In python 2.3.3 this code cause assertion:

class d(dict):
    _is_setted = None
    def __setitem__(self, k, v):
        self._is_setted = 1
 
class test:
    def __init__(self):
        self.__dict__ = d()
         
t = test()
t.x = 2
assert t.__dict__._is_setted == 1, 'no `__setitem__` calls'

Logged In: YES 
user_id=80475

Underneath the hood, test's setattr is accessing the
dictionary directly instead of going through its __setattr__
slot.

While this one is fixable, I'm not sure it's really a bug. 
There are no documentation promises that things like setattr
won't make direct accesses to the underlying dictionary. 
Python's implementation has tons of direct access code --
the reasons include clarity, speed, avoidance of hard to
debug code paths, or just not having been looked at in this way.

In any case, I would not have expected old style classes
like 'test' to know about new style subclasses.

For the use cases modeled after the OP's code, overiding
setattr in 'test' is likely the way to go.

Logged In: YES 
user_id=33168

I agree with Raymond.  I think this pretty much the design
and should not be fixed.  Therefore, I'm closing this report.

Viktor, if you disagree, you might try discussing this on
comp.lang.python or brining up this issue on python-dev. 
Feel free to comment here if you would like clarifications.