[legacy] [incorrect approach] pretty-print inherent projections "correctly"

[fmease]

Given the these anonymous constants

• Parent::<1>::Mapping::<{ 2 * 5 }>::{constant#0}, substs: [1, {2 * 5}] (1 parent subst, 1 own subst)
• <impl at dump_mir_2.rs:5:1: 5:37>::Mapping::{constant#0}, substs: [] (0 parent substs, 0 own substs)

the branch added in this PR & labeled // Inherent projections is not taken (which is good) since

if let DefPathData::TypeNs(_) = def_path_data // true && let DefKind::Impl { of_trait: false } = self.tcx().def_kind(parent_def_id) // true && let Some(self_ty) = substs.get(0).and_then(|subst| subst.as_type()) // FALSE (!)

However, when the first parent subst is a type, we incorrectly assume it to be the self type and
the branch is taken which is bad.

For example:

#![feature(inherent_associated_types, generic_const_exprs)] struct Parent<T, const A: usize>(T); impl<T, const A: usize> Parent<T, A> { type Mapping<const I: usize> = Store<{ A + I }> where [(); A + I]: ; } struct Store<const N: usize>; impl<const N: usize> Store<N> { fn reify() -> usize { N } } fn main() { let _ = Parent::<(), 3>::Mapping::<{ 2 * 5 }>::reify(); }

Here (via -Zdump-mir=all) we incorrectly print the anonymous constant corresp. to A + I (I guess) inside of Store<{…}> assigned to Mapping as <()>::Mapping::<3, { 2 * 5 }>::{constant#0} instead of Parent::<(), 3>::Mapping::<{ 2 * 5 }>::{constant#0}.

Conclusion: Don't use print_def_path for inherent projections!