[F821] Improve error message for anext/aiter

[JP-Ellis]

Summary

The lint F821 has a false positive with anext.

As pointed out by @MichaReiser, this ultimately is due to having either project.requires-python or tools.ruff.target-version set to a too lower version.

In this instances, it was discovered that Ruff was targeting Python 3.8 (from an erroneous config).

Example

async def foo(iter: AsyncGenerator[str, None]):
    assert await anext(iter)

Desired Behaviour

Where a symbol is introduced in a more recent version of Python, Ruff should:

1. Still raise F821 as it is strictly correct; but also,
2. Inform the use Ruff is assuming Python 3.x, and the symbol in question was only introduced in Python 3.y, and nudge the user to set reures-python or target-version.

Environment

❯ ruff --version
ruff 0.6.4

❯ python --version
Python 3.12.5

[evanarlian]

False positive on aiter as well.

[MichaReiser]

Have you specified requires-python in your pyproject.toml or a target-version in your ruff.toml? I'm asking because anext is only available in Python 3.10 and newer (and the default is Python 3.8)

[JP-Ellis]

Good catch. I have a monorepo and the root project has

[project]
# ...
requires-python = "~=3.12.0"

[tool.ruff]
target-version = "py312"

and one of the sub-packages has

[project]
# ...
requires-python = "~=3.12.0"

[tool.ruff]
target-version = "py38"

I can confirm that changing from py38 to py312 fixes this.

In cases like there where there is a new symbol introduced in a more recent version of Python, should ruff instead suggest/direct the end-user to update the settings?

[MichaReiser]

In cases like there where there is a new symbol introduced in a more recent version of Python, should ruff instead suggest/direct the end-user to update the settings?

That's a neat idea. I think the error message could possibly be improved by saying that the symbol is only defined in Python XX or newer.

[JP-Ellis]

I have updated the original issue to incorporate the suggestion that Ruff provide a better suggestion in this case.

[MichaReiser]

Thanks @AlexWaygood !