r? @marcoieni

@rustbot ready

I'm not sure if it's worth adding this complexity. I can imagine us easily introducing bugs because of this feature.

Aren't there other ways to speed up the compilation? E.g. cache dependencies or use a pre-compiled binary.

I'm not sure if it's worth adding this complexity. I can imagine us easily introducing bugs because of this feature.

Could you clarify what kind of bugs you have in mind? I can't imagine how this could cause any bugs.

Aren't there other ways to speed up the compilation? E.g. cache dependencies or use a pre-compiled binary.

A pre-compiled binary would IMO introduce too much implementation complexity, due to having to handle cache invalidations. An easy solution would be to use rust-cache. I was a bit vary of adding it to our CI, but as long as we would pin the hash and we only use it in the matrix computation step, I suppose that it should be fine. If you're ok with that, I will send a PR.

I prefer using the rust-cache action wrt to selecting features like this 👍

Could you clarify what kind of bugs you have in mind? I can't imagine how this could cause any bugs.

E.g. forget to add configuration flags around, which could cause compilation error. I have seen codebases that don't test all permutation of features that don't catch these errors, but probably in this CI would me more difficult.

Still I would prefer rust-cache to keep the code simpler 👍

E.g. forget to add configuration flags around, which could cause compilation error. I have seen codebases that don't test all permutation of features that don't catch these errors, but probably in this CI would me more difficult.

Right, I mean sure, there could be a compilation error, but that's not a bug =D We would compile the crate both with and without the single feature, so we would find out soon. But rust-cache should be simple.

Obsoleted by #139819.