Huite
commented
3 months ago Xarray still supports 3.9, that's convincing to me. I might have been a bit cavalier when bumping to 3.10 -- I must've had a reason, but I cannot recall at the moment...
Apparent numpy 2.0 failures seem to be indexing with a boolean array. That should be tedious but straightforward to fix, I guess.
The single failure for numpy < 2 looks like a weird numba failure. Tests do run with NUMBA_DISABLE_JIT, I guess.
Good call about the outdated lockfile. If we configure a bot to update it regularly, we have the best of all worlds. Pretty sure this works for the Ribasim repo, but I haven't quite found the option yet.
I'll take a look into the Python 3.9 pixi thing, it should just work.
This is to test if xugrid still works on Python 3.9. I think it does.
There are 75 failing testcases when running it as is: https://github.com/Deltares/xugrid/actions/runs/9891762563/job/27322884959
When requiring
numpy<2.0.0there is 1 failing testcase: https://github.com/Deltares/xugrid/actions/runs/9891771516/job/27322914549I could not get this to work in pixi, but I think it would be valuable to re-add py3.9 support since it will take >1 year before being EOL: https://endoflife.date/python. Also, if xugrid drops python 3.9 support, this means that all Deltares packages that use xugrid should drop python 3.9 support and it is still a bit early for that in my opinion.
Furthermore, I think this PR makes clear that xugrid does not support numpy>=2.0.0. This is typically an issue that does not show up when using a lockfile. I see the benefits of a lockfile, but I think it is wise to have a parallel pip-only testcase for say python 3.12 that checks with all latest dependency versions so you are always aware of new dependency versions that cause issues. Also, I think the numpy dependency restriction (upper limit) should be introduced. I consider upper limits very dangerous, but as long as it is temporary I guess it is fine.