CI: drop python 3.9, numpy 1.21 #266
Conversation
Drop python 3.9 and 3.10, drop numpy 1.21 Not adding python 3.13 because ndonnx only supports 3.12 at the moment.
|
scikit-learn have now dropped 3.9 and 1.21 |
| # Min version of dask we need dropped support for Python 3.9 | ||
| # There is no numpy git tip for Python 3.9 or 3.10 | ||
| python-version: ${{ (inputs.package-name == 'dask' && fromJson('[''3.10'', ''3.11'', ''3.12'']')) || (inputs.package-name == 'numpy' && inputs.xfails-file-extra == '-dev' && fromJson('[''3.11'', ''3.12'']')) || fromJson('[''3.9'', ''3.10'', ''3.11'', ''3.12'']') }} | ||
| python-version: ['3.11', '3.12', '3.13'] |
There was a problem hiding this comment.
Yes, as an attempt to trim the CI size a bit. Experience over the last couple of months is that there were no issues which were python version dependent.
|
Am planning to merge this next week if this looks reasonable to people. A possible contention point is the size of the CI matrix: this PR proposes to drop python 3.9 and 3.10 and numpy 1.21; this way we'll test on python 3.11, 3.12 and 3.13 x numpy 1.26, "latest released", and numpy-dev. Jax, ndonnx, dask and pytorch: no changes, we only test with the latest released version. |
|
The linked scikit-learn issue has a clear plan, which is nice. It just dropped Python 3.9, so dropping Python 3.10 here seems quite aggressive - and probably not necessary? If you want to reduce the CI matrix, just remove 3.11 instead while keeping the lowest-supported version. |
|
Similarly, I think we should be testing NumPy 1.22 as the min. supported by sklearn. |
|
I spoke to Olivier and Tim yesterday and suggested that there shouldn't be any problems with array-api-compat and array-api-extra dropping more aggressively than scikit-learn, now that they have written down some reasonable rules. This matters to them as we are very close to them vendoring both libraries: scikit-learn/scikit-learn#30340 |
|
For comparison, in array-api-extra we define the following test envs tests-numpy1 = ["py310", "tests", "numpy1"]
tests-py310 = ["py310", "tests"]
tests-py313 = ["py313", "tests"]
tests-backends = ["py310", "tests", "backends"]
tests-cuda = ["py310", "tests", "backends", "cuda-backends"]and use the non-CUDA envs in the CI matrix matrix:
environment: [tests-py310, tests-py313, tests-numpy1, tests-backends]Our |
closes gh-230
This likely affects scikit-learn, scikit-learn/scikit-learn#30895 (comment)