Re: Assert failure of the cross-check for nullingrels

[Tom Lane]

... BTW, something I'd considered in an earlier attempt at fixing this
was to change clause_is_computable_at's API to pass the clause's
RestrictInfo not just the clause_relids, along the lines of

@@ -541,9 +547,10 @@ extract_actual_join_clauses(List *restrictinfo_list,
  */
 bool
 clause_is_computable_at(PlannerInfo *root,
-                        Relids clause_relids,
+                        RestrictInfo *rinfo,
                         Relids eval_relids)
 {
+    Relids        clause_relids = rinfo->clause_relids;
     ListCell   *lc;
 
     /* Nothing to do if no outer joins have been performed yet. */

with corresponding simplifications at the call sites.  That was with
a view to examining has_clone/is_clone inside this function.  My
current proposal doesn't require that, but I'm somewhat tempted
to make this API change anyway for future-proofing purposes.
Thoughts?

                        regards, tom lane