If future returns Pending, it must store waker

What it does

If future returns Pending, it should use the provided waker to notify executor that it can be polled again.
If the waker is not stored, the future will be unable to notify it and therefore it will be never polled again. Such a future is probably incorrect.

What is verified

(Probably imprecise, so just a proposal.)
For each MIR block B0 whose terminator is return clippy should check that there is least one maybe-predecessor statement (i.e. defined in a block B1, and B0 is reachable from B1 in CFG) which somehow uses cx.

Categories (optional)

• Kind: correctness

Drawbacks

False-positive here is a future that sometimes diverges and sometimes terminates (because if a future always diverges that programmer should use futures_util::future::pending() instead). I think such features are rare but don't have any estimations or experiments.

Example

impl Future for MyFut {
    type Output = ...;
    fn poll(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Self::Output> {
        match self.state() {
            // ...
            State::Foo => {
                  // oops, we've forgotten to store waker, future will be never polled again
                  Poll::Pending
             }
        }
   }
}

Start by reading the proposed MIR check for return terminators, maybe-predecessor statements, CFG reachability, and uses of cx in Future::poll. Determine how Clippy represents these blocks and how a lint could distinguish stored wakers from other uses. Done means the lint reliably identifies the shown missing-waker case without unacceptable false positives.