We Keep Coding

Why differences in class.__slots__ assignment via decorator vs. class body?

I'm working on a decorator to implement some behaviors for an immutable class. I'd like a class to inherit from namedtuple (to have attribute immutability) and also want to add some new methods. Like this ... but correctly preventing new attributes being assigned to the new class.
When inheriting from namedtuple, you should define __new__ and set __slots__ to be an empty tuple (to maintain immutability):
def define_new(clz):
def __new(cls, *args, **kwargs):
return super(clz, cls).__new__(cls, *args, **kwargs)
clz.__new__ = staticmethod(__new) # delegate namedtuple.__new__ to namedtuple
return clz
#define_new
class C(namedtuple('Foo', "a b c")):
__slots__ = () # Prevent assignment of new vars
def foo(self): return "foo"
C(1,2,3).x = 123 # Fails, correctly
... great. But now I'd like to move the __slots__ assignment into the decorator:
def define_new(clz):
def __new(cls, *args, **kwargs):
return super(clz, cls).__new__(cls, *args, **kwargs)
#clz.__slots__ = ()
clz.__slots__ = (123) # just for testing
clz.__new__ = staticmethod(__new)
return clz
#define_new
class C(namedtuple('Foo', "a b c")):
def foo(self): return "foo"
c = C(1,2,3)
print c.__slots__ # Is the (123) I assigned!
c.x = 456 # Assignment succeeds! Not immutable.
print c.__slots__ # Is still (123)
Which is a little surprising.
Why has moving the assignment of __slots__ into the decorator caused a change in behavior?
If I print C.__slots__, I get the object I assigned. What do the x get stored?

The code doesn't work because __slots__ is not a normal class property consulted at run-time. It is a fundamental property of the class that affects the memory layout of each of its instances, and as such must be known when the class is created and remain static throughout the its lifetime. While Python (presumably for backward compatibility) allows assigning to __slots__ later, the assignment has no effect on the behavior of existing or future instances.
How __slots__ is set
The value of __slots__ determined by the class author is passed to the class constructor when the class object is being created. This is done when the class statement is executed; for example:
class X:
__slots__ = ()
The above statement is equivalent1 to creating a class object and assigning it to X:
X = type('X', (), {'__slots__': ()})
The type object is the metaclass, the factory that creates and returns a class when called. The metaclass invocation accepts the name of the type, its superclasses, and its definition dict. Most of the contents of the definition dict can also be assigned later The definition dict contains directives that affect low-level layour of the class instances. As you discovered, later assignment to __slots__ simply has no effect.
Setting __slots__ from the outside
To modify __slots__ so that it is picked up by Python, one must specify it when the class is being created. This can be accomplished with a metaclass, the type responsible for instantiating types. The metaclass drives the creation of the class object and it can make sure __slots__ makes its way into the class definition dict before the constructor is invoked:
class DefineNew(type):
def __new__(metacls, name, bases, dct):
def __new__(cls, *new_args, **new_kwargs):
return super(defcls, cls).__new__(cls, *new_args, **new_kwargs)
dct['__slots__'] = ()
dct['__new__'] = __new__
defcls = super().__new__(metacls, name, bases, dct)
return defcls
class C(namedtuple('Foo', "a b c"), metaclass=DefineNew):
def foo(self):
return "foo"
Testing results in the expected:
>>> c = C(1, 2, 3)
>>> c.foo()
'foo'
>>> c.bar = 1
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
AttributeError: 'C' object has no attribute 'bar'
Metaclass mixing pitfall
Note that the C type object will itself be an instance of DefineMeta - which is not surprising, since that follows from the definition of a metaclass. But this might cause an error if you ever inherit from both C and a type that specifies a metaclass other than type or DefineMeta. Since we only need the metaclass to hook into class creation, but are not using it later, it is not strictly necessary for C to be created as an instance of DefineMeta - we can instead make it an instance of type, just like any other class. This is achieved by changing the line:
defcls = super().__new__(metacls, name, bases, dct)
to:
defcls = type.__new__(type, name, bases, dct)
The injection of __new__ and __slots__ will remain, but C will be a most ordinary type with the default metaclass.
In conclusion...
Defining a __new__ which simply calls the superclass __new__ is always superfluous - presumably the real code will also do something different in the injected __new__, e.g. provide the default values for the namedtuple.
1
In the actual class definition the compiler adds a couple of additional items to the class dict, such as the module name. Those are useful, but they do not affect the class definition in the fundamental way that __slots__ does. If X had methods, their function objects would also be included in the dict keyed by function name - automatically inserted as a side effect of executing the def statement in the class definition namespace dict.

__slots__ has to be present during class creation. It affects the memory layout of a class's instances, which isn't something you can just change at will. (Imagine if you already had instances of the class and you tried to reassign the class's __slots__ at that point; instances would all break.) The processing that bases the memory layout on __slots__ only happens during class creation.
Assigning __slots__ in a decorator is too late to do anything. It has to happen before class creation, in the class body or a metaclass __new__.
Also, your define_new is pointless; namedtuple.__new__ already does what you need your __new__ to do.

How can I know whether an object exists or not safely? [duplicate]

I tried to return a value in a class constructor (init):
class A:
def __init__(self):
return 1
but there is a run-time error saying init should return None. If this is the case, how to understand:
a=A()
where "a" is assigned as the class instance?

Strictly speaking, it's not A.__new__() that's creating the the instance a.
When you define class A(object): (or class A: as well if you are using Python3, class A: is the old-style class that has been deprecated), it is __new__ from the inheritedobject.__new__() that is being called to create the instance a.
When a = A() is executed, what happens is:
A() is a shorthand for A.__call__
object.__new__(cls, *args, **kwargs) where cls=A, is what actually happens under hood to create instance a. It allocates the memory for the new object, and should then return a new object (the instance).
if and only if the newly created object is returned does __init__(self) then get called with the newly created object passed to it to "initilize" the object.
Consider the following demo:
when we override __new__ and no longer returns an object, __init__ will
not get called:
class A(object):
def __new__(cls, *args, **kwargs):
print cls, args, kwargs
def __init__(self):
self.x = 'init!'
print self.x
In : a = A()
<class '__main__.A'> () {}
# note that "init!" has not appeared here because __new__ didn't return an
# new instance
now, return a new instance by using object.__new__, and you will see that
after __new__, __init__ would be called as well:
class A(object):
def __new__(cls, *args, **kwargs):
print cls, args, kwargs
return object.__new__(cls, args, kwargs)
def __init__(self):
self.x = 'init!'
print self.x
In : a = A()
<class '__main__.A'> () {}
init!
Here is another demo to display the difference, note that instance a can be created without calling __init__():
class A(object):
def __init__(self):
self.x = "init!"
print self.x
In : a = object.__new__(A)
In : a
Out: <__main__.A at 0x103466450>
In : a.__dict__
Out: {}
In : aa = A()
init!
In : aa
Out: <__main__.A at 0x1033ddf50>
In : aa.__dict__
Out: {'x': 'init!'}
Now for the inquisitive (and also to refresh my own memory =]):
Roughly speaking, there are two main ways to create new objects in Python:
Create new object (type / class) by subclassing:
class statements tells Python to create a new type / class object(by
subclassing an existing type/class such as object):
class Hello(object):
pass
>>> Hello.__class__
<type 'type'>
In fact all class/type object have type type. The type of type
(type(type)) is still type.
You can subclass a type / class object.
Create new object (instance) by instantiating:
You can also create a new object by instatiating an existing type object.
This is done via the using the __call__ operator (shorthand by ()):
>>> h = hello()
>>> type(h)
<class '__main__.Hello'>
>>> type(int('1'))
<type 'int'>
You cannot subclass an instance object.
(note that you can also create a new instance object by some other means such
as using the list operator [1,2,3], in this case it creates an list instance)
You can check an object's type by type(my_obj) or my_object.__class__.
Now you know how an instance object is created, but what really creates the type / class object (that allows to create instance objects)?
In fact, these objects are created by instantiation as well, albeit it is a
slightly different kind of instantiation from what was mentioned earlier.
Aside from class statement, you can also use
type(cls_name, parent_class_tuple, attr_dict) to create a new class.
For eg:
type('Hello', (object,), {})
will create the Hello class same as the one shown earlier.
What is type? Enter metaclass.
type is a metaclass, which is the class of class, i.e., classes are
instances of metaclasses. The __class__ of type is still type.
So here is a graph that shows the relationships between metaclass, class,
instance:
instantiate instantiate
metaclass --------------> class ----------------> instance
type.__new__() object.__new__()
When metaclass type is called to create a new class, the similar flow goes:
type.__call__() is excuted
type.__new__() allocates memory and then returns new a class (a metaclass instace), and then calls type.__init__().
type.__init__() initalizes the newly created class that was passed from step 2.
You can even create a new metaclass by subclassing type:
class MyMeta(type):
def __new__(meta, name, bases, dct):
# do something
return super(MyMeta, meta).__new__(meta, name, bases, dct)
def __init__(cls, name, bases, dct):
# do something
super(MyMeta, cls).__init__(name, bases, dct)
then you can create a new class from this MyMeta metaclass just like you do
with type:
MyClass = MyMeta('MyClass', (object, ), {'x': 1})
Or, use __metaclass__ when defining your class, which has exactly the
same effect as what was shown above:
class MyClass(object):
__metaclass__ = MyMeta
x = 1

It works like this:
You do:
a = A()
A.__new__() is called, and returns an instance of the class A.
Equivalently:
a = A.__new__(A)
Python then calls
a.__init__()
Which, as the error message says, should not return a value.

