GetEnd() == GraphTimeToStreamTime(interval.mStart) (Samples missing), at /Users/mark/loop/gecko-dev/dom/media/TrackUnionStream.cpp:279
#01: mozilla::MediaStreamGraphImpl::ProduceDataForStreamsBlockByBlock(unsigned int, int, long long, long long)[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x1f3ef40]
#02: mozilla::MediaStreamGraphImpl::Process(long long, long long)[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x1f3f491]
#03: mozilla::MediaStreamGraphImpl::OneIteration(long long, long long, long long, long long)[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x1f3fba4]
#04: mozilla::AudioCallbackDriver::DataCallback(float*, long)[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x1ef727a]
#05: audiounit_output_callback[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x32664b3]
#06: AUConverterBase::RenderBus(unsigned int&, AudioTimeStamp const&, unsigned int, unsigned int)[/System/Library/Components/CoreAudio.component/Contents/MacOS/CoreAudio +0x8230]
#07: AUBase::DoRenderBus(unsigned int&, AudioTimeStamp const&, unsigned int, AUOutputElement*, unsigned int, AudioBufferList&)[/System/Library/Components/CoreAudio.component/Contents/MacOS/CoreAudio +0x5c61]
#08: AUBase::DoRender(unsigned int&, AudioTimeStamp const&, unsigned int, unsigned int, AudioBufferList&)[/System/Library/Components/CoreAudio.component/Contents/MacOS/CoreAudio +0x4515]
#09: AUHAL::AUIOProc(unsigned int, AudioTimeStamp const*, AudioBufferList const*, AudioTimeStamp const*, AudioBufferList*, AudioTimeStamp const*, void*)[/System/Library/Components/CoreAudio.component/Contents/MacOS/CoreAudio +0xb878]
#10: HALC_ProxyIOContext::IOWorkLoop()[/System/Library/Frameworks/CoreAudio.framework/Versions/A/CoreAudio +0x279db]
#11: HALC_ProxyIOContext::IOThreadEntry(void*)[/System/Library/Frameworks/CoreAudio.framework/Versions/A/CoreAudio +0x26add]
#12: HALB_IOThread::Entry(void*)[/System/Library/Frameworks/CoreAudio.framework/Versions/A/CoreAudio +0x2699d]
#13: _pthread_body[/usr/lib/system/libsystem_pthread.dylib +0x1899]
#14: _pthread_struct_init[/usr/lib/system/libsystem_pthread.dylib +0x172a]

Comment 1

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

This is audio. Are we replaceTracking audio? Byron, do your patches to remove unused streams consider audio tracks?

Comment 2

[Mark Banner (:standard8)]

The sdk has an option to replace audio, but I've just confirmed that it isn't actually being activated. I also tried commenting it out to doubly make sure.

Comment 3

[Byron Campen [:bwc]]

The code that prunes unused streams does not care what kind of track was the last to go.

Comment 4

[Mark Banner (:standard8)]

We've just had an update of the sdk, that seems to be tidying things up better with respect to the previous tracks, and managing the new ones.

Do we still want to try and find out what's going on here, or close as WFM?

Comment 5

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

WFM WFM.

Comment 6

[Eric Rescorla (:ekr)]

I don't think this can be closed yet. If we have an assert, that means
that there is presumably a bug in the C++ code. Just because the
SDK no longer exercises it doesn't mean that it can't be exercised from
content JS.

Comment 7

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

Paul can you take a look? I hear Mark is willing to make a build with the old sdk so that we can figure out what's going on here.

Comment 8

[Paul Adenot (:padenot)]

If someone can hand me a (preferably debug/non-opt if it can be reproed there) build, I'm happy to have a look. Any platform will do, just do what is easier.

Comment 9

[Paul Adenot (:padenot)]

Attachment: debug trace

This is video, not audio. It's not clear to me if it is the remote camera, local camera, or the screen sharing, because it seems like listeners have been removed already, so I can't go to the *Engine.

In any case, according to the gdb trace attached, video frames are missing from the stream, maybe that's the screen capture that forgot to add a final frame or something?

Comment 10

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

Odd, in comment 0 it's in the audio driver:
>#04: mozilla::AudioCallbackDriver::DataCallback(float*, long)[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x1ef727a]
>#05: audiounit_output_callback[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x32664b3]

Comment 11

[Paul Adenot (:padenot)]

(In reply to Jan-Ivar Bruaroey [:jib] from comment #10)
> Odd, in comment 0 it's in the audio driver:
> >#04: mozilla::AudioCallbackDriver::DataCallback(float*, long)[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x1ef727a]
> >#05: audiounit_output_callback[/Users/mark/loop/gecko-dev/objdir-ff/dist/NightlyDebug.app/Contents/MacOS/XUL +0x32664b3]

It's always an audio driver for everything when there is at least one MediaStream that has a MediaStreamTrack that has audio. The AudioCallbackDriver is just running the graph to ensure audio latency is minimal.

Comment 12

[Mark Banner (:standard8)]

I believe this doesn't currently block the tab sharing release as we're not hitting it with the newer sdk. Though platform will obviously want to fix this still.

Comment 13

[Jesse Ruderman]

Bug 1140030 has a testcase that triggers the same assertion, somewhat reliably. HTH.

Comment 14

[Bulk Bug Changes for mreavy's org]

Mass change P3->P4 to align with new Mozilla triage process.