Add pytest filtering markers for GPU to improve workflows

[liweintu]

To address issue #28 , we add pytest filtering markers for GPU based on its availability.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 9.92%. Comparing base (53382b8) to head (434932a).
Report is 442 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #72       +/-   ##
==========================================
- Coverage   33.13%   9.92%   -23.22%     
==========================================
  Files           3      10        +7     
  Lines         172     393      +221     
==========================================
- Hits           57      39       -18     
- Misses        115     354      +239     

Flag	Coverage Δ
unittests	9.92% <ø> (-23.22%)	⬇️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[liweintu]

The pytest mark filtering can work and pass now. But the CI encountered a permission issue when running the following codecov create-commit command. @alecandido Do we need to adjust some settings to grant the permission? Thanks.

==> Running command '/home/runner/work/_actions/codecov/codecov-action/v4/dist/codecov create-commit'
/home/runner/work/_actions/codecov/codecov-action/v4/dist/codecov create-commit --git-service github -Z
info - 2024-09-18 09:28:14,787 -- ci service found: github-actions
info - 2024-09-18 09:28:14,959 -- Process Commit creating complete
error - 2024-09-18 09:28:14,959 -- Commit creating failed: {"detail":"You do not have permission to perform this action."}
Error: Codecov: Failed to properly create commit: The process '/home/runner/work/_actions/codecov/codecov-action/v4/dist/codecov' failed with exit code 1

[scarrazza]

@liweintu I have reset the codecov token and now tests are working.

[liweintu]

Thanks @scarrazza! Now all checks can pass, we'll proceed to review and merge this PR.

[alecandido]

Why is Qibotn now depending on Qibojit?

[Vinitha-balachandran]

Qibotn had qibojit dependency earlier in the tests. So far we were installing qibojit seperately. Since, the previous version of workflow was skipping tests it was not an issue in github actions. But, now in the absence of qibojit tests will fail.

[alecandido]

But you should not run tests on Qibojit, since you're explicitly testing another backend (i.e. Qibotn).

Could you point me to the logs where tests are failing?

[liweintu]

I guess the question here is "Does Qibotn need to depend on Qibojit?". My quick answer is "No, it doesn't."

So I removed qibojit package, ran pytest, and encountered an error below.

        # Test qibo
>       qibo.set_backend(backend=config.qibo.backend, platform=config.qibo.platform)

tests/test_quimb_backend.py:45:
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
../../miniforge3/envs/py311/lib/python3.11/site-packages/qibo/backends/__init__.py:150: in set_backend
    GlobalBackend.set_backend(backend, **kwargs)
../../miniforge3/envs/py311/lib/python3.11/site-packages/qibo/backends/__init__.py:108: in set_backend
    cls._instance = construct_backend(backend, **kwargs)
../../miniforge3/envs/py311/lib/python3.11/site-packages/qibo/backends/__init__.py:230: in construct_backend
    raise_error(
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

exception = <class 'ValueError'>
message = "The 'qibojit' backends' provider is not available. Check that a Python package named 'qibojit' is installed, and it is exposing valid Qibo backends."

I guess it's due to qibojit being the first and default choice during backend construction. If my guess is correct, maybe we specify the backend to be numpy for test cases of quimb instead of using the default backend?

[alecandido]

I guess it's due to qibojit being the first and default choice during backend construction. If my guess is correct, maybe we specify the backend to be numpy for test cases of quimb instead of using the default backend?

This is a known problem, already solved recently qiboteam/qibo#1430

The point is that you should not encounter that problem, since you should set Qibotn as the backend, and not relying on the default one.

[Vinitha-balachandran]

Okay, will modify the tests to remove the dependency

[alecandido]

Ok, that's even fine by me, but maybe you wanted to change

qibotn/tests/config.py

Line 11 in f571fc5

qibo = Executor(backend="qibojit", platform="numpy")

Otherwise, you can decide to drop that file, and inline the values in the tests (that is a pretty good option).

[Vinitha-balachandran]

Can drop that file as it is doing nothing. Need to modify cuquantum side also.

[liweintu]

Yeah, I like the latter option because it explicitly avoids the dependency on qibojit.