Spurious unused type alias warning (regression)

[jseyfried]

This example produced a spurious unused type alias warning:

struct S<T>(T);
type Alias<T> = S<T>;

trait Tr<T> { fn f(&self) {} }
impl<T> Tr<T> for S<T> {}

impl<T> S<T> where Alias<T>: Tr<T> {
    fn g(&self) { self.f() }
}

fn main() {
    S(0).g();
}

[Mark-Simulacrum]

Bisected regression to aedb49c -- rollup #39199; not really sure what PR in there this was caused by. Nominating for compiler discussion since regression without priority, though this is in stable now so not overly important I guess.

[eddyb]

@Mark-Simulacrum What's the "before" on that bisection? I've looked through the entire rollup and I can't find anything that would change the behavior there.

[Mark-Simulacrum]

Never mind -- sorry for the runaround, my bisect was broken, missing libstdc++6; this does not bisect to any point since 2017-01-13.

[nikomatsakis]

@jseyfried says they had this working on a nightly from October without warning.

[nikomatsakis]

triage: P-high

[brson]

@jseyfried do you mind self-assigning this P-high regression and taking a look?

[jseyfried]

@brson sure.

[sanxiyn]

#38051 implementing unused type alias landed in December, so of course October nightly would work.

[brson]

cc @rust-lang/compiler another P-high bug

[Mark-Simulacrum]

I can't find an issue, but I seem to recall this being a known problem of not checking impl signatures (everything outside the body) for "use"

[jseyfried]

This was introduced between nightly-2016-12-16 (8f02c42) and nightly-2016-12-19 (71c06a5).
I suspect #38051.

[nikomatsakis]

Heh, sounds like a likely candidate!

[nikomatsakis]

@sanxiyn think you'll have time to investigate @jseyfried's hypothesis?

[Mark-Simulacrum]

Arguably this is the same/similar bug as #18290 (this is the bug I was looking for earlier).

[Mark-Simulacrum]

Closing as a duplicate of #18290.