search

scipy / oldest-supported-numpy

Meta-package providing the oldest supported Numpy for a given Python version and platform
BSD 2-Clause "Simplified" License
55 stars 33 forks source link

Determine if specific constraints for Windows are necessary #65

Closed cmarmo closed 1 year ago

cmarmo commented 2 years ago

Dear maintainers, I am opening this issue as a follow-up of the discussion started in https://github.com/scikit-learn/scikit-learn/issues/24669#issuecomment-1282981971.

I am not sure this is a real issue, but it probably needs some clarification outside the scikit-learn issue where it has been discussed. I try to sum up my understanding.

There are no specific constraints on numpy versions for Windows on setup.cfg.

They were added in https://github.com/scipy/oldest-supported-numpy/pull/44, then reverted in https://github.com/scipy/oldest-supported-numpy/pull/47, after discussions in https://github.com/scipy/oldest-supported-numpy/issues/45. Perhaps they are indeed not needed?

Feel free to close if this is not relevant. Thanks for your attention.

cmarmo commented 2 years ago

Just linking here https://github.com/scikit-learn/scikit-learn/issues/24604

        ERROR: Cannot install numpy==1.21.6 and numpy==1.22.3 because these package versions have conflicting dependencies.

        The conflict is caused by:
            The user requested numpy==1.22.3
            The user requested numpy==1.21.6

Could this be related to scikit-learn using old-supported-numpy but not scipy so different versions of numpy are requested in this particular combination on Windows?

rgommers commented 2 years ago

Could this be related to scikit-learn using old-supported-numpy but not scipy so different versions of numpy are requested in this particular combination on Windows?

@cmarmo yes indeed, that is exactly it. See https://github.com/scikit-learn/scikit-learn/issues/24604#issuecomment-1285227874

rgommers commented 1 year ago

This was discussed further and resolved on the linked scikit-learn issue. There's nothing to change here, so I'll close this issue. Thanks @cmarmo!