User guide

Using type checker functions

Two functions are provided, potentially for use with the assert statement:

These can be used to implement fine grained type checking for select functions. If the function is called with incompatible types, or check_return_type() is used and the return value does not match the return type annotation, then a TypeError is raised.

For example:

from typeguard import check_argument_types, check_return_type

def some_function(a: int, b: float, c: str, *args: str) -> bool:
    assert check_argument_types()
    ...
    assert check_return_type(retval)
    return retval

When combined with the assert statement, these checks are automatically removed from the code by the compiler when Python is executed in optimized mode (by passing the -O switch to the interpreter, or by setting the PYTHONOPTIMIZE environment variable to 1 (or higher).

Note

This method is not reliable when used in nested functions (i.e. functions defined inside other functions). This is because this operating mode relies on finding the correct function object using the garbage collector, and when a nested function is running, its function object may no longer be around anymore, as it is only bound to the closure of the enclosing function. For this reason, it is recommended to use @typechecked instead for nested functions.

Using the decorator

The simplest way to type checking of both argument values and the return value for a single function is to use the @typechecked decorator:

from typeguard import typechecked

@typechecked
def some_function(a: int, b: float, c: str, *args: str) -> bool:
    ...
    return retval

@typechecked
class SomeClass:
    # All type annotated methods (including static and class methods and properties)
    # are type checked.
    # Does not apply to inner classes!
    def method(x: int) -> int:
        ...

The decorator works just like the two previously mentioned checker functions except that it has no issues with nested functions. The drawback, however, is that it adds one stack frame per wrapped function which may make debugging harder.

When a generator function is wrapped with @typechecked, the yields, sends and the return value are also type checked against the Generator annotation. The same applies to the yields and sends of an async generator (annotated with AsyncGenerator).

Note

The decorator also respects the optimized mode setting so it does nothing when the interpreter is running in optimized mode.

Using the profiler hook

Deprecated since version 2.6: Use the import hook instead. The profiler hook will be removed in v3.0.

This type checking approach requires no code changes, but does come with a number of drawbacks. It relies on setting a profiler hook in the interpreter which gets called every time a new Python stack frame is entered or exited.

The easiest way to use this approach is to use a TypeChecker as a context manager:

from warnings import filterwarnings

from typeguard import TypeChecker, TypeWarning

# Display all TypeWarnings, not just the first one
filterwarnings('always', category=TypeWarning)

# Run your entire application inside this context block
with TypeChecker(['mypackage', 'otherpackage']):
    mypackage.run_app()

Alternatively, manually start (and stop) the checker:

checker = TypeChecker(['mypackage', 'otherpackage'])
checker.start()
mypackage.start_app()

The profiler hook approach has the following drawbacks:

Return values of None are not type checked, as they cannot be distinguished from exceptions being raised
The hook relies on finding the target function using the garbage collector which may make it miss some type violations, especially with nested functions
Generator yield types are checked, send types are not
Generator yields cannot be distinguished from returns
Async generators are not type checked at all

Hint

Some other things you can do with TypeChecker:

Display all warnings from the start with python -W always::typeguard.TypeWarning
Redirect them to logging using logging.captureWarnings()
Record warnings in your pytest test suite and fail test(s) if you get any (see the pytest documentation about that)

Using the import hook

The import hook, when active, automatically decorates all type annotated functions with @typechecked. This allows for a noninvasive method of run time type checking. This method does not modify the source code on disk, but instead modifies its AST (Abstract Syntax Tree) when the module is loaded.

Using the import hook is as straightforward as installing it before you import any modules you wish to be type checked. Give it the name of your top level package (or a list of package names):

from typeguard.importhook import install_import_hook

install_import_hook('myapp')
from myapp import some_module  # import only AFTER installing the hook, or it won't take effect

If you wish, you can uninstall the import hook:

manager = install_import_hook('myapp')
from myapp import some_module
manager.uninstall()

or using the context manager approach:

with install_import_hook('myapp'):
    from myapp import some_module

You can also customize the logic used to select which modules to instrument:

from typeguard.importhook import TypeguardFinder, install_import_hook

class CustomFinder(TypeguardFinder):
    def should_instrument(self, module_name: str):
        # disregard the module names list and instrument all loaded modules
        return True

install_import_hook('', cls=CustomFinder)

To exclude specific functions or classes from run time type checking, use the @typeguard_ignore decorator:

from typeguard import typeguard_ignore

@typeguard_ignore
def f(x: int) -> int:
    return str(x)

Unlike no_type_check(), this decorator has no effect on static type checking.

Using the pytest plugin

Typeguard comes with a pytest plugin that installs the import hook (explained in the previous section). To use it, run pytest with the appropriate --typeguard-packages option. For example, if you wanted to instrument the foo.bar and xyz packages for type checking, you can do the following:

pytest --typeguard-packages=foo.bar,xyz

There is currently no support for specifying a customized module finder.

Checking types directly

Typeguard can also be used as a beefed-up version of isinstance() that also supports checking against annotations in the typing module:

from typeguard import check_type

# Raises TypeError if there's a problem
check_type('variablename', [1234], List[int])

Support for mock objects

Typeguard handles the unittest.mock.Mock and unittest.mock.MagicMock classes specially, bypassing any type checks when encountering instances of these classes.

Supported typing.* types

The following types from the typing (and typing_extensions) package have specialized support:

Type	Notes
`AbstractSet`	Contents are typechecked
`Callable`	Argument count is checked but types are not (yet)
`Dict`	Keys and values are typechecked
`List`	Contents are typechecked
`Literal`
`NamedTuple`	Field values are typechecked
`NoReturn`
`Protocol`	Run-time protocols are checked with `isinstance()`, others are ignored
`Set`	Contents are typechecked
`Sequence`	Contents are typechecked
`Tuple`	Contents are typechecked
`Type`
`TypedDict`	Contents are typechecked; On Python 3.8 and earlier, `total` from superclasses is not respected (see #101 for more information); On Python 3.9.0 or `typing_extensions` <= 3.7.4.3, false positives can happen when constructing `TypedDict` classes using old-style syntax (see issue 42059)
`TypeVar`	Constraints, bound types and co/contravariance are supported but custom generic types are not (due to type erasure)
`Union`