GNOME Bugzilla – Bug 314615
Reproducable crash on remote server
Last modified: 2006-10-08 09:17:28 UTC
I'm using GNOME 2.12 RC. It seems that, on a ssh remote server, when I open a folder, nautilus can crash if one of the parent folder window is closed. I'm using the spatial mode. Eg: I open window.gnome.org:~/ in nautilus. I open public_html with a doubl-click (left button). I close the ~/ window. I open public_html/tmp with a doubl-click (left button). Crash. It's always reproducable (so far) Stack trace: Backtrace was generated from '/usr/bin/nautilus' Using host libthread_db library "/lib/tls/i686/cmov/libthread_db.so.1". `system-supplied DSO at 0xffffe000' has disappeared; keeping its symbols. [Thread debugging using libthread_db enabled] [New Thread -1226619200 (LWP 2972)] [New Thread -1235666000 (LWP 2981)] [New Thread -1234875472 (LWP 2977)] [New Thread -1234609232 (LWP 2976)] [New Thread -1234342992 (LWP 2975)] [New Thread -1228903504 (LWP 2973)] 0xffffe410 in __kernel_vsyscall ()
+ Trace 62642
Thread 6 (Thread -1228903504 (LWP 2973))
I don't see this myself with the latest cvs HEAD. Does this happen with all directories on the ssh share? Do you do things fast, before the windows have been fully read in, or do you let them finish all operations in each step? The end of the backtrace looks exactly like the decoded one in bug 302096. I wonder if these are the same.
I'm at work (with 2.10), so I can't verify right now. But, iirc: * it happens with many subdirectories on the ssh share (I don't know if it's all the directories, but it happened at least for 3 or 4 subdirectories) * I didn't do things fast: when reproducing, I waited for all the windows to have finished loading. I'll verify later today if you want. I'll look if it's still reproducable once I get home.
We're opening a new window, merging various ui descriptions. Then when we ask for the menu bar widget and GtkUIManager starts to create the various widgets that are part of the menus. On creating the GtkMenuItem widget for some menu item in the ui description we need to set the accel_closure on the label in the item, but it seems like the accel group returned from gtk_accel_group_from_accel_closure() on that accel closure is invalid. The accel closure is tightly bound lifetime-wise to the GtkAction, so it should be ok if the action is ok. Maybe the accel group that was connected to the closure has been freed? How can this be related to the closing of the parent window though? I don't think the accel groups are shared between windows. Also, Vincent, it would be nice if you checked for any output on the console before the crash.
The accel_group is tied to the GtkUIManager, but the GtkUIManager for this window was just created before the current call in nautilus_window_init() and is only freed on window finalization. Very strange.
It really smells like something is just overwriting memory somewhere. That would also explain why bug 302096 has slightly different behaviour just before the crash.
Also, just to check, if you can reproduce it, does the new patch in bug 170947 fix it? (Its unlikely, but I wanna make sure.)
Sorry for the delay. I can confirm that it crashes with all the folders. Also I'm noit doing things quickly (the windows have even finished loading before I start moving the mouse ;-)). One interesting point is that it seems it doesn't crash if I close one parent window that is not the "root" parent of the share. E.g.: I have this on my ssh share: ~/public_html/tmp/traduc. If I open public_html, then tmp, close ~, and open traduc, it crashes. If I open public_html, then tmp, close public_html, and open traduc, it doesn't crash. I don't see anything on the console before the crash. I just killed nautilus and launched it again from a terminal, and the crash does not happen any more for now... Weird. Maybe nautilus got upgraded since last time it was launched (possible, although I really don't think it was the case according to my logs). Or maybe it was because I had lots of other open folders in my previous tests...
ugh. Non-repeatable is bad. You didn't update to the latest gnome-vfs inbetween (with the fix from bug 170947)?
gnome-vfs was last updated at the same time than nautilus, so no.
*** Bug 319006 has been marked as a duplicate of this bug. ***
Are you still able to reproduce this one?
Just tried once and it didn't happen. I'd like to make more tests, but I can't now, so marking as NEEDINFO for now.
Hello Vincent, I know you have a lot of free time so could you try to reproduce your crash ? :) Thanks
Closing since I couldn't reproduce anymore.