Repository navigation
Introduce _typeshed.WeakUnion - #7437
not-my-profile wants to merge 2 commits into
Conversation
This comment has been minimized.
This comment has been minimized.
| from typing import AbstractSet, Any, Container, Generic, Iterable, Protocol, TypeVar, Union | ||
| from typing_extensions import Final, Literal, final | ||
|
|
||
| WeakUnion = Any |
There was a problem hiding this comment.
what if we do this:
| WeakUnion = Any | |
| class WeakUnion(Any): | |
| def __class_getitem__(cls, args: Any) -> Any: ... |
There was a problem hiding this comment.
mypy doesn't (yet) have a generalised understanding of __class_getitem__. It only understands a class as being generic if the class inherits from typing.Generic. https://mypy-play.net/?mypy=latest&python=3.10&gist=1a3ed0af028c24c591ea1e63d80954f6
There was a problem hiding this comment.
That's too bad. Putting __getitem__ on the metaclass doesn't work either. This might be easier to fix than full PEP 646 support though.
There was a problem hiding this comment.
I have reservations about teaching mypy about __class_getitem__, as I think __class_getitem__ should properly be seen as an implementation detail of the typing framework. But I agree that it would be nice for mypy to have a stronger understanding of metaclass __getitem__.
There was a problem hiding this comment.
Can you give an example of what __getitem__ with a metaclass would look like? And is there any type type checker that currently handles that? I think we should open issues with mypy, pyright, pytype & pyre if they don't support that currently.
There was a problem hiding this comment.
using it in a return type doesn't work with pyright:
variable not allowed unless it is a type alias
There was a problem hiding this comment.
Hmm. I guess maybe that's part of the "Unknown" vs "Any" distinction pyright has? The even dumber hack is from _typeshed.does_not_exist import WeakUnion as WeakUnion # type: ignore
There was a problem hiding this comment.
That doesn't help pyright still treats such types it cannot import as Unknown.
There was a problem hiding this comment.
@erictraut, how hard would it be for you to implement special-casing in pyright for a WeakUnion/UnsafeUnion symbol imported from _typeshed? It wouldn't need to be the full implementation of this proposed feature just yet; it would just need to be a special form semantically identical to Any that could be provided with arbitrary subscriptions.
There was a problem hiding this comment.
I'd prefer not to do it that way. This approach looks very hacky to me, and I don't think it should go into typeshed this way.
If it's going to be a special form, then I think we should go through a PEP process and define it formally. If we're going to use PEP 646 as I suggested, then let's do that — although we'd need to wait for some minimal support in mypy first.
If this is just going to be an Any for now, then there's no benefit in the short term. I don't see the need to rush it. Let's pick a sound approach and get this right.
|
By the way, if it's appropriate to bikeshed here, I have a big preference for a name with "Any" (or possible "Unsafe") in it. The principle being that it should be obvious to users when a type involves unsafety. So candidates include:
(my vote is currently still for AnyOf) |
I like |
8eef570 to
3d235f3
Compare
not-my-profile
changed the title
656757e to
ffc0d33
Compare
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
This comment has been minimized.
Co-authored-by: Alex Waygood <Alex.Waygood@Gmail.com>
|
According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉 |
|
The test failures indicate that this would be very difficult to do without adding special-casing to type checkers to account for this feature. As such, I'm going to close this for now, but we can maybe revisit it after further discussion in the python/typing repo or the typing-sig mailing list. |
5 participants
No description provided.