Update numpy.i for SWIG 4.3 compatibility

[musicinmybrain]

Update numpy.i to the latest version, from
https://github.com/numpy/numpy/blob/v2.1.2/tools/swig/numpy.i, plus apply the changes from numpy/numpy#27579.

An incompatibility with the upcoming SWIG 4.3 test was reported in Fedora Linux. I reported the issue to numpy in numpy/numpy#27578 and proposed a fix in numpy/numpy#27579. This PR circles back to where things started, copying in the latest numpy.i from numpy upstream, plus the changes from numpy/numpy#27579.

[traversaro]

fyi @diegoferigo

[diegoferigo]

Thanks a lot @musicinmybrain for your work and for propagating it all the way up to this project, I'm no longer following that much the SWIG-related development. It makes sense to update the vendored numpy.i file of the example. I'm wondering if there's a more robust approach, like downloading the revision of numpy.i matching the numpy version found in the system from git repository. I accept PRs if anyone wants to have a look at it.

I need to fix the PyPI pipeline before merging this PR.

[traversaro]

I'm wondering if there's a more robust approach, like downloading the revision of numpy.i matching the numpy version found in the system from git repository. I accept PRs if anyone wants to have a look at it.

That would not work for the specific case of SWIG 4.3.0 compatibility. For example, the latest numpy released does not contain this fix, so if you for example installed the latest numpy and swig from conda-forge, you will end up with swig 4.3.0 and numpy 2.1.2, but https://github.com/numpy/numpy/blob/v2.1.2/tools/swig/numpy.i does not contain the changes, that indeed are in numpy/numpy#27579 .

[diegoferigo]

I'm wondering if there's a more robust approach, like downloading the revision of numpy.i matching the numpy version found in the system from git repository. I accept PRs if anyone wants to have a look at it.

That would not work for the specific case of SWIG 4.3.0 compatibility. For example, the latest numpy released does not contain this fix, so if you for example installed the latest numpy and swig from conda-forge, you will end up with swig 4.3.0 and numpy 2.1.2, but https://github.com/numpy/numpy/blob/v2.1.2/tools/swig/numpy.i does not contain the changes, that indeed are in numpy/numpy#27579 .

Mmh it makes sense. I guess that there's no real workaround. Vendoring numpy.i and maintaining it aligned is and will remain a responsibility of downstream projects.