Understanding is vs. == in Python

Quick summary for those in a hurry

• Identity vs. equality:
  • x is y: Checks if variables x and y reference the same object. This is an identity check, equivalent to id(x) == id(y).
  • x == y: Checks if objects x and y have the same value. This is an equality check, which translates to x.__eq__(y).
• Common pitfalls:
  • Incorrect usage of is to compare numbers, strings, or other immutable types might resolve to True due to optimizations like interning.
  • Do not rely on these optimizations since they are implementation-specific (e.g., CPython, PyPy).
• Best practices:
  • Use is for checking if something is None or something is not None.
  • Use == for all other equality comparisons and implement __eq__() in custom classes if needed.
  • There are valid use cases for is, but only if you are confident about what you are doing.

For a more in-depth look at how is and == work, potential pitfalls, and best practices, keep reading.

Introduction

When learning Python after working with languages like C, Java, or JavaScript, it’s natural to assume that logical operators behave similarly. While &&, ||, and ! in other languages become and, or, and not in Python, using is for equality checks can lead to subtle bugs depending on the Python implementation. This post aims to clarify the differences between is and ==, highlight common pitfalls, and provide best practices to avoid these issues.

Fundamentals of is and ==

Let’s start with the basics by examining what these operators do and how they differ.

Identity (is)

The operators is and is not perform an identity comparison by checking if two variables point to the same object in memory. As described by the official Python documentation:

The operators is and is not test for an object’s identity: x is y is true if and only if x and y are the same object. An Object’s identity is determined using the id() function. x is not y yields the inverse truth value.

In other words, x is y is the same as evaluating id(x) == id(y). The behavior of the is operator is consistent regardless of the objects being compared because neither the is operator nor the built-in id() function can be overloaded.

a = [1, 2, 3]
b = a
c = [1, 2, 3]

print(a is b)  # True, because `b` is an alias of `a`
print(a is c)  # False, because `c` is a different list object

Equality (==)

The operators == and != compare two objects by their values or contents. The statement x == y translates to x.__eq__(y). Unlike is, this operator can be overloaded, allowing control over how objects are compared. Implementing __eq__() for custom classes can be useful for designing elegant and convenient APIs. However, the downside of overloading is that this operator might not behave consistently, as it depends on how the __eq__() method is implemented in a class.

Example

class Container:
    def __init__(self, data):
        self.data = data

container = Container("test")
print(container is Container("test"))  # False, a new object is created
print(container == Container("test"))  # False, no `__eq__()` method defined

Even though two objects are created with the same attributes, they are not automatically equal. This can be resolved by adding a custom __eq__() method:

class Container:
    def __init__(self, data):
        self.data = data

    def __eq__(self, other):
        return isinstance(other, type(self)) and self.data == other.data

container = Container("test")
print(container is Container("test"))  # False, a new object is created
print(container == Container("test"))  # True, because of custom `__eq__()`.

Even though this example is basic, it shows that proper method overloading can lead to more predictable equality checks.

Additional note on !=

The statement x != y translates to x.__ne__(y), which is mostly an historic artifact. Since Python 3, __ne__() returns not x == y by default, but Python 2 did not have such a relation between == and != and __ne__() had to be implemented explicitly. The default Python 3 behavior is usually what you want, so __ne__() is rarely implemented in practice anymore.

Pitfalls

Using is for equality comparisons can produce unexpected results due to certain optimizations in Python. For example, you might expect x is 5 to return False even if x equals 5, but under some conditions, it might return True. These optimizations are dependent on the Python implementation. The table below illustrates this with some examples.

The reason for these results is that Python implementations perform certain optimizations, such as ‘interning’, for frequently used immutable values. These optimizations can cause confusion, since is can behave like == for certain values.

Luckily, recent versions of CPython and PyPy will warn against this incorrect usage of is with a SyntaxWarning, but only when comparing to literals: - CPython: SyntaxWarning: "is" with 'int' literal. Did you mean "=="? - PyPy: SyntaxWarning: "is" with a literal. Did you mean "=="?

Reflexivity

In general, if x is y, then x == y is also True. This property is known as a reflexive relation, and it is the reason that objects without an __eq__() method only compare equal to themselves. When implementing custom __eq__() methods, reflexivity usually follows naturally. However, it is possible to break reflexivity. Consider the following example:

class BadEq:
    def __eq__(self, other):
        return False

bad_eq = BadEq()
print(bad_eq is bad_eq)  # True, it is the same instance.
print(bad_eq == bad_eq)  # False, because of the custom __eq__().

