Type Hints

Type hints are the Python native way to define the type of the objects in a program.

Traditionally, the Python interpreter handles types in a flexible but implicit way. Recent versions of Python allow you to specify explicit type hints that different tools can use to help you develop your code more efficiently.

def headline(text: str, align: bool = True) -> str:
    if align:
        return f"{text.title()}\n{'-' * len(text)}"
    else:
        return f" {text.title()} ".center(50, "o")

Type hints are not enforced on their own by python. So you won't catch an error if you try to run headline("use mypy", align="center") unless you use a static type checker like Mypy.

Advantages and disadvantages⚑

Advantages:

• Help catch certain errors if used with a static type checker.
• Help check your code. It's not trivial to use docstrings to do automatic checks.
• Help to reason about code: Knowing the parameters type makes it a lot easier to understand and maintain a code base. It can speed up the time required to catch up with a code snippet. Always remember that you read code a lot more often than you write it, so you should optimize for ease of reading.

• Help you build and maintain a cleaner architecture. The act of writing type hints force you to think about the types in your program.

• Helps refactoring: Type hints make it trivial to find where a given class is used when you're trying to refactor your code base.

• Improve IDEs and linters.

Cons:

• Type hints take developer time and effort to add. Even though it probably pays off in spending less time debugging, you will spend more time entering code.
• Introduce a slight penalty in start-up time. If you need to use the typing module, the import time may be significant, even more in short scripts.
• Work best in modern Pythons.

Follow these guidelines when deciding if you want to add types to your project:

• In libraries that will be used by others, they add a lot of value.
• In complex projects, type hints help you understand how types flow through your code and are highly recommended.
• If you are beginning to learn Python, don't use them yet.
• If you are writing throw-away scripts, don't use them.

So, Use Type hints whenever unit tests are worth writing.

Usage⚑

Function annotations⚑

def func(arg: arg_type, optarg: arg_type = default) -> return_type:
    ...

When running the code, the special .__annotations__ attribute on the function stores the typing information.

Variable annotations⚑

Sometimes the type checker needs help in figuring out the types of variables as well. The syntax is similar:

pi: float = 3.142

def circumference(radius: float) -> float:
    return 2 * pi * radius

Composite types⚑

If you need to hint other types than str, float and bool, you'll need to import the typing module.

For example to define the hint types of list, dictionaries and tuples:

>>> from typing import Dict, List, Tuple

>>> names: List[str] = ["Guido", "Jukka", "Ivan"]
>>> version: Tuple[int, int, int] = (3, 7, 1)
>>> options: Dict[str, bool] = {"centered": False, "capitalize": True}

If your function expects some kind of sequence but don't care whether it's a list or a tuple, use the typing.Sequence object. In fact, try to use Sequence if you can because using List could lead to some unexpected errors when combined with type inference. For example:

class A: ...
class B(A): ...

lst = [A(), A()]  # Inferred type is List[A]
new_lst = [B(), B()]  # inferred type is List[B]
lst = new_lst  # mypy will complain about this, because List is invariant

Possible strategies in such situations are:

• Use an explicit type annotation:

  new_lst: List[A] = [B(), B()]
  lst = new_lst  # OK
  

• Make a copy of the right hand side:

  lst = list(new_lst) # Also OK
  

• Use immutable collections as annotations whenever possible:

  def f_bad(x: List[A]) -> A:
      return x[0]
  f_bad(new_lst) # Fails
  
  def f_good(x: Sequence[A]) -> A:
      return x[0]
  f_good(new_lst) # OK
  

Dictionaries with different value types per key.⚑

TypedDict declares a dictionary type that expects all of its instances to have a certain set of keys, where each key is associated with a value of a consistent type. This expectation is not checked at runtime but is only enforced by type checkers.

