WIP add pyresample

[rickgrubin-noaa]

WIP

Addresses:

• [INSTALL]: Need pyresample in the python package of spack-stack #1732
• [INSTALL]: (Attempt to) Update Python to 3.12 #2005

Changes to common/packages.yaml that result in a successful build:

     python:
       require:
-      - '@3.11'
+      - '@3.12:'

     py-flit-core:
       require:
-      - '@3.8.0'
+      - '@3.11.0:'

     py-numpy:
       require:
-      - '@1'
+      - '@2'

     py-setuptools:
       require:
-      - '@73.0.1'
+      - '@79:

+    py-pandas:
+      require:
+      - '@3'
+      - ~performance

+    py-msgpack:
+      require:
+      - '@1.1.1'

Note that py-numexpr would build with the gcc compiler but not the oneapi compiler.

Attempting to concretize for python@3.13 resulted in a number of errors, primarily because a version of setuptools could not be satisfied in some package recipes.

[mathomp4]

Looking at my Python 3.13 attempt, I had to go to:

• py-numpy@2.1:
• py-setuptools@80:
• py-pandas@2.2.3:

So looks like py-resample needed a newer pandas at least.

But we now reach the py-pandas ~performance issue. Using +performance leads down to needing LLVM around.

I guess the question is: which one should be default?

cc @climbfuji

[rickgrubin-noaa]

But we now reach the py-pandas ~performance issue. Using +performance leads down to needing LLVM around.

I guess the question is: which one should be default?

I do not see a need for having to build LLVM for anything EPIC-related, at least at this point in time.

[climbfuji]

But we now reach the py-pandas ~performance issue. Using +performance leads down to needing LLVM around.
I guess the question is: which one should be default?

I do not see a need for having to build LLVM for anything EPIC-related, at least at this point in time.

I agree. I would make the default ~performance. If individual applications (templates) or sites (site configs) need the pandas performance, then they can enable that in the appropriate place. Maybe you can make it a prefer: +performance rather than a require.

[mathomp4]

I am getting close to getting a Python 3.13 unified-env built on discover. I hope. (Currently in Rust + Python hell).

Once I can finally get that happy, I'll start trickling in these updates as well!

If nothing else, pandas ~performance will at least negate the LLVM issue. :)