Will a version of the XP65 Conda environments maintain a NumPy version < 2

General
,
1

Hello,

I am currently using the 26.01 version of the analysis3 Conda environment on the NCI, mainly because NumPy v1.26.4 is installed (i.e. a version < 2). However, I’ve received a message that this particular version of the analysis3 environment will soon be depreciated.

Will a version of the analysis3 environment using a NumPy version < 2.x be maintained in the future?

The reason I’m not upgrading to a newer version of NumPy is because it is incompatible with a statistical package I’m using that includes Fortran libraries that must be compiled. I haven’t been able to compile the package with NumPy 2.0+.

Thank you…

2

Hi Chris,

I’ve had a bit of a dig and I think that this constraint might begin to cause you more compatibility issues in the not too distant future: I had a look through the conda environment for packages depending on numpy 2.0 or greater:

❯ pixi list --json | jq -r '
  .[] |
  .name as $parent |
  .depends[]? |
  select(test("^numpy\\s+>=2([.,\\s]|$)")) |
  "\($parent): \(.)"
'
astropy-base: numpy >=2.0
cf-python: numpy >=2.0.0
cfdm: numpy >=2.0.0
climpred: numpy >=2.0.0
cvxpy-base: numpy >=2
esmvaltool-python: numpy >=2.0.0
mo_pack: numpy >=2.3.2,<3.0a0
numdifftools: numpy >=2.0
pcmdi_metrics: numpy >=2.0.0,<3
pyfive: numpy >=2
pygmt: numpy >=2.0
pyspharm: numpy >=2.4.3,<3.0a0
python-stratify: numpy >=2
rioxarray: numpy >=2
satpy: numpy >=2.0.0
segregation: numpy >=2.0.0
shap: numpy >=2
timezonefinder: numpy >=2,<3
tobler: numpy >=2.0
virtualizarr: numpy >=2.1.0
wrf-python: numpy >=2.4.2,<3.0a0
xcdat: numpy >=2.0.0,<3.0.0
zarr: numpy >=2

There are a few packages in there which we can’t easily relax or remove, and I assume this list will only get more restrictive as time goes on. It might be worth trying to get in touch with the authors of the library to see if they plan to add numpy 2.0 compatibility?

In the meantime, we should be able to leave this package available: cc @rbeucher

3

We are following SPEC0 scientific recommendations and NumPy <2.0 is out of the support window since the third quarter of 2025. Of course, we can be flexible and keep that one around a bit longer, but it’s definitely worth chasing the developers of the package and ask if they have a plan in the future.

4

Thank you…

5

Thank you. Yes… I agree that this package needs to be upgraded.