ignored regions constraints when normalizing user-provided types

[aliemjay]

This compiles after full NLL (>= 1.63) but it shouldn't:

trait Trait {
    type Ty;
}

impl Trait for &'static () {
    type Ty = ();
}

fn extend<'a>() {
    None::<<&'a () as Trait>::Ty>;
}

HIR typeck unconditionally normalizes user-provided types modulo regions, so it's not NLL's fault this time :)

@rustbot label regression-from-stable-to-stable T-compiler T-types C-bug

[apiraino]

bisection results seems to point to #95565 (quite big pack of changes) cc @jackh726

searched nightlies: from nightly-2022-05-01 to nightly-2022-09-05
regressed nightly: nightly-2022-06-08
searched commits: from 50b0025 to 5435ed6
regressed commit: bb55bd4

bisected with cargo-bisect-rustc v0.6.0

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc ./script.sh --start 2022-05-01 --preserve 

[jackh726]

Going to nominate this for types team. This seems pretty rough, as it's a pretty simple repro. It would be nice to have someone look at this. Not sure if its implied bounds related or something else.

[lcnr]

@rustbot claim

issue with user type annotations, going to take a look and try to fix this

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-high

[jackh726]

This is moving, unnominating.