Unpickling child class fails if base class __init_subclass__ expects an argument

issue の再現プログラムから始め、cloudpickle.py の _make_skeleton_class を調べてください。ここでは、traceback に子クラス Child が再作成されていることが示されています。シリアライズ中にクラス定義のキーワード引数がどのように処理されるかを追跡し、例を再実行して修正を検証してください。cloudpickle は arg の値を保持したまま Child の unpickle に成功するはずです。

If you have a base class that defines __init_subclass__ with an argument, unpickling a child class will fail because cloudpickle does not pass along the original kwarg when re-creating the class.

This case seems to work fine in the built-in pickle package.

This usage mirrors examples shown in the docs and the original PEP that introduced the __init_subclass__ method.

Python 3.11.4
cloudpickle 2.2.1
EDIT: This also appears to be broken with cloudpickle 3.0.0.

Example:

#!/usr/bin/env python3
import pickle
import cloudpickle
print(cloudpickle.__version__)


class Base:
    def __init_subclass__(cls, /, arg, **kwargs):
        super().__init_subclass__(**kwargs)
        cls.arg = arg


class Child(Base, arg='hello'):
    pass


print(pickle.loads(pickle.dumps(Base)))
print(pickle.loads(pickle.dumps(Child)))
print(cloudpickle.loads(cloudpickle.dumps(Base)))
print(cloudpickle.loads(cloudpickle.dumps(Child)))

2.2.1
<class '__main__.Base'>
<class '__main__.Child'>
<class '__main__.Base'>
Traceback (most recent call last):
  File "/home/tsanders/test_cloudpickle_bug.py", line 20, in <module>
    print(cloudpickle.loads(cloudpickle.dumps(Child)))
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/production-3.11/lib/python3.11/site-packages/cloudpickle/cloudpickle.py", line 827, in _make_skeleton_class
    skeleton_class = types.new_class(
                     ^^^^^^^^^^^^^^^^
  File "/opt/conda/envs/production-3.11/lib/python3.11/types.py", line 75, in new_class
    return meta(name, resolved_bases, ns, **kwds)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
TypeError: Base.__init_subclass__() missing 1 required positional argument: 'arg'