379666 - crash on startup with Minefield 20070504-10

Status: VERIFIED FIXED

(Core :: Graphics, defect)

Description

[Radosław Szkodziński]

User-Agent:       Links (2.1pre28; Linux 2.6.21-nicus2 x86_64; x)
Build Identifier: Minefield 20070504-10

Minefield crashed before startup, on both clean and used profile.
I've excluded a lot of possible causes, including various libs (pango, cairo)
and mozconfig options.

Backtrace (not too enlightening):
http://wklej.org/id/b4f709e113
Debug log:
http://wklej.org/id/7f08001de0

The system is non-multilib x86-64, thus I cannot test mozilla.org builds.
Firefox 2.0.0.3 works fine for reference.


Reproducible: Always

Comment 1

[Mats Palmgren (inactive)]

# #0  0x00002aaaad72a3e5 in ?? ()
# #1  0x00002ab2c91deeea in g_slice_free1 () from /usr/lib/libglib-2.0.so.0
# #2  0x00002ab2c99c9f70 in ?? () from /usr/lib/libpango-1.0.so.0
# #3  0x00002ab2c906dc52 in g_object_unref () from /usr/lib/libgobject-2.0.so.0
# #4  0x00002aaaab210e2f in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #5  0x00002aaaab20a6e5 in gfxFontCache::DestroyFont () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #6  0x00002aaaab20c92a in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #7  0x00002aaaab20c957 in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #8  0x00002aaaab20ca48 in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #9  0x00002aaaab208bc4 in gfxFontCache::Shutdown () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #10 0x00002aaaab20e060 in gfxPlatform::Shutdown () from /usr/lib64/mozilla-firefox/components/../libthebes.so
# #11 0x00002ab2c80e4f83 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #12 0x00002ab2c8112ef1 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #13 0x00002ab2c8112970 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #14 0x00002ab2c80dcb36 in PL_DHashTableEnumerate () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #15 0x00002ab2c8112de6 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #16 0x00002ab2c811294f in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #17 0x00002ab2c810ec00 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #18 0x00002ab2c80e6a02 in NS_ShutdownXPCOM_P () from /usr/lib64/mozilla-firefox/libxpcom_core.so
# #19 0x00002ab2c7b2d8b2 in ?? () from /usr/lib64/mozilla-firefox/libxul.so
# #20 0x00002ab2c7b31831 in XRE_main () from /usr/lib64/mozilla-firefox/libxul.so 

Comment 2

[Mats Palmgren (inactive)]

Radosław, is this a recent regression?

Comment 3

[Mats Palmgren (inactive)]

The stack looks like bug 378436.

Comment 4

[Radosław Szkodziński]

Could you propose some bisection points?

I recall some build of 2007-03 working, but that might be too old for bisection purposes and the build takes a lot of time.

Comment 5

[Mats Palmgren (inactive)]

FWIW, bug 378436 is from a 2007-04-23 build.
Beginning of April seems like a good starting point.
You can use http://bonsai.mozilla.org/cvsqueryform.cgi to see what was
checked in and http://tinderbox.mozilla.org/showbuilds.cgi?tree=Firefox
to avoid checking out a broken build (red).  Thanks for helping out.

BTW, what did you mean by "non-multilib"?

Comment 6

[Radosław Szkodziński]

Non-multilib means w/o 32-bit compatibility libraries, so I can't test 32-bit builds.

I've tried build from 20070401, and it works.
Will bisect further.

Comment 7

[Radosław Szkodziński]

Seems like 20070427 and 20070501 also work, though both crash w/o a profile _once_
(with an entirely different traceback), but work correctly later on.

