Support arm64 architecture on macOS

[khaeru]

This PR responds to #531.

• Improve .backend.jdbc:

  • Pass -Djava.library.path=… when starting the JVM, giving two directories containing the required binary libraries (with the extension .so on Linux; .dll on Windows; and .dylib on macOS):

    • The …/apifiles/Java/api/ directory within the user's GAMS installation directory.
    • The ixmp/backend/jdbc/ directory within the package itself, which contains (very old) bundled copies of the same files.

    The second directory is redundant if the first one is present (i.e. if a GAMS installation exists). To confirm this, the PR currently deletes the bundled libraries entirely. These commits could also be dropped instead.

• Expand the CI matrix to include both x86_64 and arm64 architectures on the macOS.
  • Use an updated version of iiasa/actions/setup-gams per Handle macOS arm64 architecture in setup-gams actions#13. This version installs an architecture-appropriate GAMS, and handles caching automatically.
  • Exclude macos-latest (arm64) for Python 3.8 and 3.9, as there are no pre-built wheels for JPype1 for this combination.

How to review

• Read the diff.
• Note the CI checks all pass.
• message_ix usage: Install ixmp from this branch on a macOS arm64 system with an arm64 version of Python. Run the test suite or other code and confirm the code functions as expected.
• Scenario Explorer usage: Consider implications for non-message_ix deployments; particularly, for the Scenario Explorer. Is GAMS installed in these contexts? Will the absence of the .dylib, .so, and/or .dll files bundled with the ixmp Python package affect these deployments?

PR checklist

• Continuous integration checks all ✅
• Add or expand tests; coverage checks both ✅
• Add, expand, or update documentation.
• Update release notes.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.7%. Comparing base (f08447f) to head (97d9f30).

❗ Current head 97d9f30 differs from pull request most recent head 1d40c22

Please upload reports for the commit 1d40c22 to get more accurate results.

Additional details and impacted files

@@           Coverage Diff           @@
##            main    #532     +/-   ##
=======================================
- Coverage   98.8%   98.7%   -0.1%     
=======================================
  Files         44      44             
  Lines       4804    4803      -1     
=======================================
- Hits        4747    4744      -3     
- Misses        57      59      +2     

Files	Coverage Δ
ixmp/backend/jdbc.py	97.2% <100.0%> (-0.2%)	⬇️

... and 4 files with indirect coverage changes

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 98.8%. Comparing base (cf56d03) to head (caaaa3d).

Additional details and impacted files

@@          Coverage Diff          @@
##            main    #532   +/-   ##
=====================================
  Coverage   98.8%   98.8%           
=====================================
  Files         44      44           
  Lines       4796    4813   +17     
=====================================
+ Hits        4739    4756   +17     
  Misses        57      57           

Files	Coverage Δ
ixmp/backend/jdbc.py	97.4% <100.0%> (+<0.1%)	⬆️
ixmp/model/gams.py	96.2% <100.0%> (+0.3%)	⬆️

[khaeru]

@glatterf42 let me know whether you think there is urgency to include this in the 3.9 release.

We would need to invite and wait for reviews from:

• Ravit, Sidd, and/or Setu —to check behaviour on their machines.
• Philip or someone else —to address the point about impacts on Scenario Explorer deployments.

Given this might take a few days, and the first three are probably capable to use it from main in the period between the 3.9 and 3.10 releases, I would lean to "no". OTOH, if we have users in the upcoming workshop who have macOS arm64 machines, this might (not necessarily) save some support overhead.

If we wait, then I will also wait until after the workshop to request a review of iiasa/actions#13.

[glatterf42]

I'd also say it's likely not urgent. Workshop users will learn about GitHub anyway and if they need to use this branch, this will be a good opportunity to practice. They could also then provide feedback about their install process.
However, I think we can already invite review from our team members and if these come in very soon, we can maybe still include this in 3.9 :)

[khaeru]

@ravitby @SiddharthJoshi-Git @setupelz —could you please try the steps above under "How to review" / "message_ix usage" and let us know the results?

@phackstock —could you please comment on the questions above under "How to review" / "Scenario Explorer usage"?

[phackstock]

Thanks for pinging me @khaeru. As far as I know there's no GAMS installation for the Scenario Explorer.

Will the absence of the .dylib, .so, and/or .dll files bundled with the ixmp Python package affect these deployments?

I also don't think that that would be an issue.
Green light from my side to merge. If there are any issues that would arise from the Scenario Explorer side, we can always pin a previous version and open an issue here.

[khaeru]

Great, thanks for confirming.

[khaeru]

@glatterf42, since the other requested colleagues seem unable to review this over the past ~6 weeks, I propose we merge and wait for feedback that may come as they/others use it from main. If we don't get positive confirmation before we start prep for the next release, then we can ask again.

[glatterf42]

Sounds good to me, just one possible typo :)