Start PEP 728 implementation

[JelleZijlstra]

Early version but need to get this code off the laptop where it currently lives

[JelleZijlstra]

OK this is finally ready for review.

[Daraan]

A few days ago I made a PR #1 against your fork branch. Addressing this statement from the PEP:

If closed is not passed, the value of closed is None.

[Daraan]

I realized python/cpython#109544 is missing

- if not hasattr(tp_dict, '__total__'):- tp_dict.__total__ = total+ tp_dict.__total__ = total

Resulting in a different behavior when using it outside of the spec, i.e. it being the value and not True.

class TD_with_total(TypedDict):
__total__ = "I want a total value here"

[Daraan]

From my understanding, this example in the PEP should pass without any errors or warnings:

class BaseMovie(TypedDict, closed=True):
name: str
class MovieA(BaseMovie): # OK, still closed
pass

Currently this does raise a warning; so I think the relevant check should be closed is False (see suggestion below).

This will cause quite a few with self.assertWarns(DeprecationWarning): to fail afterwards which then need a look-over.

---

From another side should we even check this, or should this only be handled by type-checkers?

It is a type checker error to pass closed=False if a superclass has closed=True or sets extra_items.

[Daraan]

Looking at the runtime specifications once more. __extra_items__ currently should also reflect the constructor and closed=True => extra_items=Never is only implicitly implied/equivalent. Suggested the necessary changes for that.

Also, suggested the GrandGrandChild as an example of an implicit closed subclass. I think such a thing was still missing.

[JelleZijlstra]

I realized python/cpython#109544 is missing

Good catch, I guess when we made that change in CPython we didn't think about the fact that you could set something in the class body. Would you be interested in (1) adding a test case to CPython where there is an assignment to __total__ in the class body and (2) backporting python/cpython#109544 to typing-extensions?

[Daraan]

Is a new entry to the changelog necessary? Else looks good to me.

---

(1) you have seen already, (2) added #533

[hauntsaninja]

kind of surprised we bothered. tbh if someone wants to do type(NoDefault).asdf = 123 i could live with that

[JelleZijlstra]

I guess I did it initially for parity with the C version in CPython for NoDefault, where it's easier to not accept extra attributes, and now that I need another one it makes sense to follow the pattern.

[AlexWaygood]

Basically LGTM, a few nits

[AlexWaygood]

I think this branch may also be uncovered, but feel free to merge and tackle this as a followup!