Rollup merge of #140214 - compiler-errors:remove-hack, r=lcnr

Remove comment about handling non-global where bounds with corresponding projection

This comment is no longer relevant since we only assemble rigid projections if no param-env candidates hold.

Also remove a stray comment from the old solver.

r? lcnr

Author: Zalathar

compiler/rustc_next_trait_solver/src/solve/trait_goals.rs

@@ -1308,18 +1308,6 @@ where
                     return true;
                 }
 
-                // We don't consider a trait-bound global if it has a projection bound.
-                //
-                // See ui/traits/next-solver/normalization-shadowing/global-trait-with-project.rs
-                // for an example where this is necessary.
-                for p in goal.param_env.caller_bounds().iter() {
-                    if let ty::ClauseKind::Projection(proj) = p.kind().skip_binder() {
-                        if proj.projection_term.trait_ref(self.cx()) == trait_pred.trait_ref {
-                            return true;
-                        }
-                    }
-                }
-
                 false
             }
             _ => false,

compiler/rustc_trait_selection/src/traits/select/candidate_assembly.rs

@@ -240,8 +240,6 @@ impl<'cx, 'tcx> SelectionContext<'cx, 'tcx> {
             if !drcx.args_may_unify(obligation_args, bound_trait_ref.skip_binder().args) {
                 continue;
             }
-            // FIXME(oli-obk): it is suspicious that we are dropping the constness and
-            // polarity here.
             let wc = self.where_clause_may_apply(stack, bound_trait_ref)?;
             if wc.may_apply() {
                 candidates.vec.push(ParamCandidate(bound));