TypedDict started life as an experimental Mypy feature to wrangle typing onto the heterogeneous, structure-oriented use of dictionaries. As of Python 3.8, it was adopted into the standard library.

try:
    from typing import TypedDict  # >=3.8
except ImportError:
    from mypy_extensions import TypedDict  # <=3.7

Movie = TypedDict('Movie', {'name': str, 'year': int})

A class-based type constructor is also available:

class Movie(TypedDict):
    name: str
    year: int

By default, all keys must be present in a TypedDict. It is possible to override this by specifying totality. Usage:

class point2D(TypedDict, total=False):
    x: int
    y: int

This means that a point2D TypedDict can have any of the keys omitted. A type checker is only expected to support a literal False or True as the value of the total argument. True is the default, and makes all items defined in the class body be required.

Functions without return values⚑

Some functions aren't meant to return anything. Use the -> None hint in these cases.

def play(player_name: str) -> None:

    print(f"{player_name} plays")


ret_val = play("Filip")

The annotation help catch the kinds of subtle bugs where you are trying to use a meaningless return value.

If your function doesn't return any object, use the NoReturn type.

from typing import NoReturn

def black_hole() -> NoReturn:
    raise Exception("There is no going back ...")

Optional arguments⚑

A common pattern is to use None as a default value for an argument. This is done either to avoid problems with mutable default values or to have a sentinel value flagging special behavior.

This creates a challenge for type hinting as the argument may be of type string (for example) but it can also be None. We use the Optional type to address this case.

from typing import Optional

def player(name: str, start: Optional[str] = None) -> str:
    ...

A similar way would be to use Union[None, str].

Type aliases⚑

Type hints might become oblique when working with nested types. If it's the case, save them into a new variable, and use that instead.

from typing import List, Tuple

Card = Tuple[str, str]
Deck = List[Card]

def deal_hands(deck: Deck) -> Tuple[Deck, Deck, Deck, Deck]:

    """Deal the cards in the deck into four hands"""

    return (deck[0::4], deck[1::4], deck[2::4], deck[3::4])

Allow any subclass⚑

Every class is also a valid type. Any instance of a subclass is also compatible with all superclasses – it follows that every value is compatible with the object type (and incidentally also the Any type, discussed below). Mypy analyzes the bodies of classes to determine which methods and attributes are available in instances. For example

class A:
    def f(self) -> int:  # Type of self inferred (A)
        return 2

class B(A):
    def f(self) -> int:
         return 3
    def g(self) -> int:
        return 4

def foo(a: A) -> None:
    print(a.f())  # 3
    a.g()         # Error: "A" has no attribute "g"

foo(B())  # OK (B is a subclass of A)

Deduce returned value type from the arguments⚑

The previous approach works if you don't need to use class objects that inherit from a given class. For example:

class User:
    # Defines fields like name, email

class BasicUser(User):
    def upgrade(self):
        """Upgrade to Pro"""

class ProUser(User):
    def pay(self):
        """Pay bill"""

def new_user(user_class) -> User:
    user = user_class()
    # (Here we could write the user object to a database)
    return user

Where:

• ProUser doesn't inherit from BasicUser.
• new_user creates an instance of one of these classes if you pass it the right class object.

The problem is that right now mypy doesn't know which subclass of User you're giving it, and will only accept the methods and attributes defined in the parent class User.

buyer = new_user(ProUser)
buyer.pay()  # Rejected, not a method on User

This can be solved using Type variables with upper bounds.

UserT = TypeVar('UserT', bound=User)

def new_user(user_class: Type[UserT]) -> UserT:
    # Same  implementation as before

We're creating a new type UserT that is linked to the class or subclasses of User. That way, mypy knows that the return value is an object created from the class given in the argument user_class.

beginner = new_user(BasicUser)  # Inferred type is BasicUser
beginner.upgrade()  # OK

"Using `UserType` is [not supported by
pylint](https://github.com/PyCQA/pylint/issues/6003), use `UserT`
instead."

Keep in mind that the TypeVar is a Generic type, as such, they take one or more type parameters, similar to built-in types such as List[X].

