conda-forge / scikit-image-feedstock

A conda-smithy repository for scikit-image.
BSD 3-Clause "New" or "Revised" License
4 stars 25 forks source link

Upstream replied for skipped test(s) #79

Open step21 opened 3 years ago

step21 commented 3 years ago

FYI https://github.com/scikit-image/scikit-image/issues/4775

jakirkham commented 3 years ago

Sounds like the next step is try running all tests here again and see what fails?

step21 commented 3 years ago

Yeah I guess. I am super busy the next two weeks with finishing my thesis, so maybe someone else could open a PR?

hmaarrfk commented 3 years ago

Does conda-forge really care about releasing this patch?

I am a core dev at scikit-image, but ultimately, I feel like issues like this should be used as motivation for upstream developers to cut new release.

Unfortunately, build infrastructure is hard to maintain.

So if you cut 2 releases a year, then you can't be expected that it will always be a smooth process.

I would love to be able to cut a 0.18.x release, but I haven't had time for the last month or so.

My point is more that, i'm unsure if you, @step21, want us to release a patch that makes your code with scikit-image "unreleased" version 0.18. It will make your code very fragile to a specific build made by conda-forge, and will not be usable by those that follow upstream releases.

step21 commented 3 years ago

@hmaarrfk Others are more involved with scikit-image, but generally - no, I don't think we want a special release. It has been like this for a while, so a bit more won't hurt anything. We could run tests again to see which errors still happen - then you or whoever from scikit-image can cut a new release whenever you get to it. (or not if you cannot find the error)

hmaarrfk commented 3 years ago

If you want to just the CIs then you can make a merge request labelled as "draft" or "WIP". You definitely don't need permissions to do this.

Patching isn't always easy (especially since alot of cleanup has happened in scikit-image recently).

Are you interested in running tests on a particular platform? Or x86 64bit?

step21 commented 3 years ago

Yeah, I know that I can do a PR. I just wanted to bring it to the rest of the conda-forge teams attention, as I have zero time atm. In the recipe file, it is noted on which combination the test failed, f.e.e python 3.8 macos. But if I do a PR I'd probably just run everything.