metaclass - Writing and using metaclasses

Metaclasses are a very powerful tool in Python. You can control the entire class creation process with them.

Most of the time, however, they are too powerful. This module helps you to use some of the advantages of metaclasses, without having to know all the details. It defines a base class SubclassInit. Inheriting from this class one can modify the subclass creation process.

Initializing subclasses

A very common usecase for a metaclass is that you just want to execute some code after a class is created. This can easily done with SubclassInit. You just define a method __init_subclass__, which is implicitly considered a @classmethod and will be called after each subclass that is generated of your class. As a parameter it gets the namespace of the class. An example is a simple subclass registration:

class Register(SubclassInit):
    subclasses = []

    def __subclass_init__(cls, ns, **kwargs):
        super().__subclass_init__(ns, **kwargs)
        Register.subclasses.append(cls)

Note how you can add keyword arguments. Those are the keyword arguments given on the class definition line, as in:

class Subclass(Base, spam="ham"):
    pass

Don't forget to properly call super()! Other classes may want to initialize subclasses as well. This is also why you should pass over the keyword arguments, just taking out the ones you need.

Initializing Descriptors

Descriptors are a powerful technique to create object attributes which calculate their value on-the-fly. A property is a simple example of such a descriptor. There is a common problem with those descriptors: they do not know their name. Using SubclassInit you can add an __init_descriptor__ method to a descriptor which gets called once the class is ready and the descriptor's name is known.

As an example, we can define a descriptor which makes an attribute a weak reference:

import weakref

class WeakAttribute:
    def __get__(self, instance, owner):
        return instance.__dict__[self.name]()

    def __set__(self, instance, value):
        instance.__dict__[self.name] = weakref.ref(value)

    def __init_descriptor__(self, owner, name):
        self.name = name

Order of Attributes

Sometimes one is interested in which order the attributes were defined in the class. SubclassInit leaves a tuple with all the names of the attributes in the order they were defined as a class attribute called __attribute_order__. Note that Python already defines some class attributes, like __module__, some of which also show up in this tuple.

As an example:

class AttributeOrder(SubclassInit):
    a = 1

    def b(self):
        pass

    c = 5

assert AttributeOrder.__attribute_order__ == \
    ('__module__', '__qualname__', 'a', 'b', 'c')