Fix pre-release tests

[flying-sheep]

@ilan-gold this should unblock you

See #1434 (doesn‘t fix that one though)

[ilan-gold]

I'm fine with upper bounding packages. I will take a look at the issue. Is @ivirshup ok with upper bounding this? I can't think of a good reason not to but he often has something up his sleeve!

[flying-sheep]

I’m usually the “no upper bounds” guy. Here’s the reasoning why it’s OK in cases like this:

Scipy is unlikely to change their implementation to not use copy=True, so it’s 1. on us to make this work and 2. a !=1.13.0.rc1 constraint won‘t be sufficient.

Therefore all anndata versions that don’t have the future fix are incompatible with all future scipy versions. Therefore a temporary upper bound (with a comment linking to the issue) is the right call.

In an ideal world, all past anndata versions would have the same bound, but that’s impossible to do.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.09%. Comparing base (8f4c755) to head (fbfa3b0).

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1435      +/-   ##
==========================================
- Coverage   86.29%   84.09%   -2.21%     
==========================================
  Files          36       36              
  Lines        5612     5602      -10     
==========================================
- Hits         4843     4711     -132     
- Misses        769      891     +122     

Flag	Coverage Δ
gpu-tests	?

Flags with carried forward coverage won't be shown. Click here to find out more.

see 10 files with indirect coverage changes

[ivirshup]

Tbf, I did also write the scipy patch that introduced this.

[ivirshup]

In an ideal world, all past anndata versions would have the same bound, but that’s impossible to do.

I believe conda-forge actually allows this

[flying-sheep]

Yeah, I was thinking about PyPI. There it would also kinda maybe be possible by yanking everything and releasing .post1 releases but I don't think that's a reasonable standard to hold ourselves to and it would force people to change their lock files