Shouldn't __metaclass__ force the use of a metaclass in Python?

Question

I've been trying to learn about metaclasses in Python. I get the main idea, but I can't seem to activate the mechanism. As I understand it, you can specify M to be as the metaclass when constructing a class K by setting __metaclass__ to M at the global or class level. To test this out, I wrote the following program:

p = print

class M(type):
    def __init__(*args):
        type.__init__(*args)
        print("The rain in Spain")

p(1)
class ClassMeta:
    __metaclass__ = M

p(2)
__metaclass__ = M
class GlobalMeta: pass

p(3)
M('NotMeta2', (), {})

p(4)

However, when I run it, I get the following output:

C:\Documents and Settings\Daniel Wong\Desktop>python --version
Python 3.0.1

C:\Documents and Settings\Daniel Wong\Desktop>python meta.py
1
2
3
The rain in Spain
4

Shouldn't I see "The rain in Spain" after 1 and 2? What's going on here?

Answer 1

In Python 3 (which you are using) metaclasses are specified by a keyword parameter in the class definition:

class ClassMeta(metaclass=M):
  pass

Specifying a __metaclass__ class property or global variable is old syntax from Python 2.x and not longer supported. See also "What's new in Python 3" and PEP 2115.

Answer 2

This works as you expect in Python 2.6 (and earlier), but in 3.0 metaclasses are specified differently:

class ArgMeta(metaclass=M): ...

Answer 3

The syntax of metaclasses has changed in Python 3.0. The __metaclass__ attribute is no longer special at either the class nor the module level. To do what you're trying to do, you need to specify metaclass as a keyword argument to the class statement:

p = print

class M(type):
    def __init__(*args):
        type.__init__(*args)
        print("The rain in Spain")

p(1)
class ClassMeta(metaclass=M): pass

Yields:

1
The rain in Spain

As you'd expect.