[gst-devel] new prerelease

[Ronald S. Bultje]

Hi,

On Wed, 2004-06-02 at 11:26, Johan Dahlin wrote:
> > I'm experiencing random segfaults in threaded pipelines using any
> > scheduler. I also get a lot of warnings:
> [..]
> > Try gst-launch filesrc location=/media/clips/lego_troopers_divx.avi !
> > spider ! { queue ! ffcolorspace ! ximagesink } spider0. ! { queue !
> > audioconvert ! audioscale ! osssink } [--gst-scheduler=basicgthread] to
> > experience yourself. Refocus the xwindow and the terminal a few times
> > and enjoy the crashes.
> 
> I'm pretty sure it's a local problem. I can play that pipeline with that
> very media file. And we've been using quite a lot of threads here and it
> works for everyone.

I can confirm that it isn't. Even after a reboot, I'm able to reproduce
it, this time non-randomly. The backtrace is as follows:

(gdb) r filesrc location=/media/clips/lego_troopers_divx.avi ! spider !
{ queue ! ffcolorspace ! ximagesink } spider0. ! { queue ! audioconvert
! audioscale ! alsasink } --gst-fatal-warnings
[..]
#6  0x40304313 in g_log (log_domain=0x400737e0 "GStreamer",
    log_level=G_LOG_LEVEL_CRITICAL,
    format=0x40074000 "file %s: line %d (%s): assertion `%s' failed")
    at gmessages.c:541
#7  0x400363b0 in gst_data_unref (data=0x805d3e8) at gstdata.c:245
#8  0x4130d6a7 in gst_avi_demux_handle_src_event (pad=0x8146ca8,
    event=0x805d3e8) at gstavidemux.c:637
#9  0x4004fb89 in gst_pad_send_event (pad=0x8147ad8, event=0x805d3e8)
    at gstpad.c:4078
#10 0x4004f68e in gst_pad_event_default_dispatch (pad=0x81485b0,
    element=0x80e18a0, event=0x805d3e8) at gstpad.c:3941
#11 0x4004f754 in gst_pad_event_default (pad=0x81485b0, event=0x805d3e8)
    at gstpad.c:4004
#12 0x408d2f06 in gst_opt_scheduler_event_wrapper (srcpad=0x81485b0,
---Type <return> to continue, or q <return> to quit---
    event=0x805d3e8) at gstoptimalscheduler.c:1473
#13 0x4004fb89 in gst_pad_send_event (pad=0x81485b0, event=0x805d3e8)
    at gstpad.c:4078
#14 0x4004f68e in gst_pad_event_default_dispatch (pad=0x8130c60,
    element=0x81302f0, event=0x805d3e8) at gstpad.c:3941
#15 0x4004f754 in gst_pad_event_default (pad=0x8130c60, event=0x805d3e8)
    at gstpad.c:4004
#16 0x401256a8 in gst_spider_identity_handle_src_event (pad=0x8130c60,
    event=0x805d3e8) at gstspideridentity.c:588
#17 0x4004fb89 in gst_pad_send_event (pad=0x8130c60, event=0x805d3e8)
    at gstpad.c:4078
#18 0x4004f68e in gst_pad_event_default_dispatch (pad=0x8107ce0,
    element=0x8107b18, event=0x805d3e8) at gstpad.c:3941
#19 0x4004f754 in gst_pad_event_default (pad=0x8107ce0, event=0x805d3e8)
    at gstpad.c:4004
#20 0x400535e8 in gst_queue_handle_pending_events (queue=0x8107b18)
    at gstqueue.c:396
#21 0x40053747 in gst_queue_chain (pad=0x8107c18, data=0x806173c)
    at gstqueue.c:437
#22 0x4005041f in gst_pad_call_chain_function (pad=0x8107c18,
data=0x806173c)
    at gstpad.c:4354
#23 0x4004dc0b in gst_pad_push (pad=0x8130c60, data=0x806173c)
    at gstpad.c:3187

This happens if I click the window a few times. There's an event being
unreferenced in some place where it shouldn't, I assume this is the
event that ximagesink sends on a window click. Appears thread-related,
though I cannot surely confirm this:

gst-launch-0.8  filesrc location=/media/clips/lego_troopers_divx.avi !
spider ! { queue ! ffcolorspace ! ximagesink } --gst-fatal-warnings

works...

Some more fun:

gst-launch-0.8  filesrc location=/media/clips/lego_troopers_divx.avi !
spider ! ffcolorspace ! ximagesink spider0. ! audioconvert ! audioscale
! alsasink --gst-fatal-warnings
RUNNING pipeline ...
 
** WARNING **: deadlock detected, disabling group 0x80d7a98
aborting...
Aborted

Works using basicgthread, though it's very slow.

Note that MPEG movies work fine, so this is AVI specific. I looked at
the avidemux sources but can't find anything wrong.

More fun, try playing everwood.avi or so from the media dir on our
website. Audio fails and the movie stalls after a while, mad complains
on and on about lost sync. This movie used to work fine. I can provide
samples if needed. Note that all this appears plugins-related so it
shouldn't delay the core release.

Ronald