Python class hierarchy, extend ancestor data

Question

I have a hierarchy of entities that should contain dictionaries with metadata. Any sub-class should inherit its ancestor dictionary and be able to add some key-values (delete inherit keys is not needed). The easiest solution I can think about is this:

class A(object):
 def __init__(self):
 self.doc = A.generate_doc()
 @staticmethod
 def generate_doc():
 return {'a': 0}
class B(A):
 def __init__(self):
 super().__init__()
 self.doc.update(B.generate_doc())
 @staticmethod
 def generate_doc():
 return {'b': 0}
class C(B):
 def __init__(self):
 super().__init__()
 self.doc.update(C.generate_doc())
 @staticmethod
 def generate_doc():
 return {'c': 0}
print(A().doc) # {'a': 0}
print(B().doc) # {'a': 0, 'b': 0}
print(C().doc) # {'a': 0, 'b': 0, 'c': 0}

Is this a good design? Could the update() blocks be implicit maybe? I have more than one document in real code, so that would be nice.

Answer 1

Is this a good design?

No IMO it's quite horrible, please reconsider it.

Could the update() blocks be implicit maybe?

Yeah, use metaclasses.

I have more than one document in real code, so that would be nice.

Yeah, don't request us to write code for you. However, with a metaclass this is easy.

---

To make a cross Python 2 and Python 3 metaclass you need to use something like Class = Class('name', (object,), {}). You want to also build these dictionaries in the metaclasses __new__, and for the values to be taken from another attribute. To do this I defaulted to the uppercase of the value. I.e. A.DOC.

This works kinda like Python's default multiple inheritance, but I make no guarantees. Also doc is now a static value, so you may need to clone it at object instantiation.

def update(values=[]):
 if isinstance(values, str):
 values = values.split()
 class Meta(type):
 def __new__(meta, classname, bases, class_dict):
 for value in values:
 val = {}
 for base in reversed(bases):
 val.update(getattr(base, value, {}))
 val.update(class_dict.get(value.upper(), {}))
 class_dict[value] = val
 return super(Meta, meta).__new__(meta, classname, bases, class_dict)
 return Meta('Meta', (object,), {})
class A(update('doc')):
 DOC = {'v': 0}
class B(A):
 DOC = {'v': 1}
class C(A):
 DOC = {'c': 0}
class D(B, C):
 pass
class E(C, B):
 pass
print(D.doc)
print(E.doc)