More types

This section introduces a few additional kinds of types, including NoReturn, NewType, TypedDict, and types for async code. It also discusses how to give functions more precise types using overloads. All of these are only situationally useful, so feel free to skip this section and come back when you have a need for some of them.

Here’s a quick summary of what’s covered here:

• NoReturn lets you tell mypy that a function never returns normally.
• NewType lets you define a variant of a type that is treated as a separate type by mypy but is identical to the original type at runtime. For example, you can have UserId as a variant of int that is just an int at runtime.
• @overload lets you define a function that can accept multiple distinct signatures. This is useful for if you need to encode a relationship between the arguments and the return type that would be difficult to express normally.
• TypedDict lets you give precise types for dictionaries that represent objects with a fixed schema, such as {'id': 1, 'items': ['x']}.
• Async types let you type check programs using async and await.

The NoReturn type

Mypy provides support for functions that never return. For example, a function that unconditionally raises an exception:

from typing import NoReturn

def stop() -> NoReturn:
    raise Exception('no way')

Mypy will ensure that functions annotated as returning NoReturn truly never return, either implicitly or explicitly. Mypy will also recognize that the code after calls to such functions is unreachable and will behave accordingly:

def f(x: int) -> int:
    if x == 0:
        return x
    stop()
    return 'whatever works'  # No error in an unreachable block

In earlier Python versions you need to install typing_extensions using pip to use NoReturn in your code. Python 3 command line:

python3 -m pip install --upgrade typing-extensions

This works for Python 2:

pip install --upgrade typing-extensions

NewTypes

There are situations where you may want to avoid programming errors by creating simple derived classes that are only used to distinguish certain values from base class instances. Example:

class UserId(int):
    pass

get_by_user_id(user_id: UserId):
    ...

However, this approach introduces some runtime overhead. To avoid this, the typing module provides a helper function NewType that creates simple unique types with almost zero runtime overhead. Mypy will treat the statement Derived = NewType('Derived', Base) as being roughly equivalent to the following definition:

class Derived(Base):
    def __init__(self, _x: Base) -> None:
        ...

However, at runtime, NewType('Derived', Base) will return a dummy function that simply returns its argument:

def Derived(_x):
    return _x

Mypy will require explicit casts from int where UserId is expected, while implicitly casting from UserId where int is expected. Examples:

from typing import NewType

UserId = NewType('UserId', int)

def name_by_id(user_id: UserId) -> str:
    ...

UserId('user')          # Fails type check

name_by_id(42)          # Fails type check
name_by_id(UserId(42))  # OK

num = UserId(5) + 1     # type: int

NewType accepts exactly two arguments. The first argument must be a string literal containing the name of the new type and must equal the name of the variable to which the new type is assigned. The second argument must be a properly subclassable class, i.e., not a type construct like Union, etc.

The function returned by NewType accepts only one argument; this is equivalent to supporting only one constructor accepting an instance of the base class (see above). Example:

from typing import NewType

class PacketId:
    def __init__(self, major: int, minor: int) -> None:
        self._major = major
        self._minor = minor

TcpPacketId = NewType('TcpPacketId', PacketId)

packet = PacketId(100, 100)
tcp_packet = TcpPacketId(packet)  # OK

tcp_packet = TcpPacketId(127, 0)  # Fails in type checker and at runtime

You cannot use isinstance() or issubclass() on the object returned by NewType(), because function objects don’t support these operations. You cannot create subclasses of these objects either.

Note

Unlike type aliases, NewType will create an entirely new and unique type when used. The intended purpose of NewType is to help you detect cases where you accidentally mixed together the old base type and the new derived type.

For example, the following will successfully typecheck when using type aliases:

UserId = int

def name_by_id(user_id: UserId) -> str:
    ...

name_by_id(3)  # ints and UserId are synonymous

But a similar example using NewType will not typecheck:

from typing import NewType

UserId = NewType('UserId', int)

def name_by_id(user_id: UserId) -> str:
    ...

name_by_id(3)  # int is not the same as UserId

Function overloading

Sometimes the arguments and types in a function depend on each other in ways that can’t be captured with a Union. For example, suppose we want to write a function that can accept x-y coordinates. If we pass in just a single x-y coordinate, we return a ClickEvent object. However, if we pass in two x-y coordinates, we return a DragEvent object.

