Fix required and optional keys inheritance for TypedDict

[vemel]

I realized that the inheritance of TypedDict does not work as it should.

class _Animal(TypedDict):
     name: str

class Animal(_Animal, total=False):
     voice: str

class Cat(Animal):
     fur_color: str

I would assume that Cat should have required keys name and fur_color and optional voice.
But in reality, it will have required fur_color and optional name and voice, because Animal has total=False

Fixed

• __required_keys__ and __optional_keys__ are copied from base classes
• Own TypedDict annotations are added to __annotations__ after base annotations
• Conflicting required and optional keys from base classes are removed if TypedDict redefines them

[gvanrossum]

Should this also be updated upstream in CPython?

[gvanrossum]

LGTM except I wish to rename _Animal and get rid of tail.

[gvanrossum]

This isn't allowed by mypy (you can't change the type of an attribute in a subclass) so I recommend deleting the tail attribute altogether.

[vemel]

Makes sense, deleting it. I also will not remove conflicting keys from __required_keys__ and __optional_keys__ since conflicts are not allowed.

[vemel]

Pushed changes, updated tests.

[gvanrossum]

I don't see a reason for this class name to start with _ -- maybe rename to BaseAnimal?

[vemel]

I am not aware of a convention how to name TypedDict that has both required and optional keys. Usually I prefix a TypedDict with required keys with an underscore.

[vemel]

Renamed to BaseAnimal.

[vemel]

Travis CI build failed for Python 3.5.1, but it looks unrelated.

  File "/home/travis/virtualenv/python3.5.1/lib/python3.5/site-packages/importlib_metadata/__init__.py", line 9, in <module>
    import zipp
ImportError: No module named 'zipp'

Probably restarting Travis build will help.

[gvanrossum]

Thanks! I will land this despite the 3.5.1 failure.

[Zac-HD]

@vemel - I think this should go upstream, as a follow-up to python/cpython#17214, since CPython is still using the pre-this-PR logic.

[gvanrossum]

Yes, please open a good issue and submit a PR.

On Mon, May 4, 2020 at 05:26 Zac Hatfield-Dodds ***@***.***> wrote: @vemel <https://github.com/vemel> - I think this *should* go upstream, as a follow-up to python/cpython#17214 <python/cpython#17214>, since CPython is still using the pre-this-PR logic. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#700 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAWCWMWTHUVAENLMQLFPULLRP2X6BANCNFSM4KRQAMPA> .

-- --Guido (mobile)

[domdfcoding]

@Zac-HD Did you submit a PR to CPython for this? Having trouble finding it.

[Zac-HD]

According to https://github.com/python/cpython/pulls?q=is%3Apr+author%3AZac-HD I didn't, and remembering what May was like I believe it. If you'd like to, go for it 🙂