On Tue, Nov 13, 2007 at 01:03:28PM +0100, Josselin Mouette <[EMAIL PROTECTED]> wrote: > Hi, > > Le vendredi 05 octobre 2007 à 07:59 +0200, Mike Hommey a écrit : > > On Fri, Oct 05, 2007 at 07:14:09AM +0200, Mike Hommey wrote: > > > Package: gnome-panel > > > Version: 2.20.0.1-1 > > > Severity: important > > > > > > After a few hours running, here is what top displays: > > > 7816 mh 15 0 266m 167m 6788 S 0.7 22.1 4:25.10 gnome-panel > > > > > > There is obviously something leaking. > > > > I think i found the culprit. My battery is half dead, and the > > gnome-power-manager notification icon is set up to only show up when on > > battery. As it happens, the acpi information switches quite randomly > > between on battery an on AC, meaning the icon appears/disappears quite a > > lot, and following gnome-panel memory usage after a while seems to > > indicate the memory hogging doesn't happen when the icon is made still > > (always display it), though the icon itself changes (switches between the > > charging icon and the on AC icon). > > > > So, on first hand, there seems to be a leak when a notification icon > > disappears. > > The upstream developers asked if you could obtain a valgrind output of > gnome-panel when this happens.
I'd love to, if I knew how to launch the panel within valgrind (I actually tried unsucessfully). The sad thing is that my quite dead battery doesn't seem to be bouncing between not-present and present state anymore, so the notification icon might well just stay still :( So even if I managed to run the panel under valgrind, I wouldn't probably be able to reproduce the leak. I'll try to look at the g-p-m code to see how it displays/undisplays the notification icon to try to make a reduced testcase. Mike -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]