Option to sort re-exports in `__all__` (`sort_exports` from isort)

[Jackenmen]

Isort allows sorting __all__ when sort_reexports is used:
https://github.com/PyCQA/isort/blob/f547290178f86b6e5f0007cf1cb31d4736dcdaee/isort/core.py#L235-L241

[charliermarsh]

Is this undocumented behavior?

[Jackenmen]

isort has not had a release for over a year so it's one of the features that never ended up getting released. Actually, that reminds me of another (more important) feature that also has not been released yet, and seems like ruff isn't handling it properly either, will make an issue in a moment.

[Jackenmen]

Note that there's another functionality that is available in a released version of isort - sorting literals that are prepended with # isort: list or # isort: tuple:
PyCQA/isort#1358
sort_reexports option is actually using the same code for sorting.

One thing that both of these don't support is __all__ with multiple sections in it that should be sorted separately, e.g. something like what typing's code does (well, sort of, it doesn't seem like they're that good at keeping it alphabetical manually 😄):
https://github.com/python/cpython/blob/606adb4b891b52c8b9a53d29d594e996f117c0b3/Lib/typing.py#L42-L152
I'm actually not sure if isort's literal sorting even supports other profiles such as Black's, it seems to just sort it all in a pre-defined way.

[charliermarsh]

Oh yeah, this was brought to my attention on Twitter. I think we could support this! There are some other things I need to get done before I spend time on it though.

[Jackenmen]

I have a suggestion to extend this feature request a bit and allow sorting by the definition order of names (functions/classes/variables) that are put in __all__.

So for example:

from .submodule import reexported_name

__all__ = (
    "reexported_name",
    "function_b",
    "X",
    "function_a",
    "function_c",
    "Y",
)


def function_b():
    ...


class X:
   ...


def function_a():
    ...


def function_c():
    ...


class Y:
   ...

The reason for such a suggestion is that I think it makes it easier to find out what names of public APIs are missing in __all__ and what names shouldn't be in it (especially the latter one) since it allows one to just go through the file and look through the names in __all__ in the same order.

[kdeldycke]

For reference, there's another tool available to sort __all__: https://github.com/cpendery/asort

I guess it could be used to get rid of isort while we wait to have this feature implemented in ruff.

[kdeldycke]

For reference, there's another tool available to sort __all__: https://github.com/cpendery/asort

Hmm, forget about asort, I just stumble upon an issue: asort breaks docstring indention (asort#8)

That being said, regarding the sorting order, I also proposed to add support for another one that would be a case-insensitive lexicographical sorting. Not sure how popular or common this is.

[charliermarsh]

@cpendery - Yeah I'm open to including this. We'll need to decide how it should be categorized (i.e., does it go in the isort rule set? Or the Ruff-specific rules?) but it seems like a reasonable thing to include.

[cpendery]

@cpendery - Yeah I'm open to including this. We'll need to decide how it should be categorized (i.e., does it go in the isort rule set? Or the Ruff-specific rules?) but it seems like a reasonable thing to include.

@charliermarsh I think it probably should be categorized as Ruff specific rule since it was never an officially released isort feature, especially if we extend it with the additional functionality of multiple sorting options