bpo-43766: Implement PEP 647 (User-Defined Type Guards) in typing.py

[Fidget-Spinner]

https://bugs.python.org/issue43766

[gvanrossum]

Thank you again! I have a bunch of editorial nits.

[Fidget-Spinner]

Wow, you're on a roll with the typing reviews today! Thanks for taking the time to review the docs Guido!

Re: the comments about runtime mishaps, here's a snippet from lower down in the docs:

   .. note::

      Strict type narrowing is not enforced - ``TypeB`` need not be a narrower
      form of ``TypeA`` (it can even be a wider form) and this may lead to
      type-unsafe results.  The main reason is to allow for things like
      narrowing ``List[object]`` to ``List[str]`` which would fail under strict
      narrowing as ``List`` is invariant.  The responsibility of
      writing type-safe type guards is left to the user.

Do you think The responsibility of writing type-safe type guards is left to the user. is enough, or do you want me to expound more?

[gvanrossum]

Somehow GitHub did not make it possible to see just the changes since my last review. :-(

[gvanrossum]

Wow, you're on a roll with the typing reviews today! Thanks for taking the time to review the docs Guido!

Catching up. :-)

Are there any other reviews I still owe you? I frankly have lost track.

Re: the comments about runtime mishaps, here's a snippet from lower down in the docs:

   .. note::

      Strict type narrowing is not enforced - ``TypeB`` need not be a narrower
      form of ``TypeA`` (it can even be a wider form) and this may lead to
      type-unsafe results.  The main reason is to allow for things like
      narrowing ``List[object]`` to ``List[str]`` which would fail under strict
      narrowing as ``List`` is invariant.  The responsibility of
      writing type-safe type guards is left to the user.

Heh, I nearly completely rewrote that in my last review.

Do you think The responsibility of writing type-safe type guards is left to the user. is enough, or do you want me to expound more?

That part I think is just fine!

[Fidget-Spinner]

Are there any other reviews I still owe you? I frankly have lost track.

Technically one more: #24056 . However, please hold off on that. I've recently been internally bikeshedding on whether to convert the entire PEP 612 implementation into pure Python to reduce maintenance burden and complexity in types.GenericAlias. After I went to read the PEP again, I have some ideas for collections.abc.Callable which may make that possible. I'll update you on bpo if my efforts come to fruition.

If what I plan for works, we won't need that PR anymore. And we'll be able to revert the C code to what it was before PEP 612 - a big plus in my opinion.

[Fidget-Spinner]

Personally, I think we can merge this. IMO the implementation is 100% there. The docs look alright too (I've re-read it a bunch of times now). And we can always fix the docs in later update.

[gvanrossum]

Okay, will merge.

[bedevere-bot]

@gvanrossum: Please replace # with GH- in the commit message next time. Thanks!

[Fidget-Spinner]

Sorry Guido, this PR broke doc builds on master because it triggers false positives in suspicious.py (which was re-enabled just recently). I've created #25660 to address your docs comments and fix the build errors.