re.split has seemingly unnecessary Anys

[GrantGryczan]

I would think that the | Anys in the following four return types should be removed.

typeshed/stdlib/typing.pyi

Line 1144 in 54e11a1

def split(self: Pattern[str], string: str, maxsplit: int = ...) -> list[str | Any]: ...

typeshed/stdlib/typing.pyi

Line 1146 in 54e11a1

def split(self: Pattern[bytes], string: ReadableBuffer, maxsplit: int = ...) -> list[bytes | Any]: ...

typeshed/stdlib/re.pyi

Line 184 in 54e11a1

def split(pattern: str | Pattern[str], string: str, maxsplit: int = ..., flags: _FlagsType = ...) -> list[str | Any]: ...

typeshed/stdlib/re.pyi

Line 188 in 54e11a1

) -> list[bytes | Any]: ...

[JelleZijlstra]

They're there because the list entries can be None:

In [6]: re.split(r"(a)|(b)", "canada")
Out[6]: ['c', 'a', None, 'n', 'a', None, 'd', 'a', None, '']

Though this behavior doesn't appear to be documented: https://docs.python.org/3.10/library/re.html#re.split

[GrantGryczan]

Huh, alright then. Strange that it isn't documented.

[GrantGryczan]

Actually, would it be a bad idea to replace the Anys with None in that case? Otherwise, in my use case, I'll have to cast the return value as list[str | None] myself.

[JukkaL]

Having None in the return value would likely cause many false positives, since often the caller can predict that there will only be strings. That's why the | Any was there in the first place instead of | None.

It would be good to add a comment explaining the Any types, as otherwise this will continue to cause confusion.

[AlexWaygood]

It would be good to add a comment explaining the Any types, as otherwise this will continue to cause confusion.

@srittau has proposed #7870 to tackle this problem in a more general way; we're currently bikeshedding debating the appropriate name for the feature.