Change clifford to run with pytest

[eric-wieser]

We've dropped our nose dependency in clifford, and moved away from unittest-compatible tests in order to use pytest fixtures. Since python 2 is also not supported, the future module is no longer a dependency.

The upshot is our tests should be more granular.

[esc]

There are still some import nose errors. Have you already released the version of clifford that has nose removed?

[eric-wieser]

No, that hasn't been released yet. Do these tests run against both master and the released version, or only the released version?

If the former, I could revert my removal of nose and future from the required dependencies, and just let it install them needlessly for the master testing.

[esc]

These tests usually only run against the latest released version.

[esc]

For clifford specifically, the program will obtain all git tags from the 'blessed' remote and select the 'latest' one. The code is here:

https://github.com/numba/numba-integration-testing/blob/master/switchboard.py#L115

[eric-wieser]

It sounds then like the correct action is to leave this open until clifford pushes its next tag?

[esc]

@eric-wieser probably, yes.

[eric-wieser]

Clifford did a 1.1.0 release, although now I expect this to fail due to a missing conda-forge:sparse dependency, which I'm not sure how to add.

[esc]

@eric-wieser thanks for submitting this, I'll try to re-run this PR now.

[esc]

@eric-wieser for reference, you can add a string like "-c conda-forge sparse", to install sparse from conda-forge.

[esc]

I think the clifford tests are going through fine, but that is because sparse is listed as dependency in the setup.py and so we get:

Installed /home/circleci/repo/miniconda3/envs/clifford/lib/python3.8/site-packages/clifford-1.2.0-py3.8.egg
Processing dependencies for clifford==1.2.0
Searching for sparse
Reading https://pypi.org/simple/sparse/
Downloading https://files.pythonhosted.org/packages/08/73/34946ead922c072d7981be5b917f5871961a1b3bab4bc43dcd861de9ac23/sparse-0.9.1-py2.py3-none-any.whl#sha256=5d7e9bab68b20f63ba6e8c593e46864ebef3a0842fa26fb32b71e743e6c5c99e
Best match: sparse 0.9.1
Processing sparse-0.9.1-py2.py3-none-any.whl
Installing sparse-0.9.1-py2.py3-none-any.whl to /home/circleci/repo/miniconda3/envs/clifford/lib/python3.8/site-packages
Adding sparse 0.9.1 to easy-install.pth file

[esc]

@eric-wieser thanks for submitting a fix for this, will merge as soon as CI goes green!

[esc]

It looks like it has passed, but actually I see:

= 1 failed, 315 passed, 29 skipped, 8 xfailed, 1 xpassed, 5 warnings in 394.96s (0:06:34) =

[esc]

Looks like we may have to wait for: conda/conda#9665

[esc]

I have merged a workaround for the conda run bug: numba/texasbbq#7 @eric-wieser would it be possible to rebase this PR onto master (or merge master into it). This will hopefully show us the true state of these integration-tests.

[eric-wieser]

^ That should have the effect of re-running against master

[esc]

@eric-wieser yeah, I thought so too, but apparently CircleCI doesn't seem to operate like that.

[esc]

I triggered it manually now.

[eric-wieser]

Test failure is real, AttributeError: module 'numba' has no attribute 'numpy_support'.

Looks like the numba API changed on us.

[esc]

@eric-wieser indeed, we are in the middle of a 'refactor the world' update in Numba: numba/numba#5197

It is very likely that this broke clifford and lead to the failures you are seeing now. I would suggest to collect all the places where the Numba refactor broke clifford and submit these in an issue for discussion on the Numba issue tracker. This will allow us to decide if we broke a 'public' API and we need to fix Numba --- OR --- if clifford was using a private API in the first place and the fix needs to happen there.

And thanks very much for your continued understanding, patience and persistence --- we appreciate it!

[eric-wieser]

At any rate, this PR should be good to merge, right?

[esc]

@eric-wieser indeed! The errors are being reported again and the clifford tests no longer fail due to a missing pytest dependency!