Bug 106182 – Regression(r138786): Causes webaudio tests to crash

• Component: WebCore Misc.
• Status: RESOLVED FIXED

[Chris Dumez]

<http://trac.webkit.org/changeset/138786> is causing webaudio tests to crash when using gstreamer 0.10.36:
crash log for WebProcess (pid <unknown>):
STDOUT: Timed out waiting for initial response from web process
STDERR: #PROCESS UNRESPONSIVE - WebProcess

Tests failing:
http://build.webkit.org/results/EFL%20Linux%2064-bit%20Debug%20WK2/r138834%20%287688%29/results.html

[Chris Dumez]

backtrace:
http://pastebin.com/TB3mKWeW

[Chris Dumez]

GST_DEBUG=webkit*:5 trace

crash log for WebProcess (pid <unknown>):
STDOUT: Timed out waiting for initial response from web process
STDERR: 0:00:02.940318621 ^[[333m 4184^[[00m      0x102c990 ^[[36mDEBUG  ^[[00m ^[[00m   webkitwebaudiosrc /home/chris/unencrypted/WebKit/Source/WebCore/platform/audio/gstreamer/WebKitWebAudioSourceGStreamer.cpp:430:webKitWebAudioSrcChangeState:<webkitwebaudiosrc0>^[[00m PLAYING->PAUSED
STDERR: #PROCESS UNRESPONSIVE - WebProcess

[Chris Dumez]

Better backtrace with gstreamer debug symbols:
http://pastebin.com/naBeKqHi

[Chris Dumez]

Created attachment 181445 [details]
Patch

I found the part of the patch that caused the regression. I'm proposing a partial revert since the part in question does not seem essential.

[Chris Dumez]

Of course, feel free to reassign and make a proper fix if you prefer not to revert this part. I am also happy to test patches if needed.

[Philippe Normand]

I don't have time to test the patch now but if it works with gst 1.0 I'd be fine marking it r+.

Those changes were made based on the gst 1.0 docs of gstbasesrc IIRC so if the revert doesn't work with 1.0 you can use ifdefs.

The test timeouts were due to the source element never going to PLAYING BTW, maybe the ChangeLog can mention that. Thanks!

[Chris Dumez]

(In reply to comment #6)
> I don't have time to test the patch now but if it works with gst 1.0 I'd be fine marking it r+.
> 
> Those changes were made based on the gst 1.0 docs of gstbasesrc IIRC so if the revert doesn't work with 1.0 you can use ifdefs.
> 
> The test timeouts were due to the source element never going to PLAYING BTW, maybe the ChangeLog can mention that. Thanks!

I haven't tested with gstreamer 1.0. It would be good if someone using GTK port could.
I'll update the Changelog.

[Chris Dumez]

Created attachment 181455 [details]
Patch

Update Changelog.

[Chris Dumez]

Created attachment 181461 [details]
Patch

Update Changelog again. As it turns out, the regression does not affect gstreamer 0.10.x only. I have built webkitgtk with web audio support on and I can reproduce the same issue (unresponsive WebProcess). My patch works fine on webkitgtk / gstreamer 1.0 and webkitefl / gstreamer 0.10.

[Chris Dumez]

Temporarily skipping those test cases for WK2 EFL until this regression is fixed:
http://trac.webkit.org/changeset/138930

[Chris Dumez]

Comment on attachment 181461 [details]
Patch

Clearing cq for now. I'll update the patch to unskip the test cases.

[Chris Dumez]

Committed r138935: <http://trac.webkit.org/changeset/138935>