Bug#1099743: Wayland shell stops repainting the screen until monitor layout (or input devices?) change

[Nicolas Dandrimont]

Hi,

On Sun, Mar 9, 2025, at 00:28, Jeremy Bícha wrote:
> On Fri, Mar 7, 2025 at 7:03 AM Nicolas Dandrimont <olasd at debian.org> wrote:
>> This morning, I've upgraded the gnome-related packages from the 48 beta series
>> to the 48 RC series and rebooted my laptop.
>>
>> Since then, I've had a baffling issue where the screen would stop repainting,
>> until I unplug *or* replug my dock (which reconnects an external monitor,
>> keyboard and mouse). I'm assuming what puts gnome-shell back on track is the
>> change to the monitor layout, but I haven't really isolated the behavior yet
>> (and I've now downgraded packages back to the ~beta versions to be able to work
>> ;-)).
>
> I was experiencing similar behavior with 48 Beta. Notably, it is much
> more easily triggered during the hour-long Night Light transition. I
> see that you emailed early in the morning so maybe you were
> experiencing that. I recommend disabling Night Light until this issue
> is fixed.

I do not use Night Light.

> I also agree with Simon that sometimes the screen freezing seems to be
> triggered by high CPU load.

I have not had the RC packages installed for long enough to compare the effects
of load / no load. The bug triggered almost immediately, but I was starting my
working day and spinning up VMs (that happen to have their storage on a fast
USB-3 drive) at that point. The fact that unplugging or replugging the dock
fixed the issue made me initially believe a USB-related bug. Overall, the
correlation with I/O load sounds plausible.

> I would like to downgrade the severity of this bug since this bug is
> preventing 48 RC from reaching Testing. I believe it is probably that
> 48 RC is at least a bit better than 48 Beta in other areas. That isn't
> saying that this bug isn't important and even Release Critical, but
> just that it may not be new with 48 RC and staying on 48 Beta isn't
> necessarily better for people using Testing. Also, I believe we will
> want to reassign this bug to mutter. (Technically marking the bug as
> found in 48 Beta would be equivalent to downgrading so I think that's
> what I would do instead.)

I think it makes sense to reassign to mutter (I didn't have the time to isolate
the bug between mutter and gnome-shell, so I went with the "user-visible"
component).

To me, this bugs feels new under 48 RC, which is why I set the severity
explicitly to prevent 48 RC from reaching testing until the maintainers could
come up with an analysis of the issue.

All things being equal otherwise (and specifically the system load), the issue
would trigger within a minute or two under 48 RC, and has not triggered at all
during the time I've been using the 48~beta series (which, granted, has only
been 4 or 5 days at this point, but 4 or 5 days of doing similar work).

I of course defer to your opinion as a maintainer, but while my system had the
RC packages installed, it was all but unusable.

While composing this reply I was doing a build of mutter with the patch from
upstream merge request 4321 that Simon has pointed out upthread (phew, lots of
tests during build ;-)). I've just restarted my session with it, and replicated
the VM-related load, I have not noticed any issue with screen repaints yet.

Cheers,
-- 
Nicolas Dandrimont