ASYNC100 (ruff, upstream) warns when a timeout context manager doesn't contain any checkpoints - and therefore can't be cancelled by the async framework. However, if you're implementing a context manager, then it's entirely reasonable to wrap such a timeout around a yield, and we should treat that as a checkpoint - as in python-trio/flake8-async#228

(if you wrap a timeout around the yield in a generator, see PEP-789 and the ASYNC9xx rules, with my condolences)

from contextlib import asynccontextmanager
import anyio

async def bad_code():
    with anyio.fail_after(10):
        do_something_slow()  # can't cancel non-async work!


@asynccontextmanager
async def good_code():
    with anyio.fail_after(10):
        # There's no await keyword here, but we presume that there
        # will be in the caller we yield to, so this is safe.
        yield

The current version of Ruff will warn on both fail_after() contexts, but only the former is problematic.

See also: #8451