Restored beta warn checks

[vfdev-5]

Context:

• 4774fe3 broke warning check
• We can use pytest --noconftest and test is passing again, but I decided to get rid of pytest due to other possible side-effects.

[vfdev-5]

This prevents warnings from v2 imported in common_utils

[NicolasHug]

Good catch!

[pmeier]

Oh shoot, thanks Victor for catching that. LGTM, but lint CI is failing: https://app.circleci.com/pipelines/github/pytorch/vision/23407/workflows/1cb0d350-44cd-42b0-a402-e7c63c96cd87/jobs/1835063

[vfdev-5]

Flaky test here (Unit-tests on Linux CPU / tests (3.9) / linux-job (pull_request)):

2023-02-20T09:32:45.8999693Z �[36m�[1m=========================== short test summary info ============================�[0m
2023-02-20T09:32:45.9000296Z �[31mFAILED�[0m test/test_transforms_v2_functional.py::�[1mTestKernels::test_scripted_vs_eager[cpu-rotate_bounding_box-19]�[0m - AssertionError: Tensor-likes are not close!
2023-02-20T09:32:45.9000550Z 
2023-02-20T09:32:45.9000639Z Mismatched elements: 1 / 24 (4.2%)
2023-02-20T09:32:45.9000978Z Greatest absolute difference: 1.3532526992321436e-06 at index (1, 0, 0) (up to 1e-06 allowed)
2023-02-20T09:32:45.9001360Z Greatest relative difference: 1.4774978678428505e-05 at index (1, 0, 0) (up to 1e-06 allowed)
2023-02-20T09:32:45.9001529Z 
2023-02-20T09:32:45.9001607Z The failure occurred for item [0]
2023-02-20T09:32:45.9001844Z Failure happened for the following parameters:
2023-02-20T09:32:45.9001988Z 
2023-02-20T09:32:45.9002078Z (tensor([[[ 0.0000, 10.4270,  9.6026, 17.1515],
2023-02-20T09:32:45.9002272Z          [ 0.0000, 12.8995, 16.0000, 22.1969],
2023-02-20T09:32:45.9002449Z          [ 0.0000, 11.3209, 11.8092, 21.0413]],
2023-02-20T09:32:45.9002562Z 
2023-02-20T09:32:45.9002643Z         [[ 0.0916,  6.0138, 16.0000, 18.8870],
2023-02-20T09:32:45.9002829Z          [ 2.2459,  9.2190, 13.4401, 13.7892],
2023-02-20T09:32:45.9003039Z          [16.0000,  8.7480, 16.0000, 12.8995]]], dtype=torch.float64), (29, 16))
2023-02-20T09:32:45.9003262Z (tensor([[[ 0.0000, 10.4270,  9.6026, 17.1515],
2023-02-20T09:32:45.9003450Z          [ 0.0000, 12.8995, 16.0000, 22.1969],
2023-02-20T09:32:45.9003626Z          [ 0.0000, 11.3209, 11.8092, 21.0413]],
2023-02-20T09:32:45.9003818Z 
2023-02-20T09:32:45.9003900Z         [[ 0.0916,  6.0138, 16.0000, 18.8870],
2023-02-20T09:32:45.9004084Z          [ 2.2459,  9.2190, 13.4401, 13.7892],
2023-02-20T09:32:45.9004304Z          [16.0000,  8.7480, 16.0000, 12.8995]]], dtype=torch.float64), (29, 16))
2023-02-20T09:32:45.9004557Z format=BoundingBoxFormat.XYXY
2023-02-20T09:32:45.9004769Z spatial_size=(29, 16)
2023-02-20T09:32:45.9004965Z angle=-87
2023-02-20T09:32:45.9265486Z �[31m= �[31m�[1m1 failed�[0m, �[32m73886 passed�[0m, �[33m73524 skipped�[0m, �[33m1629 xfailed�[0m, �[33m4605 warnings�[0m�[31m in 1463.79s (0:24:23)�[0m�[31m =�[0m
2023-02-20T09:33:03.5461688Z ##[error]Process completed with exit code 1.
2023-02-20T09:33:03.5488532Z Prepare all required actions

[NicolasHug]

Instead of adding more ad-hoc checks on a separate test suite, I'd prefer going for a more robust solution which would allow to test more scenarios, as the one proposed in #7269. I'll try to open a PR today

[vfdev-5]

I do not think it worth adding new code like https://github.com/scikit-learn/scikit-learn/blob/main/sklearn/utils/_testing.py#L747 in torchvision just to run some files like python check_file.py inside a single pytest test call.

If you point is about using again pytest framework instead of manually coding with warnings.catch_warnings(record=True) as w:, it is possible and can remove 1-2 lines. The point of removing pytest was to ensure no mixing with contest.py etc

[NicolasHug]

We need a lot more checks than those written in this PR. Each of these additional checks would require a separate script to properly test it without risking side-effects. This is untractable.

The testing util I want to introduce doesn't have to be as complicated as the one in scikit-learn, we can trim it down for it to look more like https://github.com/pytorch/hub/blob/a8c0f7ccfde4b8ea6cda812f5f61dd82bdeeda65/test_run_python_code.py#L35-L41

[NicolasHug]

I followed-up with #7288, LMK if it's OK with you @vfdev-5

[pmeier]

Closing this since it is superseded by #7288 and the latter was merged.