Overloading

Magic/Dunder Methods

Magic (also called dunder as an abbreviation for double-underscore) methods in Python serve a similar purpose to operator overloading in other languages. They allow a class to define its behavior when it is used as an operand in unary or binary operator expressions. They also serve as implementations called by some built-in functions.

Consider this implementation of two-dimensional vectors.

import math

class Vector(object):
    # instantiation
    def __init__(self, x, y):
        self.x = x
        self.y = y

    # unary negation (-v)
    def __neg__(self):
        return Vector(-self.x, -self.y)

    # addition (v + u)
    def __add__(self, other):
        return Vector(self.x + other.x, self.y + other.y)

    # subtraction (v - u)
    def __sub__(self, other):
        return self + (-other)

    # equality (v == u)
    def __eq__(self, other):
        return self.x == other.x and self.y == other.y

    # abs(v)
    def __abs__(self):
        return math.hypot(self.x, self.y)

    # str(v)
    def __str__(self):
        return '<{0.x}, {0.y}>'.format(self)

    # repr(v)
    def __repr__(self):
        return 'Vector({0.x}, {0.y})'.format(self)

Now it is possible to naturally use instances of the Vector class in various expressions.

v = Vector(1, 4)
u = Vector(2, 0)

u + v           # Vector(3, 4)
print(u + v)    # "<3, 4>" (implicit string conversion)
u - v           # Vector(1, -4)
u == v          # False
u + v == v + u  # True
abs(u + v)      # 5.0

Container and sequence types

Callable types

Handling unimplemented behaviour

If your class doesn’t implement a specific overloaded operator for the argument types provided, it should return NotImplemented (note that this is a special constant, not the same as NotImplementedError). This will allow Python to fall back to trying other methods to make the operation work:

When NotImplemented is returned, the interpreter will then try the reflected operation on the other type, or some other fallback, depending on the operator. If all attempted operations return NotImplemented, the interpreter will raise an appropriate exception.

For example, given x + y, if x.__add__(y) returns unimplemented, y.__radd__(x) is attempted instead.

class NotAddable(object):

    def __init__(self, value):
        self.value = value

    def __add__(self, other):
        return NotImplemented


class Addable(NotAddable):

    def __add__(self, other):
        return Addable(self.value + other.value)

    __radd__ = __add__

As this is the reflected method we have to implement __add__ and __radd__ to get the expected behaviour in all cases; fortunately, as they are both doing the same thing in this simple example, we can take a shortcut.

In use:

>>> x = NotAddable(1)
>>> y = Addable(2)
>>> x + x
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for +: 'NotAddable' and 'NotAddable'
>>> y + y
<so.Addable object at 0x1095974d0>
>>> z = x + y
>>> z
<so.Addable object at 0x109597510>
>>> z.value
3

Operator overloading

Below are the operators that can be overloaded in classes, along with the method definitions that are required, and an example of the operator in use within an expression.

N.B. The use of other as a variable name is not mandatory, but is considered the norm.

The optional parameter modulo for __pow__ is only used by the pow built-in function.

Each of the methods corresponding to a binary operator has a corresponding “right” method which start with __r, for example __radd__:

class A:
    def __init__(self, a):
        self.a = a
    def __add__(self, other):
        return self.a + other
    def __radd__(self, other):
        print("radd")
        return other + self.a

A(1) + 2  # Out:  3
2 + A(1)  # prints radd. Out: 3

as well as a corresponding inplace version, starting with __i:

class B:
    def __init__(self, b):
        self.b = b
    def __iadd__(self, other):
        self.b += other
        print("iadd")
        return self

b = B(2)
b.b       # Out: 2
b += 1    # prints iadd
b.b       # Out: 3

Since there’s nothing special about these methods, many other parts of the language, parts of the standard library, and even third-party modules add magic methods on their own, like methods to cast an object to a type or checking properties of the object. For example, the builtin str() function calls the object’s __str__ method, if it exists. Some of these uses are listed below.

There are also the special methods __enter__ and __exit__ for context managers, and many more.