Python interface

[klamike]

Hi @apozharski, thanks for your awesome work on libMad. I was curious if I could use it to call MadNLP from Python, and it seems to be working great. Check out this example: https://github.com/klamike/madnlp_cvxpy. I'm also looking into trying it out with PyOptInterface.

Would you be interested in adding a Python interface to this project? I understand it is still WIP, so no rush of course, but I'd be happy to help maintain it if you're interested. The proof-of-concept one is here: https://github.com/klamike/libMad/tree/mk/latest/python

[apozharski]

Ah, I will put it this way, I am not disinterested 😅, but at the same time you hit the nail on the head wrt the stability of the project. I am not sure if long term it makes sense to co-locate the C and Python interfaces, but in truth this is primarily because I am not an expert on python packaging and do not know if that breaks some rules.

I haven't had a chance to really understand your proof of concept and unfortunately probably won't really have much time to work on this project with any regularity for a little while due to other work and deadlines taking priority for now. However, it seems you download the release and use it that way. I am quite happy to see another test that seems to suggest the release is actually successfully distributable. I suspect we would want in the long term to build the library as a platform specific wheel, however.

All that is to say I would be happy to look at contributions, and an efficient python interface for Mad* would certainly be within scope of what I want to do with this project.

PS: One thing to note ubuntu-aarch64 is not actually currently supported :) as I have yet to dive into cross-compiling on the github runners or setting up our own build server.

[klamike]

I am no expert on python packaging either, in fact this is my first time writing a C binding for Python. The goal was just to see if it is possible, and it turned out to be quite straightforward! Agreed that if/when it becomes an "official" feature, we should look into the proper conventions for distribution e.g. platform wheels.

Yes, the default behavior is to download the latest release from GitHub and cache it. There is also the build_from_source function if you have a local clone. Both worked on my Mac with #3 merged in, but I had to comment out the using MadNLPHSL since I don't have an HSL license on my personal machine. I haven't tested with GPU yet.

Thanks again for your work, libMad is super cool!