bpo-46981: Remove typing._TypingEmpty

[serhiy-storchaka]

• get_args(Tuple[()]) returns now () instead of ((),).
• Tuple[Unpack[Ts]][()] returns now the result equal to Tuple[()].

https://bugs.python.org/issue46981

[gvanrossum]

Nice simplification! One thing I wonder about is whether this would break 3rd party frameworks that do runtime introspection of annotations? E.g. pydantic. Who can we ask?

[JelleZijlstra]

Thanks!

I don't think this will affect most runtime type checkers; at most it allows them to remove a special case. I did find a reference to _TypingEmpty in the wild: https://github.com/CedricFR/dataenforce/blob/5dfc5f725d6a69197481061d6e6676fdb221adfd/dataenforce/__init__.py#L88. Seems like an attempt to implement their own version of Generic.

[gvanrossum]

All right!

I suppose there’s no consequences for typing_extensions either?

[JelleZijlstra]

I suppose there’s no consequences for typing_extensions either?

I just checked and typing-extensions uses _TypingEmpty only in the 3.6 implementation of Protocol.

[Fidget-Spinner]

Nice simplification! One thing I wonder about is whether this would break 3rd party frameworks that do runtime introspection of annotations? E.g. pydantic. Who can we ask?

Searching pydantic, it seems that get_args is rarely used. However, I will cc Samuel just to be sure. The last time we had a change in get_type_hints re Union, I sought his opinion too. Maybe this will help him figure things out if one of his tests breaks.

@samuelcolvin a summary of the change, for your info please:

typing.get_args(typing.Tuple[()]) now returns () instead of ((),).

[gvanrossum]

Go ahead and merge once tests pass!