As you know, everything in Python is an object, and classes are not an exception. Class A is an object with a method __call__(...) which is what allows you to write things like a = A(). Under the hood, A.__call__ will first create an object using a static method __new__ (which has a reasonable default implementation) and then calls the new object's __init__ with the proper constructor args.

The requirement that __init__ should return None is a concious implementation choice, that lets you rely on the fact that object of type class consistantly returns an object
of type instance of class. It's the same in many other languages such as C++.
If implemented differently it would create useless confusion.

Python - type(name,bases,dict)

Docs:
With three arguments, return a new type object. This is essentially a dynamic form of the class statement. The name string is the class name and becomes the __name__ attribute; the bases tuple itemizes the base classes and becomes the __bases__ attribute; and the dict dictionary is the namespace containing definitions for class body and becomes the __dict__ attribute.
While learning Python I have come across this use of type as "a dynamic form of the class statement", this seems very interesting and useful and I wish to understand it better. Can someone clearly explain the role of __name__, __bases__, and __dict__ in a class and also give example where type(name, bases, dict) comes into its own right.

When you define a class:
class Foo(Base1, Base2):
bar = 'baz'
def spam(self):
return 'ham'
Python essentially does this:
def spam(self):
return 'ham'
body = {'bar': 'baz', 'spam': spam}
Foo = type('Foo', (Base1, Base2), body)
where Foo is then added to the namespace the class statement was defined in.
The block under the class statement is executed as if it was a function with no arguments and the resulting local namespace (a dictionary) forms the class body; in this case a dictionary with 'bar' and 'spam' keys is formed.
You can achieve the same result by passing a name string, a tuple of base classes and a dictionary with string keys to the type() function.
Once you start exploring metaclasses as well (which are essentially subclasses of type()
, letting you customize how classes are created), you'll find that body doesn't have to be a plain dictionary. PEP 3115 - Metaclasses in Python 3 expanded the possible values to anything dict like, letting you implement all sorts of interesting class behaviour by letting you use augmented dictionaries for the class body. The new Python 3.4 enum module for example, uses an OrderedDict() instance instead, to preserve attribute ordering.

__name__ is a string containing the name of the class.
__bases__ is a tuple of classes from which the current class derives.
__dict__ is a dictionary of all methods and fields defined in the class.
The use case for type(name, bases, dict) is when you want to dynamically generate classes at runtime.
Just imagine that you want to create ORM for a database and you want all the classes representing database tables to be generated automatically. You have an idea how these classes should behave but you don't know their names, their fields until you inspect the database schema. Then you can use this function to generate these classes.

Just like
a = Foo()
creates an instance of Foo and assigns it to a,
Foo = type('Foo', (object,), {})
creates an instance of type and assigns it to Foo. Not only does this allow you to create classes dynamically (similar to, but not as restricted as, using lambda to create a function), but it lets you (correctly) think of type not as a function, but as a type itself. An instance of type is a class object. type.__new__ creates the object, type.__init__ initializes it, type.__call__ lets you use the class as a callable object (which in turn is how Foo() creates and initializes instances of Foo).
Understanding type is how you understand metaclasses. When you don't specify a metaclass, Python uses type to create the class, as if you had written (in Python 2.x)
class Foo(object):
__metaclass__ = type
Just like you can subclass ordinary classes, you can subclass type to create your own metaclass!
class mymetaclass(type):
def __new__(cls, name, bases, dict):
print "Creating a subclass of {0} named {1}".format(name, bases)
print "{0} will have the following attributes:".format(name)
for key in dict:
print " * {0}".format(key)
# This class statement is (roughly) identical to:
#
# def foo_init(self, x):
# self.y = x
#
# Foo = mymetaclass('Foo', (object,), { 'a': 3, '__init__': foo_init })
# (The class statement will also ensure that `__module__` and `__metaclass__`
# are in the dict passed to mymetaclass.__new__.)
class Foo(object):
__metaclass__ = mymetaclass
a = 3
def __init__(self, x):
self.y = x
If you run the above, you'll see output that is generate as the class statement is executed, because mymetaclass.__new__ is executed when the class statement is executed.

Why are django model classes not of type "type"? [duplicate]

What are metaclasses? What are they used for?

