Handle ReErased in responses in new solver

compiler/rustc_trait_selection/src/solve/canonicalize.rs

@@ -224,12 +224,20 @@ impl<'tcx> TypeFolder<TyCtxt<'tcx>> for Canonicalizer<'_, 'tcx> {
         let kind = match *r {
             ty::ReLateBound(..) => return r,
 
-            ty::ReStatic => match self.canonicalize_mode {
+            // We may encounter `ReStatic` in item signatures or the hidden type
+            // of an opaque. `ReErased` should only be encountered in the hidden
+            // type of an opaque for regions that are ignored for the purposes of
+            // captures.
+            //
+            // FIXME: We should investigate the perf implications of not uniquifying
+            // `ReErased`. We may be able to short-circuit registering region
+            // obligations if we encounter a `ReErased` on one side, for example.
+            ty::ReStatic | ty::ReErased => match self.canonicalize_mode {
                 CanonicalizeMode::Input => CanonicalVarKind::Region(ty::UniverseIndex::ROOT),
                 CanonicalizeMode::Response { .. } => return r,
             },
 
-            ty::ReErased | ty::ReFree(_) | ty::ReEarlyBound(_) => match self.canonicalize_mode {
+            ty::ReFree(_) | ty::ReEarlyBound(_) => match self.canonicalize_mode {
                 CanonicalizeMode::Input => CanonicalVarKind::Region(ty::UniverseIndex::ROOT),
                 CanonicalizeMode::Response { .. } => bug!("unexpected region in response: {r:?}"),
             },

tests/ui/impl-trait/erased-regions-in-hidden-ty.rs

@@ -0,0 +1,15 @@
+// revisions: current next
+//[next] compile-flags: -Ztrait-solver=next
+// check-pass
+
+// Make sure that the compiler can handle `ReErased` in the hidden type of an opaque.
+
+fn foo<'a: 'a>(x: &'a Vec<i32>) -> impl Fn() + 'static {
+    || ()
+}
+
+fn bar() -> impl Fn() + 'static {
+    foo(&vec![])
+}
+
+fn main() {}