quick slurm fix

[tulga-rdn]

This PR is a quick fix for the robustness of the SLURM environment check. As @DavideTisi and I observed in #422 , when running the training after SSHing into a node reserved by salloc, the code throws an error. I think this occurs because reserving resources, SSHing to the node and running a job through a terminal session on that node is not a SLURM-managed process. So, this process has a job ID (since we got the node from some job), but not a process ID, so it's basically the same as running the training on a local workstation.

Considering this, I changed the is_slurm function, which will now return True only if the environment has both a job ID and a process ID.

Contributor (creator of pull-request) checklist

• Tests updated (for new features and bugfixes)?
• Documentation updated (for new features)?
• Issue referenced (for PRs that solve an issue)?

---

📚 Documentation preview 📚: https://metatrain--435.org.readthedocs.build/en/435/

[tulga-rdn]

Oh, and can I have some more permissions please? I didn't have enough permissions to create a branch, I think

[DavideTisi]

so for me the PR is fine, for some reason I cannot approve the PR. Anyway we can wait for what @frostedoyster has to say, and also he probably can add you in the repo

[DavideTisi]

ah ok after refreshing the page he gave me the option to do the review. my comment are the same as before, we can wait for filippo's approval

[frostedoyster]

Invited to the repo, thanks for the fix!

[frostedoyster]

Tests failing due to this library which is a dependency of spherical
https://pypi.org/project/spinsfast/#history

[moble]

TL;DR: Either

1. bump 3.9 to 3.10 here, or
2. change this line to sudo apt-get install -y zsh fftw-dev.

---

The problem is that spinsfast should still function on various older versions of python and numpy, so I don't want to restrict the compatible versions, but I can no longer generate wheels (pre-built binaries) for anything older than python 3.10 — just because of the various combinations of github runners, numpy compatibility, cibuildwheel support, etc.  I had to update the code itself a couple days ago because of numpy nonsense, so the newest version just doesn't have wheels for older versions.

I would guess that this is the first time pip is actually trying to build spinsfast for you, rather than just downloading the wheels.  You can see from the error message that it can't find the FFTW library, which just means you should install it.  There's a line in your github action where you do sudo apt-get install -y zsh; just add fftw-dev to the end of that line.  (You might want to only do this if matrix.python-version=="3.9", but it wouldn't do any harm for the other ubuntu case, and would make it less likely to break next time I have to bump the lower version.)

The other possibility is to bump your minimum python version to 3.10.  Numpy itself dropped support for python <3.10 back on Apr 5 of this year.  And I can still build wheels for 3.10, so this problem shouldn't crop up.  But I don't know if that's an option for you.