GNOME Bugzilla – Bug 171528
SIGFPE in gstasf with http://joule.via.ecp.fr/~lool/debian/foo.asf
Last modified: 2006-02-17 13:59:10 UTC
Version details: Debian/sid Distribution/Version: Debian/sid Hi, I got a private mail report of a crash with an ASF of which the file: http://joule.via.ecp.fr/~lool/debian/foo.asf is an extract. I can reproduce the crash with gst-plugins 0.8.7 and 0.8.8 in Debian sarge and sid. Somehow, on my sid system, it doesn't occur. Here are some bits of a backtrace: Program received signal SIGFPE, Arithmetic exception.
+ Trace 57294
Thread 1145199536 (LWP 2764)
It's painful to get plugins built with debugging symbols with the Debian packages, so I did not install all debugging modules, I think the relevant part is shown. I don't know when I'll be able to look deeper into this. Bye,
Fixing the SIGFPE isn't so hard, but we appear to lose sync. ffmpeg does a better job. I'll look closer.
Any news on this? I'm getting other SIGFPE in gstasf, I ponder deactivating it, is this sane?
It appears amarok will SIGFPE on startup (only with gstreamer 0.8.10) if libgstasf is there, and it is the very same backtrace: Program received signal SIGFPE, Arithmetic exception. ---Type <return> to continue, or q <return> to quit---
+ Trace 61368
Thread NaN (LWP 1167)
Should I disable the gstasf plugin in a newer Debian build?
It stills segfault with: gst-launch-0.8 filesrc location="foo.asf" ! asfdemux ! fakesink
I get an FPE as well on x86-64, with the pipeline from comment #4 and gstreamer 0.8.11. asfdemux doesn't appear to be ported yet to 0.10.
FWIW, the SIGFPE is fixed in the 0.10 port, but there's still some playback issues. In case someone wants to port the fix back, I believe it involved an added } else { bracket in the stream parse function and an early bail out in the descramble_segment function, but my memory about that is very hazy.
asfdemux is now in plugins-ugly 0.10 cvs and with the latest fixes it works fine. 2006-02-17 Edward Hervey <edward@fluendo.com> * gst/asfdemux/gstasfdemux.c: (gst_asf_demux_process_file): Take into account the file properties preroll value for timestamping/newsegment. It's weird this value was commented out.