Does scipy need to be added as a required dependency due to being optional in sparse version 0.17.0?

[glemieux]

I recently stumbled upon an issue in which scipy was not found when building an environment through a conda-lock file:

  File "/home/glemieux/local/miniconda3/envs/landusedata-env/lib/python3.13/site-packages/xesmf/smm.py", line 224, in add_nans_to_weights
    m = weights.data.to_scipy_sparse().tolil()
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/home/glemieux/local/miniconda3/envs/landusedata-env/lib/python3.13/site-packages/sparse/numba_backend/_coo/core.py", line 1187, in to_scipy_sparse
    import scipy.sparse
ModuleNotFoundError: No module named 'scipy'

Going over to the sparse issues I found this comment that states that this package made scipy optional with sparse=0.17.0:

Hello, we switched to having SciPy as an optional dependency in 0.16.0, and it's only used on conversions to/from scipy. This was missed in the conda-forge feedstock; but it is fixed in the 0.17.0 feedstock.

If you're explicitly using scipy in your package/script, please include it in your requirements explicitly.

[aulemahal]

Oh, seems like it does! Thanks for the issue.