False positive of lint `missing_debug_implementations` if `Debug` is implemented for a free alias type that expands to the relevant type

[fmease]

#![feature(lazy_type_alias)]
#![deny(missing_debug_implementations)]

pub struct Local; //~ ERROR type does not implement `Debug`; consider adding `#[derive(Debug)]` or a manu

pub type Alias = Local;

impl std::fmt::Debug for Alias {
    fn fmt(&self, _: &mut std::fmt::Formatter<'_>) -> std::fmt::Result { Ok(()) }
}

For some reason missing_debug_implementations doesn't actually query the trait solver (!) (if the fast path doesn't apply), it literally does

        let has_impl = cx
            .tcx
            .non_blanket_impls_for_ty(
                debug,
                cx.tcx.type_of(item.owner_id).instantiate_identity().skip_norm_wip(),
            )
            .next()
            .is_some();

but I don't think that function should be used for "correctness" (well, it's a lint), ideally it would only be used as a fast path (barring more specific use cases).

For comparison, lint missing_copy_implementations calls tcx.type_is_copy_modulo_regions() which has a fast path but then queries the trait solver.

---

A temporarily solution would be to call expand_free_alias_tys inside non_blanket_impls_for_ty and friends? I haven't thought about correctness yet, so not sure.

I don't think we should call expand_free_alias_tys in fast_reject::simplify_type; that ideal feels very iffy and it would probably trash perf.

[fmease]

Why does that have its own issue instead of being part of parent issue #157758?

Well, it's because this way it's easier to track the potential blockers for LTA_unchecked's stabilization:
Stable code that currently uses eager type aliases in this position would regress if LTA_unchecked became the default.