1186820 - Firefox crashes at shutdown when running under xvfb-run (assert at jemalloc.c:1610)

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

xfvb-run ./mach mochitest dom/promise

At shutdown it hits

Hit MOZ_CRASH() at /home/khuey/dev/gecko-dev/memory/mozjemalloc/jemalloc.c:1610

It appears to be coming from a double free in Cairo.

Comment 1

[Milan Sreckovic [:milan] (needinfo for best results)]

I don't know what the importance of this is.  Hints?

Comment 2

[Kartikaya Gupta (email:[e-mail])]

I haven't seen this crash, and I've been running various reftests and mochitests under xvfb-run recently. The importance (for me at least) is that it's handy to run firefox headless over ssh, and this is a good way to do that. "Developer productivity" etc.

Comment 3

[Ben Kelly [:bkelly, not reviewing]]

I get this on a daily basis.  I am running in a headless linux instance in vmware.  (Which is why I need xvfb-run.)

Comment 4

[Ben Kelly [:bkelly, not reviewing]]

I guess I should note I've been getting this error for about a year... Thanks for filing Kyle!

Comment 5

[Steve Fink [:sfink] [:s:]]

This cost me a bunch of time because I hadn't seen it before, and I didn't uncover it until resolving a bunch of other bugs in my patch, so I naturally assumed it was my fault. I had no idea it was xvfb-related, so my bugzilla searches failed to find this.)

shu also reports seeing this, and that using Xnest works around it. glandium recommends Xephyr instead; I don't know if this problem occurs with Xephyr, but I would guess not.

Comment 6

[Benoit Girard (:BenWa)]

This would be handy for OrangeHunter for the same reason as Comment 2.

Comment 7

[Mike Hommey [:glandium]]

(In reply to Kyle Huey [:khuey] ([e-mail]) (UTC+8 July 17-25, expect delays) from comment #0)
> xfvb-run ./mach mochitest dom/promise
> 
> At shutdown it hits
> 
> Hit MOZ_CRASH() at
> /home/khuey/dev/gecko-dev/memory/mozjemalloc/jemalloc.c:1610
> 
> It appears to be coming from a double free in Cairo.

In-tree or system cairo? Thanks to Gtk+3, we now use both.

Comment 8

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

In-tree, before the gtk3 switch.  I haven't tested afterwards.

Comment 9

[Steve Fink [:sfink] [:s:]]

I have seen it both before and after the gtk3 switch, though I only looked closely at the pre-gtk3 version. (And there it looked like a double-free, which I poked at a bit with rr.)

Comment 10

[Kyle Huey (Exited; not receiving bugmail, old account, do not use)]

This seems to have gone away.

Comment 11

[Ben Kelly [:bkelly, not reviewing]]

I think it was the migration to gtk3.