Make replace_relid() leave argument unmodified

akorotkov · akorotkov · commit e0477837ce49 · 2023-12-27T03:57:57.000+02:00
There are a lot of situations when we share the same pointer to a Bitmapset structure across different places. In order to evade undesirable side effects replace_relid() function should always return a copy. Reported-by: Richard Guo Discussion: https://postgr.es/m/CAMbWs4_wJthNtYBL%2BSsebpgF-5L2r5zFFk6xYbS0A78GKOTFHw%40mail.gmail.com Reviewed-by: Richard Guo, Andres Freund, Ashutosh Bapat, Andrei Lepikhov
diff --git a/src/backend/optimizer/plan/analyzejoins.c b/src/backend/optimizer/plan/analyzejoins.c
@@ -1522,19 +1522,24 @@ replace_varno_walker(Node *node, ReplaceVarnoContext *ctx)
 
 /*
  * Substitute newId by oldId in relids.
+ *
+ * We must make a copy of the original Bitmapset before making any
+ * modifications, because the same pointer to it might be shared among
+ * different places.
  */
 static Bitmapset *
 replace_relid(Relids relids, int oldId, int newId)
 {
 	if (oldId < 0)
 		return relids;
 
+	/* Delete relid without substitution. */
 	if (newId < 0)
-		/* Delete relid without substitution. */
-		return bms_del_member(relids, oldId);
+		return bms_del_member(bms_copy(relids), oldId);
 
+	/* Substitute newId for oldId. */
 	if (bms_is_member(oldId, relids))
-		return bms_add_member(bms_del_member(relids, oldId), newId);
+		return bms_add_member(bms_del_member(bms_copy(relids), oldId), newId);
 
 	return relids;
 }
diff --git a/src/test/regress/expected/join.out b/src/test/regress/expected/join.out
@@ -6853,6 +6853,21 @@ on true;
          Filter: (t3.id IS NOT NULL)
 (8 rows)
 
+-- Check that SJE replaces join clauses involving the removed rel correctly
+explain (costs off)
+select * from emp1 t1
+   inner join emp1 t2 on t1.id = t2.id
+    left join emp1 t3 on t1.id > 1 and t1.id < 2;
+                  QUERY PLAN                  
+----------------------------------------------
+ Nested Loop Left Join
+   Join Filter: ((t2.id > 1) AND (t2.id < 2))
+   ->  Seq Scan on emp1 t2
+         Filter: (id IS NOT NULL)
+   ->  Materialize
+         ->  Seq Scan on emp1 t3
+(6 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
@@ -2610,6 +2610,12 @@ select * from generate_series(1,10) t1(id) left join
     lateral (select t1.id as t1id, t2.id from emp1 t2 join emp1 t3 on t2.id = t3.id)
 on true;
 
+-- Check that SJE replaces join clauses involving the removed rel correctly
+explain (costs off)
+select * from emp1 t1
+   inner join emp1 t2 on t1.id = t2.id
+    left join emp1 t3 on t1.id > 1 and t1.id < 2;
+
 -- 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.
