Numpy v2 support

[maresb]

I would like to be able to use the Python API in Python environments that I have already upgraded to Numpy v2. Currently I can't do this with a release version due to

copernicus-marine-toolbox/pyproject.toml

Line 27 in 3dee4b5

numpy = ">=1.23.0,<2.0.0"

The version notes for v1.3.1 mention pydata/xarray#9179 as blocking, but it has since been merged and released.

Before finding this repo on GitHub I chatted with Leïla on the website. She pointed me to pydata/xarray#9545, which seems to be a performance regression.

Question: If I were to unpin numpy<2 on my fork (using v1.3.3 by default, or possibly another tag), pin to the latest xarray, and accept the performance regression, are there any other blockers or potential for incorrect data that I should be aware of?

[veenstrajelmer]

Thanks for creating this issue. I think it is good if copernicusmarine would support numpy v2 soon. This is also one of the blockers for python 3.13 support: #158

[renaudjester]

Hi @maresb,

Indeed it's a performance regression that should be fixed in the next release of xarray. It basically add 5 seconds every time you do: xarray.open_dataset

We haven't noticed any other problems. However, it hasn't been extensively tested as we discovered the two bugs one after the other. And the first one was on the data itself (no errors or anything).

So theoretically, yeah you are good to go if you accept the performance regression!

Don't hesitate to report any other issues you find :D

[maresb]

I haven't tested this so thoroughly, but so far it seems like it's working. In case anyone else wants to give this a try:

pip install maresb-copernicusmarine-unpin-numpy