GNOME Bugzilla – Bug 316096
gnome panel crashes starting Mathematica from Command Line applet
Last modified: 2006-03-29 15:54:36 UTC
add command line applet to panel start Mathematica from command line applet Mathematica gets started wnck-applet uses 100% of my system resources I can't select any windows on my panel at the bottom but can still access menu and applications on my top panel. I start gnome-terminal kill wnck-applet I get these messages: "Show Desktop" has quit unexpectedly "Window List" has quit unexpectedly "Workspace Switcher" has quit unexpectedly this happens every time I start Mathematica. it doesn't happen for other programs.
Created attachment 52131 [details] Mathematica script calling the mathematica kernel Maybe it helps to see what the Mathematica command does
Ubuntu bugzilla has a bug about that: http://bugzilla.ubuntu.com/show_bug.cgi?id=14173 "... (gdb) thread apply all bt
+ Trace 63005
Thread 1 (Thread -1223653696 (LWP 10757))
Debug backtrace: (gdb) bt
+ Trace 63060
Sébastien: is "task->was_active = FALSE;" line 2165 in tasklist.c for you? I'm also not sure I understand: is it crashing or is it a stack trace of the applet during a 100%CPU loop?
correct for l2165 (libwnck 2.12.0), it doesn't crash that a bt from a loop when starting mathematica
anything else that could be useful for you guys on that?
time? :-)
Distro comment: "A workaround is starting Mathematica using -noSplashScreen. This would prevent the splash screen to show up (looks like the splash is the window causing trouble). ..."
What's the output of running xprop and then clicking on that Mathematica splashscreen?
XKLAVIER_STATE(INTEGER) = 0, 2 WM_STATE(WM_STATE): window state: Normal icon window: 0x0 _NET_FRAME_EXTENTS(CARDINAL) = 5, 5, 26, 5 _NET_WM_DESKTOP(CARDINAL) = 0 _NET_WM_STATE(ATOM) = _NET_WM_STATE_SKIP_TASKBAR _NET_WM_ALLOWED_ACTIONS(ATOM) = _NET_WM_ACTION_MOVE, _NET_WM_ACTION_RESIZE, _NET_WM_ACTION_FULLSCREEN, _NET_WM_ACTION_SHADE, _NET_WM_ACTION_CHANGE_DESKTOP, _NET_WM_ACTION_CLOSE WM_TRANSIENT_FOR(WINDOW): window id # 0x420002d WM_NAME(STRING) = " "
Dang. I don't see anything wrong there... The stack traces don't make any sense to me either (wnck_task_finalize() doesn't call wnck_task_popup_menu() though it does call wnck_tasklist_change_active_task which could toggle the togglebutton calling wnck_task_button_toggled() which could call wnck_task_popup_menu() or wnck_tasklist_activate_task_window()....also, wnck_task_popup_menu() does not call wnck_tasklist_activate_task_window() and I don't see any path it could take to get there either). I wonder if memory is totally hosed or something since the stack trace looks so bad (maybe someone with mathematica could run libwnck under valgrind?) Also, is there any chance you could run under a profiler like oprofile or sysprof? Output from those are often a lot better at tracking down CPU spikes than single-backtrace-snapshots from gdb.
from valgrind: ==25368== Conditional jump or move depends on uninitialised value(s) ==25368== at 0x1B94C853: _wnck_read_icons (xutils.c:2072) ==25368== by 0x1B930D65: get_icons (application.c:290) ==25368== by 0x1B930EFE: wnck_application_get_icon (application.c:347) ==25368== by 0x1B932276: get_icons_from_applications (class-group.c:380) ==25368== by 0x1B9323BC: set_icon (class-group.c:436) ==25368== by 0x1B932653: _wnck_class_group_add_window (class-group.c:483) ==25368== by 0x1B937E20: update_client_list (screen.c:946) ==25368== by 0x1B939004: do_update_now (screen.c:1511) ==25368== by 0x1B937573: wnck_screen_force_update (screen.c:652) ==25368== by 0x804E227: window_list_applet_fill (window-list.c:676) ==25368== by 0x1B924474: panel_applet_marshal_BOOLEAN__STRING (panel-applet-marshal.c:128) ==25368== by 0x1BE373A7: g_closure_invoke (gclosure.c:492)
Yeah, isn't that the same valgrind error that's been showing up in valgrind logs for months/years now? We really ought to look at it... Doesn't seem to be filed either, which is weird, especially since I'm sure I've seen it several times now.
*** Bug 323883 has been marked as a duplicate of this bug. ***
*** Bug 326841 has been marked as a duplicate of this bug. ***
I have this too on gentoo. I just want to add that killall wnck-applet makes those popups appear that ask if the panel applets and the panel need to be restarted. If you choose to do so everything works normal (with Mathematica), so the bug is definetely related to the splash. The symptoms here are just a freeze (and grey-out when overlapped with something) of my bar that contains applets such as the windowswitcher. Alt+tab and the panel with the menu's keep working. Used to work fine with gnome 2.10. Mathematica 5.2.
Just wanted to note, that the problem went away for me using latest gnome 2.14 rc. so it's probably not worth digging around in this here anymore.
Bug 336264 reports that this was the self-transiency problem (for which there is a patch) in bug 332493; I find it odd that 2.14 fixed it if that was indeed the problem. Andreas: Are you using a distro that patched libwnck?
(In reply to comment #18) I'm using gentoo here and I'm using the pure, unpatched version of libwnck-2.14.0. I just restarted Mathematica and can confirm again that the bug is gone.
Did you start it from a terminal? If so, the window wouldn't get focus and would prevent this bug from happening, despite the bug still being there. Could you try clicking on the splashscreen before it disappears and verify whether the bug occurs or not when you do that?
I did start it from the terminal. and running mathematica from the deskbar applet actually brings the bug back (but I remember that starting mathematica from terminal did use to crash the wnck-applet - so this at least is somehow gone). to your second question: clicking on the splashscreen while running mathematica from the terminal brings back the bug aswell.
Awesome, thanks for the help Andreas. Is there any chance I could get you to also verify whether the two patches in bug 332493 fix this bug?
I can confirm that the two patches do solve both problems. starting mathematica from the deskbar runs without a problem as well as running mathematica from terminal and clicking on the splashscreen.
Great, thanks for all the detective work, Andreas. I'll mark as a duplicate then, and since the patches have been committed, this'll be fixed in 2.14.1. :) *** This bug has been marked as a duplicate of 332493 ***
no problem. could you please commit this fix to the 2.12.x tree aswell? I know a lot of people who are forced to use gnome 2.12 and who would be happy if this issue would get fixed.
Yeah, I can do that. Not sure how much good it'll do since there aren't any more 2.12.x releases scheduled. But, I guess it can't hurt just in case distributors check out the branch or in case we do make another release and distributors happen to look for it.