A static type analyzer for Python code

User guide
Developer guide
Error classes
Typing FAQ
Supported features
Mailing list
File a bug

View the Project on GitHub google/pytype

Hosted on GitHub Pages — Theme by orderedlist


How is pytype different from other type checkers?

pytype has the ability to infer types for unannotated code. For more information, check out our typing FAQ.

Can I find out what pytype thinks the type of my expression is?

Yes, insert reveal_type(expr) as a statement inside your code. This will cause pytype to emit an error that will describe the type of expr.

If you would like to ensure that pytype’s view of a type matches what you expect it to be, use assert_type(expr, expected-type) or assert_type(expr, 'expected-type'). Note that the string version matches on the string pytype uses to display the type, so you might have to tweak your expected type a bit to eliminate false positives (e.g. assert_type(x, 'foo.A') might fail because pytype thinks x is of type, due to fully qualifying imports and resolving aliases, but assert_type(x, foo.A) should work even if foo is an alias for

To simply verify that pytype has inferred some type for an expression, and not fallen back to Any, use assert_type(x) without the second argument.

If you would like to leave the assert_type statement in your code (rather than adding it, running pytype, and removing it), add from pytype_extensions import assert_type to your module.

How do I reference a type from within its definition? (Forward References)

To reference a type from within its definition (e.g. when a method’s return type is an instance of the class to which the method belongs), specify the type as a string. This will be resolved later by pytype. For example:

class Person(object):
  def CreatePerson(name: str) -> 'Person':

I’m dynamically populating a class / module using setattr or by modifying locals() / globals(). Now pytype complains about missing attributes or module members. How do I fix this?

Add _HAS_DYNAMIC_ATTRIBUTES = True to your class or module.

Why didn’t pytype catch that my program (might) pass an invalid argument to a function?

pytype accepts a function call if there’s at least one argument combination that works. For example,

def f(x: float):
  return x
f(random.random() or 'foo')

is not considered an error, because f() works for float. I.e., the str argument isn’t considered. (This will change at some point in the future.) Note that this is different to attribute checking, where e.g.

(random.random() or 'foo').as_integer_ratio()

will indeed result in a type error.

How do I declare that something can be either byte string or unicode?

Use str if it is conceptually a text object and bytes | str otherwise. See the style guide for more information.

I’m trying to use a mixin, but pytype raises errors about it. What should I do?

This happens when a mixin expects the classes it is mixed into to define particular functions. Let’s say we have a LoggerMixin class that expects a name method to be used in the log message:

class LoggerMixin:
  ...  # Other initialization.
  def log(self, msg: str):
    self._log.print(f'{}: {msg}')

When pytype checks LoggerMixin, it will raise an error that LoggerMixin has no method name. The solution is to make the mixin class have all the methods it needs.

One way to do this is to create an abstract base class that defines the expected API for the mixin:

import abc
class LoggerMixinInterface(metaclass=abc.ABCMeta):
  def name(self) -> str:
    raise NotImplementedError

class LoggerMixin(LoggerMixinInterface):
  ...  # Other initialization
  def log(self, msg: str):
    self._log.print(f'{}: {msg}')

class Person(LoggerMixinInterface):
  ...  # Other initialization
  def name(self):
    return self._name

With this setup, pytype won’t complain about, and it’s clear that LoggerMixin should only be mixed into classes that implement name.

Why is pytype taking so long?

If pytype is taking a long time on a file, the easiest workaround is to disable it with a skip-file directive. Otherwise, there are a few things you can try to speed up the analysis:

How do I disable all pytype checks for a particular file?

You can use

# pytype: skip-file

at the start of the file to disable all checking for a particular file. Callers will also see the APIs from this file having the type Any, but the rest of the blaze target is still type-checked.

How do I disable all pytype checks for a particular import?

You can use

from typing import Any
import foo  # type: Any

to disable checking for module foo. Note that pytype will still verify that foo is present among your target’s dependencies. To disable that check as well, replace # type: Any with # type: ignore.

How do I write code that is seen by pytype but ignored at runtime?

You can nest it inside an if typing.TYPE_CHECKING: block. This is occasionally needed to, for instance, conditionally import a module that is only used to provide type annotations.

