In my previous blog post Python rich comparison I looked at simplifying the comparison of objects.

Using NotImplemented in boolean context has been deprecated since Python 3.9: As bool(NotImplemented) is True this resulted in many wong implementations, including mine.

So how do you correctly implement rich comparison?

While listening to the episode Talk Python 441: Python = Syntactic Sugar? I learned something new: CPython already does some internal optimizations which saved you from re-implementing this again in Python yourself. Brett Cannon wrote several blog post about Python’s syntactic sugar. There are two very important posts about this:

• Unravelling binary arithmetic operations in Python
• Unravelling rich comparison operators

My most important learnings from them are:

• if left hand side (LHS) and right hand side (RHS) are of the same type, then CPython will only ever call the comparion method if the LHS, e.g. 1 < 2 will result in only int.__lt__(1,2) being called, not also int.__ge__(2, 1).

• if one argument is a true sub-type of the other, CPython will ask the sub-type first to compare itself to the super-type, e.g. 1 < my_int(2) is automatically translated to my_int.__ge__(2, 1). So if you derive your class my_subclass from the super-class super_class, make sure your code is able to compare to super_class.

Depending on what you compare, the function should raise TypeError when incompatible types are compared, but return NotImplemented if the types are compatible.

For example comparing appels with oranges should return a TypeError:

>>> 1 == None
False
>>> 1 < None
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: '<' not supported between instances of 'int' and 'NoneType'

>>> "" == 0
False
>>> "" < 0
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: '<' not supported between instances of 'str' and 'int'

>>> object() == True
False
>>> object() < True
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: '<' not supported between instances of 'object' and 'bool'

On the other hand some floating point numbers are integers, but in Python neither is a sub-type of the other:

>>> int(5) == float(5.0)
True
>>> int.__eq__(5, 5.0)
NotImplemented
>>> float.__eq__(5.0, 5)
True

Please note that namedtuple are a sub-class of tuple, so namedtuple().__eq__ is used when comparing an instance to a tuple:

>>> from collections import namedtuple
>>> Named = namedtuple("Named", "x y")
>>> Named.__bases__
(<class 'tuple'>,)
>>> plain = (1, 2)
>>> named = Named(*plain)
>>> Named.__eq__(named, plain)
True