Trait aliases should not bring supertraits into scope

[eggyal]

In-scope trait aliases bring into scope (methods of) the aliased trait and all of its supertraits. I can understand the rationale for bringing (methods of) the aliased trait into scope (namely to fix #56485), but also bringing into scope (methods of) that trait's supertraits is both inconsistent (compared with bringing the aliased trait itself into scope) and surprising:

#![feature(trait_alias)]

use std::fmt;

trait Foo: fmt::Debug {}

// uncommenting the following line brings Debug into scope, which
// in turn causes the (unrelated) Display impl below to become ambiguous

// trait Bar = Foo;

#[derive(Debug)]
struct Qux(bool);

impl fmt::Display for Qux {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        self.0.fmt(f)
    }
}

Playground.

(For the avoidance of doubt, only the methods are brought into scope—not the trait itself: thus, in the above example with the Bar alias uncommented, one cannot use an unqualified Debug).

cc #41517

@rustbot label A-resolve A-traits C-bug F-trait_alias

[eggyal]

From cursory inspection, the following (originally from #59166, cc @seanmonstar) looks suspect:

rust/compiler/rustc_hir_typeck/src/method/probe.rs

Lines 954 to 971 in e1eaa2d

	// For trait aliases, assume all supertraits are relevant.
	let bounds = iter::once(ty::Binder::dummy(trait_ref));
	self.elaborate_bounds(bounds, |this, new_trait_ref, item| {
	let new_trait_ref = this.erase_late_bound_regions(new_trait_ref);
	let (xform_self_ty, xform_ret_ty) =
	this.xform_self_ty(&item, new_trait_ref.self_ty(), new_trait_ref.substs);
	this.push_candidate(
	Candidate {
	xform_self_ty,
	xform_ret_ty,
	item,
	import_ids: import_ids.clone(),
	kind: TraitCandidate(new_trait_ref),
	},
	false,
	);
	});

[eggyal]

@rustbot claim