Python Metaclasses

Copy it to a fixed width font terminal for a good reading. 

This is what I gathered and guessed about Python metaclasses from the following pages: 

https://mail.python.org/pipermail/python-3000/2006-December/005030.html

https://www.geeksforgeeks.org/python-metaclasses/

https://realpython.com/python-metaclasses/

https://www.geeksforgeeks.org/metaprogramming-metaclasses-python/

Python is a work in progress and so is my understanding of this programming language. 

dynamically create a new class: 

'name' = type('name', 'bases', 'dict') 

  differs from class 'name''bases' 

    by the fact that the name, bases and attributes of the newly created class are decided at runtime 

  'name' - the name of the new class, memorized in 'name'. __name__ 

  'bases' - a tuple of the base classes from which the class inherits, memorized in 'name'.__bases__ 

  'dict' - a namespace dictionary containing definitions for the class body, memorized in 'name'.__dict__

metaclasses configure classes when they are created 

    this is different from inheritance 

  all classes inherit the class 'object' or its subclasses 

  all classes are configured by the metaclass 'type' or a subclass of it 

    (maybe it has to be a direct subclass) 

  'myclass'.__bases__, 'myclass'.__base__ - show directly inherited classes 

  'myclass'.__class__, type('myclass') - show the metaclass that configured 'myclass' 

   >>> type.__bases__ 

   (<class 'object'>,)

   >>> type(type) 

   <class 'type'>

   >>> int.__bases__ 

   (<class 'object'>,)

   >>> type(int) 

   <class 'type'>

   >>> type(7) 

   <class 'int'>

objects have a __call__() method  

  which calls the __new__() and __init__() methods

  __new__() creates the new object and returns it

    'cls' in __new__() is the new object

  __init__() initializes the newly created object passed as a parameter

    'self' in __init__() is the new object

  these methods can be overridden to configure the newly created object

  help(type.__call__), help(object.__call__)

  help(type) has a __call__ method, but help(object) does not

metaclasses are a “solution in search of a problem” 

examples of metaclass usage 

$ cat meta.py 

# MyMeta  is an instance of type   and a subclass of type 

# MyClass is an instance of MyMeta and a subclass of object 

def prints(): 

  print('   type(MyMeta) =', type(MyMeta), '\n   type(MyClass) =', type(MyClass), '\n') 

class MyClass: 

  pass 

# dynamic creation (definition) of MyClass, parameters can be decided at runtime 

MyClass = type('MyClass', (object,), {}) 

class MyMeta(type): 

  def __new__(cls, cls_name, bases, attrib): 

    print('>> MyMeta.__new__()') 

    return super(MyMeta, cls).__new__(cls, cls_name, bases, attrib) 

#          super() arguments are not required 

print('\nMyMeta.__bases__ =', MyMeta.__bases__, '\nMyMeta.__class__ =', MyMeta.__class__, '\n') 

class MyClass(metaclass = MyMeta): 

  pass 

prints() 

MyClass = MyMeta('MyClass', (object, ), {}) 

prints() 

# MyClass = type('MyClass', (object, metaclass = MyMeta,), {}) 

#                             invalid syntax   ^ 

#   so type() cannot create a class of a MyMeta type 

# prints() 

# MyClass = MyMeta() 

#   new() missing 3 required positional arguments: 'cls_name', 'bases', and 'attrib' 

# prints() 

def new(cls, cls_name, bases, attrib): 

  print('>> new()') 

  return super(MyMeta, cls).__new__(cls, cls_name, bases, attrib) 

#        super() arguments are required 

MyMeta = type('MyMeta', (type,), {'__new__' : new}) 

class MyClass(metaclass = MyMeta): 

  pass 

prints() 

$ python meta.py 

MyMeta.__bases__ = (<class 'type'>,) 

MyMeta.__class__ = <class 'type'> 

>> MyMeta.__new__()

   type(MyMeta) = <class 'type'> 

   type(MyClass) = <class '__main__.MyMeta'> 

>> MyMeta.__new__()

   type(MyMeta) = <class 'type'> 

   type(MyClass) = <class '__main__.MyMeta'> 

>> new()

   type(MyMeta) = <class 'type'> 

   type(MyClass) = <class '__main__.MyMeta'> 

$ cat meta.py 

class Meta(type):

  def __new__(cls, name, bases, dict):

    x = super().__new__(cls, name, bases, dict)

    x.attr = 100

    return x

class Foo(metaclass=Meta):

  pass

print('Foo.attr =', Foo.attr)

$ python meta.py 

Foo.attr = 100

$ cat meta.py 

class Meta(type):

  def __init__(cls, name, bases, dict):

    cls.attr = 100

class X(metaclass=Meta):

  pass

print('X.attr =', X.attr)

$ python meta.py 

X.attr = 100