28360 - linux hangs-on-exit when "Tasks" menu is opened

Status: VERIFIED FIXED

(SeaMonkey :: General, defect, P1)

Description

[Chris Waterson]

To reproduce:

1. Start browser
2. Open "Tasks" menu
3. Close browser

Results: browser hangs on exit.
Expected: browser closes cleanly.

Alternatively,

1. Start browser
2. Open "Open windows" menu at the bottom of the screen
3. Close browser

This makes me pretty sure something is going wrong with the window mediator. Can
you take a look?

Comment 1

[don]

Yes, this is definitely a beta 1 stopper.

Comment 2

[leger]

Putting on PDT+ radar for beta1.

Comment 3

[davidm]

I can take a look but I really doubt the problem is in my code since I haven't 
touched it in several months.

Comment 4

[jim_nance]

I have observed this for the last few days.  Another good way to reproduce it is
to use the pulldown menus a few times.  I traced it in gdb yesterday.  We are
blocked on a pthread lock inside a call to free().  This should obviously never
happen.  If I had to guess I would say someone is writting to an invalid pointer
and scribbling over things.  Can anyone run purify under Solaris and see if it
picks anything up?

Comment 5

[Chris Waterson]

hey, dmose, mcafee. can on of you guys run purify on a solaris build to test 
jim's theory? (or help me to do it?)

Comment 6

[jim_nance]

OK, I ran it in the debugger again, and its not as simple as I thought.  It is
not simply stuck in free (though it may have been last night.  I have no way of
knowing).  I am going to try and get better tracebacks in a few minutes.

Comment 7

[jim_nance]

Here is a stack trace where we are stuck.  This is with glib and gtk
1.2.7, but we get stuck with 1.2.6 as well.  I just got 1.2.7 to see
if it fixed the problem, and thats what happens to be in the debugger
at the moment.  I stopped/started the program several times and it
always seems to be in some function down in glib.

(gdb) bt
#0  0x40964c25 in g_main_iterate (block=1, dispatch=1)
    at /usr/local/home/jlnance/tmp/glib-1.2.7/gmain.c:843
#1  0x40964ec9 in g_main_run (loop=0x8198578)
    at /usr/local/home/jlnance/tmp/glib-1.2.7/gmain.c:935
#2  0x4089396a in gtk_main ()
    at /usr/local/home/jlnance/tmp/gtk+-1.2.7/gtk/gtkmain.c:476
#3  0x407bbcf7 in nsAppShell::Run (this=0x8136888) at nsAppShell.cpp:304
#4  0x40652e4d in ?? ()
   from /mnt/tmp/mozilla/dist/bin/components/libnsappshell.so
#5  0x804e31e in main1 (argc=1, argv=0xbffffb44, splashScreen=0x0)
    at nsAppRunner.cpp:651
#6  0x804e8b0 in main (argc=1, argv=0xbffffb44) at nsAppRunner.cpp:770

Comment 8

[David Baron :dbaron: (⌚️UTC-5, no longer working on Mozilla)]

I've seen occasional hang-on-exit with similar stack traces recently.  I first
started seeing them when I setenv'd XPCOM_MEM_LEAK_LOG=1, so I wasn't sure if it
was a known problem or not.  There were certain classes where, if I set
XPCOM_MEM_REFCNT_LOG and XPCOM_MEM_LOG_CLASSES=class, I would hang 100% of the
time.  (I got decent logs by going into the debugger and just forcing g_main_run
(or was it g_main_iterate?) to return.)  This has happened in cases as simple as
just clicking "Exit" when I see the profile manager.

Comment 9

[davidm]

I put a couple of printfs and what appears to me to be happening is that the 
ExitCallback never gets called. 

Comment 10

[Sitsofe Wheeler]

I think this may have started happening when the history was made to appear in
the tasks menu in addition to the go menu...

Comment 11

[Syd Logan]

I'm noticing on my startup hang bug that the ExitApp Exit callback in appshell
is not being called either when the profile wizard is closed. Also, on linux
when I bring up the tasks menu, it draws wider than need be, i.e., incorrectly.
Whether or not that has any bearing on the problem I can't say.

Comment 12

[Syd Logan]

Probably is a duplicate, yes. Since this is not affected by chris's soon to be
landed appshellservice change (afaik), I'll take ownership since my tests
indicate it is related to timer priorities.

Comment 13

[Syd Logan]

Fixed

Comment 14

[leger]

I no longer see a crash with 2000-02-28-09-M15 Linux build.  Marking Verified.