Bug#1072512: mutter: flaky autopgktest

[Simon McVittie]

On Fri, 06 Sep 2024 at 23:40:15 +0200, Paul Gevers wrote:
> On Mon, 3 Jun 2024 10:12:07 -0400 =?UTF-8?Q?Jeremy_B=C3=ADcha?=
> <jeremy.bicha at canonical.com> wrote:
> > The tests passed on retry and libxcursor is no longer blocked from
> > migrating. Sorry that the autopkgtests are flaky. :(
>
> The ratio at which mutter (at least on amd64) is failing on ci.d.n is
> unacceptable. (The ratio where I'm starting to doubt about severity is about
> 1/6 to 1/8 failures, but recent history is more like 1/2.)

Looking at the recent failure logs, the failures all seem to be in the
"stacking" test suite, which has two semi-common failure modes:

- sometimes mutter tries to create a texture with a negative size for
  reasons that I tried to investigate but do not understand
  (#1077800 aka #1078359)

- and sometimes the stacking (window management) tests just aren't reliable
  (#1050022, #1050023, similar random failures)

None of the individual test failures are particularly frequent, but there
are enough affected test cases that there's a high probability of at least
one of them failing in any given buildd build or autopkgtest run.

To mitigate this, I'm going to try something I had been considering for
a while: treat the whole "stacking" test suite as flaky. Unfortunately it
just isn't stable enough to be used as a QA gate.

    smcv