That means that when you create type aliases, you'll need to give the type parameter. So:

UserT = TypeVar("UserT", bound=User)
UserTs = List[Type[UserT]]


def new_users(user_class: UserTs) -> UserT: # Type error!
    pass

Will give a Missing type parameters for generic type "UserTs" error. To solve it use:

def new_users(user_class: UserTs[UserT]) -> UserT: # OK!
    pass

Define a TypeVar with restrictions⚑

By default, a type variable can be replaced with any type. However, sometimes it’s useful to have a type variable that can only have some specific types as its value. A typical example is a type variable that can only have values str and bytes:

from typing import TypeVar

AnyStr = TypeVar('AnyStr', str, bytes)

This is actually such a common type variable that AnyStr is defined in typing and we don’t need to define it ourselves.

We can use AnyStr to define a function that can concatenate two strings or bytes objects, but it can’t be called with other argument types:

from typing import AnyStr

def concat(x: AnyStr, y: AnyStr) -> AnyStr:
    return x + y

concat('a', 'b')    # Okay
concat(b'a', b'b')  # Okay
concat(1, 2)        # Error!

Note that this is different from a union type, since combinations of str and bytes are not accepted:

concat('string', b'bytes')   # Error!

In this case, this is exactly what we want, since it’s not possible to concatenate a string and a bytes object! The type checker will reject this function:

def union_concat(x: Union[str, bytes], y: Union[str, bytes]) -> Union[str, bytes]:
    return x + y  # Error: can't concatenate str and bytes

Overloading the methods⚑

Sometimes the types of several variables are related, such as “if x is type A, y is type B, else y is type C”. Basic type hints cannot describe such relationships, making type checking cumbersome or inaccurate. We can instead use @typing.overload to represent type relationships properly.

from __future__ import annotations

from collections.abc import Sequence
from typing import overload


@overload
def double(input_: int) -> int:
    ...


@overload
def double(input_: Sequence[int]) -> list[int]:
    ...


def double(input_: int | Sequence[int]) -> int | list[int]:
    if isinstance(input_, Sequence):
        return [i * 2 for i in input_]
    return input_ * 2

This looks a bit weird at first glance—we are defining double three times! Let’s take it apart.

The first two @overload definitions exist only for their type hints. Each definition represents an allowed combination of types. These definitions never run, so their bodies could contain anything, but it’s idiomatic to use Python’s ... (ellipsis) literal.

The third definition is the actual implementation. In this case, we need to provide type hints that union all the possible types for each variable. Without such hints, Mypy will skip type checking the function body.

When Mypy checks the file, it collects the @overload definitions as type hints. It then uses the first non-@overload definition as the implementation. All @overload definitions must come before the implementation, and multiple implementations are not allowed.

When Python imports the file, the @overload definitions create temporary double functions, but each is overridden by the next definition. After importing, only the implementation exists. As a protection against accidentally missing implementations, attempting to call an @overload definition will raise a NotImplementedError.

@overload can represent arbitrarily complex scenarios. For a couple more examples, see the function overloading section of the Mypy docs.

Use a constrained TypeVar in the definition of a class attributes.⚑

If you try to use a TypeVar in the definition of a class attribute:

class File:
    """Model a computer file."""

    path: str
    content: Optional[AnyStr] = None # mypy error!

mypy will complain with Type variable AnyStr is unbound [valid-type], to solve it, you need to make the class inherit from the Generic[AnyStr].

class File(Generic[AnyStr]):
    """Model a computer file."""

    path: str
    content: Optional[AnyStr] = None

Why you ask? I have absolutely no clue. I've asked that question in the gitter python typing channel but the kind answer that @ktbarrett gave me sounded like Chinese.

You can't just use a type variable for attributes or variables, you have to create some generic context, whether that be a function or a class, so that you can instantiate the generic context (or the analyzer can infer it) (i.e. context[var]). That's not possible if you don't specify that the class is a generic context. It also ensure than all uses of that variable in the context resolve to the same type.

