Access base class attribute in derived class - in “class scope”

Question

class Outer(object):
    class InnerBase(object): _var = {'foo', 'bar'}
    class Derived(InnerBase):
        _var = _var | {'baz'} # NameError: name '_var' is not defined
        _var = InnerBase._var | {'baz'} #  name 'InnerBase' is not defined
        _var = Outer.InnerBase._var | {'baz'} #  free variable 'Outer'
        # referenced before assignment in enclosing scope

Moving _var in Outer does not help - moving it in module scope would work but defeats the purpose of having classes. So how to go about that ?

EDIT: coming from Java so the scoping rules of classes are a head scratcher for me - a briefing would be appreciated. This works btw:

    class Derived(InnerBase): pass
    Derived._var = InnerBase._var | {'baz'}

but it's not the pinnacle of elegance.

Related: Nested classes' scope? - but here we specifically want to access our parent class (rather than the Outer type)

EDIT2: What I am actually after is a _var = __class__._var-like syntax (or hack), or an explanation as to why it's not there

Answer 1

You can bypass the class statement and use an explicit call to type.

class Outer(object):
    class InnerBase(object): _var = {'foo', 'bar'}
    Derived = type('Derived',
                   (InnerBase,),
                   {'_var': InnerBase._var | {'baz'}}
                  )

This works because Derived._var is not being set via an assignment statement in the class statement defining Derived, but is imported from a dictionary you create in the same environment as InnerBase itself.

Answer 2

Python never searches for a name in enclosing class statements. Mark Lutz uses the acronym LEGB to summarize scope in his introduction to Python (Learning Python): Python searches the local scope, then the local scope of any enclosing def statements, then the global scope, and finally the built-in scope. Class statements are excluded from this scope list; Python does not search enclosing class statements for a name.

One solution is to un-nest your classes. In Python, using un-nested classes is often preferred for its simplicity. But, of course, there are good reasons to nest classes as well. Why have you nested InnerBase? I wonder if you might have nested that class because of your experience in Java. Would the following work for you just as well?

class InnerBase(object):
    _var = {'foo', 'bar'}

class Derived(InnerBase):
    _var = InnerBase._var | {'baz'}

>>> Derived._var
set(['baz', 'foo', 'bar'])

The moment you nest these two class statements under another class statement they will be excluded from name searches, since they have become part of the larger class statement and are thus excluded from the searching of the various scopes.

Answer 3

When you put it in a method it works.

    def __init__(self):
        _var = Outer.InnerBase._var | {'baz'}
        #_var = super(Outer.Derived, self)._var | {'baz'}    # or with super
        print(_var)

So my impression is that it's all about initialization.