@bouweandela all done the Pixi work here - as expected, pys2index is not playing ball, but there are ways we can replace it (I really don't want to spend the work to rebuild it with the new abseil TBF); one alternative is sklearn - @schlunma likes that lib:

from sklearn.neighbors import BallTree
import numpy as np

points = np.radians([
    [51.5074, -0.1278],  # London
    [48.8566, 2.3522],   # Paris
])

tree = BallTree(points, metric='haversine')

query = np.radians([[51.5, -0.1]])

dist, ind = tree.query(query, k=1)
print(ind)

I suggest we mark that cmorizer Not Working until we find a replacement for pys2index, and skip the test here, so we can crack on with support for Python 3.14? 🍺

as expected, pys2index is not playing ball, but there are ways we can replace it (I really don't want to spend the work to rebuild it with the new abseil TBF)

We might also be able to replace this with our existing unstructured regridding features, which have grown really powerful.

I suggest we mark that cmorizer Not Working until we find a replacement for pys2index, and skip the test here, so we can crack on with support for Python 3.14? 🍺

Agreed. We should definitely not postpone Python 3.14 any longer because of a single CMORizer. If we want to be super thorough, we could provide a pixi environment that includes pys2index for users that need to run this CMORizer (that pulls in Python 3.13).

thanks @schlunma 🍺 I'll crack on here with an Xfail on that test and a mention in the docs; TBF I think having a separate Pixi env just for that cmorizer may be overkill, but then again am also quite new to Pixi mentality - @bouweandela what you reckon?

It would be nice to at least open an issue about replacing pys2index with the proposed steps to do so, if we plan to do that.

The problem seems to be that the main dependency of pys2index, s2geometry needs a rather specific version of abseil conda-forge/s2geometry-feedstock#35 (comment), but abseil is globally pinned on conda-forge so it may not be possible to build s2geometry at all with the most recent conda-forge global pins. A possible solution would be to try static linking instead.

It would be nice to at least open an issue about replacing pys2index with the proposed steps to do so, if we plan to do that.

The problem seems to be that the main dependency of pys2index, s2geometry needs a rather specific version of abseil conda-forge/s2geometry-feedstock#35 (comment), but abseil is globally pinned on conda-forge so it may not be possible to build s2geometry at all with the most recent conda-forge global pins. A possible solution would be to try static linking instead.

given how poorly maintained pys2index (so the top stack) is I doubt we should pay any effort, buds - unless we take over its maintenance, but frankly, am not keen on that, am already maintaining too many packages that I can afford 😁