If you don't mind helping me understand it, please contact me.

Specify the type of the class in it's method and attributes⚑

If you are using Python 3.10 or later, it just works. Python 3.7 introduces PEP 563: postponed evaluation of annotations. A module that uses the future statement from __future__ import annotations to store annotations as strings automatically:

from __future__ import annotations

class Position:
    def __add__(self, other: Position) -> Position:
        ...

But pyflakes will still complain, so I've used strings.

from __future__ import annotations

class Position:
    def __add__(self, other: 'Position') -> 'Position':
        ...

Type hints of Generators⚑

from typing import Generator

def generate() -> Generator[int, None, None]:

Where the first argument of Generator is the type of the yielded value.

Usage of ellipsis on Tuple type hints⚑

The ellipsis is used to specify an arbitrary-length homogeneous tuples, for example Tuple[int, ...].

Using typing.cast⚑

Sometimes the type hints of your program don't work as you expect, if you've given up on fixing the issue you can # type: ignore it, but if you know what type you want to enforce, you can use typing.cast() explicitly or implicitly from Any with type hints. With casting we can force the type checker to treat a variable as a given type.

The simplest cast()⚑

When we call cast(), we pass it two arguments: a type, and a value. cast() returns value unchanged, but type checkers will treat the return value as the given type instead of the input type. For example, we can make Mypy treat an integer as a string:

from typing import cast

x = 1
reveal_type(x)
y = cast(str, x)
reveal_type(y)
y.upper()

Checking this program with Mypy, it doesn't report any errors, but it does debug the types of x and y for us:

$ mypy example.py

example.py:6: note: Revealed type is "builtins.int"
example.py:8: note: Revealed type is "builtins.str"

But, if we remove the reveal_type() calls and run the code, it crashes:

$ python example.py
Traceback (most recent call last):
  File "/.../example.py", line 7, in <module>
    y.upper()
AttributeError: 'int' object has no attribute 'upper'

Usually Mypy would detect this bug, as it knows int objects do not have an upper() method. But our cast() forced Mypy to treat y as a str, so it assumed the call would succeed.

Use cases⚑

The main case to reach for cast() are when the type hints for a module are either missing, incomplete, or incorrect. This may be the case for third party packages, or occasionally for things in the standard library.

Take this example:

import datetime as dt
from typing import cast

from third_party import get_data

data = get_data()
last_import_time = cast(dt.datetime, data["last_import_time"])

Imagine get_data() has a return type of dict[str, Any], rather than using stricter per-key types with a TypedDict. From reading the documentation or source we might find that the last_import_time key always contains a datetime object. Therefore, when we access it, we can wrap it in a cast(), to tell our type checker the real type rather than continuing with Any.

When we encounter missing, incomplete, or incorrect type hints, we can contribute back a fix. This may be in the package itself, its related stubs package, or separate stubs in Python’s typeshed. But until such a fix is released, we will need to use cast() to make our code pass type checking.

Implicit Casting From Any⚑

It’s worth noting that Any has special treatment: when we store a variable with type Any in a variable with a specific type, type checkers treat this as an implicit cast. We can thus write our previous example without cast():

import datetime as dt

from third_party import get_data

data = get_data()
last_import_time: dt.datetime = data["last_import_time"]

This kind of implicit casting is the first tool we should reach for when interacting with libraries that return Any. It also applies when we pass a variable typed Any as a specifically typed function argument or return value.

Calling cast() directly is often more useful when dealing with incorrect types other than Any.

Mypy’s warn_redundant_casts option⚑

When we use cast() to override a third party function’s type, that type be corrected in a later version (perhaps from our own PR!). After such an update, the cast() is unnecessary clutter that may confuse readers.

We can detect such unnecessary casts by activating Mypy’s warn_redundant_casts option. With this flag turned on, Mypy will log an error for each use of cast() that casts a variable to the type it already has.

