PEP-563: Fixed a nested class example in

[superbobry]

The "Backward Compatibility" section in PEP-563 contains an example which
violates Python scoping rules. This PR updates the PEP to say that the example
is not OK and should fail.

See discussion in typing-sig for details *.

[the-knights-who-say-ni]

Hello, and thanks for your contribution!

I'm a bot set up to make sure that the project can legally accept this contribution by verifying everyone involved has signed the PSF contributor agreement (CLA).

CLA Missing

Our records indicate the following people have not signed the CLA:

@superbobry

For legal reasons we need all the people listed to sign the CLA before we can look at your contribution. Please follow the steps outlined in the CPython devguide to rectify this issue.

If you have recently signed the CLA, please wait at least one business day
before our records are updated.

You can check yourself to see if the CLA has been received.

Thanks again for the contribution, we look forward to reviewing it!

[JelleZijlstra]

Thanks! You'll have to sign the CLA and I'd also like to see signoff from @ambv.

[superbobry]

@JelleZijlstra I have submitted the CLA form, but IIUC the processing will take ~24h. Look forward to hearing from @ambv!

[gvanrossum]

I'd really like @ambv's review of this.

[gvanrossum]

Why mention this here and not in the previous example? And why not show the correct way (C.field) like you did there?

[ambv]

Ah, this is an example of a bad practice in PEP writing: aspirational truths. At the time reviewers of the PEP were disappointed with nested class scoping not working, so somebody had the idea that typing.get_type_hints() should special-case methods and classes by injecting their name to globalns, essentially making this automatic:

>>> from __future__ import annotations
>>> from typing import *
>>> class C:
...   class D:
...     def meth(self) -> D:
...       ...
...
>>> C.D
<class '__main__.C.D'>
>>> get_type_hints(C.D.meth)
Traceback (most recent call last):
  ...
NameError: name 'D' is not defined
>>> get_type_hints(C.D.meth, globalns=dict(C.__dict__))
{'return': <class '__main__.C.D'>}

In practice implementing this wasn't trivial and I never gotten to fixing this case. That it contradicts scoping rules, sure, but would still be a useful addition to get_type_hints() because there's a more important case which the example in the PEP doesn't mention: functions which are class factories. So, to make the following work:

>>> class B:
...   ...
...
>>> def f() -> B:
...   class A(B):
...     def meth(self) -> A:
...       ...
...   return A
...
>>> A_ = f()
>>> get_type_hints(A_.meth)
Traceback (most recent call last):
  ...
NameError: name 'A' is not defined
>>> g2 = dict(globals())
>>> g2[A_.__name__] = A_
>>> get_type_hints(A_.meth, globalns=g2)
{'return': <class '__main__.f.<locals>.A'>}

WDYT @gvanrossum, is this still worth touching? If so, I'll prepare a fix to 3.10 get_type_hints(). If not, then I guess updating the PEP with this PR here is the way to go.

[gvanrossum]

I don't think it is worth fixing -- it will just proliferate questions about scopes here, and AFAIK all the static checkers implement the scope rules as used by the runtime, so get_type_hints() would be the odd one out.

[superbobry]

Thanks for the context, Łukasz!