Class basics

Instance and class attributes

Mypy type checker detects if you are trying to access a missing attribute, which is a very common programming error. For this to work correctly, instance and class attributes must be defined or initialized within the class. Mypy infers the types of attributes:

class A:
    def __init__(self, x: int) -> None:
        self.x = x     # Attribute x of type int

a = A(1)
a.x = 2       # OK
a.y = 3       # Error: A has no attribute y

This is a bit like each class having an implicitly defined __slots__ attribute. This is only enforced during type checking and not when your program is running.

You can declare types of variables in the class body explicitly using a type comment:

class A:
    x = None  # type: List[int]  # Declare attribute x of type List[int]

a = A()
a.x = [1]     # OK

As in Python, a variable defined in the class body can used as a class or an instance variable.

Similarly, you can give explicit types to instance variables defined in a method:

class A:
    def __init__(self) -> None:
        self.x = []  # type: List[int]

    def f(self) -> None:
        self.y = 0  # type: Any

You can only define an instance variable within a method if you assign to it explicitly using self:

class A:
    def __init__(self) -> None:
        self.y = 1   # Define y
        a = self
        a.x = 1      # Error: x not defined

Overriding statically typed methods

When overriding a statically typed method, mypy checks that the override has a compatible signature:

class A:
    def f(self, x: int) -> None:

class B(A):
    def f(self, x: str) -> None:   # Error: type of x incompatible

class C(A):
    def f(self, x: int, y: int) -> None:  # Error: too many arguments

class D(A):
    def f(self, x: int) -> None:   # OK


You can also vary return types covariantly in overriding. For example, you could override the return type object with a subtype such as int.

You can also override a statically typed method with a dynamically typed one. This allows dynamically typed code to override methods defined in library classes without worrying about their type signatures.

There is no runtime enforcement that the method override returns a value that is compatible with the original return type, since annotations have no effect at runtime:

class A:
    def inc(self, x: int) -> int:
        return x + 1

class B(A):
    def inc(self, x):       # Override, dynamically typed
        return 'hello'

b = B()
print(   # hello
a = b # type: A
print(   # hello

Abstract base classes and multiple inheritance

Mypy uses Python abstract base classes for protocol types. There are several built-in abstract base classes types (for example, Sequence, Iterable and Iterator). You can define abstract base classes using the abc.ABCMeta metaclass and the abc.abstractmethod function decorator.

from abc import ABCMeta, abstractmethod
import typing

class A(metaclass=ABCMeta):
    def foo(self, x: int) -> None: pass

    def bar(self) -> str: pass

class B(A):
    def foo(self, x: int) -> None: ...
    def bar(self) -> str:
        return 'x'

a = A() # Error: A is abstract
b = B() # OK

Unlike most Python code, abstract base classes are likely to play a significant role in many complex mypy programs.

A class can inherit any number of classes, both abstract and concrete. As with normal overrides, a dynamically typed method can implement a statically typed abstract method defined in an abstract base class.

Protocols and structural subtyping


Structural subtyping is experimental. Some things may not work as expected. Mypy may pass unsafe code or it can reject valid code.

Mypy supports two ways of deciding whether two classes are compatible as types: nominal subtyping and structural subtyping. Nominal subtyping is strictly based on the class hierarchy. If class D inherits class C, it’s also a subtype of C. This form of subtyping is used by default in mypy, since it’s easy to understand and produces clear and concise error messages, and since it matches how the native isinstance() check works – based on class hierarchy. Structural subtyping can also be useful. Class D is a structural subtype of class C if the former has all attributes and methods of the latter, and with compatible types.

Structural subtyping can be seen as a static equivalent of duck typing, which is well known to Python programmers. Mypy provides an opt-in support for structural subtyping via protocol classes described below. See PEP 544 for the detailed specification of protocols and structural subtyping in Python.

Simple user-defined protocols

You can define a protocol class by inheriting the special typing_extensions.Protocol class:

from typing import Iterable
from typing_extensions import Protocol

class SupportsClose(Protocol):
    def close(self) -> None:
       ...  # Explicit '...'

class Resource:  # No SupportsClose base class!
    # ... some methods ...

    def close(self) -> None:

def close_all(items: Iterable[SupportsClose]) -> None:
    for item in items:

close_all([Resource(), open('some/file')])  # Okay!

Resource is a subtype of the SupportClose protocol since it defines a compatible close method. Regular file objects returned by open() are similarly compatible with the protocol, as they support close().


The Protocol base class is currently provided in the typing_extensions package. Once structural subtyping is mature and PEP 544 has been accepted, Protocol will be included in the typing module. Several library types such as typing.Sized and typing.Iterable will also be changed into protocols. They are currently treated as regular ABCs by mypy.

Defining subprotocols

You can also define subprotocols. Existing protocols can be extended and merged using multiple inheritance. Example:

# ... continuing from the previous example

class SupportsRead(Protocol):
    def read(self, amount: int) -> bytes: ...

class TaggedReadableResource(SupportsClose, SupportsRead, Protocol):
    label: str

class AdvancedResource(Resource):
    def __init__(self, label: str) -> None:
        self.label = label

    def read(self, amount: int) -> bytes:
        # some implementation

resource: TaggedReadableResource
resource = AdvancedResource('handle with care')  # OK

Note that inheriting from an existing protocol does not automatically turn the subclass into a protocol – it just creates a regular (non-protocol) ABC that implements the given protocol (or protocols). The typing_extensions.Protocol base class must always be explicitly present if you are defining a protocol:

class NewProtocol(SupportsClose):  # This is NOT a protocol
    new_attr: int

class Concrete:
   new_attr: int = 0

   def close(self) -> None:

# Error: nominal subtyping used by default
x: NewProtocol = Concrete()  # Error!


You can use Python 3.6 variable annotations (PEP 526) to declare protocol attributes. On Python 2.7 and earlier Python 3 versions you can use type comments and properties.

Recursive protocols

Protocols can be recursive (self-referential) and mutually recursive. This is useful for declaring abstract recursive collections such as trees and linked lists:

from typing import TypeVar, Optional
from typing_extensions import Protocol

class TreeLike(Protocol):
    value: int

    def left(self) -> Optional['TreeLike']: ...

    def right(self) -> Optional['TreeLike']: ...

class SimpleTree:
    def __init__(self, value: int) -> None:
        self.value = value
        self.left: Optional['SimpleTree'] = None
        self.right: Optional['SimpleTree'] = None

root = SimpleTree(0)  # type: TreeLike  # OK

Using isinstance() with protocols

You can use a protocol class with isinstance() if you decorate it with the typing_extensions.runtime class decorator. The decorator adds support for basic runtime structural checks:

from typing_extensions import Protocol, runtime

class Portable(Protocol):
    handles: int

class Mug:
    def __init__(self) -> None:
        self.handles = 1

mug = Mug()
if isinstance(mug, Portable):
   use(mug.handles)  # Works statically and at runtime


isinstance() with protocols is not completely safe at runtime. For example, signatures of methods are not checked. The runtime implementation only checks that all protocol members are defined.