TST: test CPython 3.13 (free-threading flavor)

[neutrinoceros]

No description provided.

[neutrinoceros]

I thought numpy dev would build easily already, but maybe I'm doing something wrong. Will re-assess in a couple weeks to months.

[neutrinoceros]

numpy now builds nightly wheels for both 3.13 and 3.13t, but Cython generation is currently failing

[neutrinoceros]

Building works but I really wasn't expecting tests to pass too. I need to better understand what's going on before I can confidently merge this.

[neutrinoceros]

This macro is not documented yet, so I'll wait until cython/cython#6298 is merged before I can call it stable enough

[da-woods]

It's documented in https://cython.readthedocs.io/en/latest/src/userguide/source_files_and_compilation.html#compiler-directives already.

(With that said, you should the official line is that everything free-threaded is experimental for the foreseeable future and may change as needed. But that particular directive is probably fairly unlikely to change)

[neutrinoceros]

thanks, I missed that ! And dully noted, all of this is still experimental, I'm just experimenting along :)

[neutrinoceros]

I think the reason why tests pass already is that all loops run in single threads (i.e., I use range instead of prange everywhere), and I'm not using more than one thread anywhere in my test setup. Overall, particle deposition is extremely thread-unsafe, but it shouldn't matter in practice as I don't see how users could be affected.

[neutrinoceros]

I understand now that I need to exercise multi-threading in CI for this to have value beyond just knowing that gpgi builds in 3.13t. I got started on that in #233