compat: move check for closure cellvar update wrapper function

[torque]

This sanity checking code ended up getting decoupled from the old implementation in a18b395 when a new, much simpler method was added. This check can fail spuriously in certain environments where it doesn't even need to be run because the newer approach is used.

This is noticeable when using python scripts compiled by nuitka. Since next-gen attrs uses slots on everything by default, the use of this fixup function is now pervasive, as well.

Here's the simplest reproduction I came up with (tested on Python 3.10.9 and nuitka 1.4)

# problem.py

# pip install attrs nuitka ordered-set
# nuitka3 --standalone --onefile problem.py
# running `python problem.py` without this patch succeeds
# running nuitka's output `problem.bin` without this patch fails. With this patch it succeeds.

import attrs

@attrs.define
class Parent:
    a: int

    def __attrs_post_init__(self):
        assert self.__class__ is Child, f"{self.__class__} is not {Child} ({id(self.__class__)}, {id(Child)})"

@attrs.define
class Child(Parent):
    b: float

    def __attrs_post_init__(self):
        assert __class__ is Child, f"{__class__} is not {Child} ({id(__class__)}, {id(Child)})"

        super().__attrs_post_init__()


Child(a=1, b=0.0)

Because nuitka compiles the python code to C, I guess it loses the introspection through the __code__ attribute, which is why the sanity check causes a problem. Fortunately, that's apparently entirely unneeded with any python version 3.8 or above.

(Annotated) Pull Request Check List

• Added tests for changed code.

  I am not sure how to reasonably add a test case for this in-tree because my reproduction requires a heavyweight external process (compiling with nuitka).

• Updated documentation for changed code.

  the changed code is an internal function with minimal documentation. There doesn't seem to be anything to add here

• Changes (and possible deprecations) have news fragments in changelog.d.

[hynek]

Thanks!