>>> class A(object): pass
...
>>> A.__dict__
<dictproxy object at 0x173ef30>
>>> A.__dict__.__dict__
Traceback (most recent call last):
File "<string>", line 1, in <fragment>
AttributeError: 'dictproxy' object has no attribute '__dict__'
>>> A.__dict__.copy()
{'__dict__': <attribute '__dict__' of 'A' objects> ... }
>>> A.__dict__['__dict__']
<attribute '__dict__' of 'A' objects> # What is this object?
If I do A.something = 10
, this goes into A.__dict__
. What is this <attribute '__dict__' of 'A' objects>
found in A.__dict__.__dict__
, and when does it contain something?
First of all A.__dict__.__dict__
is different from A.__dict__['__dict__']
, and the former doesn't exist. The latter is the __dict__
attribute that the instances of the class would have. It's a descriptor object that returns the internal dictionary of attributes for the specific instance. In short, the __dict__
attribute of an object can't be stored in object's __dict__
, so it's accessed through a descriptor defined in the class.
To understand this, you'd have to read the documentation of the descriptor protocol.
The short version:
A
, access to instance.__dict__
is provided by A.__dict__['__dict__']
which is the same as vars(A)['__dict__']
.A.__dict__
is provided by type.__dict__['__dict__']
(in theory) which is the same as vars(type)['__dict__']
.The long version:
Both classes and objects provide access to attributes both through the attribute operator (implemented via the class or metaclass's __getattribute__
), and the __dict__
attribute/protocol which is used by vars(ob)
.
For normal objects, the __dict__
object creates a separate dict
object, which stores the attributes, and __getattribute__
first tries to access it and get the attributes from there (before attempting to look for the attribute in the class by utilizing the descriptor protocol, and before calling __getattr__
). The __dict__
descriptor on the class implements the access to this dictionary.
x.name
is equivalent to trying those in order: x.__dict__['name']
, type(x).name.__get__(x, type(x))
, type(x).name
x.__dict__
does the same but skips the first one for obvious reasonsAs it's impossible for the __dict__
of instance
to be stored in __dict__
of the instance, it is accessed through the descriptor protocol directly instead, and is stored in a special field in the instance.
A similar scenario is true for classes, although their __dict__
is a special proxy object that pretends to be a dictionary (but might not be internally), and doesn't allow you to change it or replace it with another one. This proxy allows you, among all else, to access the attributes of a class that are specific to it, and not defined in one of its bases.
By default, a vars(cls)
of an empty class carries three descriptors - __dict__
for storing the attributes of the instances, __weakref__
which is used internally by weakref
, and the docstring of the class. The first two might be gone if you define __slots__
. Then you wouldn't have __dict__
and __weakref__
attributes, but instead you'd have a single class attribute for each slot. The attributes of the instance then wouldn't be stored in a dictionary, and access to them will be provided by the respective descriptors in the class.
And lastly, the inconsistency that A.__dict__
is different from A.__dict__['__dict__']
is because the attribute __dict__
is, by exception, never looked up in vars(A)
, so what is true for it isn't true for practically any other attribute you'd use. For example, A.__weakref__
is the same thing as A.__dict__['__weakref__']
. If this inconsistency didn't exist, using A.__dict__
would not work, and you'd have to always use vars(A)
instead.