Using mypy with an existing codebase⚑

These steps will get you started with mypy on an existing codebase:

• Start small: Pick a subset of your codebase to run mypy on, without any annotations.

  You’ll probably need to fix some mypy errors, either by inserting annotations requested by mypy or by adding # type: ignore comments to silence errors you don’t want to fix now.

  Get a clean mypy build for some files, with some annotations. * Write a mypy runner script to ensure consistent results. Here are some steps you may want to do in the script: * Ensure that you install the correct version of mypy. * Specify mypy config file or command-line options. * Provide set of files to type check. You may want to configure the inclusion and exclusion filters for full control of the file list. * Run mypy in Continuous Integration to prevent type errors:

  Once you have a clean mypy run and a runner script for a part of your codebase, set up your Continuous Integration (CI) system to run mypy to ensure that developers won’t introduce bad annotations. A small CI script could look something like this:

  python3 -m pip install mypy==0.600  # Pinned version avoids surprises
  scripts/mypy  # Runs with the correct options
  

  * Gradually annotate commonly imported modules: Most projects have some widely imported modules, such as utilities or model classes. It’s a good idea to annotate these soon, since this allows code using these modules to be type checked more effectively. Since mypy supports gradual typing, it’s okay to leave some of these modules unannotated. The more you annotate, the more useful mypy will be, but even a little annotation coverage is useful. * Write annotations as you change existing code and write new code: Now you are ready to include type annotations in your development workflows. Consider adding something like these in your code style conventions:
  • Developers should add annotations for any new code.
  • It’s also encouraged to write annotations when you change existing code.

If you need to ignore a linter error and a type error use first the type and then the linter. For example, # type: ignore # noqa: W0212.

Reveal the type of an expression⚑

You can use reveal_type(expr) to ask mypy to display the inferred static type of an expression. This can be useful when you don't quite understand how mypy handles a particular piece of code. Example:

reveal_type((1, 'hello'))  # Revealed type is 'Tuple[builtins.int, builtins.str]'

You can also use reveal_locals() at any line in a file to see the types of all local variables at once. Example:

a = 1
b = 'one'
reveal_locals()
# Revealed local types are:
#     a: builtins.int
#     b: builtins.str

reveal_type and reveal_locals are only understood by mypy and don't exist in Python. If you try to run your program, you’ll have to remove any reveal_type and reveal_locals calls before you can run your code. Both are always available and you don't need to import them.

Solve cyclic imports due to typing⚑

You can use a conditional import that is only active in "type hinting mode", but doesn't interfere at run time. The typing.TYPE_CHECKING constant makes this easily possible. For example:

# thing.py
from typing import TYPE_CHECKING
if TYPE_CHECKING:
    from connection import ApiConnection

class Thing:
    def __init__(self, connection: 'ApiConnection'):
        self._conn = connection

The code will now execute properly as there is no circular import issue anymore. Type hinting tools on the other hand should still be able to resolve the ApiConnection type hint in Thing.__init__.

Make your library compatible with mypy⚑

PEP 561 notes three main ways to distribute type information. The first is a package that has only inline type annotations in the code itself. The second is a package that ships stub files with type information alongside the runtime code. The third method, also known as a “stub only package” is a package that ships type information for a package separately as stub files.

If you would like to publish a library package to a package repository (e.g. PyPI) for either internal or external use in type checking, packages that supply type information via type comments or annotations in the code should put a py.typed file in their package directory. For example, with a directory structure as follows

setup.py
package_a/
    __init__.py
    lib.py
    py.typed

the setup.py might look like:

from distutils.core import setup

setup(
    name="SuperPackageA",
    author="Me",
    version="0.1",
    package_data={"package_a": ["py.typed"]},
    packages=["package_a"]
)

Reference⚑

• Bernat gabor article on the state of type hints in python
• Real python article on type checking