Bug#1138999: mutter: frequent test failures on armhf

[Simon McVittie]

Control: found -1 49.4-1
Control: tags -1 + help

On Sat, 06 Jun 2026 at 13:33:13 +0300, Adrian Bunk wrote:
>https://buildd.debian.org/status/logs.php?pkg=mutter&arch=armhf
>
>There is no clear pattern why/how it fails, but builds are
>supposed to be non-flaky.

This is not a new problem, so I'm marking it as found in testing to 
avoid it blocking migration. #1121518 is essentially the same thing.

Anyone who can help to debug/improve the tests is more than welcome to 
do so, especially the porting teams for the affected architectures (I 
don't think anyone in the GNOME team is running it on armhf). 
Unfortunately upstream only runs the test suite on relatively powerful 
x86 PCs, so it's probably making timing assumptions that don't hold on 
slower CPUs.

On one hand, yes, of course we want tests that reliably pass. On the 
other hand, Debian's self-imposed requirements say that every package 
must work on every architecture where it's successfully built, and for 
the less-widely-used architectures (particularly armhf) the build-time 
tests are the only evidence we have that a package can work (nobody is 
routinely testing the full GUI on 32-bit ARM). Realistically our choices 
are tests that aren't fully reliable, or no tests at all; neither of 
these is a desirable situation to be in, so it's a choice between two 
bad options.

I've queued a commit for the next upload to disable the 
wayland-subsurface test, which seems to be particularly bad on armhf for 
whatever reason.

     smcv