Update PMP version

[lee1043]

Update pcmdi_metrics version to 3.4.1 in meta.yaml

[xylar]

@lee1043, thanks for the suggestion! We won't have another E3SM-Unified release until November but I have made a note of this change to include it then. There's no problem with keeping this PR open until then but just to note that the change won't happen for a while.

[lee1043]

@xylar thanks. No problem with keeping this PR open, just wanted to leave a note for any future update.

[xylar]

@lee1043, in the latest pcmdi_metrics, you are pinning a specific numpy verison:
https://github.com/conda-forge/pcmdi_metrics-feedstock/blob/main/recipe/meta.yaml#L37
That won't work for us in the next E3SM-Unified, where we will have numpy >=2.0,<3.0, and it is not very fair to other packages to have such a specific pinning of such a common package. What is the timeline for numpy 2.x support and can you make sure you don't pin such a specific version in the future to play nicer with other packages?

Much appreciated!

[lee1043]

@xylar thank you for pining me. I recall this numpy version restriction was mostly because of cdms. We are working on removing cdms dependency from PMP, and once completed, I think that will release the numpy version restriction. We are aiming to get this done by mid Jan 2025 to be better aligned with the next E3SM unified env version timeline. Would this work for you?

[xylar]

@lee1043 that's what I was hoping to hear! Thanks!

Also, I patched cdms2 and genutil for numpy 2.0 on Friday so you may want to see how that goes even in the interim before you move away from cdat.

[lee1043]

@xylar that is great to hear, thank you for letting me know! Is the patched cdms2 available through conda channel?

[xylar]

@lee1043, yes:
conda-forge/cdms2-feedstock#92

[xylar]

Please let me know if you run into trouble. I patched every issue I came across but the testing on conda-forge may not cover everything.

[lee1043]

@xylar thank you! I will test it and let you know how it works. Just a note, a while ago I opened an issue on CDMS regarding its conflict with numpy version CDAT/cdms#449, which I believe that must be resolved by your patch. Did I get that right? Is there specific CDMS version number assigned after for your patch?

[xylar]

I believe that issue was fixed in earlier patches by me:
conda-forge/cdms2-feedstock#81
conda-forge/cdms2-feedstock#85

No one is maintaining CDAT so I think your issues are unfortunately being lost in the ether. You could try to make issues on the conda-forge feedstock but I also have really limited time to maintain a dying and very frustrating package. Still, better than nothing...

[xylar]

Is there specific CDMS version number assigned after for your patch?

Not a version number (that would require the developers to create a new release) but a build number. If you install cdms2 and you have a constraint numpy >=2.0, you should get build 23, e.g.:

$ conda create -y -n test "python=3.10" "numpy>=2.0" cdms2
...
 cdms2              conda-forge/linux-64::cdms2-3.1.5-py310h4f3777b_23
...

The end of the build string (py310h4f3777b_23) is the build number, 23.

[lee1043]

@xylar great to know, thank you! I agree it is not exciting to maintain a dying package.

[lee1043]

@tomvothecoder it looks like the same numpy version issue applies to xcdat (https://github.com/xCDAT/xcdat/blob/96bc649ad59750ec37b959d9d4890891b54c62c7/conda-env/dev.yml#L14) when @xylar plans to upgrade numpy version in E3SM unified.

[tomvothecoder]

@tomvothecoder it looks like the same numpy version issue applies to xcdat (xCDAT/xcdat@96bc649/conda-env/dev.yml#L14) when @xylar plans to upgrade numpy version in E3SM unified.

Thanks, yes I plan on removing the constraint soon. It was added because xESMF was not compatible with numpy>=2.0.0 but that has since been fixed.