Issue 23572: functools.singledispatch fails when "not BaseClass" is True

I admit this case is rather convoluted, but I have been debugging a few hours so I think I should share my findings.

I have a metaclass MetaA that provides to the classes constructed with it have a dictionary interface. Of all the functions only __len__ is important. In some cases, the result of __len__ is 0 and that is what triggers the error

class MetaA(type):
    def __len__(self):
        return 0

class A(metaclass=MetaA):
    pass

class AA(A):
    pass

Now, I construct a function with single dispatch and register the case of class A but not the class AA

@singledispatch
def fun(a):
    print('base case')

@fun.register(A)
def _(a):
    print('fun A')

And then, call fun with an object of class AA

fun(AA())

This should call the function for the class up in the hierarchy, A
Instead it raises an exception 

RuntimeError: Inconsistent hierarchy

in function functools._c3_merge
because in the line

if not candidate:
    raise RuntimeError("Inconsistent hierarchy")

"not candidate" evaluates to True due to __len__ returning 0


This can be avoided by:

* adding __nonzero__ to MetaA

* not adding the map interface to a class in the first place

* changing the code in _c3_merge 

I'm not really sure, but instead of:

if not candidate:
    raise RuntimeError("Inconsistent hierarchy")

would this work?

if candidate is None:
    raise RuntimeError("Inconsistent hierarchy")


I attach a test case

Yes, this needs addressing.

An example is Enum.

Attached is patch and test case.

I was also bitten by this via Enum. Is there any chance this will be fixed in Python 3.5?

New changeset 73984e665bf5 by Yury Selivanov in branch '3.5':
Issue #23572: Fixed functools.singledispatch on classes with falsy metaclasses.
https://hg.python.org/cpython/rev/73984e665bf5

New changeset 94d0c219d46f by Yury Selivanov in branch 'default':
Merge 3.5 (issue #23572)
https://hg.python.org/cpython/rev/94d0c219d46f

> I was also bitten by this via Enum. Is there any chance this will be fixed in Python 3.5?

It will be fixed in 3.5.1.

Thanks for the patch, Ethan!

New changeset 586195685aaf by Yury Selivanov in branch '3.4':
Issue #23572: Fixed functools.singledispatch on classes with falsy metaclasses.
https://hg.python.org/cpython/rev/586195685aaf

Thank you for fixing this and sorry for not being responsive sooner!

nosy: + yselivanov
messages: + msg248783

resolution: fixed

stage: patch review