use pytest rather than python -m pytest

[jeromedockes]

part of changes discussed in #738

[jeromedockes]

the failure is not related to this but to #748

[Vincent-Maladiere]

Do we also need to discuss adding a "python" or "src" folder to avoid importing the local files rather than using the installed package?

[jeromedockes]

Do we also need to discuss adding a "python" or "src" folder to avoid importing the local files rather than using the installed package?

this PR already prevents importing the local files when testing in the CI. adding a src folder would be an extra safety net to catch this kind of mistake in case we mess up the CI in the future, so it is not strictly necessary if the CI remains configured correctly. I still think it could be marginally useful because a misconfigured CI that passes when it should fail can be hard to detect before users report defects after a release. However, Gaël was not a fan because it would make the layout of skrub more different from scikit-learn. I don't have a very strong opinion, happy to discuss it some more. Locally during development, the package is likely to be installed in editable mode so the files in the repo will be imported during tests anyway, unless we use something like tox. It could be useful to add a tox config file for contributors who wish to run the tests in isolated virtual envs (with different python and dependencies versions) and empty test directories locally (contributors who don't install or run tox wouldn't be affected). Here also I don't think it is strictly necessary nor have a strong opinion about it.

[Vincent-Maladiere]

Sounds fair! Thank you for these insights. I guess we can safely merge this.