ASYNC102 await-in-finally-or-cancelled false alarms on nurseries

python-trio / flake8-async

Highly opinionated linter for Trio code

https://flake8-async.readthedocs.io

MIT License

19 stars 2 forks source link

ASYNC102 await-in-finally-or-cancelled false alarms on nurseries #283

Open jakkdl opened 2 months ago

jakkdl commented 2 months ago

https://github.com/python-trio/flake8-async/blob/225f15a98db4bb8ce4600127fd0fe90bc5c9c04b/tests/eval_files/async102.py#L117-L125

TODO:

[ ] Special-case async with open_nursery(...), and async with anyio.create_task_group(...) as not being checkpoints. (also affects other rules)
[x] handle [nursery_name].cancel_scope.[deadline/shield]

jakkdl commented 1 month ago

The other rules that could be relevant are:

ASYNC100 cancel-scope-no-checkpoint
- A cancel scope with only a nursery can never be cancelled, so this should ignore open_nursery/create_task_group
ASYNC120 await-in-except
- Same as 102, and also already fixed by #289
ASYNC91x
- nurseries can checkpoint, so these should not ignore nurseries, and do not need to be changed.

async def foo(): # should not give ASYNC910
  async with trio.open_nursery() as nursery():
    nursery.start_soon(bar)

tl;dr, ASYNC100 also needs changing

jakkdl commented 1 month ago

Hrm, neither of these statements are cancellation sources

my_nursery = await trio.open_nursery()
async with my_nursery:
    ...

So it's not just async with trio.open_nursery() that needs to be checked against. Both of those cases are definitely less common, but far from unheard of. The former is an easy fix, the latter is a bit more tricky but we're already type-tracking nurseries for other reasons so shouldn't be too bad.