This updates PEP 718 to answer the open points from the Discourse thread (mainly @carljm's review) and @gvanrossum's comments on @Gobot1234's commit 14882ca. It ports parts of that commit; James is credited as co-author.

A note on methodology. I used an LLM to summarise every pending technical question from the Discourse thread and Guido's review of James's commit, and turned them into a questionnaire that I answered myself: all questions, options and my decisions. The decisions are mine, aligned with the original proposal and with Guido's, Carl's and other core devs' input; where core devs disagreed, I picked what felt most natural. The LLM then helped write them into the PEP, and I checked the result. It's a draft and I'll do a full cleanup pass soon, so if parts read a bit LLM-like, the decisions behind them are human, not unread slop.

Main changes

• Runtime: __getitem__ on functions doesn't validate and returns a types.FunctionGenericAlias (#123).
• Callable values: checkers allow subscribing any value with a generic callable type (the lenient prototype); the strict variant goes to Rejected Ideas.
• Type checking: same rules as generic classes (ParamSpec, TypeVarTuple examples); unused type params count (#131); unsolvable params are reported at the call (#142); PEP 696 defaults per #121; error below Python 3.16.
• Overloads: a subscription applies to each overload, following the mypy prototype.
• Kinds of functions: generators and async functions are subscriptable (with a send() example); lambdas can't be generic; staticmethod and decorator rules (functools.wraps, @contextmanager).
• New sections: Use Cases, Runtime Performance, Backwards Compatibility, Rejected Ideas, Deferred Ideas (reification, generic lambdas).
• Other: Motivation workarounds fixed, reference implementation links updated, Python-Version: 3.16.

Guido's comments on 14882ca

All 11 addressed: quotes around "function objects"; reification moved to Deferred Ideas; "requested many times" softened; Foo[int]() output shown; class-only examples removed; current runtime error shown; "roots of runtime type checking" and the .T sketch dropped; blank lines fixed; foo_2 split into .py/.pyi; plain assignment instead of type X = foo_1[int].

Open question (@gvanrossum)

The spec lists types.MethodWrapperType (e.g. object().__str__) as subscriptable, but python/cpython#156201 doesn't cover it. Keep it and add support there, or drop it from the spec? The text keeps it for now.

Follow-up PR

A Motivation rewrite with real-world examples; How to Teach This and Security Implications.

• Change is either:
  • To a Draft PEP
  • To an Accepted or Final PEP, with Steering Council approval
  • To fix an editorial issue (markup, typo, link, header, etc)
• PR title prefixed with PEP number (e.g. PEP 123: Summary of changes)