On main, the pyperformance bm_typing_runtime_protocols benchmark takes 177us +- 5us on my machine. With this PR, it takes 192us +- 5us. So, a slight slowdown. But that's okay if we avoid incorrect behaviour.

Oh, we will still hit the problematic super().__instancecheck__ sooner or later.

Consider this

from typing import runtime_checkable, Generic, Protocol, TypeVar

T = TypeVar("T")


@runtime_checkable
class Spam(Protocol[T]):
    x: T


class Eggs(Generic[T]):
    def __init__(self, x: T) -> None:
        self._x = x

    def __getattr__(self, attr: str) -> T:
        if attr == "x":
            return self._x
        raise AttributeError(attr)


print(isinstance(Eggs(42), Spam))
print(issubclass(Eggs, Spam))

Running on 3.10, 3.11 -> TypeError
Running on current main and this PR -> no TypeError

I think one possible solution is that "do not let super().__subclasscheck()__ cache __subclasscheck__ result in this case".

@sunmy2019, I think we were thinking the same thought simultaneously :)

How does it look to you after 5f72b82? (It's basically a completely different approach now.)

This new approach also has the advantage that it seems like it might provide a speedup relative to main, for isinstance checks that don't involve ABCMeta.register()? (But might just be noise.)

This new approach looks better. Great Job!

Thanks @JelleZijlstra for helping debug this, @sunmy2019 for spotting the flaws in my original approach, and everybody for reviewing!

⚠️⚠️⚠️ Buildbot failure ⚠️⚠️⚠️

Hi! The buildbot AMD64 Windows11 Bigmem 3.x has failed when building commit b27fe67.

What do you need to do:

1. Don't panic.
2. Check the buildbot page in the devguide if you don't know what the buildbots are or how they work.
3. Go to the page of the buildbot that failed (https://buildbot.python.org/all/#builders/1079/builds/1325) and take a look at the build logs.
4. Check if the failure is related to this commit (b27fe67) or if it is a false positive.
5. If the failure is related to this commit, please, reflect that on the issue and make a new Pull Request with a fix.

You can take a look at the buildbot page here:

https://buildbot.python.org/all/#builders/1079/builds/1325

Failed tests:

• test_hashlib

Failed subtests:

• test_case_md5_uintmax - test.test_hashlib.HashLibTestCase.test_case_md5_uintmax
• test_case_md5_huge - test.test_hashlib.HashLibTestCase.test_case_md5_huge

Summary of the results of the build (if available):

== Tests result: FAILURE then FAILURE ==

428 tests OK.

10 slowest tests:

• test_bigmem: 43 min 23 sec
• test_lzma: 34 min 43 sec
• test_bz2: 21 min 5 sec
• test_array: 7 min 22 sec
• test_zlib: 4 min 58 sec
• test_multiprocessing_spawn: 2 min 2 sec
• test_math: 1 min 27 sec
• test_hashlib: 1 min 22 sec
• test_pickle: 1 min 14 sec
• test_concurrent_futures: 1 min 13 sec

1 test failed:
test_hashlib

36 tests skipped:
test.test_asyncio.test_unix_events test_clinic test_curses
test_dbm_gnu test_dbm_ndbm test_devpoll test_epoll test_fcntl
test_fork1 test_gdb test_grp test_ioctl test_kqueue
test_multiprocessing_fork test_multiprocessing_forkserver test_nis
test_openpty test_ossaudiodev test_peg_generator
test_perf_profiler test_pipes test_poll test_posix test_pty
test_pwd test_readline test_resource test_spwd test_syslog
test_threadsignals test_wait3 test_wait4 test_xxlimited
test_xxtestfuzz test_zipfile64 test_zoneinfo

1 re-run test:
test_hashlib

Total duration: 47 min 52 sec

Traceback (most recent call last):
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\support\__init__.py", line 979, in wrapper
    return f(self, maxsize)
           ^^^^^^^^^^^^^^^^
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\test_hashlib.py", line 540, in test_case_md5_uintmax
    self.check('md5', b'A'*size, '28138d306ff1b8281f1a9067e1a1a2b3')
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\test_hashlib.py", line 394, in check
    self.check_file_digest(name, data, hexdigest)
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\test_hashlib.py", line 407, in check_file_digest
    f.write(data)
OSError: [Errno 28] No space left on device


Traceback (most recent call last):
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\support\__init__.py", line 979, in wrapper
    return f(self, maxsize)
           ^^^^^^^^^^^^^^^^
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\test_hashlib.py", line 535, in test_case_md5_huge
    self.check('md5', b'A'*size, 'c9af2dff37468ce5dfee8f2cfc0a9c6d')
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\test_hashlib.py", line 394, in check
    self.check_file_digest(name, data, hexdigest)
  File "R:\buildarea\3.x.ambv-bb-win11.bigmem\build\Lib\test\test_hashlib.py", line 407, in check_file_digest
    f.write(data)
OSError: [Errno 28] No space left on device