Updating numpy.i (cvxcore) swig interface

[Transurgeon]

Description

This PR updates the cvxpy build to allow compatible wheels with NumPy 2.0.
The changes come in the following stages:

• updating the numpy.i file in cvxcore which is used for the swig interface. (As pointed out by Riley in the corresponding issue linked below). The new file is taken from here and was last updated 2 years ago.
• Rebuild cvxcore using the swig command in the corresponding README located here.

So I believe the updates to cvxcore are necessary for wheels compatibility with NumPy 2.0.
Any feedback/questions are welcome!

P.S: I am quite new to the python build ecosystem so I may have missed some important steps.

Issue link (if applicable): #2389

Type of change

• New feature (backwards compatible)
• New feature (breaking API changes)
• Bug fix
• Other (Documentation, CI, ...)

Contribution checklist

• Add our license to new files.
• Check that your code adheres to our coding style.
• Write unittests.
• Run the unittests and check that they’re passing.
• Run the benchmarks to make sure your change doesn’t introduce a regression.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Transurgeon seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[Transurgeon]

I tried to follow the steps outlined in the NumPy migration guide, but could have certainly missed some steps.
I think the next steps for this would be to update the CI to test both on the oldest numpy version and the nightly version of NumPy.
Another option could just be to wait until NumPy 2.0 is officially released through pip. (see image below)

[PTNobel]

Another option could just be to wait until NumPy 2.0 is officially released through pip. (see image below)

I think that's my preference tbh... I don't think we're in a huge rush to get this merged and I feel like we'll be far from the only package that is broken right after 2.0 drops.

[Transurgeon]

@PTNobel @SteveDiamond @rileyjmurray
I propose that we first merge this PR without the changes to thepyproject.toml setup file, that way the CI will pass and we can figure out later how we want to go about testing the new builds.
I'll also change the title of the PR to something like "updating numpy.I swig interface".

This could also maybe allow the cvxpy-feedstock to rebuild without any problems.

What are your thoughts?

[PTNobel]

That sounds like a great first step. I think I agree with h-vetenari in another thread that we should release 1.5.2 with NumPy 2.0 support before 2.0 comes out. That'll mean some changes to the GitHub action runners

[Transurgeon]

Ok, I just reverted the changes to pyproject.toml. CI failure seems to be unrelated to the PR.

[PTNobel]

I want to make sure that the cvxpy-base build is working before merging... Rerunning CI, since the error seemed spurious

[PTNobel]

I'm not a huge fan of downgrading our swig version. I had thought we had updated to fix a bug; William can you update the swig on your computer and rerun this?

[Transurgeon]

I tried rerunning the command, but to no avail. I also couldn't find a way to upgrade the swig version (sorry! I am very unfamiliar with swig).

I decided to just revert the changes to the file since it seems like it was already generated using the desired swig 4.1.1.

[Transurgeon]

Ok I just managed to install swig 4.2.1 (the latest). And regenerated the cvxcore wrap file. Everything seems to work now :).

[PTNobel]

Upgrading to Swig 4.1.1 was quite intentional. I don't think we should backout #2273

[Transurgeon]

Upgrading to Swig 4.1.1 was quite intentional. I don't think we should backout #2273

Sorry, I wasn't aware of that PR.

I just tried rebuilding cvxpy with numpy 2.0, but it didn't work. So we do need to rerun the given command with a newer version of swig.

Anaconda only provides swig 4.0.2, see here: https://anaconda.org/anaconda/swig. Same thing for homebrew. (This means we should probably update the instructions in the cvxcore README as well).

I found the releases on the swig GitHub, see here: https://github.com/swig/swig/tags. Would we potentially want to update to the latest version which is 4.2.1?

[Transurgeon]

@PTNobel could you have another look at this PR please?

I managed to install the newest swig version and regenerated the cvxcore wrap file (it is now version 4.2.1).
I then retried to build cvxpy with numpy 2.0 and it seems to work without any problems.
Finally, I updated the cvxcore/README.md to reflect anaconda being outdated on the swig version.

[PTNobel]

Looks good to me

[phschiele]

Let's see if this works with the pre-release CI!