Fix Numpy 2.0 compatibility issues

[speth]

Changes proposed in this pull request

Fix usage of functions that were removed or deprecated in NumPy 2.0.

We can make sure that at least some of the CI tests are using NumPy 2.0 once the official release is made on June 16 (numpy/numpy#24300), and then merge this PR.

If applicable, fill in the issue number this pull request is fixing

Partially resolves #1675. I think the issues with matching versions of NumPy for packaged versions of Cantera will need to be resolved in the various packaging scripts, but we can leave the issue open until those changes are made.

If applicable, provide an example illustrating new features this pull request is introducing

Checklist

• The pull request includes a clear description of this code change
• Commit messages have short titles and reference relevant issues
• Build passes (scons build & scons test) and unit tests address code coverage
• Style & formatting of contributed code follows contributing guidelines
• The pull request is ready for review

[bryanwweber]

Is this backwards compatible? In other words, will Cantera 3.1 require NumPy >=2.0, or can we support 1.x and 2.x simultaneously?

[speth]

This is meant to be backwards compatible, at least in terms of the source -- all the changes for us are pretty minor. I think the conda and PyPI packages will have to be specific to one version or another. I was hoping for your help in figuring some of that out. 😁

[bryanwweber]

I think the conda and PyPI packages will have to be specific to one version or another. I was hoping for your help in figuring some of that out. 😁

Yeah, the NumPy headers are different, right? I haven't looked into how that's meant to be handled. I'll see if there's anything out there, I assume NumPy have gotten out ahead of this.

[speth]

This may be easier than I thought. According to the NumPy Docs:

When you build wheels for your package using a NumPy 2.x version at build time, those will work with NumPy 1.xx.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 72.83%. Comparing base (9f48b07) to head (8e2e1ce).

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1706   +/-   ##
=======================================
  Coverage   72.82%   72.83%           
=======================================
  Files         379      379           
  Lines       53593    53593           
  Branches     9123     9123           
=======================================
+ Hits        39031    39032    +1     
+ Misses      11610    11609    -1     
  Partials     2952     2952           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[bryanwweber]

I think this should be coming from setup.cfg, is it not?

[speth]

Just trying things. I'm not finding the supposed ABI compatibility of the compiled module with older NumPy versions.

[bryanwweber]

I'll experiment with this tomorrow too

[speth]

Well, from the latest runs, it seems like backwards compatibility is working as expected when building with NumPy 2.0.0. The problem is with the builds for Python 3.8 using NumPy 1.24 and then running with NumPy 1.19. And now I've finally read this page enough times to understand that this is expected not to work and we have to go the other way around in this case.

[bryanwweber]

Ok, got it. The dependency, at least in the sdist at wheel build time, is that "oldest-supported-numpy" option they mention. But good to know that it needs to depend on the Python versions that NumPy supports

[speth]

Yep. It sounds like this should be a lot easier going forward -- a build-time dependency on NumPy 2.x will provide backwards compatibility with NumPy 1.x and also forward compatibility with future 2.x releases. If we want to get ahead of things, I suppose we could set the version requirement to numpy<3.

[bryanwweber]

One tricky thing I found here is that Pint 0.24 supports Python 3.10 and up, and only Pint >=0.24 supports NumPy 2.0. hgrecco/pint@749d77c for the Python 3.10 stuff and I'm having test failures on Python 3.9 with cibuildwheel because it's installing 0.23.2 of Pint and 2.0 of NumPy.

I think the solution is to edit the setup.cfg to force numpy<2.0 if the units extra is selected and Python is < 3.9

[bryanwweber]

Thanks @speth I'm going to merge this and move any further work to #1713