Our first attempt at writing this function might look like this:

from typing import Union, Optional

def mouse_event(x1: int,
                y1: int,
                x2: Optional[int] = None,
                y2: Optional[int] = None) -> Union[ClickEvent, DragEvent]:
    if x2 is None and y2 is None:
        return ClickEvent(x1, y1)
    elif x2 is not None and y2 is not None:
        return DragEvent(x1, y1, x2, y2)
    else:
        raise TypeError("Bad arguments")

While this function signature works, it’s too loose: it implies mouse_event could return either object regardless of the number of arguments we pass in. It also does not prohibit a caller from passing in the wrong number of ints: mypy would treat calls like mouse_event(1, 2, 20) as being valid, for example.

We can do better by using overloading which lets us give the same function multiple type annotations (signatures) to more accurately describe the function’s behavior:

from typing import Union, overload

# Overload *variants* for 'mouse_event'.
# These variants give extra information to the type checker.
# They are ignored at runtime.

@overload
def mouse_event(x1: int, y1: int) -> ClickEvent: ...
@overload
def mouse_event(x1: int, y1: int, x2: int, y2: int) -> DragEvent: ...

# The actual *implementation* of 'mouse_event'.
# The implementation contains the actual runtime logic.
#
# It may or may not have type hints. If it does, mypy
# will check the body of the implementation against the
# type hints.
#
# Mypy will also check and make sure the signature is
# consistent with the provided variants.

def mouse_event(x1: int,
                y1: int,
                x2: Optional[int] = None,
                y2: Optional[int] = None) -> Union[ClickEvent, DragEvent]:
    if x2 is None and y2 is None:
        return ClickEvent(x1, y1)
    elif x2 is not None and y2 is not None:
        return DragEvent(x1, y1, x2, y2)
    else:
        raise TypeError("Bad arguments")

This allows mypy to understand calls to mouse_event much more precisely. For example, mypy will understand that mouse_event(5, 25) will always have a return type of ClickEvent and will report errors for calls like mouse_event(5, 25, 2).

As another example, suppose we want to write a custom container class that implements the __getitem__ method ([] bracket indexing). If this method receives an integer we return a single item. If it receives a slice, we return a Sequence of items.

We can precisely encode this relationship between the argument and the return type by using overloads like so:

from typing import Sequence, TypeVar, Union, overload

T = TypeVar('T')

class MyList(Sequence[T]):
    @overload
    def __getitem__(self, index: int) -> T: ...

    @overload
    def __getitem__(self, index: slice) -> Sequence[T]: ...

    def __getitem__(self, index: Union[int, slice]) -> Union[T, Sequence[T]]:
        if isinstance(index, int):
            # Return a T here
        elif isinstance(index, slice):
            # Return a sequence of Ts here
        else:
            raise TypeError(...)

Note

If you just need to constrain a type variable to certain types or subtypes, you can use a value restriction.

Runtime behavior

An overloaded function must consist of two or more overload variants followed by an implementation. The variants and the implementations must be adjacent in the code: think of them as one indivisible unit.

The variant bodies must all be empty; only the implementation is allowed to contain code. This is because at runtime, the variants are completely ignored: they’re overridden by the final implementation function.

This means that an overloaded function is still an ordinary Python function! There is no automatic dispatch handling and you must manually handle the different types in the implementation (e.g. by using if statements and isinstance checks).

If you are adding an overload within a stub file, the implementation function should be omitted: stubs do not contain runtime logic.

Note

While we can leave the variant body empty using the pass keyword, the more common convention is to instead use the ellipsis (...) literal.

