I've noticed a few places where await trio.sleep(0) is added in places that don't actually need a checkpoint, and so I've thought up a two-part plan to mitigate cargo-cult copy-paste coding.

• Instead of await trio.sleep(0), suggest await trio.lowlevel.checkpoint(). It's an identical runtime effect, but the name is much more suggestive of what's actually going on: "we need a checkpoint here for some low-level reason" ✅
• Refactoring the increasingly-large plugin file as suggested below.
• In each async function, for each await trio.lowlevel.checkpoint(), check if there would be no 107/108 warnings emitted if that checkpoint was removed (in addition to any already-suggested-removals, of course).

I initially marked this as an "idea" issue rather than for implementation because while I'm fairly confident (1) is a good plan, (2) seems a little harder to implement. It's also triggering my "this feels pretty tedious" detector, but building a libCST-based autofixer also seems like a little too much duplication of effort, right? Even if we could autofix 105, 107, 108, 112, and 113; and hook it in to shed --refactor... tempting but probably not worth it 😕

Thoughts?