must_not_suspend lint regression with -Zdrop-tracking

[eholk]

The test case src/test/ui/lint/must_not_suspend/ref.rs fails when -Zdrop-tracking is enabled.

We are not recognizing the Umm type as held across a borrow. Here's the revelant code from the test:

rust/src/test/ui/lint/must_not_suspend/ref.rs

Lines 17 to 25 in 7f997f5

	async fn uhoh(&mut self) {
	let guard = &mut self.u; //~ ERROR `Umm` held across
	other().await;
	*guard = Umm {
	i: 2
	}
	}

In a way, that's correct. We aren't holding Umm across the borrow, we are holding &mut Umm. The question is whether this should trigger the lint? If not, we need some way to annotate &mut Umm as also not being able to be held across the borrow.

It looks like in check_must_not_suspend_ty, we already recurse through a few types, so maybe we should look through references too?

[eholk]

@guswynn - What do you think is the right behavior here?

[guswynn]

@eholk this was always an issue with must_not_suspend: how deep do we recurse into types to check. I think the conclusion was that the cases covered by the existing tests were adequate? I think its just T, `Box, and references?

[eholk]

I'm spending some time looking at this one again this afternoon to see if it's still an issue after @jyn514's changes. It seems like yes, although we aren't getting duplicate warnings now, we're just getting now warnings. I think seeing through references is probably the right option, but I'm getting kind of lost between the different PRs I have open so I'm going to rebase some things and figure out where we're at.

[jyn514]

Sounds good :) @eholk let me know if there's any way I can help, I've been repeatedly running into places at work that would be fixed by drop tracking haha

[eholk]

This should be fixed as of #97962!