CI job matrix differ

[psychocoderHPC]

The job matrix for 48a66b0 (job_work.txt)
and
73642da (job_broken.txt)
differ but there is only one small fix which should not affect the CI matrix.

Due to the fact that the CI matrix is not equal we hit sometimes a configuration where nvcc + clang jobs failing. Will be addressed in PR #2295.

The problem is not that we found a broken combination the problem is that two CI runs are not reproduceable.

I added the job generator output to this issue. I did a diff and what I saw is that the alpine linux version differ for the runner, could it be that the version of the parwise testing library differ?

job_works.txt
job_broken.txt

[SimeonEhrig]

There is randomness somewhere in the allpairspy library, which generates the combinations. I didn't found out, where setting an initial seed does not solve the problem:

import random

random.seed(42)

[SimeonEhrig]

The problem will be solved by bashi: alpaka-group/bashi#25
I'm not sure, if I find a solution for the current implementation