tolerate region vars in implied bounds

See #109628.

Author: aliemjay

compiler/rustc_infer/src/infer/outlives/env.rs

@@ -142,7 +142,10 @@ impl<'tcx> OutlivesEnvironmentBuilder<'tcx> {
                         ty::ReStatic | ty::ReEarlyBound(_) | ty::ReFree(_),
                     ) => self.region_relation.add(r_a, r_b),
                     (ty::ReError(_), _) | (_, ty::ReError(_)) => {}
-                    _ => bug!("add_outlives_bounds: unexpected regions"),
+                    // FIXME(#109628): We shouldn't have existential variables in implied bounds.
+                    // Panic here once the linked issue is resolved!
+                    (ty::ReVar(_), _) | (_, ty::ReVar(_)) => {}
+                    _ => bug!("add_outlives_bounds: unexpected regions: ({r_a:?}, {r_b:?})"),
                 },
             }
         }

tests/ui/implied-bounds/ice-unbound-region-vars.rs

@@ -0,0 +1,24 @@
+// Because of #109628, we can have unbounded region vars in implied bounds.
+// Make sure we don't ICE in this case!
+//
+// check-pass
+
+pub trait MapAccess {
+    type Error;
+    fn next_key_seed(&mut self) -> Option<Self::Error>;
+}
+
+struct Access<'a> {
+    _marker: std::marker::PhantomData<&'a ()>,
+}
+
+// implied_bounds(Option<Self::Error>) = ['?1: 'a, ]
+// where '?1 is a fresh region var.
+impl<'a, 'b: 'a> MapAccess for Access<'a> {
+    type Error = ();
+    fn next_key_seed(&mut self) -> Option<Self::Error> {
+        unimplemented!()
+    }
+}
+
+fn main() {}