Type checking calls to overloads

When you call an overloaded function, mypy will infer the correct return type by picking the best matching variant, after taking into consideration both the argument types and arity. However, a call is never type checked against the implementation. This is why mypy will report calls like mouse_event(5, 25, 3) as being invalid even though it matches the implementation signature.

If there are multiple equally good matching variants, mypy will select the variant that was defined first. For example, consider the following program:

from typing import List, overload

@overload
def summarize(data: List[int]) -> float: ...

@overload
def summarize(data: List[str]) -> str: ...

def summarize(data):
    if not data:
        return 0.0
    elif isinstance(data[0], int):
        # Do int specific code
    else:
        # Do str-specific code

# What is the type of 'output'? float or str?
output = summarize([])

The summarize([]) call matches both variants: an empty list could be either a List[int] or a List[str]. In this case, mypy will break the tie by picking the first matching variant: output will have an inferred type of float. The implementor is responsible for making sure summarize breaks ties in the same way at runtime.

There are however are two exceptions to the “pick the first match” rule. First, if multiple variants match due to an argument being of type Any, mypy will make the inferred type also be Any:

dynamic_var: Any = some_dynamic_function()

# output2 is of type 'Any'
output2 = summarize(dynamic_var)

Second, if multiple variants match due to one or more of the arguments being a union, mypy will make the inferred type be the union of the matching variant returns:

some_list: Union[List[int], List[str]]

# output3 is of type 'Union[float, str]'
output3 = summarize(some_list)

Note

Due to the “pick the first match” rule, changing the order of your overload variants can change how mypy type checks your program.

To minimize potential issues, we recommend that you:

1. Make sure your overload variants are listed in the same order as the runtime checks (e.g. isinstance checks) in your implementation.
2. Order your variants and runtime checks from most to least specific. (See the following section for an example).

Type checking the variants

Mypy will perform several checks on your overload variant definitions to ensure they behave as expected. First, mypy will check and make sure that no overload variant is shadowing a subsequent one. For example, consider the following function which adds together two Expression objects, and contains a special-case to handle receiving two Literal types:

from typing import overload, Union

class Expression:
    # ...snip...

class Literal(Expression):
    # ...snip...

# Warning -- the first overload variant shadows the second!

@overload
def add(left: Expression, right: Expression) -> Expression: ...

@overload
def add(left: Literal, right: Literal) -> Literal: ...

def add(left: Expression, right: Expression) -> Expression:
    # ...snip...

While this code snippet is technically type-safe, it does contain an anti-pattern: the second variant will never be selected! If we try calling add(Literal(3), Literal(4)), mypy will always pick the first variant and evaluate the function call to be of type Expression, not Literal. This is because Literal is a subtype of Expression, which means the “pick the first match” rule will always halt after considering the first overload.

Because having an overload variant that can never be matched is almost certainly a mistake, mypy will report an error. To fix the error, we can either 1) delete the second overload or 2) swap the order of the overloads:

# Everything is ok now -- the variants are correctly ordered
# from most to least specific.

@overload
def add(left: Literal, right: Literal) -> Literal: ...

@overload
def add(left: Expression, right: Expression) -> Expression: ...

def add(left: Expression, right: Expression) -> Expression:
    # ...snip...

Mypy will also type check the different variants and flag any overloads that have inherently unsafely overlapping variants. For example, consider the following unsafe overload definition:

from typing import overload, Union

@overload
def unsafe_func(x: int) -> int: ...

@overload
def unsafe_func(x: object) -> str: ...

def unsafe_func(x: object) -> Union[int, str]:
    if isinstance(x, int):
        return 42
    else:
        return "some string"

On the surface, this function definition appears to be fine. However, it will result in a discrepency between the inferred type and the actual runtime type when we try using it like so:

some_obj: object = 42
unsafe_func(some_obj) + " danger danger"  # Type checks, yet crashes at runtime!

Since some_obj is of type object, mypy will decide that unsafe_func must return something of type str and concludes the above will type check. But in reality, unsafe_func will return an int, causing the code to crash at runtime!

To prevent these kinds of issues, mypy will detect and prohibit inherently unsafely overlapping overloads on a best-effort basis. Two variants are considered unsafely overlapping when both of the following are true:

1. All of the arguments of the first variant are compatible with the second.
2. The return type of the first variant is not compatible with (e.g. is not a subtype of) the second.

So in this example, the int argument in the first variant is a subtype of the object argument in the second, yet the int return type not is a subtype of str. Both conditions are true, so mypy will correctly flag unsafe_func as being unsafe.

However, mypy will not detect all unsafe uses of overloads. For example, suppose we modify the above snippet so it calls summarize instead of unsafe_func:

some_list: List[str] = []
summarize(some_list) + "danger danger"  # Type safe, yet crashes at runtime!

We run into a similar issue here. This program type checks if we look just at the annotations on the overloads. But since summarize(...) is designed to be biased towards returning a float when it receives an empty list, this program will actually crash during runtime.

The reason mypy does not flag definitions like summarize as being potentially unsafe is because if it did, it would be extremely difficult to write a safe overload. For example, suppose we define an overload with two variants that accept types A and B respectively. Even if those two types were completely unrelated, the user could still potentially trigger a runtime error similar to the ones above by passing in a value of some third type C that inherits from both A and B.

Thankfully, these types of situations are relatively rare. What this does mean, however, is that you should exercise caution when designing or using an overloaded function that can potentially receive values that are an instance of two seemingly unrelated types.

Type checking the implementation

The body of an implementation is type-checked against the type hints provided on the implementation. For example, in the MyList example up above, the code in the body is checked with argument list index: Union[int, slice] and a return type of Union[T, Sequence[T]]. If there are no annotations on the implementation, then the body is not type checked. If you want to force mypy to check the body anyways, use the --check-untyped-defs flag (more details here).

The variants must also also be compatible with the implementation type hints. In the MyList example, mypy will check that the parameter type int and the return type T are compatible with Union[int, slice] and Union[T, Sequence] for the first variant. For the second variant it verifies the parameter type slice and the return type Sequence[T] are compatible with Union[int, slice] and Union[T, Sequence].

Note

The overload semantics documented above are new as of mypy 0.620.

Previously, mypy used to perform type erasure on all overload variants. For example, the summarize example from the previous section used to be illegal because List[str] and List[int] both erased to just List[Any]. This restriction was removed in mypy 0.620.

Mypy also previously used to select the best matching variant using a different algorithm. If this algorithm failed to find a match, it would default to returning Any. The new algorithm uses the “pick the first match” rule and will fall back to returning Any only if the input arguments also contain Any.

Typing async/await

Mypy supports the ability to type coroutines that use the async/await syntax introduced in Python 3.5. For more information regarding coroutines and this new syntax, see PEP 492.

Functions defined using async def are typed just like normal functions. The return type annotation should be the same as the type of the value you expect to get back when await-ing the coroutine.

import asyncio

async def format_string(tag: str, count: int) -> str:
    return 'T-minus {} ({})'.format(count, tag)

async def countdown_1(tag: str, count: int) -> str:
    while count > 0:
        my_str = await format_string(tag, count)  # has type 'str'
        print(my_str)
        await asyncio.sleep(0.1)
        count -= 1
    return "Blastoff!"

loop = asyncio.get_event_loop()
loop.run_until_complete(countdown_1("Millennium Falcon", 5))
loop.close()

The result of calling an async def function without awaiting will be a value of type typing.Coroutine[Any, Any, T], which is a subtype of Awaitable[T]:

my_coroutine = countdown_1("Millennium Falcon", 5)
reveal_type(my_coroutine)  # has type 'Coroutine[Any, Any, str]'

Note

reveal_type() displays the inferred static type of an expression.

If you want to use coroutines in Python 3.4, which does not support the async def syntax, you can instead use the @asyncio.coroutine decorator to convert a generator into a coroutine.