Comment 8

[Radosław Szkodziński]

No, the traceback is actually similar:

#0  0x00002aaaab5e13e5 in ?? ()
No symbol table info available.
#1  0x00002ade9e6f3eea in g_slice_free1 () from /usr/lib/libglib-2.0.so.0
No symbol table info available.
#2  0x00002ade9efe0099 in ?? () from /usr/lib/libpango-1.0.so.0
No symbol table info available.
#3  0x00002ade9e582c52 in g_object_unref () from /usr/lib/libgobject-2.0.so.0
No symbol table info available.
#4  0x00002adea3386615 in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#5  0x00002adea337ed27 in gfxFontCache::DestroyFont () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#6  0x00002adea33821b3 in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#7  0x00002adea33821e5 in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#8  0x00002adea33822c6 in ?? () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#9  0x00002adea337e3ce in gfxFontCache::Shutdown () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#10 0x00002adea3383931 in gfxPlatform::Shutdown () from /usr/lib64/mozilla-firefox/components/../libthebes.so
No symbol table info available.
#11 0x00002ade9d1f7f4e in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#12 0x00002ade9d225b0d in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#13 0x00002ade9d225568 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#14 0x00002ade9d1efa25 in PL_DHashTableEnumerate () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#15 0x00002ade9d225a03 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#16 0x00002ade9d225547 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#17 0x00002ade9d2219a8 in ?? () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#18 0x00002ade9d1f99b4 in NS_ShutdownXPCOM_P () from /usr/lib64/mozilla-firefox/libxpcom_core.so
No symbol table info available.
#19 0x00002ade9ca357b5 in ?? () from /usr/lib64/mozilla-firefox/libxul.so
No symbol table info available.
#20 0x00002ade9ca397a6 in XRE_main () from /usr/lib64/mozilla-firefox/libxul.so
No symbol table info available.
#21 0x00002ade9cf79323 in __libc_start_main () from /lib/libc.so.6
---Type <return> to continue, or q <return> to quit---
No symbol table info available.
#22 0x0000000000400681 in ?? ()
No symbol table info available.
#23 0x00007fff0e2d32c8 in ?? ()
No symbol table info available.
#24 0x0000000000000000 in ?? ()
No symbol table info available.

So, it's probably the same bug as bug 378436, but at different time.
These versions also crash at exit.

Comment 9

[Radosław Szkodziński]

20070415 is unaffected by any of these bugs

Comment 10

[Radosław Szkodziński]

20070420 is also affected by the bug 
(only shows when started w/o any profile, as with 20070501)

Comment 11

[Radosław Szkodziński]

Ok, I've narrowed down the bug to have appeared between builds from 20070415 UTC and 20070416 UTC.

The most suspect patch is this:
https://bugzilla.mozilla.org/show_bug.cgi?id=374332

Comment 12

[Mats Palmgren (inactive)]

Radosław, thank you so much for narrowing down the regression range.
That's really helpful to us.   So, to be absolutely sure it's bug 374332,
can you either apply the patch from that bug to your 0415 build (which should
make it crash) or undo it in the 0416 build (crash should go away).

Comment 13

[Michael Wu [:mwu]]

Attachment: Remove incorrect PR_UnloadLibrary call

GConf should not be unloaded once it is loaded. The comments in the Init function say:

    // Don't unload past this point, since GConf's initialization of ORBit
    // causes atexit handlers to be registered.

It causes crashes when libgconf-2 is found but libgnome-2 is not. This is because having libgnome-2 installed also pins libgconf-2 and conceals the problem.

Comment 14

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Comment on attachment 266961 [details] [diff] [review]
Remove incorrect PR_UnloadLibrary call

Please comment this out rather than removing it so it's clear what we're not doing that we really ought to be.  With that, r=dbaron.

(Note that it's possible restoring this may block  not doing a restart for extension manager installation/update on startup.)

Comment 15

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

And thanks for fixing this.

Comment 16

[Michael Wu [:mwu]]

(In reply to comment #14)
> (From update of attachment 266961 [details] [diff] [review])
> Please comment this out rather than removing it so it's clear what we're not
> doing that we really ought to be.  With that, r=dbaron.
> 
Sure. I'll add a comment about it too.

Comment 17

[Michael Wu [:mwu]]

Attachment: Remove incorrect PR_UnloadLibrary call, v2

PR_UnloadLibrary commented out with a comment explaining why.

Comment 18

[Michael Wu [:mwu]]

Checking in nsSystemPrefService.cpp;
/cvsroot/mozilla/extensions/pref/system-pref/src/gconf/nsSystemPrefService.cpp,v  <--  nsSystemPrefService.cpp
new revision: 1.12; previous revision: 1.11
done

Comment 19

[Radosław Szkodziński]

And it is so.