Python Type Hints - How to Use typing.cast()

Cast iron horseshoe

Python’s dynamism means that, although support continues to expand, type hints will never cover every situation. For edge cases we need to use an “escape hatch” to override the type checker.

One such escape hatch is the # type: ignore comment, which disables a type checking error for a given line. I previously covered managing such comments, and making them more specific so they do.

Another, preferable escape hatch we can use is casting. We can cast explicitly with typing.cast(), or implicitly from Any with type hints. With casting we can force the type checker to treat a variable as a given type.

Let’s look at how we can use explicit and implicit casting, and a Mypy feature for managing calls to cast().

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 __future__ import annotations

from typing import cast

x = 1
y = cast(str, x)

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

$ mypy note: Revealed type is "" note: Revealed type is "builtins.str"

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

$ python
Traceback (most recent call last):
  File "/.../", line 7, in <module>
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.

At runtime, cast() really does do nothing - its special behaviour is only in how type checkers interpret it. Python’s source code reveals cast() is a simple no-op function call:

def cast(typ, val):
    """Cast a value to a type.
    This returns the value unchanged.  To the type checker this
    signals that the return value has the designated type, but at
    runtime we intentionally don't check anything (we want this
    to be as fast as possible).
    return val

As cast() is a normal Python function, calling it does carry a very slight runtime performance penalty. This will very rarely be an issue. As usual, you should profile your code before making any assumptions about performance.

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.

(This provides a similar type-cleanliness check to warn_unused_ignores, which I covered previously.)

For example, take this unnecessary cast():

from typing import cast

x = 1
y = cast(int, x)

Running Mypy with the option active, we see this error:

$ mypy --warn-redundant-casts error: Redundant cast to "int"
Found 1 error in 1 file (checked 1 source file)

Activating this option is a great guard for keeping our types clean.


Now don’t go cast()ing lots!


Make your development more pleasant with Boost Your Django DX.

Subscribe via RSS, Twitter, or email:

One summary email a week, no spam, I pinky promise.

Related posts:

Tags: ,