Run Python 3.13 testbank

[veenstrajelmer]

First pre-release is available at the end of October according to https://peps.python.org/pep-0719/#schedule, final in October 2024:

• 3.13.0 beta 1: Wednesday, 2024-05-08 (No new features beyond this point.)
• 3.13.0 beta 2: Wednesday, 2024-06-05
• 3.13.0 beta 3: Thursday, 2024-06-27
• now (4-7-2024)
• 3.13.0 beta 4: Tuesday, 2024-07-16
• 3.13.0 candidate 1: Tuesday, 2024-07-30
• 3.13.0 candidate 2: Tuesday, 2024-09-03
• 3.13.0 final: Tuesday, 2024-10-01

Github available versions: https://raw.githubusercontent.com/actions/python-versions/main/versions-manifest.json

Testbank fails because of:

• ERROR: No matching distribution found for numpy>=2.0.0.dev0: Pip module fails to build with Python 3.13 numpy/numpy#24318 and cannot install numpy on python 3.13 numpy/numpy#25917 >> numpy will support python 3.13 approximately from may 2024 >> ENH: Support free-threaded python build (tracking issue) numpy/numpy#26157
• retried on 1-5-2024, now failing on scipy with ../scipy/meson.build:163:9: ERROR: Dependency "OpenBLAS" not found, tried pkgconfig and cmake >> Tracker: Support free-threaded CPython builds scipy/scipy#20669
• pandas 2.2.3 can be installed on py3.13 on windows, all older versions raise errors
• copernicusmarine requires zarr which requires numcodecs which does not yet support py3.13: Installation fails on Python 3.13 zarr-developers/zarr-python#2317 and Installation fails on Python 3.13 zarr-developers/numcodecs#593
• [BUG] installation fails on python 3.13 pydata/bottleneck#459
• cgi deprecated in python 3.13 ecmwf/multiurl#14 (comment)
• upon installation of netCDF4: ValueError: did not find HDF5 headers >> New post release to create Python 3.13 wheels Unidata/netcdf4-python#1370
• Python 3.13 is not supported mercator-ocean/copernicus-marine-toolbox#158
• copernicusmarine (and hydrolib-core and others) require numpy<2 but this does not support python 3.13
• xugrid requires numba: Python 3.13 numba/numba#9413
• check meshkernel: Future proofing package MeshKernelPy#190
• and probably more

If py3.13 testbank is green:

• add to general pytest workflow file (runs on every PR and merge)
• remove dedicated py313 workflow file (or maybe keep since it is prerelease)
• add py3.13 to supported versions in pyproject.toml
• updated readme/installation/contributing docs

For future reference: https://scientific-python.org/specs/spec-0000/

[rgommers]

Hi, I noticed this from an incoming link on a scipy issue. I just wanted to point out that there are numpy and scipy nightlies at https://anaconda.org/scientific-python-nightly-wheels/numpy/ and https://anaconda.org/scientific-python-nightly-wheels/scipy/. You should use those if you want 3.13 support without having to build from source. And never build releases from PyPI for a future Python version; build from main instead if you do choose to build from source.

[veenstrajelmer]

Hi @rgommers thanks for the heads up. I read something like that indeed, but I am not in a hurry. This issue is just meant to track the development of Python 3.13 support of the dfm_tools dependencies. Every now and then I run the testbank to see if it can already install all dependencies or not (with released versions). Once it can install without issues, I will check if all tests pass and then add 3.13 to the supported versions in the pyproject.toml.

Also I enjoy following the discussion on numpy/scipy boards, since it seems to be quite a thing to add 3.13 support. Thanks for all the hard work!