How to tell mypy that I am explicitly testing for an incorrect type?

Consider the following toy example:

import pytest


def add(a: float) -> float:
    if not isinstance(a, float):
        raise ValueError("a must be of type float")
    return 10 + a


def test_add_wrong_type() -> None:
    with pytest.raises(ValueError) as err:
        add("foo")  # mypy is complaining here

    assert str(err.value) == "a must be of type float"

mypy is complaining as follows:
Argument 1 to "add" has incompatible type "str"; expected "float" [arg-type]

Well, mypy is correct. However, in that case I put in an incorrect type on purpose. How can I tell mypy to ignore this line`?

Put it differently, what is a pythonic way to test for an incorrect input type?

The Python typing community has at least 2 ways of unit-testing type annotations. Note that # type: ignore and typing.cast are not testing them - they're ignoring them and overriding them, respectively.

Depending on your purpose, you'd want to go for one of the following:

• stubtest, which ships with mypy. This is extensively used by typeshed, but is more suited to check that any given type annotations in .pyi files match with the runtime implementation.
• pytest-mypy-plugins, which sounds like what you should go for in your example. Taking an example from README.md, you'd probably want to write something like this:

  # test_add.yml
  - case: test_add_wrong_type
    main: |
      from my_module import add
  
      add("")  # E: Argument 1 to "add" has incompatible type "str"; expected "float" [arg-type]    
  

  (I haven't tested this - their documentation doesn't mention inline-testing errors. I took the # E: ... syntax from the examples in mypy/test-data/unit).

You can also make mypy a runtime dependency for your pytest test suite, in which case you probably don't need pytest-mypy-plugins - their test suite runs on their own non-exposed pytest hooks, which you can import directly in your conftest.py. Just write out examples like their check-*.test files.

Use typing.cast to assert that "foo" is a float.

from typing import cast


def test_add_wrong_type() -> None:
    with pytest.raises(ValueError) as err:
        add(cast(float, "foo"))

    assert str(err.value) == "a must be of type float"

cast doesn't do type conversions; at runtime, it simply returns its second argument. Its purpose is to provide a standard way to assert a static type for a given value, independent of whatever type checker you might be using.