Should we move away from Cython?

[Joao-Dionisio]

It takes a long time to compile PySCIPOpt, which makes development very annoying. This was exacerbated (I assume) by the upgrade to Cython 3. You can see that installing PySCIPOpt in PR 163 takes 3m 18s, while it took PR 148 1m 41s, which is still quite long. Is there something we can do to mitigate this?

@mmghannam also suggested using something that does the bindings automatically to the C header files, akin to what's happening with russcip. This is really not the sort of thing I'm knowledgeable about, but it seems like it would facilitate communication with SCIP, and I'm all for it. I have no clue about the impact it would have on the compilation time, maybe these are two entirely separate problems.

This would take considerable effort, but it's something that we could be slowly working on on the side.

Anyway, I'd love to hear your thoughts, please chime in!

EDIT: Okay, Mo is a genius and figured out that export CFLAGS="-O0 -ggdb" helps a ton. It tells Cython not to optimize the code, which makes compilation faster, but pyscipopt slower. We just need to take care with using unset CFLAGS before actually optimizing. See PR #823.