338343 - Firefox does not shut down when closing the last window

Status: RESOLVED DUPLICATE of bug 337648

(Toolkit :: Startup and Profile System, defect)

Description

[Dennis Jacobfeuerborn]

User-Agent:       Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9a1) Gecko/20060516 Minefield/3.0a1
Build Identifier: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.9a1) Gecko/20060516 Minefield/3.0a1

When I close the last Firefox window the firefox-bin process doesn't quit. When I do a kill on the process and restart Firefox the Talkback window appears. I submitted the following Talkback: TB18802400M

Reproducible: Always




OS:Fedora Core release 5.89 (Rawhide)

Comment 1

[Philip Withnall (unavailable)]

How long did you wait for the process to end?

Comment 2

[Dennis Jacobfeuerborn]

I waited for several minutes. After some experimentation I found the following:

This occurs on all nightly builds after the 10th of May (checkin for bug 326273 ?) but not in the build before. What triggers this is a crash in the Nightly Tester Tools extension version 1.0.3 when closing Firefox. When I disable this extension everything is fine. When I enable it Firefox hangs on shutdown. When I enable it but disable the Talkback extension Firefox segfaults on shutdown.

Comment 3

[Philip Withnall (unavailable)]

You say Firefox segfaults if Talkback isn't enabled but that extension is? Could you build with debug symbols and get a backtrace?

Comment 4

[Philip Withnall (unavailable)]

Mozilla/5.0 (X11; U; Linux x86_64; en-US; rv:1.9a1) Gecko/20060517 Minefield/3.0a1 - Build ID: 0000000000
Confirmed, but I don't have the Nightly Tester Tools extension installed.

Comment 5

[Dennis Jacobfeuerborn]

I think the Nightly Tester Tools extension only serves as a catalyst in this case and that basically anything that causes Firefox to crash on shutdown will trigger the real problem which is the hanging of the Talkback extension.

When I attach gdb to the process (which is a non-debug build) I see Talkback on the callstack so it gets started but then apparently so confused that it locks up. Could this be some sort of race condition in which the threadmanager tries to shutdown threads at the same time as Talkback tries to gather whatever information it needs for its reports from them?

Comment 6

[Dave Townsend [:mossop]]

This is almost certainly caused by Nightly Tester Tools and a workaround for that extension can be seen at: http://users.blueprintit.co.uk/~dave/web/firefox/nightly/linuxcrash.html

The actual cause, to my best estimation is basically the same as bug 337648 though while it seems to only be an assertion on Windows, it is actually causing problems on Linux.

Essentially NTT uses some objects on a background thread, then the garbage collector cleans them up in a different thread.

For info I have some extra talkback incidents, though they are all pretty much the same:

TB19037388W
TB19037360Q
TB19037339Z
TB19037319G
TB19037215G

The hanging only started since bug 326273 but the actual problem existed before then.

Comment 7

[Boris Zbarsky [:bzbarsky]]

Cleaning things up via GC on the wrong threads is bad.  :(  I _know_ we have bugs on that, though.

Comment 8

[Darin Fisher]

So, can this bug be closed now that the actual cause is known to be with the nightly tester tools?

Comment 9

[Dave Townsend [:mossop]]

Well I think this is Nightly Tester Tools exposing a problem with threading in Mozilla. If those are already filed then this can just be duped there.

Comment 10

[Wayne Mery (:wsmwk)]

Dennis appears to be gone.
I never see a hang with NTT - using trunk or branch.

Comment 11

[Dave Townsend [:mossop]]

NTT was changed long ago to avoid using threading since it seems to cause so much problems. I think this is basically just bug 337648 though.