Assigning attribute to instance method causes AttributeError

Bug report

Assigning (and reassigning) attributes to functions works as expected as with any object.

def bar(self):
    ...


bar.something = 1
print(bar.something)
# prints: 1

bar.something = 2
print(bar.something)
# prints: 2

However, the experience is not consistent with class methods. When dealing with the class, you can assign and reassign similar to regular functions.

class Foo:
    def bar(self):
        ...


Foo.bar.something = 1
print(Foo.bar.something)
# prints: 1

Foo.bar.something = 2
print(Foo.bar.something)
# prints: 2

This even carries through after instantiation:

foo = Foo()
print(foo.bar.something)
# prints: 2

We can clearly see the value exists in the __dict__:

print(foo.bar.__dict__)
# {'something': 2}

However, trying to assign to the method from the instance fails:

foo.bar.something = 2
# AttributeError: 'method' object has no attribute 'something'

While I guess I can see why this might be (since you would be modifying an object on the class and not the instance--if this indeed is the rationale), this is not intuitive. It seems inconsistent with setting attributes otherwise.

If this behavior is intended and not a bug, then perhaps the error should be more explicit. The error stating that the method has no attribute named something is not correct. It does have the attribute and we can otherwise interact with it, just not set it.

Either we should be able to assign back to that attribute from the instance, or we should receive a more explicit error why this is not possible.

Your environment

• CPython versions tested on: 3.7.12, 3.8.11, 3.9.9, 3.10.8, 3.11.0rc1
• Operating system and architecture:

▶ uname -a           
Linux hopkins 6.0.2-arch1-1 #1 SMP PREEMPT_DYNAMIC Sat, 15 Oct 2022 14:00:49 +0000 x86_64 GNU/Linux

Commencez par reproduire les exemples d’affectation d’attributs aux méthodes d’instance de l’issue avec les versions de Python concernées. Examinez le comportement des attributs de méthode et des descripteurs avant de décider si l’affectation doit être prise en charge ou si l’erreur doit être clarifiée ; le travail est considéré comme terminé lorsqu’un comportement convenu et la couverture de régression correspondante sont en place.