An in-depth analysis of the usage of super keyword in Python programming

The official document does not say much about the definition of super. It roughly means that it returns a proxy object so that you can call some inherited methods. The search mechanism follows the mro rules. The most commonly used situation is as shown in the following example:

class C(B):
  def method(self, arg):
    super(C, self).method(arg)

Subclass C rewrites the method with the same name in parent class B. In the overwritten implementation, the method with the same name of the parent class is called through the proxy object instantiated by super.

The initial method signature of the super class is as follows:

def __init__(self, type1, type2=None): # known special case of super.__init__
    """
    super(type, obj) -> bound super object; requires isinstance(obj, type)
    super(type) -> unbound super object
    super(type, type2) -> bound super object; requires issubclass(type2, type)
    Typical use to call a cooperative superclass method:

Except for self, it accepts one or two parameters. As stated in the annotation, when two parameters are accepted, a bound super instance is returned. When the second parameter is omitted, an unbound super object is returned. .

Under normal circumstances, when calling an inherited class method or static method, there is no need to bind a specific instance. At this time, using super(type, type2).some_method can achieve the purpose. Of course, super(type, obj) is It can also be used in this case. The super object has a custom implemented getattribute method that can also be processed. However, the latter is generally used to call instance methods, so that the corresponding instance can be passed in when searching for a method, thereby obtaining the bound instance method:

class A(object):
  def __init__(self):
    pass

  @classmethod
  def klass_meth(cls):
    pass

  @staticmethod
  def static_meth():
    pass

  def test(self):
    pass

class B(A):
  pass

>>> b = B()
>>> super(B, b).test
<bound method B.test of <__main__.B object at 0x02DA3570>>
>>> super(B, b).klass_meth
<bound method type.klass_meth of <class '__main__.B'>>
>>> super(B, b).static_meth
<function static_meth at 0x02D9CC70>
>>> super(B, B).test
<unbound method B.test>
>>> super(B, B).klass_meth
<bound method type.klass_meth of <class '__main__.B'>>
>>> super(B,B).satic_meth
>>> super(B,B).static_meth
<function static_meth at 0x02D9CC70>

When initializing the super object, the second parameter passed is actually the bound object. The first parameter can be roughly understood as the starting point of the mark search, such as the situation in the above example: super(B, b ).test will find the method test in the classes listed in B.__mro__ except B itself. Because the methods are non-data descriptors, the custom getattribute of the super object will actually be converted into A.__dict[' test'].__get__(b, B).

Super is used in many places. In addition to making the program more dynamic without having to hardcode the specified type, there are other specific places where it must be used, such as using super in metaclasses to find new in baseclass to generate custom type templates; in When customizing getattribute, it is used to prevent infinite loops, etc.

Regarding super, it is recommended that readers understand it together with python descriptors, because super implements the descriptor protocol and is a non-data descriptor, which can help everyone better understand the use and working principle of super.

At the same time, there are the following 4 points worthy of your attention:
1. In single inheritance, the functions implemented by super() and __init__() are similar

class Base(object):
  def __init__(self):
    print 'Base create'

class childA(Base):
  def __init__(self):
    print 'creat A ',
    Base.__init__(self)


class childB(Base):
  def __init__(self):
    print 'creat B ',
    super(childB, self).__init__()

base = Base()

a = childA()
b = childB()

Output result:

Base create
creat A Base create
creat B Base create

There is no need to explicitly reference the base class when using super() inheritance.

2. super() can only be used in new-style classes

Change the base class to an old-style class, that is, do not inherit any base class

class Base():
  def __init__(self):
    print 'Base create'

When executed, an error will be reported when initializing b:

  super(childB, self).__init__()
TypeError: must be type, not classobj

3. super is not the parent class, but the next class in the inheritance sequence

In multiple inheritance, the inheritance order is involved. super() is equivalent to returning the next class in the inheritance order, not the parent class, similar to this function:

def super(class_name, self):
  mro = self.__class__.mro()
  return mro[mro.index(class_name) + 1]

mro() is used to obtain the inheritance order of classes.

For example:

class Base(object):
  def __init__(self):
    print 'Base create'

class childA(Base):
  def __init__(self):
    print 'enter A '
    # Base.__init__(self)
    super(childA, self).__init__()
    print 'leave A'


class childB(Base):
  def __init__(self):
    print 'enter B '
    # Base.__init__(self)
    super(childB, self).__init__()
    print 'leave B'

class childC(childA, childB):
  pass

c = childC()
print c.__class__.__mro__

The input results are as follows:

enter A 
enter B 
Base create
leave B
leave A
(<class '__main__.childC'>, <class '__main__.childA'>, <class '__main__.childB'>, <class '__main__.Base'>, <type 'object'>)

supder is not related to the parent class, so the execution order is A —> B—>—>Base

The execution process is equivalent to: when initializing childC(), it will first call super(childA, self).__init__() in the constructor of childA, super(childA, self) returns the inheritance order of the current class after childA A class childB; then execute childB().__init()__, and continue in this order.

In multiple inheritance, if you replace super(childA, self).__init__() in childA() with Base.__init__(self), during execution, after inheriting childA, it will jump directly to the Base class. And childB is skipped:

enter A 
Base create
leave A
(<class '__main__.childC'>, <class '__main__.childA'>, <class '__main__.childB'>, <class '__main__.Base'>, <type 'object'>)

As can be seen from the super() method, the first parameter of super() can be the name of any class in the inheritance chain,

If it is itself, it will inherit the next class in turn;

If it is the previous class in the inheritance chain, it will recurse infinitely;

If it is a class later in the inheritance chain, the class between the inheritance chain summary itself and the incoming class will be ignored;

For example, if you change super in childA() to: super(childC, self).__init__(), the program will recurse infinitely.

Such as:

 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
 File "C:/Users/Administrator/Desktop/crawler/learn.py", line 10, in __init__
  super(childC, self).__init__()
RuntimeError: maximum recursion depth exceeded while calling a Python object

4. super() can avoid repeated calls

If childA is based on Base, childB inherits childA and Base, and if childB needs to call the __init__() method of Base, it will cause __init__() to be executed twice:

class Base(object):
  def __init__(self):
    print 'Base create'

class childA(Base):
  def __init__(self):
    print 'enter A '
    Base.__init__(self)
    print 'leave A'


class childB(childA, Base):
  def __init__(self):
    childA.__init__(self)
    Base.__init__(self)

b = childB()
  Base的__init__()方法被执行了两次

enter A 
Base create
leave A
Base create
使用super()是可避免重复调用

class Base(object):
  def __init__(self):
    print 'Base create'

class childA(Base):
  def __init__(self):
    print 'enter A '
    super(childA, self).__init__()
    print 'leave A'


class childB(childA, Base):
  def __init__(self):
    super(childB, self).__init__()

b = childB()
print b.__class__.mro()

enter A 
Base create
leave A
[<class '__main__.childB'>, <class '__main__.childA'>, <class '__main__.Base'>, <type 'object'>]