Reject type variable whose upper bound is Self - #21967
Conversation
| fullname = self.qualified_name(type_param.name) | ||
| if type_param.upper_bound: | ||
| upper_bound = self.anal_type(type_param.upper_bound, allow_placeholder=True) | ||
| # TODO: we should validate the upper bound is valid for a given kind. |
There was a problem hiding this comment.
I’m not entirely sure if this is the kind of validity that this comment is referring to.
|
I filed #21961 for this issue with the narrower The check covers more than class D[U]:
def g[T: U](self, x: T) -> None: ...
class F:
def nested[T: list[Self]](self, x: T) -> None: ...The first is the plain "bound parametrized by a type variable" case the spec On fallout: with the check applied to Minor: |
|
According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅ |
`mailman contributions --refresh` read only our own pull request's state. python/mypy#21967 was opened against the issue python/mypy#21961 fixed and the ledger said nothing for a day; it was found by hand in the PR thread. While a pull request is open, the refresh now reads its issue's timeline, where GitHub records every cross-reference, and keeps the other pull requests from the same repository that are open or merged. The listing names each one, says "no competing pull request" when the read was clean, and says why when it could not be read, so an unchecked row never reads like an unchallenged one. The command exits non-zero while a competitor exists. The parser is tested against the recorded timeline of python/mypy#21960, which holds ours, the competitor and a cross-repository reference to filter out. Closes #86. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
The repeated cross-linking and follow-up comments are making this a bit harder to follow than it needs to be. I also don't think the extra AI-generated activity is helping much here, especially given mypy's stance on LLM contributions from first-time contributors. |
|
Fair point, and sorry for the noise. The cross-posting and the long review comment were more than this thread needed, and posting them was my call. I'll step back and leave the PR to you and the maintainers. |
…g reply The comment that drew the complaint on python/mypy#21967 went out twelve seconds before our own pull request closed, so a gate keyed on the close could not have caught it. `handoff` now refuses an issue comment aimed at a pull request another author opened, in any run state; `--closing-reply` reaches only the superseding one, after provenance names it. A reply over 120 words gets a length warning in the block. The procedure says which threads a filed run writes to, and in what order a close is recorded. Follows #87. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2 participants
Fixes #21960
First attempt by an ai agent, See: #21961