Note that regardless of whether you use TYPE_CHECKING, if you’re using a build system, you’ll need to list all modules you import as dependencies of your target. That can lead to cycles in your build graph. Typically, that means that, short of rearranging your source tree, you won’t be able to annotate with that specific type. You can typically work around the “inexpressible” type by inserting Any where you would have used it. See the style guide for more information.

How do I silence overzealous pytype errors when adding multiple types to a dict (or list, set, etc.)?

A common pattern is to use a dictionary as a container for values of many types, for example:

    "slot1": Class1,
    "slot2": Class2,

This will often cause pytype to produce errors for any operation that is not valid on all of the types. To fix this, annotate the value type as Any:

MY_REGISTRY: Dict[str, Any] = {

Note that if you modify the dictionary in a different scope from the one in which it is defined, you may need to re-annotate it at the modification site to indicate to pytype that you are intentionally doing something it deems unsafe.

How do I get type information for third-party libraries?

The open-source version of pytype gets type information from the typeshed project. Pytype treats all imports from third-party (that is, pip-installed) libraries that do not have stubs in typeshed as having type Any. Note that pytype does not yet support the PEP 561 conventions for distributing and packaging type information.

Why doesn’t str match against string iterables?

As of early August 2021, Pytype introduced a check that forbids matching str against the following types to prevent a common accidental bug:

NOTE: str continues to match against general iterables of type Any (e.g., Iterable[Any], Sequence[Any], etc.).

If you wish to pass a string s into a function that expects a string iterable:

If you are annotating a function parameter that expects both iterating over a single string and multiple strings, you can use a union (expressed with |) to explicitly allow this. For example,

def f(x: str | Iterable[str]): ...

# Alternatively, if your function expects any kind of Iterable
def g(x: Iterable[Any]): ...

How can I automatically generate type annotations for an existing codebase?

Rather than using generated type annotations, we suggest you embrace an incremental approach of adding type annotations. Don’t let the perfect be the enemy of the good. While fully annotating your code will better realize the full benefits of pytype, pytype’s inferencer is pretty powerful even with few or no type annotations.

When starting out, you can add some type annotations now and others later, so if you feel like adding some, don’t let a feeling of needing to add all stop you from adding whichever few you want. In many cases, you don’t need to annotate everything and will have the most success annotating public code elements and complicated private code elements.

How do I annotate *args and **kwargs?

Varargs (*args) and keyword arguments (**kwargs) should be annotated with the type of each individual argument.


def f(*args: int) -> int:
  return sum(args)

def g(**kwargs: int) -> int:
  return sum(kwargs.values())


from typing import Mapping, Sequence

def f(*args: Sequence[int]) -> int:
  return sum(args)

def g(**kwargs: Mapping[str, int]) -> int:
  return sum(kwargs.values())

Why are signature mismatches in subclasses bad?

A mismatch in the signatures of an overridden method in a superclass and an overriding method in a subclass can cause the following problems:

class A:
  def func(self, x: int) -> int:
    return x

class B(A):
  def func(self) -> int:  # signature-mismatch
    return 0

def f(a: A, x: int) -> int:
  return a.func(x)

a = B()
f(a, 0)

Fails with an error:

TypeError: func() missing 1 required positional argument: 'x'

class A:
  def func(self, x: int, y: int) -> int:
    return 0

class B(A):
  def func(self, y: int, x: int) -> int:  # signature-mismatch
    return x - y

def f(a: A, x: int, y: int) -> None:
  print(a.func(x, y))
  print(a.func(x=x, y=y))
  print(a.func(x, y=y))

a = B()
f(a, 2, 1)


Traceback (most recent call last):
TypeError: func() got multiple values for argument 'y'

What is the nothing type?

In error messages and type stubs generated by pytype, you may occasionally come across a type called nothing. For example:

def f() -> str:
  return []  # [bad-return-type]
             # Expected: str
             # Actually returned: List[nothing]

nothing represents an unknown type that has not yet been filled in. It typically appears in the context of empty containers. It differs from Any in that the union of Any with another type is Any, but the union of nothing with another type is the other type.

Do not use nothing in type annotations; it is an internal detail of pytype’s inference engine.

What does ... mean in a type annotation?

... in a type annotation has several possible meanings: