fix_estimate_num_groups_with_self_join_removal.patch

diff --git a/src/backend/optimizer/prep/prepunion.c b/src/backend/optimizer/prep/prepunion.c
index 0c68ec011be..d19908ad60e 100644
--- a/src/backend/optimizer/prep/prepunion.c
+++ b/src/backend/optimizer/prep/prepunion.c
@@ -324,7 +324,7 @@ recurse_set_operations(Node *setOp, PlannerInfo *root,
 		 * anyway; otherwise do statistical estimation.
 		 *
 		 * XXX you don't really want to know about this: we do the estimation
-		 * using the subquery's original targetlist expressions, not the
+		 * using the subroot->parse's original targetlist expressions, not the
 		 * subroot->processed_tlist which might seem more appropriate.  The
 		 * reason is that if the subquery is itself a setop, it may return a
 		 * processed_tlist containing "varno 0" Vars generated by
@@ -332,6 +332,9 @@ recurse_set_operations(Node *setOp, PlannerInfo *root,
 		 * mightily.  We ought to get rid of the "varno 0" hack, but that
 		 * requires a redesign of the parsetree representation of setops, so
 		 * that there can be an RTE corresponding to each setop's output.
+		 * Note, we use this not subquery's targetlist but subroot->parse's
+		 * targetlist, because it was revised by self-join removal.  subquery's
+		 * targetlist might contain the references to the removed relids.
 		 */
 		if (pNumGroups)
 		{
@@ -341,7 +344,7 @@ recurse_set_operations(Node *setOp, PlannerInfo *root,
 				*pNumGroups = subpath->rows;
 			else
 				*pNumGroups = estimate_num_groups(subroot,
-												  get_tlist_exprs(subquery->targetList, false),
+												  get_tlist_exprs(subroot->parse->targetList, false),
 												  subpath->rows,
 												  NULL,
 												  NULL);
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
index 446959e3c5f..892ea5f1702 100644
--- a/src/test/regress/expected/join.out
+++ b/src/test/regress/expected/join.out
@@ -6759,6 +6759,23 @@ WHERE c3.id=c2.id AND c3.id*c2.id=c1.id*c1.id;
          Filter: ((id IS NOT NULL) AND ((id * id) IS NOT NULL))
 (3 rows)
 
+-- Check the usage of a parse tree by the set operations (bug #18170)
+explain (costs off)
+SELECT c1.code FROM emp1 c1 LEFT JOIN emp1 c2 ON c1.id = c2.id
+WHERE c2.id IS NOT NULL
+EXCEPT ALL
+SELECT c3.code FROM emp1 c3;
+                  QUERY PLAN                  
+----------------------------------------------
+ HashSetOp Except All
+   ->  Append
+         ->  Subquery Scan on "*SELECT* 1"
+               ->  Seq Scan on emp1 c2
+                     Filter: (id IS NOT NULL)
+         ->  Subquery Scan on "*SELECT* 2"
+               ->  Seq Scan on emp1 c3
+(7 rows)
+
 -- We can remove the join even if we find the join can't duplicate rows and
 -- the base quals of each side are different.  In the following case we end up
 -- moving quals over to s1 to make it so it can't match any rows.
diff --git a/src/test/regress/sql/join.sql b/src/test/regress/sql/join.sql
index 66dd03dd88b..559c36dc074 100644
--- a/src/test/regress/sql/join.sql
+++ b/src/test/regress/sql/join.sql
@@ -2573,6 +2573,13 @@ explain (costs off)
 SELECT count(*) FROM emp1 c1, emp1 c2, emp1 c3
 WHERE c3.id=c2.id AND c3.id*c2.id=c1.id*c1.id;
 
+-- Check the usage of a parse tree by the set operations (bug #18170)
+explain (costs off)
+SELECT c1.code FROM emp1 c1 LEFT JOIN emp1 c2 ON c1.id = c2.id
+WHERE c2.id IS NOT NULL
+EXCEPT ALL
+SELECT c3.code FROM emp1 c3;
+
 -- We can remove the join even if we find the join can't duplicate rows and
 -- the base quals of each side are different.  In the following case we end up
 -- moving quals over to s1 to make it so it can't match any rows.