Classes as objects
Before understanding metaclasses, it helps to understand Python classes more deeply. Python has a very peculiar idea of what classes are, which it borrows from the Smalltalk language.
In most languages, classes are just pieces of code that describe how to produce an object. That is somewhat true in Python too:
>>> class ObjectCreator(object):
... pass
>>> my_object = ObjectCreator()
>>> print(my_object)
<__main__.ObjectCreator object at 0x8974f2c>
But classes are more than that in Python. Classes are objects too.
Yes, objects.
When a Python script runs, every line of code is executed from top to bottom. When the Python interpreter encounters the class keyword, Python creates an object out of the "description" of the class that follows. Thus, the following instruction
>>> class ObjectCreator(object):
... pass
...creates an object with the name ObjectCreator!
This object (the class) is itself capable of creating objects (called instances).
But still, it's an object. Therefore, like all objects:
you can assign it to a variable1
JustAnotherVariable = ObjectCreator
you can attach attributes to it
ObjectCreator.class_attribute = 'foo'
you can pass it as a function parameter
print(ObjectCreator)
1 Note that merely assigning it to another variable doesn't change the class's __name__, i.e.,
>>> print(JustAnotherVariable)
<class '__main__.ObjectCreator'>
>>> print(JustAnotherVariable())
<__main__.ObjectCreator object at 0x8997b4c>
Creating classes dynamically
Since classes are objects, you can create them on the fly, like any object.
First, you can create a class in a function using class:
>>> def choose_class(name):
... if name == 'foo':
... class Foo(object):
... pass
... return Foo # return the class, not an instance
... else:
... class Bar(object):
... pass
... return Bar
...
>>> MyClass = choose_class('foo')
>>> print(MyClass) # the function returns a class, not an instance
<class '__main__.Foo'>
>>> print(MyClass()) # you can create an object from this class
<__main__.Foo object at 0x89c6d4c>
But it's not so dynamic, since you still have to write the whole class yourself.
Since classes are objects, they must be generated by something.
When you use the class keyword, Python creates this object automatically. But as
with most things in Python, it gives you a way to do it manually.
Remember the function type? The good old function that lets you know what
type an object is:
>>> print(type(1))
<type 'int'>
>>> print(type("1"))
<type 'str'>
>>> print(type(ObjectCreator))
<type 'type'>
>>> print(type(ObjectCreator()))
<class '__main__.ObjectCreator'>
Well, type has also a completely different ability: it can create classes on the fly. type can take the description of a class as parameters,
and return a class.
(I know, it's silly that the same function can have two completely different uses according to the parameters you pass to it. It's an issue due to backward
compatibility in Python)
type works this way:
type(name, bases, attrs)
Where:
name: name of the class
bases: tuple of the parent class (for inheritance, can be empty)
attrs: dictionary containing attributes names and values
e.g.:
>>> class MyShinyClass(object):
... pass
can be created manually this way:
>>> MyShinyClass = type('MyShinyClass', (), {}) # returns a class object
>>> print(MyShinyClass)
<class '__main__.MyShinyClass'>
>>> print(MyShinyClass()) # create an instance with the class
<__main__.MyShinyClass object at 0x8997cec>
You'll notice that we use MyShinyClass as the name of the class
and as the variable to hold the class reference. They can be different,
but there is no reason to complicate things.
type accepts a dictionary to define the attributes of the class. So:
>>> class Foo(object):
... bar = True
Can be translated to:
>>> Foo = type('Foo', (), {'bar':True})
And used as a normal class:
>>> print(Foo)
<class '__main__.Foo'>
>>> print(Foo.bar)
True
>>> f = Foo()
>>> print(f)
<__main__.Foo object at 0x8a9b84c>
>>> print(f.bar)
True
And of course, you can inherit from it, so:
>>> class FooChild(Foo):
... pass
would be:
>>> FooChild = type('FooChild', (Foo,), {})
>>> print(FooChild)
<class '__main__.FooChild'>
>>> print(FooChild.bar) # bar is inherited from Foo
True
Eventually, you'll want to add methods to your class. Just define a function
with the proper signature and assign it as an attribute.
>>> def echo_bar(self):
... print(self.bar)
...
>>> FooChild = type('FooChild', (Foo,), {'echo_bar': echo_bar})
>>> hasattr(Foo, 'echo_bar')
False
>>> hasattr(FooChild, 'echo_bar')
True
>>> my_foo = FooChild()
>>> my_foo.echo_bar()
True
And you can add even more methods after you dynamically create the class, just like adding methods to a normally created class object.
>>> def echo_bar_more(self):
... print('yet another method')
...
>>> FooChild.echo_bar_more = echo_bar_more
>>> hasattr(FooChild, 'echo_bar_more')
True
You see where we are going: in Python, classes are objects, and you can create a class on the fly, dynamically.
This is what Python does when you use the keyword class, and it does so by using a metaclass.
What are metaclasses (finally)
Metaclasses are the 'stuff' that creates classes.
You define classes in order to create objects, right?
But we learned that Python classes are objects.
Well, metaclasses are what create these objects. They are the classes' classes,
you can picture them this way:
MyClass = MetaClass()
my_object = MyClass()
You've seen that type lets you do something like this:
MyClass = type('MyClass', (), {})
It's because the function type is in fact a metaclass. type is the
metaclass Python uses to create all classes behind the scenes.
Now you wonder "why the heck is it written in lowercase, and not Type?"
Well, I guess it's a matter of consistency with str, the class that creates
strings objects, and int the class that creates integer objects. type is
just the class that creates class objects.
You see that by checking the __class__ attribute.
Everything, and I mean everything, is an object in Python. That includes integers,
strings, functions and classes. All of them are objects. And all of them have
been created from a class:
>>> age = 35
>>> age.__class__
<type 'int'>
>>> name = 'bob'
>>> name.__class__
<type 'str'>
>>> def foo(): pass
>>> foo.__class__
<type 'function'>
>>> class Bar(object): pass
>>> b = Bar()
>>> b.__class__
<class '__main__.Bar'>
Now, what is the __class__ of any __class__ ?
>>> age.__class__.__class__
<type 'type'>
>>> name.__class__.__class__
<type 'type'>
>>> foo.__class__.__class__
<type 'type'>
>>> b.__class__.__class__
<type 'type'>
So, a metaclass is just the stuff that creates class objects.
You can call it a 'class factory' if you wish.
type is the built-in metaclass Python uses, but of course, you can create your
own metaclass.
The __metaclass__ attribute
In Python 2, you can add a __metaclass__ attribute when you write a class (see next section for the Python 3 syntax):
class Foo(object):
__metaclass__ = something...
[...]
If you do so, Python will use the metaclass to create the class Foo.
Careful, it's tricky.
You write class Foo(object) first, but the class object Foo is not created
in memory yet.
Python will look for __metaclass__ in the class definition. If it finds it,
it will use it to create the object class Foo. If it doesn't, it will use
type to create the class.
Read that several times.
When you do:
class Foo(Bar):
pass
Python does the following:
Is there a __metaclass__ attribute in Foo?
If yes, create in-memory a class object (I said a class object, stay with me here), with the name Foo by using what is in __metaclass__.
If Python can't find __metaclass__, it will look for a __metaclass__ at the MODULE level, and try to do the same (but only for classes that don't inherit anything, basically old-style classes).
Then if it can't find any __metaclass__ at all, it will use the Bar's (the first parent) own metaclass (which might be the default type) to create the class object.
Be careful here that the __metaclass__ attribute will not be inherited, the metaclass of the parent (Bar.__class__) will be. If Bar used a __metaclass__ attribute that created Bar with type() (and not type.__new__()), the subclasses will not inherit that behavior.
Now the big question is, what can you put in __metaclass__?
The answer is something that can create a class.
And what can create a class? type, or anything that subclasses or uses it.
Metaclasses in Python 3
The syntax to set the metaclass has been changed in Python 3:
class Foo(object, metaclass=something):
...
i.e. the __metaclass__ attribute is no longer used, in favor of a keyword argument in the list of base classes.
The behavior of metaclasses however stays largely the same.
One thing added to metaclasses in Python 3 is that you can also pass attributes as keyword-arguments into a metaclass, like so:
class Foo(object, metaclass=something, kwarg1=value1, kwarg2=value2):
...
Read the section below for how Python handles this.
Custom metaclasses
The main purpose of a metaclass is to change the class automatically,
when it's created.
You usually do this for APIs, where you want to create classes matching the
current context.
Imagine a stupid example, where you decide that all classes in your module
should have their attributes written in uppercase. There are several ways to
do this, but one way is to set __metaclass__ at the module level.
This way, all classes of this module will be created using this metaclass,
and we just have to tell the metaclass to turn all attributes to uppercase.
Luckily, __metaclass__ can actually be any callable, it doesn't need to be a
formal class (I know, something with 'class' in its name doesn't need to be
a class, go figure... but it's helpful).
So we will start with a simple example, by using a function.
# the metaclass will automatically get passed the same argument
# that you usually pass to `type`
def upper_attr(future_class_name, future_class_parents, future_class_attrs):
"""
Return a class object, with the list of its attribute turned
into uppercase.
"""
# pick up any attribute that doesn't start with '__' and uppercase it
uppercase_attrs = {
attr if attr.startswith("__") else attr.upper(): v
for attr, v in future_class_attrs.items()
}
# let `type` do the class creation
return type(future_class_name, future_class_parents, uppercase_attrs)
__metaclass__ = upper_attr # this will affect all classes in the module
class Foo(): # global __metaclass__ won't work with "object" though
# but we can define __metaclass__ here instead to affect only this class
# and this will work with "object" children
bar = 'bip'
Let's check:
>>> hasattr(Foo, 'bar')
False
>>> hasattr(Foo, 'BAR')
True
>>> Foo.BAR
'bip'
Now, let's do exactly the same, but using a real class for a metaclass:
# remember that `type` is actually a class like `str` and `int`
# so you can inherit from it
class UpperAttrMetaclass(type):
# __new__ is the method called before __init__
# it's the method that creates the object and returns it
# while __init__ just initializes the object passed as parameter
# you rarely use __new__, except when you want to control how the object
# is created.
# here the created object is the class, and we want to customize it
# so we override __new__
# you can do some stuff in __init__ too if you wish
# some advanced use involves overriding __call__ as well, but we won't
# see this
def __new__(upperattr_metaclass, future_class_name,
future_class_parents, future_class_attrs):
uppercase_attrs = {
attr if attr.startswith("__") else attr.upper(): v
for attr, v in future_class_attrs.items()
}
return type(future_class_name, future_class_parents, uppercase_attrs)
Let's rewrite the above, but with shorter and more realistic variable names now that we know what they mean:
class UpperAttrMetaclass(type):
def __new__(cls, clsname, bases, attrs):
uppercase_attrs = {
attr if attr.startswith("__") else attr.upper(): v
for attr, v in attrs.items()
}
return type(clsname, bases, uppercase_attrs)
You may have noticed the extra argument cls. There is
nothing special about it: __new__ always receives the class it's defined in, as the first parameter. Just like you have self for ordinary methods which receive the instance as the first parameter, or the defining class for class methods.
But this is not proper OOP. We are calling type directly and we aren't overriding or calling the parent's __new__. Let's do that instead:
class UpperAttrMetaclass(type):
def __new__(cls, clsname, bases, attrs):
uppercase_attrs = {
attr if attr.startswith("__") else attr.upper(): v
for attr, v in attrs.items()
}
return type.__new__(cls, clsname, bases, uppercase_attrs)
We can make it even cleaner by using super, which will ease inheritance (because yes, you can have metaclasses, inheriting from metaclasses, inheriting from type):
class UpperAttrMetaclass(type):
def __new__(cls, clsname, bases, attrs):
uppercase_attrs = {
attr if attr.startswith("__") else attr.upper(): v
for attr, v in attrs.items()
}
# Python 2 requires passing arguments to super:
return super(UpperAttrMetaclass, cls).__new__(
cls, clsname, bases, uppercase_attrs)
# Python 3 can use no-arg super() which infers them:
return super().__new__(cls, clsname, bases, uppercase_attrs)
Oh, and in Python 3 if you do this call with keyword arguments, like this:
class Foo(object, metaclass=MyMetaclass, kwarg1=value1):
...
It translates to this in the metaclass to use it:
class MyMetaclass(type):
def __new__(cls, clsname, bases, dct, kwargs1=default):
...
That's it. There is really nothing more about metaclasses.
The reason behind the complexity of the code using metaclasses is not because
of metaclasses, it's because you usually use metaclasses to do twisted stuff
relying on introspection, manipulating inheritance, vars such as __dict__, etc.
Indeed, metaclasses are especially useful to do black magic, and therefore
complicated stuff. But by themselves, they are simple:
intercept a class creation
modify the class
return the modified class
Why would you use metaclasses classes instead of functions?
Since __metaclass__ can accept any callable, why would you use a class
since it's obviously more complicated?
There are several reasons to do so:
The intention is clear. When you read UpperAttrMetaclass(type), you know
what's going to follow
You can use OOP. Metaclass can inherit from metaclass, override parent methods. Metaclasses can even use metaclasses.
Subclasses of a class will be instances of its metaclass if you specified a metaclass-class, but not with a metaclass-function.
You can structure your code better. You never use metaclasses for something as trivial as the above example. It's usually for something complicated. Having the ability to make several methods and group them in one class is very useful to make the code easier to read.
You can hook on __new__, __init__ and __call__. Which will allow you to do different stuff, Even if usually you can do it all in __new__,
some people are just more comfortable using __init__.
These are called metaclasses, damn it! It must mean something!
Why would you use metaclasses?
Now the big question. Why would you use some obscure error-prone feature?
Well, usually you don't:
Metaclasses are deeper magic that
99% of users should never worry about it.
If you wonder whether you need them,
you don't (the people who actually
need them know with certainty that
they need them, and don't need an
explanation about why).
Python Guru Tim Peters
The main use case for a metaclass is creating an API. A typical example of this is the Django ORM. It allows you to define something like this:
class Person(models.Model):
name = models.CharField(max_length=30)
age = models.IntegerField()
But if you do this:
person = Person(name='bob', age='35')
print(person.age)
It won't return an IntegerField object. It will return an int, and can even take it directly from the database.
This is possible because models.Model defines __metaclass__ and
it uses some magic that will turn the Person you just defined with simple statements
into a complex hook to a database field.
Django makes something complex look simple by exposing a simple API
and using metaclasses, recreating code from this API to do the real job
behind the scenes.
The last word
First, you know that classes are objects that can create instances.
Well, in fact, classes are themselves instances. Of metaclasses.
>>> class Foo(object): pass
>>> id(Foo)
142630324
Everything is an object in Python, and they are all either instance of classes
or instances of metaclasses.
Except for type.
type is actually its own metaclass. This is not something you could
reproduce in pure Python, and is done by cheating a little bit at the implementation
level.
Secondly, metaclasses are complicated. You may not want to use them for
very simple class alterations. You can change classes by using two different techniques:
monkey patching
class decorators
99% of the time you need class alteration, you are better off using these.
But 98% of the time, you don't need class alteration at all.

