Abstract Base Class (abc) in Python

Have you ever thought about checking whether the objects you are using adheres to a particular specification? It is necessary to verify whether an object implements a given method or property, especially while creating a library where other developers make use of it. A developer can use hasattr or isinstance methods to check whether the input conforms to a particular identity. But sometimes it is inconvenient to use those methods to check a myriad of different properties and methods.    

As a solution to this inconvenience, Python introduced a concept called abstract base class (abc). In this section, we will discuss the abstract base class and its importance.  

• Abstract Base Class
• Declaring an Abstract Base Class
• Why declare an Abstract Base Class?
• Abstract Properties
• Built-In Abstract Classes

Abstract Base Class

The main goal of the abstract base class is to provide a standardized way to test whether an object adheres to a given specification. It can also prevent any attempt to instantiate a subclass that doesn't override a particular method in the superclass. And finally, using an abstract class, a class can derive identity from another class without any object inheritance.

Declaring an Abstract Base Class

Python has a module called abc (abstract base class) that offers the necessary tools for crafting an abstract base class. First and foremost, you should understand the ABCMeta metaclass provided by the abstract base class. The rule is every abstract class must use ABCMeta metaclass.

ABCMeta metaclass provides a method called register method that can be invoked by its instance. By using this register method, any abstract base class can become an ancestor of any arbitrary concrete class. Let's understand this process by considering an example of an abstract base class that registers itself as an ancestor of dict.

import abc


class AbstractClass(metaclass=abc.ABCMeta):
    def abstractfunc(self):
        return None


print(AbstractClass.register(dict))

Output:

<class 'dict'>

Here, dict identifies itself as a subclass of AbstractClass. Let's do a check.

import abc


class AbstractClass(metaclass=abc.ABCMeta):
    def abstractfunc(self):
        return None


AbstractClass.register(dict)
print(issubclass(dict, AbstractClass))

Output:

True

Why Declare an Abstract Base Class?

To understand the need to declare a virtual subclass, we need to consider the example of a list-like object where you don't want to put a restriction of only considering list or tuple. Before that let's see how to use isinstance to check against a list or tuple of class. 

isinstance([], (list, tuple))

This isinstance check meets the purpose if you are accepting only a list or tuple. But here the case is different, there is no such restriction. So, this solution is not extensible for a developer who uses your library to send something else other than a list or tuple. Here comes the importance of abstract class. Let's understand through the below code.

import abc


class MySequence(metaclass=abc.ABCMeta):
    pass

MySequence.register(list)
MySequence.register(tuple)

a = [1, 2, 3]
b = ('x', 'y', 'z')

print('List instance:', isinstance(a, MySequence))
print('Tuple instance:', isinstance(b, MySequence))
print('Object instance:', isinstance(object(), MySequence))

Output:

List instance: True
Tuple instance: True
Object instance: False

As you can see, when you do isinstance check, it returns true for both the list and tuple; for the other objects, it returns false. Let's consider a scenario where a developer expects a class object itself. In the above case, the isinstance will return false. But it can be achieved by creating a custom class and registering it with the abstract base class. 

Here 'MySequence' is an abstract class within the library. A developer can import it and register a custom class. Let's have a look at the below code.

import abc


class MySequence(metaclass=abc.ABCMeta):
    pass

class CustomListLikeObjCls(object):
    pass

MySequence.register(CustomListLikeObjCls)
print(issubclass(CustomListLikeObjCls, MySequence))

Output:

True

Here, CustomListLikeObjCls instance is passed to the library by registering it with MySequence. Therefore, the instance check returns True. Apart from the above method, you can also use the register method as a decorator to register a custom class. Let's see how to use the register method as a decorator.

import abc


class MySequence(metaclass=abc.ABCMeta):
    pass

@MySequence.register
class CustomListLikeObjCls(object):
    pass

print(issubclass(CustomListLikeObjCls, MySequence))

Output:

True

Registering a class using the above-implemented method meets the purpose. However, you have to do manual registration for every intended subclass. How about automatic subclassing based on a particular method?. An abstract class has a concept called __subclasshook__   to subclass the classes.  

