into_iterator not resolved or completed on Vec<{unknown}>

[lnicola]

It feels like we could offer into_iter here, since the impl is not constrained in any way:

fn main() {
    let mut x = Vec::new();
    x.into_iter();
}

// impl<T, A: Allocator> IntoIterator for Vec<T, A>

[iDawer]

Reproduced (update: minicore: from => minicore: sized )

#[test]
fn issue_10297() {
    check(
        r#"
//- minicore: iterator, sized
struct Vec<T>;
impl<T> IntoIterator for Vec<T> {
    type Item = ();
    type IntoIter = ();
    fn into_iter(self);
}
fn main() {
    let x: Vec<_>;
    x.$0;
}
"#,
        expect![[r#"
                me into_iter() (as IntoIterator) fn(self) -> <Self as IntoIterator>::IntoIter
            "#]],
    )
}

Without minicore: sized the test is fine

[iDawer]

Some logs:

iterate_trait_method_candidates: trait.name=IntoIterator
[crates/hir_ty/src/method_resolution.rs:696] &goal = Canonical {
    value: InEnvironment {
        environment: Env([]),
        goal: Implemented(AdtId(StructId(StructId(0)))<[?0 := {error}]>: TraitId(2)),
    },
    binders: [],
}
chalk program:
#[upstream]
#[non_enumerable]
#[object_safe]
#[lang(sized)]
trait Sized {}
#[upstream]
#[non_enumerable]
#[object_safe]
trait Iterator {
  type Item: Sized;
}
#[upstream]
#[non_enumerable]
#[object_safe]
trait IntoIterator {
  type Item: Sized;
  type IntoIter: Iterator + Iterator<Item=<Self as IntoIterator>::Item> + Sized;
}
impl<_1_0> IntoIterator for Vec<_1_0>
where
  _1_0: Sized
{
  type Item = ();
  type IntoIter = ();
}
struct Vec<_1_0>
where
  _1_0: Sized
{}

[crates/hir_ty/src/method_resolution.rs:698] db.trait_solve(krate, goal.cast(&Interner)) = None

As far I can interpret the goal is Vec<{error}>: IntoIterator. This drills down to solving {error}: Sized which then results in No possible solution. I'm not sure where is the problem: in passing error types to Chalk or Chalk failing to report ambiguity?

[flodiebold]

I think that's this FIXME:
https://github.com/rust-analyzer/rust-analyzer/blob/009e6ceb1ddcd27a9ced3bcb7d0ef823379185a1/crates/hir/src/lib.rs#L2604
It should probably just do this:
https://github.com/rust-analyzer/rust-analyzer/blob/009e6ceb1ddcd27a9ced3bcb7d0ef823379185a1/crates/hir/src/lib.rs#L2658

The error type is legitimate here, it comes from fallback for the uninferred variable. But we need to handle that when using the type for completion etc. (In theory we'd want something like the 'type before fallback' (but with canonicalized variables).)

[lnicola]

Yeah, that works ^.