Re: [Patch] Fix for #96915, #105290

[Alexander Larsson <alexl redhat com>]

On Wed, 2003-09-17 at 04:24, Martin Wehner wrote:
> Hello,
> 
> here's a patch that fixes nautilus crashing on startup when it is
> shutdown immediately (like with -q, --no-default-window &&
> show_desktop=off).
> 
> gnome_vfs_async_job_map_get_job doesn't handle the case that there's no
> job map yet/anymore gracefully. All of its callers assume it returns
> NULL if you try to remove a non-existing job. But it asserts the job_map
> is not null and bails out via g_assert() otherwise. But a non_existing
> job_map is just a special case of a non-existing job in it.
> 
> If Nautilus is immediately shutdown after startup remove_trash_volume()
> in nautilus-trash-volume tries to cancel the
> gnome_vfs_async_find_directory jobs it fired up. But at that point the
> jobs are already finished and the map destroyed - leading to the
> assertion failure. This cancellation takes only place if the handle is
> not null. This handle is set to null in the callback that is called when
> an operation is finished, so no cancel should be executed for a finished
> operation. That works fine most of time, but if a shutdown is in
> progress it doesn't. The callbacks to be called after the job are
> registered with g_idle_add() and executed when there's nothing more
> important to do. Well, as it turns out, during shutdown of Nautilus
> there's always more important stuff to do until the application exits
> and the main loop is destroyed. So the callbacks are never called,
> handle never set to null and cancel called on all trash volume jobs at a
> time where there's no job_map anymore. That's bug #96915, which I can
> reproduce by setting show_desktop=off and starting nautilus with
> nautilus --no-default-window (as gnome-session does).
> 
> Another way to fix this would be to register the callbacks in a way that
> execution is guarantied in a timely manner for the finish callbacks of
> the async operations and make the trash code work as intendend.
> Actually, the whole of nautilus-directory-async.c uses the same
> mechanism of flags which get nullified in the callbacks (see the stack
> of #105290).

Yes, its possible we should use g_idle_add_full() instead, with a
priority higher than G_PRIORITY_DEFAULT_IDLE. Maybe just use
G_PRIORITY_HIGH_IDLE? This needs some though and looking at other
priorities.

> But in any case job_map_get_job should not crap out when there's no job
> map.

Yeah. I'm applying this on HEAD and gnome-2-4.

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
 Alexander Larsson                                            Red Hat, Inc 
                   alexl redhat com    alla lysator liu se 
He's a maverick guitar-strumming astronaut on a search for his missing sister. 
She's a sarcastic Buddhist detective descended from a line of powerful 
witches. They fight crime!