A metaclass is the class of a class. A class defines how an instance of the class (i.e. an object) behaves while a metaclass defines how a class behaves. A class is an instance of a metaclass.
While in Python you can use arbitrary callables for metaclasses (like Jerub shows), the better approach is to make it an actual class itself. type is the usual metaclass in Python. type is itself a class, and it is its own type. You won't be able to recreate something like type purely in Python, but Python cheats a little. To create your own metaclass in Python you really just want to subclass type.
A metaclass is most commonly used as a class-factory. When you create an object by calling the class, Python creates a new class (when it executes the 'class' statement) by calling the metaclass. Combined with the normal __init__ and __new__ methods, metaclasses therefore allow you to do 'extra things' when creating a class, like registering the new class with some registry or replace the class with something else entirely.
When the class statement is executed, Python first executes the body of the class statement as a normal block of code. The resulting namespace (a dict) holds the attributes of the class-to-be. The metaclass is determined by looking at the baseclasses of the class-to-be (metaclasses are inherited), at the __metaclass__ attribute of the class-to-be (if any) or the __metaclass__ global variable. The metaclass is then called with the name, bases and attributes of the class to instantiate it.
However, metaclasses actually define the type of a class, not just a factory for it, so you can do much more with them. You can, for instance, define normal methods on the metaclass. These metaclass-methods are like classmethods in that they can be called on the class without an instance, but they are also not like classmethods in that they cannot be called on an instance of the class. type.__subclasses__() is an example of a method on the type metaclass. You can also define the normal 'magic' methods, like __add__, __iter__ and __getattr__, to implement or change how the class behaves.
Here's an aggregated example of the bits and pieces:
def make_hook(f):
"""Decorator to turn 'foo' method into '__foo__'"""
f.is_hook = 1
return f
class MyType(type):
def __new__(mcls, name, bases, attrs):
if name.startswith('None'):
return None
# Go over attributes and see if they should be renamed.
newattrs = {}
for attrname, attrvalue in attrs.iteritems():
if getattr(attrvalue, 'is_hook', 0):
newattrs['__%s__' % attrname] = attrvalue
else:
newattrs[attrname] = attrvalue
return super(MyType, mcls).__new__(mcls, name, bases, newattrs)
def __init__(self, name, bases, attrs):
super(MyType, self).__init__(name, bases, attrs)
# classregistry.register(self, self.interfaces)
print "Would register class %s now." % self
def __add__(self, other):
class AutoClass(self, other):
pass
return AutoClass
# Alternatively, to autogenerate the classname as well as the class:
# return type(self.__name__ + other.__name__, (self, other), {})
def unregister(self):
# classregistry.unregister(self)
print "Would unregister class %s now." % self
class MyObject:
__metaclass__ = MyType
class NoneSample(MyObject):
pass
# Will print "NoneType None"
print type(NoneSample), repr(NoneSample)
class Example(MyObject):
def __init__(self, value):
self.value = value
#make_hook
def add(self, other):
return self.__class__(self.value + other.value)
# Will unregister the class
Example.unregister()
inst = Example(10)
# Will fail with an AttributeError
#inst.unregister()
print inst + inst
class Sibling(MyObject):
pass
ExampleSibling = Example + Sibling
# ExampleSibling is now a subclass of both Example and Sibling (with no
# content of its own) although it will believe it's called 'AutoClass'
print ExampleSibling
print ExampleSibling.__mro__

Note, this answer is for Python 2.x as it was written in 2008, metaclasses are slightly different in 3.x.
Metaclasses are the secret sauce that make 'class' work. The default metaclass for a new style object is called 'type'.
class type(object)
| type(object) -> the object's type
| type(name, bases, dict) -> a new type
Metaclasses take 3 args. 'name', 'bases' and 'dict'
Here is where the secret starts. Look for where name, bases and the dict come from in this example class definition.
class ThisIsTheName(Bases, Are, Here):
All_the_code_here
def doesIs(create, a):
dict
Lets define a metaclass that will demonstrate how 'class:' calls it.
def test_metaclass(name, bases, dict):
print 'The Class Name is', name
print 'The Class Bases are', bases
print 'The dict has', len(dict), 'elems, the keys are', dict.keys()
return "yellow"
class TestName(object, None, int, 1):
__metaclass__ = test_metaclass
foo = 1
def baz(self, arr):
pass
print 'TestName = ', repr(TestName)
# output =>
The Class Name is TestName
The Class Bases are (<type 'object'>, None, <type 'int'>, 1)
The dict has 4 elems, the keys are ['baz', '__module__', 'foo', '__metaclass__']
TestName = 'yellow'
And now, an example that actually means something, this will automatically make the variables in the list "attributes" set on the class, and set to None.
def init_attributes(name, bases, dict):
if 'attributes' in dict:
for attr in dict['attributes']:
dict[attr] = None
return type(name, bases, dict)
class Initialised(object):
__metaclass__ = init_attributes
attributes = ['foo', 'bar', 'baz']
print 'foo =>', Initialised.foo
# output=>
foo => None
Note that the magic behaviour that Initialised gains by having the metaclass init_attributes is not passed onto a subclass of Initialised.
Here is an even more concrete example, showing how you can subclass 'type' to make a metaclass that performs an action when the class is created. This is quite tricky:
class MetaSingleton(type):
instance = None
def __call__(cls, *args, **kw):
if cls.instance is None:
cls.instance = super(MetaSingleton, cls).__call__(*args, **kw)
return cls.instance
class Foo(object):
__metaclass__ = MetaSingleton
a = Foo()
b = Foo()
assert a is b

Others have explained how metaclasses work and how they fit into the Python type system. Here's an example of what they can be used for. In a testing framework I wrote, I wanted to keep track of the order in which classes were defined, so that I could later instantiate them in this order. I found it easiest to do this using a metaclass.
class MyMeta(type):
counter = 0
def __init__(cls, name, bases, dic):
type.__init__(cls, name, bases, dic)
cls._order = MyMeta.counter
MyMeta.counter += 1
class MyType(object): # Python 2
__metaclass__ = MyMeta
class MyType(metaclass=MyMeta): # Python 3
pass
Anything that's a subclass of MyType then gets a class attribute _order that records the order in which the classes were defined.

One use for metaclasses is adding new properties and methods to an instance automatically.
For example, if you look at Django models, their definition looks a bit confusing. It looks as if you are only defining class properties:
class Person(models.Model):
first_name = models.CharField(max_length=30)
last_name = models.CharField(max_length=30)
However, at runtime the Person objects are filled with all sorts of useful methods. See the source for some amazing metaclassery.

I think the ONLamp introduction to metaclass programming is well written and gives a really good introduction to the topic despite being several years old already.
http://www.onlamp.com/pub/a/python/2003/04/17/metaclasses.html (archived at https://web.archive.org/web/20080206005253/http://www.onlamp.com/pub/a/python/2003/04/17/metaclasses.html)
In short: A class is a blueprint for the creation of an instance, a metaclass is a blueprint for the creation of a class. It can be easily seen that in Python classes need to be first-class objects too to enable this behavior.
I've never written one myself, but I think one of the nicest uses of metaclasses can be seen in the Django framework. The model classes use a metaclass approach to enable a declarative style of writing new models or form classes. While the metaclass is creating the class, all members get the possibility to customize the class itself.
Creating a new model
The metaclass enabling this
The thing that's left to say is: If you don't know what metaclasses are, the probability that you will not need them is 99%.

What are metaclasses? What do you use them for?
TLDR: A metaclass instantiates and defines behavior for a class just like a class instantiates and defines behavior for an instance.
Pseudocode:
>>> Class(...)
instance
The above should look familiar. Well, where does Class come from? It's an instance of a metaclass (also pseudocode):
>>> Metaclass(...)
Class
In real code, we can pass the default metaclass, type, everything we need to instantiate a class and we get a class:
>>> type('Foo', (object,), {}) # requires a name, bases, and a namespace
<class '__main__.Foo'>
Putting it differently
A class is to an instance as a metaclass is to a class.
When we instantiate an object, we get an instance:
>>> object() # instantiation of class
<object object at 0x7f9069b4e0b0> # instance
Likewise, when we define a class explicitly with the default metaclass, type, we instantiate it:
>>> type('Object', (object,), {}) # instantiation of metaclass
<class '__main__.Object'> # instance
Put another way, a class is an instance of a metaclass:
>>> isinstance(object, type)
True
Put a third way, a metaclass is a class's class.
>>> type(object) == type
True
>>> object.__class__
<class 'type'>
When you write a class definition and Python executes it, it uses a metaclass to instantiate the class object (which will, in turn, be used to instantiate instances of that class).
Just as we can use class definitions to change how custom object instances behave, we can use a metaclass class definition to change the way a class object behaves.
What can they be used for? From the docs:
The potential uses for metaclasses are boundless. Some ideas that have been explored include logging, interface checking, automatic delegation, automatic property creation, proxies, frameworks, and automatic resource locking/synchronization.
Nevertheless, it is usually encouraged for users to avoid using metaclasses unless absolutely necessary.
You use a metaclass every time you create a class:
When you write a class definition, for example, like this,
class Foo(object):
'demo'
You instantiate a class object.
>>> Foo
<class '__main__.Foo'>
>>> isinstance(Foo, type), isinstance(Foo, object)
(True, True)
It is the same as functionally calling type with the appropriate arguments and assigning the result to a variable of that name:
name = 'Foo'
bases = (object,)
namespace = {'__doc__': 'demo'}
Foo = type(name, bases, namespace)
Note, some things automatically get added to the __dict__, i.e., the namespace:
>>> Foo.__dict__
dict_proxy({'__dict__': <attribute '__dict__' of 'Foo' objects>,
'__module__': '__main__', '__weakref__': <attribute '__weakref__'
of 'Foo' objects>, '__doc__': 'demo'})
The metaclass of the object we created, in both cases, is type.
(A side-note on the contents of the class __dict__: __module__ is there because classes must know where they are defined, and __dict__ and __weakref__ are there because we don't define __slots__ - if we define __slots__ we'll save a bit of space in the instances, as we can disallow __dict__ and __weakref__ by excluding them. For example:
>>> Baz = type('Bar', (object,), {'__doc__': 'demo', '__slots__': ()})
>>> Baz.__dict__
mappingproxy({'__doc__': 'demo', '__slots__': (), '__module__': '__main__'})
... but I digress.)
We can extend type just like any other class definition:
Here's the default __repr__ of classes:
>>> Foo
<class '__main__.Foo'>
One of the most valuable things we can do by default in writing a Python object is to provide it with a good __repr__. When we call help(repr) we learn that there's a good test for a __repr__ that also requires a test for equality - obj == eval(repr(obj)). The following simple implementation of __repr__ and __eq__ for class instances of our type class provides us with a demonstration that may improve on the default __repr__ of classes:
class Type(type):
def __repr__(cls):
"""
>>> Baz
Type('Baz', (Foo, Bar,), {'__module__': '__main__', '__doc__': None})
>>> eval(repr(Baz))
Type('Baz', (Foo, Bar,), {'__module__': '__main__', '__doc__': None})
"""
metaname = type(cls).__name__
name = cls.__name__
parents = ', '.join(b.__name__ for b in cls.__bases__)
if parents:
parents += ','
namespace = ', '.join(': '.join(
(repr(k), repr(v) if not isinstance(v, type) else v.__name__))
for k, v in cls.__dict__.items())
return '{0}(\'{1}\', ({2}), {{{3}}})'.format(metaname, name, parents, namespace)
def __eq__(cls, other):
"""
>>> Baz == eval(repr(Baz))
True
"""
return (cls.__name__, cls.__bases__, cls.__dict__) == (
other.__name__, other.__bases__, other.__dict__)
So now when we create an object with this metaclass, the __repr__ echoed on the command line provides a much less ugly sight than the default:
>>> class Bar(object): pass
>>> Baz = Type('Baz', (Foo, Bar,), {'__module__': '__main__', '__doc__': None})
>>> Baz
Type('Baz', (Foo, Bar,), {'__module__': '__main__', '__doc__': None})
With a nice __repr__ defined for the class instance, we have a stronger ability to debug our code. However, much further checking with eval(repr(Class)) is unlikely (as functions would be rather impossible to eval from their default __repr__'s).
An expected usage: __prepare__ a namespace
If, for example, we want to know in what order a class's methods are created in, we could provide an ordered dict as the namespace of the class. We would do this with __prepare__ which returns the namespace dict for the class if it is implemented in Python 3:
from collections import OrderedDict
class OrderedType(Type):
#classmethod
def __prepare__(metacls, name, bases, **kwargs):
return OrderedDict()
def __new__(cls, name, bases, namespace, **kwargs):
result = Type.__new__(cls, name, bases, dict(namespace))
result.members = tuple(namespace)
return result
And usage:
class OrderedMethodsObject(object, metaclass=OrderedType):
def method1(self): pass
def method2(self): pass
def method3(self): pass
def method4(self): pass
And now we have a record of the order in which these methods (and other class attributes) were created:
>>> OrderedMethodsObject.members
('__module__', '__qualname__', 'method1', 'method2', 'method3', 'method4')
Note, this example was adapted from the documentation - the new enum in the standard library does this.
So what we did was instantiate a metaclass by creating a class. We can also treat the metaclass as we would any other class. It has a method resolution order:
>>> inspect.getmro(OrderedType)
(<class '__main__.OrderedType'>, <class '__main__.Type'>, <class 'type'>, <class 'object'>)
And it has approximately the correct repr (which we can no longer eval unless we can find a way to represent our functions.):
>>> OrderedMethodsObject
OrderedType('OrderedMethodsObject', (object,), {'method1': <function OrderedMethodsObject.method1 at 0x0000000002DB01E0>, 'members': ('__module__', '__qualname__', 'method1', 'method2', 'method3', 'method4'), 'method3': <function OrderedMet
hodsObject.method3 at 0x0000000002DB02F0>, 'method2': <function OrderedMethodsObject.method2 at 0x0000000002DB0268>, '__module__': '__main__', '__weakref__': <attribute '__weakref__' of 'OrderedMethodsObject' objects>, '__doc__': None, '__d
ict__': <attribute '__dict__' of 'OrderedMethodsObject' objects>, 'method4': <function OrderedMethodsObject.method4 at 0x0000000002DB0378>})

Python 3 update
There are (at this point) two key methods in a metaclass:
__prepare__, and
__new__
__prepare__ lets you supply a custom mapping (such as an OrderedDict) to be used as the namespace while the class is being created. You must return an instance of whatever namespace you choose. If you don't implement __prepare__ a normal dict is used.
__new__ is responsible for the actual creation/modification of the final class.
A bare-bones, do-nothing-extra metaclass would like:
class Meta(type):
def __prepare__(metaclass, cls, bases):
return dict()
def __new__(metacls, cls, bases, clsdict):
return super().__new__(metacls, cls, bases, clsdict)
A simple example:
Say you want some simple validation code to run on your attributes -- like it must always be an int or a str. Without a metaclass, your class would look something like:
class Person:
weight = ValidateType('weight', int)
age = ValidateType('age', int)
name = ValidateType('name', str)
As you can see, you have to repeat the name of the attribute twice. This makes typos possible along with irritating bugs.
A simple metaclass can address that problem:
class Person(metaclass=Validator):
weight = ValidateType(int)
age = ValidateType(int)
name = ValidateType(str)
This is what the metaclass would look like (not using __prepare__ since it is not needed):
class Validator(type):
def __new__(metacls, cls, bases, clsdict):
# search clsdict looking for ValidateType descriptors
for name, attr in clsdict.items():
if isinstance(attr, ValidateType):
attr.name = name
attr.attr = '_' + name
# create final class and return it
return super().__new__(metacls, cls, bases, clsdict)
A sample run of:
p = Person()
p.weight = 9
print(p.weight)
p.weight = '9'
produces:
9
Traceback (most recent call last):
File "simple_meta.py", line 36, in <module>
p.weight = '9'
File "simple_meta.py", line 24, in __set__
(self.name, self.type, value))
TypeError: weight must be of type(s) <class 'int'> (got '9')
Note: This example is simple enough it could have also been accomplished with a class decorator, but presumably an actual metaclass would be doing much more.
The 'ValidateType' class for reference:
class ValidateType:
def __init__(self, type):
self.name = None # will be set by metaclass
self.attr = None # will be set by metaclass
self.type = type
def __get__(self, inst, cls):
if inst is None:
return self
else:
return inst.__dict__[self.attr]
def __set__(self, inst, value):
if not isinstance(value, self.type):
raise TypeError('%s must be of type(s) %s (got %r)' %
(self.name, self.type, value))
else:
inst.__dict__[self.attr] = value

