Bug 386459 – [gnome-overlay] gnome-base/gnome-shell-3.2.0-r1 crashes in libmozjs185 (dev-lang/spidermonkey-1.8.5)

• Status: RESOLVED WORKSFORME
• Product: Gentoo Linux

[Description Tilo Prütz 2011-10-08 21:15:32 UTC]

After upgrading from Gnome 3.0 to Gnome 3.2 my Gnome-Shell stopped working. It always crashes on startup.

I will append the backtrace.

Reproducible: Always

[Comment 1 Tilo Prütz 2011-10-08 21:17:07 UTC]

Created attachment 289265 [details]
Backtrace

[Comment 2 Alexandre Rostovtsev (RETIRED) 2011-10-09 18:31:33 UTC]

Could you please
1. provide a backtrace from all threads (use the "t a a bt" command in gdb); and
2. say in what function the crash is occurring (e.g. if gdb prints something like "Program received signal SIGSEGV [...] in foo_bar ()", then the crash is in function foo_bar()).

[Comment 3 Tilo Prütz 2011-10-09 18:50:17 UTC]

for 1.: will attach

for 2.:

Core was generated by `gnome-shell --replace'.
Program terminated with signal 11, Segmentation fault.
#0  0xb6732884 in array_extra (cx=0x830e760, mode=MAP, argc=<value optimized out>, vp=0xac8fd140) at jsarray.cpp:2838
2838	jsarray.cpp: Datei oder Verzeichnis nicht gefunden.
	in jsarray.cpp

[Comment 4 Tilo Prütz 2011-10-09 18:51:04 UTC]

Created attachment 289401 [details]
Gnome-Shell backtrace for all threads

[Comment 5 Alexandre Rostovtsev (RETIRED) 2011-10-11 02:47:06 UTC]

Mozilla team, is there a way for line 2838 of jsarray.cpp in dev-lang/spidermonkey-1.8.5 to result in a segfault?

The only way that I can think of is if tvr.addr() returned an invalid pointer. However, looking at the definition of the AutoValueRooter class, that seems to be impossible :/

[Comment 6 Tilo Prütz 2011-10-11 06:49:44 UTC]

Something weird is happening. I put some debug code into jsarray.cpp (since I am not a C++ hacker it’s dumb) and now it does no longer crash :(.

In Objective-C (where I am from) this would be a good indicator for freed memory which is wrongfully still in use.

I will apply the patch just for the sake of completeness.

[Comment 7 Tilo Prütz 2011-10-11 06:50:39 UTC]

Created attachment 289509 [details, diff]
Debug-Output-Patch

[Comment 8 Tilo Prütz 2011-10-20 12:39:45 UTC]

The problem vanished with the last update (which included gnome-shell-3.2.1).
It now works without problems with the original spidermonkey-1.8.5 ebuild.

[Comment 9 Nico R. 2011-12-14 16:55:56 UTC]

Created attachment 295823 [details]
Backtrace for all threads with gnome-shell-3.2.1-r1 and spidermonkey-1.8.5-r1

I am seeing this problem with gnome-base/gnome-shell-3.2.1-r1 and dev-lang/spidermonkey-1.8.5-r1. Backtrace attached, but it looks like attachment 289401 [details].

I’d suggest to set the status of this bug report to REOPENED.

[Comment 10 Ian Stakenvicius (RETIRED) 2011-12-14 17:11:48 UTC]

(In reply to comment #9)
> Created attachment 295823 [details]
> Backtrace for all threads with gnome-shell-3.2.1-r1 and spidermonkey-1.8.5-r1
> 
> I am seeing this problem with gnome-base/gnome-shell-3.2.1-r1 and
> dev-lang/spidermonkey-1.8.5-r1. Backtrace attached, but it looks like
> attachment 289401 [details].
> 
> I’d suggest to set the status of this bug report to REOPENED.

Please re-emerge spidermonkey with USE="debug" and try again?  This will enable a bunch of 'assert' logic that will give a useful message to the segfault instead of just a failure.

[Comment 11 Nico R. 2011-12-14 22:34:38 UTC]

When I build spidermonkey with USE="debug", I get another segfault, which is described in bug 388521.