I think this is a good solution. The goal of #29935 is still sufficiently served by users' local tests and but other workflow runs (such as the ci-linux runs on release tags). @kliem any objections?

There were also two runs for the push on the dev branch (one for the push, and one for the tag), so I guess we will only see after merging this if it really works then.

I think this is a good solution. The goal of #29935 is still sufficiently served by users' local tests and but other workflow runs (such as the ci-linux runs on release tags). @kliem any objections?

I think the suggested changes are an improvement.

Overall, I still think that if people write a test that claims that something works for any random input of a type, it should. E.g. matrix addition should be entry-wise addition.

However, when I pushed for enforcing the random seed for CI, we still used patchbots that would occasionally fail anyway. Now, we use a more stable CI, I believe. If we are able to eliminate almost all noise from the CI, we should I think. And it's also good, if you don't get affected by problems not related to your PR.

Documentation preview for this PR is ready! 🎉
Built with commit: aba447a