Role of a metaclass' __call__() method when creating a class instance
If you've done Python programming for more than a few months you'll eventually stumble upon code that looks like this:
# define a class
class SomeClass(object):
# ...
# some definition here ...
# ...
# create an instance of it
instance = SomeClass()
# then call the object as if it's a function
result = instance('foo', 'bar')
The latter is possible when you implement the __call__() magic method on the class.
class SomeClass(object):
# ...
# some definition here ...
# ...
def __call__(self, foo, bar):
return bar + foo
The __call__() method is invoked when an instance of a class is used as a callable. But as we've seen from previous answers a class itself is an instance of a metaclass, so when we use the class as a callable (i.e. when we create an instance of it) we're actually calling its metaclass' __call__() method. At this point most Python programmers are a bit confused because they've been told that when creating an instance like this instance = SomeClass() you're calling its __init__() method. Some who've dug a bit deeper know that before __init__() there's __new__(). Well, today another layer of truth is being revealed, before __new__() there's the metaclass' __call__().
Let's study the method call chain from specifically the perspective of creating an instance of a class.
This is a metaclass that logs exactly the moment before an instance is created and the moment it's about to return it.
class Meta_1(type):
def __call__(cls):
print "Meta_1.__call__() before creating an instance of ", cls
instance = super(Meta_1, cls).__call__()
print "Meta_1.__call__() about to return instance."
return instance
This is a class that uses that metaclass
class Class_1(object):
__metaclass__ = Meta_1
def __new__(cls):
print "Class_1.__new__() before creating an instance."
instance = super(Class_1, cls).__new__(cls)
print "Class_1.__new__() about to return instance."
return instance
def __init__(self):
print "entering Class_1.__init__() for instance initialization."
super(Class_1,self).__init__()
print "exiting Class_1.__init__()."
And now let's create an instance of Class_1
instance = Class_1()
# Meta_1.__call__() before creating an instance of <class '__main__.Class_1'>.
# Class_1.__new__() before creating an instance.
# Class_1.__new__() about to return instance.
# entering Class_1.__init__() for instance initialization.
# exiting Class_1.__init__().
# Meta_1.__call__() about to return instance.
Observe that the code above doesn't actually do anything more than logging the tasks. Each method delegates the actual work to its parent's implementation, thus keeping the default behavior. Since type is Meta_1's parent class (type being the default parent metaclass) and considering the ordering sequence of the output above, we now have a clue as to what would be the pseudo implementation of type.__call__():
class type:
def __call__(cls, *args, **kwarg):
# ... maybe a few things done to cls here
# then we call __new__() on the class to create an instance
instance = cls.__new__(cls, *args, **kwargs)
# ... maybe a few things done to the instance here
# then we initialize the instance with its __init__() method
instance.__init__(*args, **kwargs)
# ... maybe a few more things done to instance here
# then we return it
return instance
We can see that the metaclass' __call__() method is the one that's called first. It then delegates creation of the instance to the class's __new__() method and initialization to the instance's __init__(). It's also the one that ultimately returns the instance.
From the above it stems that the metaclass' __call__() is also given the opportunity to decide whether or not a call to Class_1.__new__() or Class_1.__init__() will eventually be made. Over the course of its execution it could actually return an object that hasn't been touched by either of these methods. Take for example this approach to the singleton pattern:
class Meta_2(type):
singletons = {}
def __call__(cls, *args, **kwargs):
if cls in Meta_2.singletons:
# we return the only instance and skip a call to __new__()
# and __init__()
print ("{} singleton returning from Meta_2.__call__(), "
"skipping creation of new instance.".format(cls))
return Meta_2.singletons[cls]
# else if the singleton isn't present we proceed as usual
print "Meta_2.__call__() before creating an instance."
instance = super(Meta_2, cls).__call__(*args, **kwargs)
Meta_2.singletons[cls] = instance
print "Meta_2.__call__() returning new instance."
return instance
class Class_2(object):
__metaclass__ = Meta_2
def __new__(cls, *args, **kwargs):
print "Class_2.__new__() before creating instance."
instance = super(Class_2, cls).__new__(cls)
print "Class_2.__new__() returning instance."
return instance
def __init__(self, *args, **kwargs):
print "entering Class_2.__init__() for initialization."
super(Class_2, self).__init__()
print "exiting Class_2.__init__()."
Let's observe what happens when repeatedly trying to create an object of type Class_2
a = Class_2()
# Meta_2.__call__() before creating an instance.
# Class_2.__new__() before creating instance.
# Class_2.__new__() returning instance.
# entering Class_2.__init__() for initialization.
# exiting Class_2.__init__().
# Meta_2.__call__() returning new instance.
b = Class_2()
# <class '__main__.Class_2'> singleton returning from Meta_2.__call__(), skipping creation of new instance.
c = Class_2()
# <class '__main__.Class_2'> singleton returning from Meta_2.__call__(), skipping creation of new instance.
a is b is c # True

A metaclass is a class that tells how (some) other class should be created.
This is a case where I saw metaclass as a solution to my problem:
I had a really complicated problem, that probably could have been solved differently, but I chose to solve it using a metaclass. Because of the complexity, it is one of the few modules I have written where the comments in the module surpass the amount of code that has been written. Here it is...
#!/usr/bin/env python
# Copyright (C) 2013-2014 Craig Phillips. All rights reserved.
# This requires some explaining. The point of this metaclass excercise is to
# create a static abstract class that is in one way or another, dormant until
# queried. I experimented with creating a singlton on import, but that did
# not quite behave how I wanted it to. See now here, we are creating a class
# called GsyncOptions, that on import, will do nothing except state that its
# class creator is GsyncOptionsType. This means, docopt doesn't parse any
# of the help document, nor does it start processing command line options.
# So importing this module becomes really efficient. The complicated bit
# comes from requiring the GsyncOptions class to be static. By that, I mean
# any property on it, may or may not exist, since they are not statically
# defined; so I can't simply just define the class with a whole bunch of
# properties that are #property #staticmethods.
#
# So here's how it works:
#
# Executing 'from libgsync.options import GsyncOptions' does nothing more
# than load up this module, define the Type and the Class and import them
# into the callers namespace. Simple.
#
# Invoking 'GsyncOptions.debug' for the first time, or any other property
# causes the __metaclass__ __getattr__ method to be called, since the class
# is not instantiated as a class instance yet. The __getattr__ method on
# the type then initialises the class (GsyncOptions) via the __initialiseClass
# method. This is the first and only time the class will actually have its
# dictionary statically populated. The docopt module is invoked to parse the
# usage document and generate command line options from it. These are then
# paired with their defaults and what's in sys.argv. After all that, we
# setup some dynamic properties that could not be defined by their name in
# the usage, before everything is then transplanted onto the actual class
# object (or static class GsyncOptions).
#
# Another piece of magic, is to allow command line options to be set in
# in their native form and be translated into argparse style properties.
#
# Finally, the GsyncListOptions class is actually where the options are
# stored. This only acts as a mechanism for storing options as lists, to
# allow aggregation of duplicate options or options that can be specified
# multiple times. The __getattr__ call hides this by default, returning the
# last item in a property's list. However, if the entire list is required,
# calling the 'list()' method on the GsyncOptions class, returns a reference
# to the GsyncListOptions class, which contains all of the same properties
# but as lists and without the duplication of having them as both lists and
# static singlton values.
#
# So this actually means that GsyncOptions is actually a static proxy class...
#
# ...And all this is neatly hidden within a closure for safe keeping.
def GetGsyncOptionsType():
class GsyncListOptions(object):
__initialised = False
class GsyncOptionsType(type):
def __initialiseClass(cls):
if GsyncListOptions._GsyncListOptions__initialised: return
from docopt import docopt
from libgsync.options import doc
from libgsync import __version__
options = docopt(
doc.__doc__ % __version__,
version = __version__,
options_first = True
)
paths = options.pop('<path>', None)
setattr(cls, "destination_path", paths.pop() if paths else None)
setattr(cls, "source_paths", paths)
setattr(cls, "options", options)
for k, v in options.iteritems():
setattr(cls, k, v)
GsyncListOptions._GsyncListOptions__initialised = True
def list(cls):
return GsyncListOptions
def __getattr__(cls, name):
cls.__initialiseClass()
return getattr(GsyncListOptions, name)[-1]
def __setattr__(cls, name, value):
# Substitut option names: --an-option-name for an_option_name
import re
name = re.sub(r'^__', "", re.sub(r'-', "_", name))
listvalue = []
# Ensure value is converted to a list type for GsyncListOptions
if isinstance(value, list):
if value:
listvalue = [] + value
else:
listvalue = [ None ]
else:
listvalue = [ value ]
type.__setattr__(GsyncListOptions, name, listvalue)
# Cleanup this module to prevent tinkering.
import sys
module = sys.modules[__name__]
del module.__dict__['GetGsyncOptionsType']
return GsyncOptionsType
# Our singlton abstract proxy class.
class GsyncOptions(object):
__metaclass__ = GetGsyncOptionsType()

The tl;dr version
The type(obj) function gets you the type of an object.
The type() of a class is its metaclass.
To use a metaclass:
class Foo(object):
__metaclass__ = MyMetaClass
type is its own metaclass. The class of a class is a metaclass-- the body of a class is the arguments passed to the metaclass that is used to construct the class.
Here you can read about how to use metaclasses to customize class construction.

type is actually a metaclass -- a class that creates another classes.
Most metaclass are the subclasses of type. The metaclass receives the new class as its first argument and provide access to class object with details as mentioned below:
>>> class MetaClass(type):
... def __init__(cls, name, bases, attrs):
... print ('class name: %s' %name )
... print ('Defining class %s' %cls)
... print('Bases %s: ' %bases)
... print('Attributes')
... for (name, value) in attrs.items():
... print ('%s :%r' %(name, value))
...
>>> class NewClass(object, metaclass=MetaClass):
... get_choch='dairy'
...
class name: NewClass
Bases <class 'object'>:
Defining class <class 'NewClass'>
get_choch :'dairy'
__module__ :'builtins'
__qualname__ :'NewClass'
Note:
Notice that the class was not instantiated at any time; the simple act of creating the class triggered execution of the metaclass.

