I'd like to point out that this is still a bug in pylint, which is quite annoying when you use assert_never relatively often. The patch above fixes the problem, so it would be nice to have it in the code.

In addition, handling self._never_returning_function after the first if seems wrong: If one explicitly specifies that a function never returns, this should override any inference by pylint, so this test should come first. At the current state, this configuration option is quite useless, one can't e.g. specify that assert_never never returns.

Do you want to open a PR with the patch @svenpanne ?

If one explicitly specifies that a function never returns, this should override any inference by pylint

This is not the case right now, see #4813. Changing that will be quite a journey

At the current state, this configuration option is quite useless, one can't e.g. specify that assert_never never returns.

What configuration options are we talking about ?

Do you want to open a PR with the patch @svenpanne ?

It's basically the patch above, let's see when I find some time...

If one explicitly specifies that a function never returns, this should override any inference by pylint

This is not the case right now, see #4813. Changing that will be quite a journey

Actually pylint already figures out enough, verified by print()ing in the above method. It's just that pylint doesn't handle the Never type introduced in Python 3.11 last year, see e.g. https://typing.readthedocs.io/en/latest/source/unreachable.html. In addition to checking for NoReturn, it needs to check fo Never, too. This would already solve the problem for assert_never.

At the current state, this configuration option is quite useless, one can't e.g. specify that assert_never never returns.

What configuration options are we talking about ?

https://pylint.readthedocs.io/en/latest/user_guide/configuration/all-options.html#never-returning-functions My point is: If the user explicitly configures that some functions never return, this should override any logic in pylint, so the order of tests in the function above should be reversed. If this was the case, adding typing.assert_never to that configuration option would work even without the Never change above, but currently it doesn't.

So in effect, there are 2 bugs in the method.

Repro for the second issue @svenpanne describes:

bug7565.py:

# pylint: disable=missing-function-docstring,missing-module-docstring
from __future__ import annotations

from typing_extensions import (
        assert_never,
        Union,
        )

def some_func(arg: Union[int, str]) ->  int:

    if isinstance(arg, int):
        return 1
    if isinstance(arg, str):
        return 2

    assert_never(arg)

pylint --rcfile=/dev/null --never-returning-functions=typing_extensions.assert_never bug7565.py:

************* Module bug7565
bug7565.py:9:0: R1710: Either all return statements in a function should return an expression, or none of them should. (inconsistent-return-statements)

-----------------------------------
Your code has been rated at 8.75/10

Any update on this?

OK, I've uploaded a PR for this, fixing both problems: Never was unknown and explicitly stating non-returning functions had the wrong priority compared to the incomplete inference of astroid/pylint.

With this change, things work out-of-the-box for Python 3.11 and 3.12 with both typing and typing_extensions. Older Python versions have to use the latter, which exposes its own implementation of assert_never then. Somehow astroid can't infer anything about that then, but this is a different problem. For a home-grown & visible version, things work, too, even for such old Pythons.