with glib 2.37 from git master, unity-panel-service crashes when windows close
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
DBus Menu |
Fix Released
|
Critical
|
Charles Kerr | ||
libdbusmenu (Ubuntu) |
Fix Released
|
Undecided
|
Unassigned |
Bug Description
I run Ubuntu 13.04. I generally install glib built from git master, which is either brave or foolish, depending on your perspective. :) Recently I updated glib and found a pretty significant regression in my desktop environment: whenever I close a terminal window or most other windows, unity-panel-service crashes. It restarts automatically, but I see all indicators on the system panel vanish and then reappear.
I bisected glib and found that this broke with a recent change to gobject:
commit 31fde567a95ff8f
Author: Ryan Lortie <email address hidden>
Date: Mon Apr 22 12:33:30 2013 -0400
gtype: put private data before the instance
For more details, see https:/
Related branches
- Mathieu Trudel-Lapierre: Approve
- PS Jenkins bot (community): Approve (continuous-integration)
-
Diff: 58 lines (+21/-8)1 file modifiedlibdbusmenu-gtk/client.c (+21/-8)
affects: | unity → libdbusmenu |
Changed in libdbusmenu: | |
status: | New → Confirmed |
importance: | Undecided → Critical |
status: | Confirmed → In Progress |
assignee: | nobody → Charles Kerr (charlesk) |
Changed in libdbusmenu: | |
status: | Fix Committed → Fix Released |
hi Adam,
Not sure what the issue is here but I have a sneaky suspicion that this is something that was already broken before the change and the change had the effect of turning some undefined behaviour into an outright crash. We've already found a couple of cases where the change of the memory setup caused by this patch has turned 'harmless' memory errors into outright crashers.
That being said, if you're also running git-master Gtk, it could have been this highly related patch which went in right about the same time:
https:/ /git.gnome. org/browse/ gtk+/commit/ ?id=ca0a18918c4 737327a47f8180f 064084ae5a6af1
I don't _really_ think this is the case, but this patch changed how the accel label was created and associated with the menu (which is what the backtrace is talking all about) so I thought I'd mention it as a first rule-out.