According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅

@davidfstr you've helped me with mypy PRs before and this is related (but much smaller/easier). Can you help shepherd this?

Thanks for providing the PR! Looking like it implements part of the PEP 655 (NotRequired) that was missed.

There are a lot of items on my plate right now, so the earliest I can take a look is this weekend starting 8/19.

This change is quite straightforward. LGTM.

An explanation for other reviewers:

• (Not)Required[] is only permitted in certain contexts within type annotation expressions, and is disallowed everywhere else by default. In particular (Not)Required[] is generally only permitted at the top-level.
  • However PEP 655 also explicitly allows (Not)Required[] within Annotated[] expressions, which the rest of this PR enables.
• (1) This PR alters TypeAnalyser.try_analyze_special_unbound_type's behavior when it is examining an Annotated[T] so that when it sub-analyzes T using TypeAnalyser.anal_type(), it passes allow_required=True, so that T is allowed to be a (Not)Required[] expression.
• (2) This PR alters TypeAnalyser.anal_type() with a new allow_required parameter that can override anal_type() to allow (Not)Required[] one level deeper (but no further).
  • Previously anal_type() would never allow (Not)Required[] at deeper nesting levels.

@bsmedberg-xometry , could you rebase this PR on the tip of master?

Afterwards I'd suggest tagging the recent mergers (hauntsaninja, ilevkivskyi, sobolevn), saying this PR is ready to merge.