Type inference and type annotations

Type inference

Mypy considers the initial assignment as the definition of a variable. If you do not explicitly specify the type of the variable, mypy infers the type based on the static type of the value expression:

i = 1           # Infer type "int" for i
l = [1, 2]      # Infer type "List[int]" for l

Type inference is not used in dynamically typed functions (those without a function type annotation) — every local variable type defaults to Any in such functions. Any is discussed later in more detail.

Explicit types for variables

You can override the inferred type of a variable by using a variable type annotation:

from typing import Union

x: Union[int, str] = 1

Without the type annotation, the type of x would be just int. We use an annotation to give it a more general type Union[int, str] (this type means that the value can be either an int or a str). Mypy checks that the type of the initializer is compatible with the declared type. The following example is not valid, since the initializer is a floating point number, and this is incompatible with the declared type:

x: Union[int, str] = 1.1  # Error!

The variable annotation syntax is available starting from Python 3.6. In earlier Python versions, you can use a special comment after an assignment statement to declare the type of a variable:

x = 1  # type: Union[int, str]

We’ll use both syntax variants in examples. The syntax variants are mostly interchangeable, but the variable annotation syntax allows defining the type of a variable without initialization, which is not possible with the comment syntax:

x: str  # Declare type of 'x' without initialization


The best way to think about this is that the type annotation sets the type of the variable, not the type of the expression. To force the type of an expression you can use cast(<type>, <expression>).

Explicit types for collections

The type checker cannot always infer the type of a list or a dictionary. This often arises when creating an empty list or dictionary and assigning it to a new variable that doesn’t have an explicit variable type. Here is an example where mypy can’t infer the type without some help:

l = []  # Error: Need type annotation for "l"

In these cases you can give the type explicitly using a type annotation:

l: List[int] = []       # Create empty list with type List[int]
d: Dict[str, int] = {}  # Create empty dictionary (str -> int)

Similarly, you can also give an explicit type when creating an empty set:

s: Set[int] = set()

Compatibility of container types

The following program generates a mypy error, since List[int] is not compatible with List[object]:

def f(l: List[object], k: List[int]) -> None:
    l = k  # Type check error: incompatible types in assignment

The reason why the above assignment is disallowed is that allowing the assignment could result in non-int values stored in a list of int:

def f(l: List[object], k: List[int]) -> None:
    l = k
    print(k[-1])  # Ouch; a string in List[int]

Other container types like Dict and Set behave similarly. We will discuss how you can work around this in Invariance vs covariance.

You can still run the above program; it prints x. This illustrates the fact that static types are used during type checking, but they do not affect the runtime behavior of programs. You can run programs with type check failures, which is often very handy when performing a large refactoring. Thus you can always ‘work around’ the type system, and it doesn’t really limit what you can do in your program.

Context in type inference

Type inference is bidirectional and takes context into account. For example, the following is valid:

def f(l: List[object]) -> None:
    l = [1, 2]  # Infer type List[object] for [1, 2], not List[int]

In an assignment, the type context is determined by the assignment target. In this case this is l, which has the type List[object]. The value expression [1, 2] is type checked in this context and given the type List[object]. In the previous example we introduced a new variable l, and here the type context was empty.

Declared argument types are also used for type context. In this program mypy knows that the empty list [] should have type List[int] based on the declared type of arg in foo:

def foo(arg: List[int]) -> None:
    print('Items:', ''.join(str(a) for a in arg))

foo([])  # OK

However, context only works within a single statement. Here mypy requires an annotation for the empty list, since the context would only be available in the following statement:

def foo(arg: List[int]) -> None:
    print('Items:', ', '.join(arg))

a = []  # Error: Need type annotation for "a"

Working around the issue is easy by adding a type annotation:

a: List[int] = []  # OK

Declaring multiple variable types at a time

You can declare more than a single variable at a time, but only with a type comment. In order to nicely work with multiple assignment, you must give each variable a type separately:

i, found = 0, False  # type: int, bool

You can optionally use parentheses around the types, assignment targets and assigned expression:

i, found = 0, False  # type: (int, bool)      # OK
(i, found) = 0, False  # type: int, bool      # OK
i, found = (0, False)  # type: int, bool      # OK
(i, found) = (0, False)  # type: (int, bool)  # OK

Starred expressions

In most cases, mypy can infer the type of starred expressions from the right-hand side of an assignment, but not always:

a, *bs = 1, 2, 3   # OK
p, q, *rs = 1, 2   # Error: Type of rs cannot be inferred

On first line, the type of bs is inferred to be List[int]. However, on the second line, mypy cannot infer the type of rs, because there is no right-hand side value for rs to infer the type from. In cases like these, the starred expression needs to be annotated with a starred type:

p, q, *rs = 1, 2  # type: int, int, List[int]

Here, the type of rs is set to List[int].

Silencing type errors

You might want to disable type checking on specific lines, or within specific files in your codebase. To do that, you can use a # type: ignore comment.

For example, say that the web framework that you use now takes an integer argument to run(), which starts it on localhost on that port. Like so:

# Starting app on http://localhost:8000

However, the type stubs that the package uses is not up-to-date, and it still expects only str types for run(). This would give you the following error:

error: Argument 1 to "run" of "A" has incompatible type "int"; expected "str"

If you cannot directly fix the type stubs yourself, you can temporarily disable type checking on that line, by adding a # type: ignore:

# Starting app on http://localhost:8000
app.run(8000)  # type: ignore

This will suppress any mypy errors that would have raised on that specific line.

You should probably add some more information on the # type: ignore comment, to explain why the ignore was added in the first place. This could be a link to an issue on the repository responsible for the type stubs, or it could be a short explanation of the bug. To do that, use this format:

# Starting app on http://localhost:8000
app.run(8000)  # type: ignore  # `run()` now accepts an `int`, as a port

If your error displays an error code, like so:

error: "str" has no attribute "trim"  [attr-defined]

It is possible to add a specific error-code in your ignore comment, like # type: ignore[attr-defined], to clarify what’s being silenced. You can find more information about error codes here: Silencing errors based on error codes.

Similarly, you can also ignore all mypy checks in a file, by adding a # type: ignore on the top of the file:

# type: ignore
# This is a test file, skipping type checking in it.
import unittest