Cristian Le

Is this with the default loader or the new one. If it's the same on the default, then we need to rethink it. One reference would be what do setuptools...

> @LecrisUT any other thoughts on how we ought to proceed here? I've committed the test to the repo so hopefully it's easy enough to reproduce the issue if you...

About 2nd and 3rd point, it is subtle but basically it involves how the `Loader`, `submodule_search_locations`, etc. interact. You remember the last PR on this, it basically addressed that.

I think you should go ahead with the `xfail` approach. If needed I will open a branch and you can move the target to there.

Yes, I need to look again at the `importlib` code over the weekend, but as far as I remember: - The `Loader`/`Reader` approach would require full implementation of these classes....

Hmm, wouldn't `cpack` be more appropriate for such `sdist` step? Granted I never use cpack myself and that one is a weird one for users to get right. But then,...

Ah, I guess you are making `site_package` folders part of the `CMAKE_PREFIX_PATH` and relying on the search using: ``` /*/(lib/|lib*|share)/cmake/*/ ``` Well, it seems like `lib` vs `lib64` is irrelevant...

https://github.com/scikit-build/scikit-build-core/blob/9ec17518f2917fee19e2efccb8ae046e3ada3740/.github/workflows/ci.yml#L48 You still have a few `sys.version_info < (3, 8)` stragglers. Also change the `pyproject.toml` itself

Not sure what policy is for this. Last year when python 3.7 was EOL we still kept compatibility since academic software are notoriously slow to adopt. Should we adopt dropping...

> I'd probably like to wait till 1.0 (which is probably ~3 months away) 🤩 > One issue is that bumping here causes cmake and ninja Oh, `ninja` got ported...