Created attachment 348290 [details] [review]
check whether primary logical monitor is available

Check for null pointers and rather terminate properly,
instead of a segfault.

Review of attachment 348290 [details] [review]:

::: src/backends/meta-backend.c
@@ +132,3 @@
+    {
+      meta_fatal("Unable to get primary logical monitor\n");
+    }

Does this happen on hotplug without any prior configuration, or is mutter deliberately configured to have no active outputs?

Either way, this should not be fatal. Instead, change the "meta_monitor_manager_is_headless()" to check "logical_monitors" instead of "monitors", to also include the case where the "headlessness" is a result of an explicit configuration, and not the physical state.

At the moment this is fatal anyway, because if primary == nullptr, accessing it's members will cause segfault here (next line):
https://github.com/GNOME/mutter/blob/master/src/backends/meta-backend.c#L132

Regarding the problem itself:
This is not a headless mode and the crash happens during post_init with x11 backend when Xorg is started in the root-less mode with nvidia proprietary drivers. Apparently Xorg and/or nvidia driver fails to detect/report display properly (this is what I don't know exactly).

Anyway, the ticket is rather about proper handling of null pointers, so mutter exits cleanly.

(In reply to Igor Ingultsov from comment #3)
> At the moment this is fatal anyway, because if primary == nullptr, accessing
> it's members will cause segfault here (next line):
> https://github.com/GNOME/mutter/blob/master/src/backends/meta-backend.c#L132

Right, but changing a crash to an abort is not right here, since we should handle this case better.

> 
> Regarding the problem itself:
> This is not a headless mode and the crash happens during post_init with x11
> backend when Xorg is started in the root-less mode with nvidia proprietary
> drivers. Apparently Xorg and/or nvidia driver fails to detect/report display
> properly (this is what I don't know exactly).
>

Could you check what it is that is reported? This crash *should* only be able to happen if Xorg reports that there are monitors connected, but none are activated. This could probably happen (by using xrandr directly for example) and should be handled. Changing meta_monitor_manager_is_headless() to check the logical monitors instead of monitors would fix this. Does that fix the issue?

> Right, but changing a crash to an abort is not right here, since we should handle this case better.
I don't see how fixing of logic is related to the null poitner check. It is still possible, that this pointer could be null, then it has to be checked.

Also I don't see how meta_monitor_manager_is_headless is related, because center_pointer is getting called from here in my case:
https://github.com/GNOME/mutter/blob/master/src/backends/meta-backend.c#L362
and NOT from here:
https://github.com/GNOME/mutter/blob/master/src/backends/meta-backend.c#L153

(In reply to Igor Ingultsov from comment #5)
> > Right, but changing a crash to an abort is not right here, since we should handle this case better.
> I don't see how fixing of logic is related to the null poitner check. It is
> still possible, that this pointer could be null, then it has to be checked.
> 
> Also I don't see how meta_monitor_manager_is_headless is related, because
> center_pointer is getting called from here in my case:
> https://github.com/GNOME/mutter/blob/master/src/backends/meta-backend.c#L362
> and NOT from here:
> https://github.com/GNOME/mutter/blob/master/src/backends/meta-backend.c#L153

Ah, I see, sorry about that. I guess the easiest thing would be to just if (!primary) return; after getting the primary in center_pointer().

It does not make sense to work around this problem here (even though I've tried). Because then gnome-shell crashes a little bit later in other place, because there are not logical monitors:

+ Trace 237283

I'd rather add also another check in meta_backend_real_post_init somewhere before calling center_pointer, to check for logical monitors and if there are none, mutter can't continue and should fail.

Marking as duplicate of bug 730551. See the patches attached there.

*** This bug has been marked as a duplicate of bug 730551 ***