AttributeError for Literal from typing_extensions package on Python 3.6

We’re using beartype on Python 3.6 with some Literal hints. We’ve run into an error with code hinted using this flavour of Literal.

(env) PS C:\Users\pbourke\Documents\projects\beartype_literal_error> python --version
Python 3.6.8

(env) PS C:\Users\pbourke\Documents\projects\beartype_literal_error> pip list
Package           Version
----------------- --------
beartype          0.8.1
pip               18.1
setuptools        40.6.2
typing-extensions 3.10.0.2
You are using pip version 18.1, however version 21.2.4 is available.
You should consider upgrading via the 'python -m pip install --upgrade pip' command.

(env) PS C:\Users\pbourke\Documents\projects\beartype_literal_error> cat .\test.py
from beartype import beartype
from typing_extensions import Literal


@beartype
def example(a: Literal["a"]) -> None:
    pass


if __name__ == '__main__':
    example("a")

(env) PS C:\Users\pbourke\Documents\projects\beartype_literal_error> python .\test.py
Traceback (most recent call last):
  File ".\test.py", line 6, in <module>
    def example(a: Literal["a"]) -> None:
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_decor\main.py", line 159, in beartype
    func_wrapper_code = generate_code(func_data)
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_decor\_code\codemain.py", line 162, in generate_code
    code_params = _code_check_params(data)
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_decor\_code\codemain.py", line 355, in _code_check_params
    param_index=param_index,
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_decor\_code\_pep\pepcode.py", line 157, in pep_code_check_param
    func=data.func, param_name=param.name),
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_util\cache\utilcacheerror.py", line 200, in reraise_exception_cached
    raise exception.with_traceback(exception.__traceback__)
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_decor\_code\_pep\pepcode.py", line 137, in pep_code_check_param
    ) = pep_code_check_hint(hint)
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_util\cache\utilcachecall.py", line 351, in _callable_cached
    raise exception
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_util\cache\utilcachecall.py", line 344, in _callable_cached
    *args, **kwargs)
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_decor\_code\_pep\_pephint.py", line 1883, in pep_code_check_hint
    die_unless_hint_pep586(hint_curr)
  File "C:\Users\pbourke\Documents\projects\beartype_literal_error\env\lib\site-packages\beartype\_util\hint\pep\proposal\utilpep586.py", line 89, in die_unless_hint_pep586
    hint_literals = hint.__args__
AttributeError: '_Literal' object has no attribute '__args__'

I think the root issue is that the typing package version stores arguments in __args__, but typing_extensions uses __values__:

https://github.com/python/typing/blob/5cc1d2c840d9a99aeeb442ea6bf71d7ce2dee523/typing_extensions/src_py3/typing_extensions.py#L682

[daxpryce]

@leycec I may not be able to rip out an incredibly comprehensive and dense usage guide filled with bearpuns, but I still tried a little in our latest release notes: https://graspologic.readthedocs.io/en/latest/release.html#graspologic-0-3-0

[leycec]

Fascinating… and horrifying. I’ve been hoping without evidence that typing_extensions would routinely rebase its runtime implementation against typing from the official Python repo, but… those hopes and more have now been dashed. I may be crying inside.

It could be argued that any divergence from typing is an upstream bug in typing_extensions. In this case, __values__ is surprisingly non-standard; popular annotation standards like PEP 484 and 585 institutionalized the use of __args__, __parameters__, and __origin__ for runtime introspection. Even NumPy’s third-party numpy.typing API sticks to that nomenclature. I may be grumbling inside.

It could also be argued that the utmost satisfaction of our glorious userbase comes before my own long-standing laziness and desire to continually shift blame elsewhere – and that is the better argument to make.

I was moments away from releasing beartype 0.9.0 when this and #51 delightfully landed. Naturally, the happiness of you personally and Microsoft takes precedence. I’ll get this immediately resolved for you, Pat.

Oh – and thanks so much for the detailed issue complete with nostalgic PowerShell logs and low-level diagnostic analysis. Have some fireworks!

✨ 🎇 🎆 🎇 ✨