Standalone python package for IRBEM

[sahiljhawar]

Would you be open to the idea to individually build IRBEM python wrapper (outside of this repo) with different build setups for Linux 64 and Apple Silicon and distribute via PyPi? We (https://github.com/GFZ/EL_PASO/) in future want our library to be on PyPi, and for that reason I think it will be easy for us to distribute IRBEM along with it, as IRBEM is one of the main dependency. I think, this can streamline the IRBEM Python wrappers for others as well, as they will not have to anymore build IRBEM from source. If so, I can try working on this.

[mshumko]

Hi @sahiljhawar,

That is a good idea that I've been thinking of for some time now. I also have similar annoyance with using IRBEM as a dependency, and I was considering adding a ibermpy package under the PRBEM organization.

The big hurdle that got me stuck is how to get pip/setuptools to compile the Fortran source code. If you have ideas/implementations I am happy to explore them! The other packaging tasks seem to be comparatively trivial to the automatic compilation. I've briefly tried (and failed) with f2py. Also I know that meson is a new build system used by scipy, although I have no experience with it.

[sahiljhawar]

@mshumko Thank you for welcoming this idea. I have implemented this build feature in (GFZ/EL_PASO) by setting up build commands. Maybe a similar thing can be done to package irbempy.

[mshumko]

This looks like exactly what I need! Thanks. I will copy your setup.py file over into my irbempy repo and try to make it work in the next two weeks (please poke me mid-October if you don't hear back!).

How does your library ensure that you have the latest version of the Fortran source code? Is this something that we, as irbempy devs, have to manually copy from IRBEM to irbempy? Or does your _init_submodule() handle that?

[sahiljhawar]

For the current setup to work IRBEM needs to be included as a submodule so that _init_submodule() can fetch the submodule. Alternatively, instead of bloating the new irbempy with submodule, you can also git clone at install time. In cloning case, you may have to change the subprocess.check_call(... cwd="IRBEM")

[sahiljhawar]

@mshumko I have implemented a basic working version of irbempy. You can check it here: https://github.com/sahiljhawar/irbempy

[AntoineBrunet]

Hi, thanks for taking this initiative. That's indeed something we have wanted to do for quite some time. I'll help as much as I can, but it's great that we have some people willing to participate!

When everything is worked out, I would suggest to use this repository as the main source for the pip (eventually conda) packages, so that any future update is propagated easily. We've had some troubles getting the IRBEM version embedded in Spacepy up to date in the past. We would have to think about the implications on versioning of the library, but I think it would be beneficial overall.

On another note, as we've discussed with Mike previously, I'm not entirely satisfied with the IRBEM python wrappers (either irbempy from Spacepy or the IRBEM wrapper in this repository). We have a somewhat functioning python interface at ONERA which we would be happy to share. This could come either in replacement or in addition to the available wrapper, depending on how this wrapper is actually used by the community currently.

[sahiljhawar]

@AntoineBrunet Sure. let me know what other code you want to put, alternatively you can open a PR on my repo.

In EL_PASO we are using a slightly changed version of IRBEM.py with custom type hints, annotations, etc and this is working well for us.

[DoctorRabbit55]

Hi, let me add some details to the wrapper we are using for EL-PASO. You can find the code here. Overall, I tried to add type hints to be compliant with mypy standards and also applied linting and formatting using the Ruff guidelines.

The wrapper we use in EL-PASO returns NamedTuples for every function call, which is a bit cleaner in my point of view compared to dictionaries, but is not compatible with the wrapper of this repository. We can easily change this to TypedDicts, which would be compatible with the current version.

I did not bother adding the Python-only functions of IRBEM.py since we do not need them for our specific use case. But I wrote a wrapper for make_lstar_shell_splitting.

We would also be happy to improve the current test suite for the wrapper to upgrade using pytest or to add more tests in general.