Yup, good call. Turns out I missed two objects in np.testing.
numpy/numpy#18919 should fix the issue.

Thanks @V0lantis, and thanks for the quick fix @BvB93.

The numpy nightlies are built once a week, on Friday night typically, so a build which includes the fix will show up at https://pypi.anaconda.org/scipy-wheels-nightly/simple/numpy/ on Saturday. Easiest way to check is to use the version hash (everything after .g in the wheel name, don't ask me why the g is there ....). @V0lantis can you update this PR on Saturday, removing the # type: ignore's? And then I think this is good to merge.

I have made the changes and tested upon the commit from @BvB93. it works. I pushed and next Saturday I will merge master to trigger a new workflows.

Why is the mypy test being run before the actual test run itself? I'd like to see it the other way around so that we can see any test failures first, any typing failures are less consequential.

In addition, 3.10 has now been released to beta. This CI run was originally written to identify issues with as-yet unreleased Python. Therefore, I think we need to amend the "Nightly" Action item to run off Python 3.10. If this is going to cause issues with the mypy run, then either we need to create a new Action, or move the mypy run to another Action item.

Why is the mypy test being run before the actual test run itself? I'd like to see it the other way around so that we can see any test failures first, any typing failures are less consequential.

Probably because mypy is >10x faster to run. Both need to pass, so it doesn't matter much.

In addition, 3.10 has now been released to beta. This CI run was originally written to identify issues with as-yet unreleased Python. Therefore, I think we need to amend the "Nightly" Action item to run off Python 3.10. If this is going to cause issues with the mypy run, then either we need to create a new Action, or move the mypy run to another Action item.

+1 for upgrading to 3.10. It's orthogonal to this PR, but if GitHub Actions has 3.10 available it's a one-liner change. Can you try to include it here @V0lantis? If it doesn't work easily, let's do it in a separate PR.

+1 for upgrading to 3.10. It's orthogonal to this PR, but if GitHub Actions has 3.10 available it's a one-liner change. Can you try to include it here @V0lantis? If it doesn't work easily, let's do it in a separate PR.

Yeap, going to try this

Why is the mypy test being run before the actual test run itself? I'd like to see it the other way around so that we can see any test failures first, any typing failures are less consequential.

In addition, 3.10 has now been released to beta. This CI run was originally written to identify issues with as-yet unreleased Python. Therefore, I think we need to amend the "Nightly" Action item to run off Python 3.10. If this is going to cause issues with the mypy run, then either we need to create a new Action, or move the mypy run to another Action item.

Unfortunately, we cannot because numpy still doesn't have a release for Python3.10 :

ERROR: Could not find a version that satisfies the requirement numpy (from versions: none)
ERROR: No matching distribution found for numpy
Error: Process completed with exit code 1.

You can check the workflow here

Maybe this should be added in the Nightly release @rgommers ?

Ah yes, let's open a separate issue for upgrading the nightly to py3.10, and then work on first getting NumPy nightly wheels for 3.10 up. Not difficult, but it's an extra CI job there I think.

The nightly CI used to build numpy from source, I don't think it took too long.

Ah yes, going back to building from source should be okay as well. Indeed much faster than building scipy itself, so not a big issue.

Ok so should I do an inplace build of numpy in the Nightly python job ?

No, a regular pip install . is preferred. Or, given you don't have a numpy clone yet:

pip install https://github.com/numpy/numpy/archive/refs/heads/main.zip

pip install https://github.com/numpy/numpy/archive/refs/heads/main.zip

This doesn't work, so I made a workaround. It is not pretty nor fast but it works fine. I tried with
wget -r -np -R "index.html*" --cut-dirs=6 https://github.com/numpy/numpy/archive/refs/heads/main.zip but it added a lot of overhead so I went back to this solution which works fine.

The failure with wget laid in the version of numpy with wget was :

_get_build_requires
2046
      self.run_setup()
2047
    File "/tmp/pip-build-env-15otxrx6/overlay/lib/python3.10/site-packages/setuptools/build_meta.py", line 248, in run_setup
2048
      super(_BuildMetaLegacyBackend,
2049
    File "/tmp/pip-build-env-15otxrx6/overlay/lib/python3.10/site-packages/setuptools/build_meta.py", line 142, in run_setup
2050
      exec(compile(code, __file__, 'exec'), locals())
2051
    File "setup.py", line 50, in <module>
2052
      MAJOR, MINOR, MICRO = FULLVERSION.split('.')[:3]
2053
  ValueError: not enough values to unpack (expected 3, got 1)

Okay I was being lazy, somehow the right invocation isn't easy to find. You need to use git+https://github.com/.... says https://stackoverflow.com/questions/20101834/pip-install-from-git-repo-branch.

Ok so for me, it ready to be merged