Python itself makes several assumptions about reflexivity. For example, a list is always equal to itself, regardless of its contents:

class BadEq:
    def __eq__(self, other):
        return False

lst = [bad_eq]
print(lst is lst)  # True, same object
print(lst == lst)  # True, because of reflexivity
print(all(x == x for x in lst))  # False, because of BadEq.__eq__()

Custom classes aside, reflexivity holds for all built-in classes and types, with one notable exception: NaN. The IEEE Standard for Floating-Point Arithmetic (IEEE 754) dictates that NaN should not be equal to anything, including itself. For this reason, to check if a variable is NaN, use math.isnan().

from math import nan, isnan  # or `nan = float("nan")`

print(nan is nan)  # True
print(nan == nan)  # False
print(isnan(nan))  # True

Best practices for using is and ==

Now that we have discussed both operators in more detail, how do we use them effectively?

When in doubt, use ==

Generally speaking, there are not many cases where you would want to compare the identity of two objects. Most of the time, you want to compare objects based on their value instead, for which you should use ==.

Implement __eq__() for custom classes to have more meaningful value comparisons. You can also consider using the standard library module dataclasses, which automatically implements __eq__() based on the class’s attributes.

Always use is for None (and other singletons)

In Python, None is guaranteed to be a singleton, all occurrences of None reference the same object. As such, you can use is and is not to check if a variable is None.

But you might wonder “why the exception if == None and != None work as well? This has multiple reasons: - The == operator can be overloaded. So in theory it is possible to implement __eq__() is such a way that x == None would evaluate to True even though x might not be None at all. Using is ensures that you are checking what you want. - Using is has a small performance benefit over == since it does not have look up and invoke the correct __eq__() for the object being compared.

There are more singletons in Python, the most notable being True and False. It is therefore perfectly valid to do if something is False. This pattern is rarely used, and often if not something suffices, but it is not the same. Empty collections, empty strings, and the number 0 are all treated as False in an if statement, but they are not identical to False:

print(not [])  # True, because `bool([])` returns `False`.
print([] is False)  # False, because these are different objects.

Little known fact: there are more builtin singletons in Python, but those are never really treated as such. Some examples are the elipses ... and NotImplemented. You can check this for yourself:

print(type(...))  # <class 'ellipsis'>
print(type(...)() is ...)  # True
print(type(NotImplemented)() is NotImplemented)  # True

Use is for object caches

Sometimes you truly need to know if two variables point to the same object. One example is when caching values and wanting to verify that something is indeed cached.

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

def get_expensive_object(key):
    if cache.get(key) is None:
        cache[key] = ExpensiveObject(key)
    return cache[key]

cache = {}
obj1 = get_expensive_object(42)
obj2 = get_expensive_object(42)

print(obj1 is obj2)  # True, both variables point to the cached object.

Avoid using is or == for comparing types

Types, especially built-in types, can generally be treated as singletons. As such, it is possible to use both is and == to compare them. However, this is not recommended, and you should use isinstance() instead. These are some examples that illustrate why.

isinstance() handles subclasses:

class CustomString(str):
    pass

custom_text = CustomString("some text")

print(type(custom_text) is CustomString)  # True, type is exact match
print(type(custom_text) is str)  # False, subclass is not an exact match
print(isinstance(custom_text, CustomString))  # True
print(isinstance(custom_text, str))  # True, `isinstance` handles subclasses

isinstance() can check against multiple types:

x = 5
print(type(x) is int or type(x) is float)  # Requires chained `type` and `or`
print(isinstance(x, (int, float))  # Can check against multiple types.

There are often ‘abstract base classes’ available for common patterns, and these can only be used with isinstance(), since no object will have the abstract base class as its actual type. In this example, using isinstance with numbers.Number will match any kind of number type, including the other built-in number type complex.

import numbers

print(isinstance(5, numbers.Number))  # True
print(isinstance(complex(1.0, 1.0), numbers.Number))  # True

Conclusion

While the is and == operators may appear similar at first glance, they serve distinct roles and are not interchangeable. Of the two, == is used most often since an object’s value is usually what matters. The is operator should be reserved for identity checks only, such as verifying that a variable is None or ensuring that two variables point to the same object.

The pitfalls discussed illustrate how implementation-specific optimizations can yield unexpected results when these operators are misused. These optimizations should never be relied upon, but you must be aware of them to avoid subtle bugs.

All in all, it shows the importance of understanding Python’s fundamentals. By adhering to the best practices, your code will be clearer, more predictable, and ultimately more maintainable.