search

kinnala / scikit-fem

Simple finite element assemblers
https://scikit-fem.readthedocs.io
BSD 3-Clause "New" or "Revised" License
470 stars 76 forks source link

fix split_bases and split_indices for ElementVector #1091

Closed HelgeGehring closed 6 months ago

HelgeGehring commented 6 months ago

At those two lines it should be the number of dimensions of the ElementVector not the number of dimensions of the mesh, right? (I'm having a 3D vector in a 2D mesh)

kinnala commented 6 months ago

Yes, ElementVector dim kwarg is not used too much in the examples/tests.

HelgeGehring commented 6 months ago

Yay, thanks for merging so quickly! Could you do a minor release? I'd like to use this :)

kinnala commented 6 months ago

I'm still considering whether we should also add numpy<2 at the same time because the release is in 2-3 months and I expect there to be issues with "pip install scikit-fem[all]". I already fixed most of the issues within scikit-fem in #1090 but matplotlib and meshio are currently broken so it might affect UX here unless it is done. What do you think?

HelgeGehring commented 6 months ago

Pinning now would also lead to quite a bit of trouble - if someone has numpy pre-installed, pinning now would lead to an older version of scikit-fem being installed, which would cause even more trouble :/ Thus, it would probably lead for most users to have more problems (as for example anaconda has numpy pre-installed) :/

I'd guess the best would be to just go on releasing and add a warning at import which lists the current problems and suggests downgrading numpy in case? Matplotlib will probably be fixed when numpy 2 is released, but meshio I don't know :(

kinnala commented 6 months ago

I don't think that should happen. I won't drop support for numpy<2 within scikit-fem any time soon as the changes are very minimal and the users of anaconda or other older numpy version should be able to install whatever version, no matter what numpy version they use.

I'm proposing to add numpy<2 for the short time that the issues with the dependencies are solved. Something along the lines: https://pythonspeed.com/articles/numpy-2/

kinnala commented 6 months ago

Perhaps it is not necessary and I'm just overthinking. I've been also considering to fork meshio because there are some things I'd like to fix there but I am not sure whether I have the time to maintain such a fork. Perhaps I'd just fork the parts relevant to scikit-fem to make it less popular for other users.

kinnala commented 6 months ago

There is now a release 9.0.1 coming in few hours.