Python classes are themselves objects - as in instance - of their meta-class.
The default metaclass, which is applied when when you determine classes as:
class foo:
...
meta class are used to apply some rule to an entire set of classes. For example, suppose you're building an ORM to access a database, and you want records from each table to be of a class mapped to that table (based on fields, business rules, etc..,), a possible use of metaclass is for instance, connection pool logic, which is share by all classes of record from all tables. Another use is logic to to support foreign keys, which involves multiple classes of records.
when you define metaclass, you subclass type, and can overrided the following magic methods to insert your logic.
class somemeta(type):
__new__(mcs, name, bases, clsdict):
"""
mcs: is the base metaclass, in this case type.
name: name of the new class, as provided by the user.
bases: tuple of base classes
clsdict: a dictionary containing all methods and attributes defined on class
you must return a class object by invoking the __new__ constructor on the base metaclass.
ie:
return type.__call__(mcs, name, bases, clsdict).
in the following case:
class foo(baseclass):
__metaclass__ = somemeta
an_attr = 12
def bar(self):
...
#classmethod
def foo(cls):
...
arguments would be : ( somemeta, "foo", (baseclass, baseofbase,..., object), {"an_attr":12, "bar": <function>, "foo": <bound class method>}
you can modify any of these values before passing on to type
"""
return type.__call__(mcs, name, bases, clsdict)
def __init__(self, name, bases, clsdict):
"""
called after type has been created. unlike in standard classes, __init__ method cannot modify the instance (cls) - and should be used for class validaton.
"""
pass
def __prepare__():
"""
returns a dict or something that can be used as a namespace.
the type will then attach methods and attributes from class definition to it.
call order :
somemeta.__new__ -> type.__new__ -> type.__init__ -> somemeta.__init__
"""
return dict()
def mymethod(cls):
""" works like a classmethod, but for class objects. Also, my method will not be visible to instances of cls.
"""
pass
anyhow, those two are the most commonly used hooks. metaclassing is powerful, and above is nowhere near and exhaustive list of uses for metaclassing.

The type() function can return the type of an object or create a new type,
for example, we can create a Hi class with the type() function and do not need to use this way with class Hi(object):
def func(self, name='mike'):
print('Hi, %s.' % name)
Hi = type('Hi', (object,), dict(hi=func))
h = Hi()
h.hi()
Hi, mike.
type(Hi)
type
type(h)
__main__.Hi
In addition to using type() to create classes dynamically, you can control creation behavior of class and use metaclass.
According to the Python object model, the class is the object, so the class must be an instance of another certain class.
By default, a Python class is instance of the type class. That is, type is metaclass of most of the built-in classes and metaclass of user-defined classes.
class ListMetaclass(type):
def __new__(cls, name, bases, attrs):
attrs['add'] = lambda self, value: self.append(value)
return type.__new__(cls, name, bases, attrs)
class CustomList(list, metaclass=ListMetaclass):
pass
lst = CustomList()
lst.add('custom_list_1')
lst.add('custom_list_2')
lst
['custom_list_1', 'custom_list_2']
Magic will take effect when we passed keyword arguments in metaclass, it indicates the Python interpreter to create the CustomList through ListMetaclass. new (), at this point, we can modify the class definition, for example, and add a new method and then return the revised definition.

In addition to the published answers I can say that a metaclass defines the behaviour for a class. So, you can explicitly set your metaclass. Whenever Python gets a keyword class then it starts searching for the metaclass. If it's not found – the default metaclass type is used to create the class's object. Using the __metaclass__ attribute, you can set metaclass of your class:
class MyClass:
__metaclass__ = type
# write here other method
# write here one more method
print(MyClass.__metaclass__)
It'll produce the output like this:
class 'type'
And, of course, you can create your own metaclass to define the behaviour of any class that are created using your class.
For doing that, your default metaclass type class must be inherited as this is the main metaclass:
class MyMetaClass(type):
__metaclass__ = type
# you can write here any behaviour you want
class MyTestClass:
__metaclass__ = MyMetaClass
Obj = MyTestClass()
print(Obj.__metaclass__)
print(MyMetaClass.__metaclass__)
The output will be:
class '__main__.MyMetaClass'
class 'type'

Note that in python 3.6 a new dunder method __init_subclass__(cls, **kwargs) was introduced to replace a lot of common use cases for metaclasses. Is is called when a subclass of the defining class is created. See python docs.

Here's another example of what it can be used for:
You can use the metaclass to change the function of its instance (the class).
class MetaMemberControl(type):
__slots__ = ()
#classmethod
def __prepare__(mcs, f_cls_name, f_cls_parents, # f_cls means: future class
meta_args=None, meta_options=None): # meta_args and meta_options is not necessarily needed, just so you know.
f_cls_attr = dict()
if not "do something or if you want to define your cool stuff of dict...":
return dict(make_your_special_dict=None)
else:
return f_cls_attr
def __new__(mcs, f_cls_name, f_cls_parents, f_cls_attr,
meta_args=None, meta_options=None):
original_getattr = f_cls_attr.get('__getattribute__')
original_setattr = f_cls_attr.get('__setattr__')
def init_getattr(self, item):
if not item.startswith('_'): # you can set break points at here
alias_name = '_' + item
if alias_name in f_cls_attr['__slots__']:
item = alias_name
if original_getattr is not None:
return original_getattr(self, item)
else:
return super(eval(f_cls_name), self).__getattribute__(item)
def init_setattr(self, key, value):
if not key.startswith('_') and ('_' + key) in f_cls_attr['__slots__']:
raise AttributeError(f"you can't modify private members:_{key}")
if original_setattr is not None:
original_setattr(self, key, value)
else:
super(eval(f_cls_name), self).__setattr__(key, value)
f_cls_attr['__getattribute__'] = init_getattr
f_cls_attr['__setattr__'] = init_setattr
cls = super().__new__(mcs, f_cls_name, f_cls_parents, f_cls_attr)
return cls
class Human(metaclass=MetaMemberControl):
__slots__ = ('_age', '_name')
def __init__(self, name, age):
self._name = name
self._age = age
def __getattribute__(self, item):
"""
is just for IDE recognize.
"""
return super().__getattribute__(item)
""" with MetaMemberControl then you don't have to write as following
#property
def name(self):
return self._name
#property
def age(self):
return self._age
"""
def test_demo():
human = Human('Carson', 27)
# human.age = 18 # you can't modify private members:_age <-- this is defined by yourself.
# human.k = 18 # 'Human' object has no attribute 'k' <-- system error.
age1 = human._age # It's OK, although the IDE will show some warnings. (Access to a protected member _age of a class)
age2 = human.age # It's OK! see below:
"""
if you do not define `__getattribute__` at the class of Human,
the IDE will show you: Unresolved attribute reference 'age' for class 'Human'
but it's ok on running since the MetaMemberControl will help you.
"""
if __name__ == '__main__':
test_demo()
The metaclass is powerful, there are many things (such as monkey magic) you can do with it, but be careful this may only be known to you.

The top answer is correct.
But readers may be coming here searching answers about similarly named inner classes. They are present in popular libraries, such as Django and WTForms.
As DavidW points out in the comments beneath this answer, these are library-specific features and are not to be confused with the advanced, unrelated Python language feature with a similar name.
Rather, these are namespaces within classes' dicts. They are constructed using inner classes for sake of readability.
In this example special field, abstract is visibly separate from fields of Author model.
from django.db import models
class Author(models.Model):
name = models.CharField(max_length=50)
email = models.EmailField()
class Meta:
abstract = True
Another example is from the documentation for WTForms:
from wtforms.form import Form
from wtforms.csrf.session import SessionCSRF
from wtforms.fields import StringField
class MyBaseForm(Form):
class Meta:
csrf = True
csrf_class = SessionCSRF
name = StringField("name")
This syntax does not get special treatment in the python programming language. Meta is not a keyword here, and does not trigger metaclass behavior. Rather, third-party library code in packages like Django and WTForms reads this property in the constructors of certain classes, and elsewhere.
The presence of these declarations modifies the behavior of the classes that have these declarations. For example, WTForms reads self.Meta.csrf to determine if the form needs a csrf field.

In object-oriented programming, a metaclass is a class whose instances are classes. Just as an ordinary class defines the behavior of certain objects, a metaclass defines the behavior of certain class and their instances
The term metaclass simply means something used to create classes. In other words, it is the class of a class. The metaclass is used to create the class so like the object being an instance of a class, a class is an instance of a metaclass. In python classes are also considered objects.

