gh-130827: Support typing.Self annotations in singledispatchmethod()#130829
gh-130827: Support typing.Self annotations in singledispatchmethod()#130829ZeroIntensity wants to merge 8 commits into
typing.Self annotations in singledispatchmethod()#130829Conversation
ZeroIntensity
added
topic-typing
needs backport to 3.12
Co-authored-by: Bénédikt Tran <10796600+picnixz@users.noreply.github.com>
| # But, it shouldn't work on singledispatch() | ||
| with self.assertRaises(TypeError): | ||
| @test.register | ||
| def silly(self: typing.Self, arg: int | str) -> int | str: |
There was a problem hiding this comment.
Will 'typing.Self' annotation work the same way?
There was a problem hiding this comment.
Good question! It should, yeah. get_type_hints handles forward references.
There was a problem hiding this comment.
Should we add a test case for that, maybe?
There was a problem hiding this comment.
Possibly, but as long as get_type_hints is working it's probably fine.
Co-authored-by: Tomas R. <tomas.roun8@gmail.com>
|
@vstinner I'd like to get this fixed before 3.12 goes security-only next week. Would you mind reviewing? |
|
Wait why would we want to change this in 3.12 at all? Seems like a new feature to me. |
sobolevn
removed
needs backport to 3.12
|
I agree, this is a new feature and should not be backported. |
|
Agreed with Jelle, this is a feature best left for a new release. Otherwise it's too hard to guard against in user code. |
|
Makes sense. I've updated the issue. |
|
Bump. I'd like to land this one before the beta freeze. |
JelleZijlstra
left a comment
JelleZijlstra
left a comment
There was a problem hiding this comment.
I think special-casing the Self annotation is the wrong approach. The self argument may be annotated with Self, but it may also be annotated with e.g. a TypeVar or Annotated.
We should be ignoring annotations not on the basis of what the annotation is, but on the basis of what parameter they're on. I think the correct approach would be to figure out the name of the first non-self argument (i.e., the second argument for a method), then retrieve the annotation for that argument.
If I'm not mistaken, singledispatch also incorrectly (or at least unintuitively) handles cases where the annotation is on some later argument, but the first one is unannotated.
@register
def f(x, random_arg: str = ...):
|
A Python core developer has requested some changes be made to your pull request before we can consider merging it. If you could please address their requests along with any other requests in other reviews from core developers that would be appreciated. Once you have made the requested changes, please leave a comment on this pull request containing the phrase |
Wouldn't that break if it's named something other than |
|
I mean the first argument that's not conceptually |
|
This PR is stale because it has been open for 30 days with no activity. |
|
I think @johnslavik has a different solution that he's been working on -- I'm going to close this in favor of his work. |
7 participants
singledispatchmethod.registerfails withtyping.Selfannotation #130827