GNOME panel crash/segfault on logout

[Sitsofe Wheeler]

Description of the problem:
I recently enabled core dumps on my system after trying to track down a gedit crash on logout. After some testing it appears that gedit isn't the only program that segfaults during a GNOME logout (although it is the one which does it 100% of the time).

Steps to reproduce:
1. Create a ~/.gnomerc file that contains the following:
ulimit -c unlimited
export GNOME_DISABLE_CRASH_DIALOG=1
2. Run
echo /tmp/core.%e-%p > /proc/sys/kernel/core_pattern
3. If you are logged into GNOME logout and then back in again.
4. Run the two scripts
start-internetsoundvideoapps.sh
start-accessoriesgamesapps.sh
waiting for the first script to finish before running the second.
5. After all apps have loaded play a video in totem and some music in rhythmbox.
6. Click the logout button at the top right hand corner of the screen.
7. Click the log out button in the window that appears.
8. Once logout has finally finished switch to VT1, log in and inspect /tmp/ for core files.

Expected results:
No core files to be found.

Actual results:
Following core files were found:
core.deskbar-applet-4465
core.evolution-alarm-5446
core.evolution-excha-4482
core.gedit-5520
core.gnome-panel-4402
core.python-4684
core.rhythmbox-4683
core.sound-juicer-4685
core.totem-4682

gnome-panel stack trace and scripts to come.

[Sitsofe Wheeler (sitsofe)]

Script to start programs in the accessories and games menus

[Sitsofe Wheeler (sitsofe)]

Script to start programs in the internet and sound & video menus

[Sitsofe Wheeler (sitsofe)]

gnome panel backtrace

[Sitsofe Wheeler (sitsofe)]

Oops. One oversight is that
echo /tmp/core.%e-%p > /proc/sys/kernel/core_pattern
needs to be
sudo echo /tmp/core.%e-%p > /proc/sys/kernel/core_pattern

[Andy Price (andy-price)]

Thanks for your bug report. I've tried this on an up-to-date dapper i386 install a few times but I can't reproduce the bug. I checked that the .gnomerc and core dump format settings had applied properly and installed packages so that as many of the programs listed in your scripts were on my system as possible but no luck.

Is the bug still present after updating your system? If so, is it possible to simplify the test down and still produce the same results (e.g. only running the programs that you get the core dumps for)?

[Sitsofe Wheeler (sitsofe)]

My system is fairly up to date and the core dumps are reproducible with a clean profile. More than anything else I thnk it is the fact that the system is put under swap pressure that helps this problem show up. If you have a lot of memory try booting with the mem=384m kernel parameter.

[Sitsofe Wheeler (sitsofe)]

I forgot to mention that attempts to narrow things down only resulted in the diagnosis of two programs that always leave core dumps (there should always have been a core dump from gedit and rhythmbox) but these dumps are not similar to this one.

[Andy Price (andy-price)]

I have 1GB of RAM so I repeated the steps with the mem=384m kernel parameter and the bug did occur so I am confirming this.

The cores that got dumped were:
core.ekiga-5345
core.evolution-alarm-5559
core.rhythmbox-5352
core.sound-juicer-5354

[Sitsofe Wheeler (sitsofe)]

I just thought I'd add that on my system nautilus and deskbar often segfault too.

[Sebastien Bacher (seb128)]

reassigning to orbit2 since the backtrace mentions it and that happens for different apps over the desktop. Setting as minor since it makes no difference to the users and crash are not noticed with a normal use

[Sitsofe Wheeler (sitsofe)]

I can't reproduce this in edgy (up to date as of 30 September 2006). Resolve fixed?

[Sebastien Bacher (seb128)]

Thank you for the comment. Marking as fixed for now then. Feel free to reopen if you get the issue again