Interrupting class creation in __init_subclass__ may lead to incorrect isinstance() and issubclass() results

[xitop]

BPO	38085
Nosy	@rhettinger, @cfbolz, @ambv, @hongweipeng
PRs	bpo-38085: Fix throw exception in __init__subclass__ causes wrong isinstance() a… #30112
Files	x.py

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = None
created_at = <Date 2019-09-10.09:09:41.066>
labels = ['interpreter-core', '3.8', 'type-bug', '3.7', '3.9']
title = 'Interrupting class creation in __init_subclass__ may lead to incorrect isinstance() and issubclass() results'
updated_at = <Date 2021-12-23.20:42:44.132>
user = 'https://bugs.python.org/xitop'

bugs.python.org fields:

activity = <Date 2021-12-23.20:42:44.132>
actor = 'Carl.Friedrich.Bolz'
assignee = 'none'
closed = False
closed_date = None
closer = None
components = ['Interpreter Core']
creation = <Date 2019-09-10.09:09:41.066>
creator = 'xitop'
dependencies = []
files = ['50515']
hgrepos = []
issue_num = 38085
keywords = ['patch']
message_count = 9.0
messages = ['351599', '355037', '358966', '381210', '383796', '401784', '401998', '409106', '409107']
nosy_count = 5.0
nosy_names = ['rhettinger', 'Carl.Friedrich.Bolz', 'lukasz.langa', 'hongweipeng', 'xitop']
pr_nums = ['30112']
priority = 'normal'
resolution = None
stage = 'patch review'
status = 'open'
superseder = None
type = 'behavior'
url = 'https://bugs.python.org/issue38085'
versions = ['Python 3.6', 'Python 3.7', 'Python 3.8', 'Python 3.9']

[xitop]

An exception in __init__subclass__ leads under certain circumstances to wrong isinstance() and issubclass() results. The exception probably leaves Python internal data in inconsistent state.

Here is a demonstration program from Stack Overflow:

from abc import ABCMeta

class Animal(metaclass=ABCMeta):
    pass

class Plant(metaclass=ABCMeta):
    def __init_subclass__(cls):
        assert not issubclass(cls, Animal), "Plants cannot be Animals"

class Dog(Animal):
    pass

try:
    class Triffid(Animal, Plant):
        pass
except Exception:
    pass

print("Dog is Animal?", issubclass(Dog, Animal))
print("Dog is Plant?", issubclass(Dog, Plant))

Result is:

Dog is Animal? True
Dog is Plant? True

Changing the order of the print statements will result in:

Dog is Plant? False
Dog is Animal? False

Another ill-behaving program and a partial analysis can be found at SO: https://stackoverflow.com/q/57848663/5378816

[xitop]

Same problem also in Python 3.6.8 and the new 3.8.0.

[xitop]

Please, could some experienced Python developer take a look at this issue reported more than 3 month ago?

[hongweipeng]

Class Triffid failed to create in __init_subclass__, so Triffid._abc_cache uses its base class _abc_cache.

[xitop]

Python 3.9.1 is affected too.

[xitop]

2nd anniversary. Any reaction from developers would be appreciated. Thank you.

[rhettinger]

Confirmed that this weird behavior is still present. Am not sure when someone will have the time and inclination to drill into the cause.

[cfbolz]

hm, I think I figured it out. The root cause is that even though the creation of the class Triffid fails, it can still be found via Animal.__subclasses__(), which the special subclass logic for ABCs is looking at. Triffid fills its _abc_impl data with some content, but Triffid._abc_impl was never successfully initialized, therefore it mutates the _abc_impl of its first base class Animal.

My conclusion would be that if a class is not successfully created, it shouldn't appear in the .__subclasses__() list of its bases.

See attached script for some illuminating prints.

[cfbolz]

Or, in other words, in my opinion this is the root cause of the bug:

class Base:
    def __init_subclass__(cls):
        global broken_class
        broken_class = cls
        assert 0
try:
    class Broken(Base): pass
except: pass
assert broken_class not in Base.__subclasses__()

The assert fails, which imo it shouldn't.