A class, in Python, is an object, and just like any other object, it is an instance of "something". This "something" is what is termed as a Metaclass. This metaclass is a special type of class that creates other class's objects. Hence, metaclass is responsible for making new classes. This allows the programmer to customize the way classes are generated.
To create a metaclass, overriding of new() and init() methods is usually done. new() can be overridden to change the way objects are created, while init() can be overridden to change the way of initializing the object. Metaclass can be created by a number of ways. One of the ways is to use type() function. type() function, when called with 3 parameters, creates a metaclass. The parameters are :-
Class Name
Tuple having base classes inherited by class
A dictionary having all class methods and class variables
Another way of creating a metaclass comprises of 'metaclass' keyword. Define the metaclass as a simple class. In the parameters of inherited class, pass metaclass=metaclass_name
Metaclass can be specifically used in the following situations :-
when a particular effect has to be applied to all the subclasses
Automatic change of class (on creation) is required
By API developers

I saw an interesting use case for metaclasses in a package called classutilities. It checks if all class variables are in upper case format (it is convenient to have unified logic for configuration classes), and checks if there are no instance level methods in class.
Another interesting example for metaclases was deactivation of unittests based on complex conditions (checking values of multiple environmental variables).

In Python, a metaclass is a subclass of a subclass that determines how a subclass behaves. A class is an instance of another metaclass. In Python, a class specifies how the class's instance will behave.
Since metaclasses are in charge of class generation, you can write your own custom metaclasses to change how classes are created by performing additional actions or injecting code. Custom metaclasses aren't always important, but they can be.

look this:
Python 3.10.0rc2 (tags/v3.10.0rc2:839d789, Sep 7 2021, 18:51:45) [MSC v.1929 64 bit (AMD64)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> class Object:
... pass
...
>>> class Meta(type):
... test = 'Worked!!!'
... def __repr__(self):
... return 'This is "Meta" metaclass'
...
>>> class ObjectWithMetaClass(metaclass=Meta):
... pass
...
>>> Object or type(Object())
<class '__main__.Object'>
>>> ObjectWithMetaClass or type(ObjectWithMetaClass())
This is "Meta" metaclass
>>> Object.test
AttributeError: ...
>>> ObjectWithMetaClass.test
'Worked!!!'
>>> type(Object)
<class 'type'>
>>> type(ObjectWithMetaClass)
<class '__main__.Meta'>
>>> type(type(ObjectWithMetaClass))
<class 'type'>
>>> Object.__bases__
(<class 'object'>,)
>>> ObjectWithMetaClass.__bases__
(<class 'object'>,)
>>> type(ObjectWithMetaClass).__bases__
(<class 'type'>,)
>>> Object.__mro__
(<class '__main__.Object'>, <class 'object'>)
>>> ObjectWithMetaClass.__mro__
(This is "Meta" metaclass, <class 'object'>)
>>>
In other words, when an object was not created (type of object), we looking MetaClass.

i want to add a little on why type.__new__() over type()
first, take a look at these classes
In [1]: class MyMeta(type):
...: def __new__(cls, cls_name, bases, attrs):
...: print(cls, cls_name, bases, attrs)
...: return super().__new__(cls, cls_name, bases, attrs)
...:
In [2]: class AClass(metaclass=MyMeta):
...: pass
...:
<class '__main__.MyMeta'> AClass () {'__module__': '__main__', '__qualname__': 'AClass'}
In [3]: class BClass:
...: pass
...:
In [4]: AClass.__class__
Out[4]: __main__.MyMeta
In [5]: BClass.__class__
Out[5]: type
In [6]: class SubAClass(AClass):
...: pass
...:
<class '__main__.MyMeta'> SubAClass (<class '__main__.AClass'>,) {'__module__': '__main__', '__qualname__': 'SubAClass'}
when we were trying to create SubAClass a subclass of AClass, Python would take a look at the metaclass we designated to be used to create SubAClass
and in this case, we did not pass a metaclass for SubAClass, so Python got a None for metaclass.
and Python would try to pick up the first base class of SubAClass, which is AClass, has a metaclass, which is MyMeta, then it would call MyMeta.__new__ to create SubAClass
for simplicity, we can say
type.__new__ just assigned our metaclass to the parent.__class__(AClass.__class__).
and when a subclass was defined(SubAClass) by class statement, Python
would just use the parent.__class__ to create that subclass unless
we've manually passed a metaclass to our subclass
apparently, if you called type() instead of type.__new__, the class
would have type as its __class__ attr.
that means AClass would be equivalent of BClass, both of them have type
as their __class__ attr
how the searching of metaclass works in C code?
it works pretty much like what we've just mentioned
the function builtin___build_class__ would be called when you defined a class
and code is just so straightforward
static PyObject *
builtin___build_class__(PyObject *self, PyObject *const *args, Py_ssize_t nargs,
PyObject *kwnames){
if (meta == NULL) {
/* if there are no bases, use type: */
if (PyTuple_GET_SIZE(bases) == 0) {
meta = (PyObject *) (&PyType_Type);
}
/* else get the type of the first base */
else {
PyObject *base0 = PyTuple_GET_ITEM(bases, 0);
meta = (PyObject *)Py_TYPE(base0);
}
Py_INCREF(meta);
isclass = 1; /* meta is really a class */
}
PyObject *margs[3] = {name, bases, ns};
cls = PyObject_VectorcallDict(meta, margs, 3, mkw);
}
basically, meta = (PyObject *)Py_TYPE(base0); is everything we want to know
it can be translated to be
meta = MyMeta = AClass.__class__ = Py_TYPE(AClass)

In Python or in any other language we have a type for every variable or object we declare. For getting type of anything(variable,object,etc.) in Python we can use type() function.
Bypassing the metaclass keyword in the class definition we can customize the class creation process.
class meta(type):
pass
class baseclass(metaclass=meta): # This is Mestaclass
pass
class derivedclass(baseclass):
pass
print(type(meta))
print(type(baseclass))
print(type(derivedclass))
When defining a new class if no metaclass is defined the default type metaclass is used. If a given metaclass is not the object(instance) of type(), in that situation it is used directly as a metaclass.

How to create a new instance from a class object in Python

I need to dynamically create an instance of a class in Python. Basically I am using the load_module and inspect module to import and load the class into a class object, but I can't figure out how to create an instance of this class object.

I figured out the answer to the question I had that brought me to this page. Since no one has actually suggested the answer to my question, I thought I'd post it.
class k:
pass
a = k()
k2 = a.__class__
a2 = k2()
At this point, a and a2 are both instances of the same class (class k).

Just call the "type" built in using three parameters, like this:
ClassName = type("ClassName", (Base1, Base2,...), classdictionary)
update
as stated in the comment bellow this is not the answer to this question at all. I will keep it undeleted, since there are hints some people get here trying to dynamically create classes - which is what the line above does.
To create an object of a class one has a reference too, as put in the accepted answer, one just have to call the class:
instance = ClassObject()
The mechanism for instantiation is thus:
Python does not use the new keyword some languages use - instead it's data model explains the mechanism used to create an instantance of a class when it is called with the same syntax as any other callable:
Its class' __call__ method is invoked (in the case of a class, its class is the "metaclass" - which is usually the built-in type). The normal behavior of this call is to invoke the (pseudo) static __new__ method on the class being instantiated, followed by its __init__. The __new__ method is responsible for allocating memory and such, and normally is done by the __new__ of object which is the class hierarchy root.
So calling ClassObject() invokes ClassObject.__class__.call() (which normally will be type.__call__) this __call__ method will receive ClassObject itself as the first parameter - a Pure Python implementation would be like this: (the cPython version is of course, done in C, and with lots of extra code for cornercases and optimizations)
class type:
...
def __call__(cls, *args, **kw):
constructor = getattr(cls, "__new__")
instance = constructor(cls) if constructor is object.__new__ else constructor(cls, *args, **kw)
instance.__init__(cls, *args, **kw)
return instance
(I don't recall seeing on the docs the exact justification (or mechanism) for suppressing extra parameters to the root __new__ and passing it to other classes - but it is what happen "in real life" - if object.__new__ is called with any extra parameters it raises a type error - however, any custom implementation of a __new__ will get the extra parameters normally)

This is how you can dynamically create a class named Child in your code, assuming Parent already exists... even if you don't have an explicit Parent class, you could use object...
The code below defines __init__() and then associates it with the class.
>>> child_name = "Child"
>>> child_parents = (Parent,)
>>> child body = """
def __init__(self, arg1):
# Initialization for the Child class
self.foo = do_something(arg1)
"""
>>> child_dict = {}
>>> exec(child_body, globals(), child_dict)
>>> childobj = type(child_name, child_parents, child_dict)
>>> childobj.__name__
'Child'
>>> childobj.__bases__
(<type 'object'>,)
>>> # Instantiating the new Child object...
>>> childinst = childobj()
>>> childinst
<__main__.Child object at 0x1c91710>
>>>

If you have a module with a class you want to import, you can do it like this.
module = __import__(filename)
instance = module.MyClass()
If you do not know what the class is named, you can iterate through the classes available from a module.
import inspect
module = __import__(filename)
for c in module.__dict__.values():
if inspect.isclass(c):
# You may need do some additional checking to ensure
# it's the class you want
instance = c()

If you have some class object, you can instantiate it by just calling it (with parenthesis).
class MyClass: pass
some_class = MyClass
some_instance = some_class() # -> instance of MyClass

I think the neat way would be using type. Here is an example:
>>> class Foo:
... def __init__(self, s):
... self.s = s
...
>>> a = Foo("hello")
>>> a.s
'hello'
>>> b = type(a)("world")
>>> b.s
'world'
>>> assert isinstance(a, Foo)
>>> assert isinstance(b, Foo)
b is an instance which has the same type as a.