We broke scanpy’s tests

[flying-sheep]

Please make sure these conditions are met

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest version of anndata.
• (optional) I have confirmed this bug exists on the master branch of anndata.

Report

The tests were broken in #1266

pip install anndata  # anndata[testing] doesn‘t exist

import anndata.tests.helpers

Traceback:

ModuleNotFoundError: No module named 'zarr'

The fix is not to add zarr to scanpy’s test dependencies, but to make the zarr import conditional

Afterwards, it’s time to finally fix #646

Versions

anndata 0.10.5

[ivirshup]

I don't know that it's a bug to import from a private module that has optional dependencies (which are specified under anndata[test]) and for that to error.

Seems related to why I had to pin anndata fairly high here: scverse/scanpy#2816

There really is an issue of needing a similar set of helpers, but not wanting these helpers to be public because they're not considered stable.

Also, I really think the immediate answer is to add zarr with a TODO comment to the scanpy test dependencies.

[flying-sheep]

@ilan-gold is on it. Let’s not have zarr feature-creep its way into scanpy’s minimal test deps. It’s the day of hotfixes it seems

[flying-sheep]

Once pytest 8.0 is out, we can finally tackle the whole reorganization:

1. make the anndata repo use a src layout
2. create the public test module (anndata.test_utils or anndata.testing or testing.anndata) with an associated extra
3. have scanpy depend on that
4. move anndata’s tests out of the package

[ivirshup]

I'm not against doing a testing module, though I do think we have to be selective about what goes in it and probably need to test it more thoroughly.

I just don't think that this problem neccesarily needs an immediate bug fix release.

[flying-sheep]

I'm not against doing a testing module, though I do think we have to be selective about what goes in it and probably need to test it more thoroughly.

totally! those are all reasons why we should do it!

I just don't think that this problem neccesarily needs an immediate bug fix release.

I don’t like accumulating hacks, I guess. Doing it the other way means we can forget about it, and can’t merge the minimum-deps branch before we have removed zarr again.

[ivirshup]

Could do a .post release?

I don’t like accumulating hacks, I guess. Doing it the other way means we can forget about it, and can’t merge the minimum-deps branch before we have removed zarr again.

I'm not sure how that PR would be more affected by this?

[ivirshup]

I've checked the current 0.10.x branch against scanpy, I think a post branch seems appropriate here. I can make one from 08e7b94.

Any objection?

[flying-sheep]

I think I don’t fully understand. If you just mean calling the release .post instead of giving it a new patch version number, sure!

[ivirshup]

Huh: https://github.com/scverse/anndata/actions/runs/7670757953

[ivirshup]

Guess it's just flaky somehow?

https://github.com/scverse/anndata/actions/runs/7670870992/job/20907984190

Anyways, release made