scikit-build
why is __all__ required
We want to avoid the maintaineance overhead of listing all symbols in __init__.py's __all__, but it seems to be a special scikit-build thing? We do not use the from module import * way of importing, which should be the only reason to use __all__ at all.
__all__ is from Python, not scikit-build. It is how you tell Python and your reader what is "public" in your module. __all__ is also used by static analysis tools to detect usages of things that are not "public" (like mypy and others). It's also a "table of contents" to the reader. If you define def __dir__(): return __all__, then you get correct tab completion for your public API, things like module imports (!) will stop showing up (Python 3.7+).
Scikit-build does not read __all__ as far as I can tell: https://github.com/scikit-build/scikit-build/search?q=__all__
What I meant was: with setuptools, I do not need __all__, but with scikit-build it must be defined, otherwise the library will not be consumable. So the build parameters seem to be off from what setuptools does?
Why must it be defined? I've already pointed out it's not accessed anywhere in scikit-build's code base, and I'm pretty sure I have never had to add an __all__ (though it's always a good idea, I'm often lazy and forget it).