Note that we set the YieldType of the generator to be Any in the following example. This is because the exact yield type is an implementation detail of the coroutine runner (e.g. the asyncio event loop) and your coroutine shouldn’t have to know or care about what precisely that type is.

from typing import Any, Generator
import asyncio

@asyncio.coroutine
def countdown_2(tag: str, count: int) -> Generator[Any, None, str]:
    while count > 0:
        print('T-minus {} ({})'.format(count, tag))
        yield from asyncio.sleep(0.1)
        count -= 1
    return "Blastoff!"

loop = asyncio.get_event_loop()
loop.run_until_complete(countdown_2("USS Enterprise", 5))
loop.close()

As before, the result of calling a generator decorated with @asyncio.coroutine will be a value of type Awaitable[T].

Note

At runtime, you are allowed to add the @asyncio.coroutine decorator to both functions and generators. This is useful when you want to mark a work-in-progress function as a coroutine, but have not yet added yield or yield from statements:

import asyncio

@asyncio.coroutine
def serialize(obj: object) -> str:
    # todo: add yield/yield from to turn this into a generator
    return "placeholder"

However, mypy currently does not support converting functions into coroutines. Support for this feature will be added in a future version, but for now, you can manually force the function to be a generator by doing something like this:

from typing import Generator
import asyncio

@asyncio.coroutine
def serialize(obj: object) -> Generator[None, None, str]:
    # todo: add yield/yield from to turn this into a generator
    if False:
        yield
    return "placeholder"

You may also choose to create a subclass of Awaitable instead:

from typing import Any, Awaitable, Generator
import asyncio

class MyAwaitable(Awaitable[str]):
    def __init__(self, tag: str, count: int) -> None:
        self.tag = tag
        self.count = count

    def __await__(self) -> Generator[Any, None, str]:
        for i in range(n, 0, -1):
            print('T-minus {} ({})'.format(i, tag))
            yield from asyncio.sleep(0.1)
        return "Blastoff!"

def countdown_3(tag: str, count: int) -> Awaitable[str]:
    return MyAwaitable(tag, count)

loop = asyncio.get_event_loop()
loop.run_until_complete(countdown_3("Heart of Gold", 5))
loop.close()

To create an iterable coroutine, subclass AsyncIterator:

from typing import Optional, AsyncIterator
import asyncio

class arange(AsyncIterator[int]):
    def __init__(self, start: int, stop: int, step: int) -> None:
        self.start = start
        self.stop = stop
        self.step = step
        self.count = start - step

    def __aiter__(self) -> AsyncIterator[int]:
        return self

    async def __anext__(self) -> int:
        self.count += self.step
        if self.count == self.stop:
            raise StopAsyncIteration
        else:
            return self.count

async def countdown_4(tag: str, n: int) -> str:
    async for i in arange(n, 0, -1):
        print('T-minus {} ({})'.format(i, tag))
        await asyncio.sleep(0.1)
    return "Blastoff!"

loop = asyncio.get_event_loop()
loop.run_until_complete(countdown_4("Serenity", 5))
loop.close()

For a more concrete example, the mypy repo has a toy webcrawler that demonstrates how to work with coroutines. One version uses async/await and one uses yield from.

TypedDict

Note

TypedDict is an officially supported feature, but it is still experimental.

Python programs often use dictionaries with string keys to represent objects. Here is a typical example:

movie = {'name': 'Blade Runner', 'year': 1982}

Only a fixed set of string keys is expected ('name' and 'year' above), and each key has an independent value type (str for 'name' and int for 'year' above). We’ve previously seen the Dict[K, V] type, which lets you declare uniform dictionary types, where every value has the same type, and arbitrary keys are supported. This is clearly not a good fit for movie above. Instead, you can use a TypedDict to give a precise type for objects like movie, where the type of each dictionary value depends on the key:

from mypy_extensions import TypedDict

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

movie = {'name': 'Blade Runner', 'year': 1982}  # type: Movie

Movie is a TypedDict type with two items: 'name' (with type str) and 'year' (with type int). Note that we used an explicit type annotation for the movie variable. This type annotation is important – without it, mypy will try to infer a regular, uniform Dict type for movie, which is not what we want here.

Note

If you pass a TypedDict object as an argument to a function, no type annotation is usually necessary since mypy can infer the desired type based on the declared argument type. Also, if an assignment target has been previously defined, and it has a TypedDict type, mypy will treat the assigned value as a TypedDict, not Dict.

Now mypy will recognize these as valid:

name = movie['name']  # Okay; type of name is str
year = movie['year']  # Okay; type of year is int

Mypy will detect an invalid key as an error:

director = movie['director']  # Error: 'director' is not a valid key

Mypy will also reject a runtime-computed expression as a key, as it can’t verify that it’s a valid key. You can only use string literals as TypedDict keys.

The TypedDict type object can also act as a constructor. It returns a normal dict object at runtime – a TypedDict does not define a new runtime type:

toy_story = Movie(name='Toy Story', year=1995)

This is equivalent to just constructing a dictionary directly using { ... } or dict(key=value, ...). The constructor form is sometimes convenient, since it can be used without a type annotation, and it also makes the type of the object explicit.

Like all types, TypedDicts can be used as components to build arbitrarily complex types. For example, you can define nested TypedDicts and containers with TypedDict items. Unlike most other types, mypy uses structural compatibility checking (or structural subtyping) with TypedDicts. A TypedDict object with extra items is compatible with a narrower TypedDict, assuming item types are compatible (totality also affects subtyping, as discussed below).

Note

You need to install mypy_extensions using pip to use TypedDict:

python3 -m pip install --upgrade mypy-extensions

Or, if you are using Python 2:

pip install --upgrade mypy-extensions

Totality

By default mypy ensures that a TypedDict object has all the specified keys. This will be flagged as an error:

# Error: 'year' missing
toy_story = {'name': 'Toy Story'}  # type: Movie

Sometimes you want to allow keys to be left out when creating a TypedDict object. You can provide the total=False argument to TypedDict(...) to achieve this:

GuiOptions = TypedDict(
    'GuiOptions', {'language': str, 'color': str}, total=False)
options = {}  # type: GuiOptions  # Okay
options['language'] = 'en'

You may need to use get() to access items of a partial (non-total) TypedDict, since indexing using [] could fail at runtime. However, mypy still lets use [] with a partial TypedDict – you just need to be careful with it, as it could result in a KeyError. Requiring get() everywhere would be too cumbersome. (Note that you are free to use get() with total TypedDicts as well.)

Keys that aren’t required are shown with a ? in error messages:

# Revealed type is 'TypedDict('GuiOptions', {'language'?: builtins.str,
#                                            'color'?: builtins.str})'
reveal_type(options)

Totality also affects structural compatibility. You can’t use a partial TypedDict when a total one is expected. Also, a total TypedDict is not valid when a partial one is expected.

Class-based syntax

An alternative, class-based syntax to define a TypedDict is supported in Python 3.6 and later:

from mypy_extensions import TypedDict

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

The above definition is equivalent to the original Movie definition. It doesn’t actually define a real class. This syntax also supports a form of inheritance – subclasses can define additional items. However, this is primarily a notational shortcut. Since mypy uses structural compatibility with TypedDicts, inheritance is not required for compatibility. Here is an example of inheritance:

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

class BookBasedMovie(Movie):
    based_on: str

Now BookBasedMovie has keys name, year and based_on.

Mixing required and non-required items

In addition to allowing reuse across TypedDict types, inheritance also allows you to mix required and non-required (using total=False) items in a single TypedDict. Example:

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

class Movie(MovieBase, total=False):
    based_on: str

Now Movie has required keys name and year, while based_on can be left out when constructing an object. A TypedDict with a mix of required and non-required keys, such as Movie above, will only be compatible with another TypedDict if all required keys in the other TypedDict are required keys in the first TypedDict, and all non-required keys of the other TypedDict are also non-required keys in the first TypedDict.