__subclasshook___

It is a special magic method defined by ABCMeta. The __subclasshook__ must be defined as a class method using  @classmethod decorator. It takes one additional positional argument other than the class and can return either of the three values - True, False, or NotImplemented. Let's look at the below implementation.

import abc


class AbstractClass(metaclass=abc.ABCMeta):
    @classmethod
    def __subclasshook__(cls, other):
        print('subclass hook:', other)
        hookmethod = getattr(other, 'hookmethod', None)
        return callable(hookmethod)

class SubClass(object):
    def hookmethod(self):
        pass

class NormalClass(object):
    hookmethod = 'hook'


print(issubclass(SubClass, AbstractClass))
print(issubclass(NormalClass, AbstractClass))

Output:

subclass hook: <class '__main__.SubClass'>
True
subclass hook: <class '__main__.NormalClass'>
False

From the above discussion, you understood how to hook subclasses automatically. Now we will look into how to avoid instantiating a subclass that doesn't override a particular method in the superclass. This feature can be achieved using @abc.abstractmethod.

@abc.abstractmethod

@abc.abstractmethod prevents any attempt to instantiate a subclass that doesn't override a particular method in the superclass. Let's have a look at the below code:

import abc


class AbstractClass(metaclass=abc.ABCMeta):
    @abc.abstractmethod
    def abstractName(self):
        pass

class InvalidSubClass(AbstractClass):
    pass

isc = InvalidSubClass()

Since the InvalidSubclass doesn't override the method abstractName, the @abc.abstractmethod prevents the subclass from instantiation and throws the below error.

Traceback (most recent call last):  File "/home/553d5199a662239eae3ff58efb37b6ec.py", line 11, in <module>    isc = InvalidSubClass() TypeError: Can't instantiate abstract class InvalidSubClass with abstract methods abstractName

Let's look into another example where the subclass overrides the abstract method.

import abc

class AbstractClass(metaclass=abc.ABCMeta):
    @abc.abstractmethod
    def abstractName(self):
        pass

class ValidSubClass(AbstractClass):
    def abstractName(self):
        return 'Abstract 1'

vc = ValidSubClass()
print(vc.abstractName())

Output:

Abstract 1

Next, let's see how to declare properties as an abstract class.

Abstract Properties

We can use @property decorator and @abc.abstractmethod  to declare properties as an abstract class. Let's look into the below code.

import abc


class AbstractClass(metaclass=abc.ABCMeta):
    @property
    @abc.abstractmethod
    def abstractName(self):
        pass


class ValidSubClass(AbstractClass):
    @property
    def abstractName(self):
        return 'Abstract 1'


vc = ValidSubClass()
print(vc.abstractName)

Output:

Abstract 1

Built-in Abstract classes

Python 3 standard library provides a few built-in abstract classes for both abstract and non-abstract methods. These include sequence, mutable sequence, iterable, and so on. It often serves as an alternative to subclassing a built-in Python class. For example, subclassing the MutableSequence can substitute the subclassing of list or str. The main purpose of using Abstract class is that it allows you to consider a common type of collection rather than coding for each type of collection. Here we will discuss Single-Method ABCs and Alternative-Collection ABCs.

• Single-Method ABCs
• Alternative-Collection ABCs

Single-Method ABCs

Python has five abstract base classes. They are as follows:

• Callable (__call__)
• Container (__contains__)
• Hashable (__hash__)
• Iterable (__iter__)
• Sized (__len__)

These abstract base classes contain one abstract method each. Let's consider an example of the __len__ method.

from collections.abc import Sized


class SingleMethod(object):
    def __len__(self):
        return 10


print(issubclass(SingleMethod, Sized))

Output:

True

Any class that has the appropriate method is considered as the subclass of the abstract base class. Out of the above five abstract base classes, the Iterator is slightly different. It provides an implementation for __iter__ and adds an abstract method called __next__.

Alternative-Collection ABCs

Alternative-Collection ABCs are built-in abstract base classes that identify subclasses, which serve similar purposes. They can be divided into three categories. Let's go through one by one.

• Sequence and Mutable Sequence: Sequence and Mutable Sequence are abstract base classes that generally behaves like tuples or list. A sequence abstract  base class requires __getitem__ and __len__ , whereas mutable sequence needs __setitem__ and __getitem__.
• Mapping: Mapping comes with mutable mapping, which is mainly for dictionary-like objects
• Set: The set comes with a mutable set that is intended for unordered collections.

Summary

The key purpose of the abstract class is to check whether an object conforms to a particular protocol.  It is a valuable class for testing certain attributes of a class or testing class itself. However, there are many other things that the abstract class does not check. Some of them are signatures, return type, etc. Another advantage is it provides a flexible way for developers to test common types of collections.  

sonugeorge

Follow

Improve

Article Tags :

• Python
• python-oop-concepts

Practice Tags :

• python