Remove explicit declare of package-data

[DanielYang59]

Summary

• Remove explicit declare of package data, sse Git plugin instead: https://pypi.org/project/setuptools-scm/, also see document the MANIFEST.in-replacing capabilities of setuptools_scm pypa/setuptools-scm#199
  This would include "everything" tracked by Git, including "docs", "tests", ... And need to exclude them via MANIFEST.in
• Really make sure the setuptools-scm plugin would NOT change the versioning of pymatgen
• Vave Use binary distribution (wheels) to install pymatgen in test suite #4073 in place before this as a guardrail
• Fix py.typed marker file placement:

  For namespace packages (see PEP 420), the py.typed file should be in the submodules of the namespace, to avoid conflicts and for clarity.

[DanielYang59]

UPDATE: ignore the following, it turns out I didn't understand "package-data" correctly before. In "package-data", we declare files to be included into the source distribution, and after that, with include-package-data = true we include them into the wheels.

@janosh Would appreciate it if you could give me a hand on this :)

[janosh]

@janosh Would appreciate it if you could give me a hand on this :)

happy to help. what's the problem exactly? that some cython files aren't included in the wheel?

[DanielYang59]

Thanks for your reply @janosh :)

I originally opened this PR because I thought declaring all files was a bit inefficient (also prone to error if someone add a new file but forgot to update pyproject.toml), and I would expect all files included inside the src to be packed (really?).

The main cause of my headache is: the sdist need to include *.pyx but not *.so (the latter should not exist with python -m build --sdist so not a big issue), but the wheels should include *.so but not *.pyx (not sure if this is possible because with include-package-data = true everything from sdist would be copied to the wheel?).

But I didn't manage to find a more elegant yet safe way to handle file inclusion/exclusion at this moment (I hope I'm wrong).

As of my knowledge, to declare package data:

1. (Current method) Explicitly declare all non-.py/pyx files.
2. Add a MANIFEST.in file to declare package data file, but I don't think it's offering much benefit over current method.
3. Use the following to include everything inside the src directory, but this would also include *.c and *.pyx files inside the wheels.

[tool.setuptools.package-data]
 "*" = ["*", ]

4. Use setuptools-scm to include everything tracked by Git, but this would include tests as well.

I sneaked a peek at Jinzhe's DeepMD-kit and it looks like he's using a custom backend altogether.

[janosh]

but this would also include *.c and *.pyx files inside the wheels.

could you explain what the downsides of this are? is it a problem/bad practice to have those files in a wheel? seems like it could help with debugging to have the source files around if compiled code is ever causing issues. and the files are small so wouldn't noticeably increase wheel size.

[DanielYang59]

I would rather accidentally include some useless files than missing some necessary ones, so I should be really careful using wildcard for exclusion.

Is there any hidden file needed (at least no from current distributions)?