Fix some issues with improper placement of outer join clauses.

After applying outer-join identity 3 in the forward direction,
it was possible for the planner to mistakenly apply a qual clause
from above the two outer joins at the now-lower join level.
This can give the wrong answer, since a value that would get nulled
by the now-upper join might not yet be null.

To fix, when we perform such a transformation, consider that the
now-lower join hasn't really completed the outer join it's nominally
responsible for and thus its relid set should not include that OJ's
relid (nor should its output Vars have that nullingrel bit set).
Instead we add those bits when the now-upper join is performed.
The existing rules for qual placement then suffice to prevent
higher qual clauses from dropping below the now-upper join.
There are a few complications from needing to consider transitive
closures in case multiple pushdowns have happened, but all in all
it's not a very complex patch.

This is all new logic (from 2489d76c4) so no need to back-patch.
The added test cases all have the same results as in v15.

Tom Lane and Richard Guo

Discussion: https://postgr.es/m/0b819232-4b50-f245-1c7d-c8c61bf41827@postgrespro.ru

src/backend/optimizer/path/costsize.c

@@ -4789,7 +4789,8 @@ compute_semi_anti_join_factors(PlannerInfo *root,
 	norm_sjinfo.ojrelid = 0;
 	norm_sjinfo.commute_above_l = NULL;
 	norm_sjinfo.commute_above_r = NULL;
-	norm_sjinfo.commute_below = NULL;
+	norm_sjinfo.commute_below_l = NULL;
+	norm_sjinfo.commute_below_r = NULL;
 	/* we don't bother trying to make the remaining fields valid */
 	norm_sjinfo.lhs_strict = false;
 	norm_sjinfo.semi_can_btree = false;
@@ -4957,7 +4958,8 @@ approx_tuple_count(PlannerInfo *root, JoinPath *path, List *quals)
 	sjinfo.ojrelid = 0;
 	sjinfo.commute_above_l = NULL;
 	sjinfo.commute_above_r = NULL;
-	sjinfo.commute_below = NULL;
+	sjinfo.commute_below_l = NULL;
+	sjinfo.commute_below_r = NULL;
 	/* we don't bother trying to make the remaining fields valid */
 	sjinfo.lhs_strict = false;
 	sjinfo.semi_can_btree = false;

src/backend/optimizer/path/equivclass.c

@@ -1366,19 +1366,20 @@ generate_base_implied_equalities_broken(PlannerInfo *root,
  * commutative duplicates, i.e. if the algorithm selects "a.x = b.y" but
  * we already have "b.y = a.x", we return the existing clause.
  *
- * If we are considering an outer join, ojrelid is the associated OJ relid,
- * otherwise it's zero.
+ * If we are considering an outer join, sjinfo is the associated OJ info,
+ * otherwise it can be NULL.
  *
  * join_relids should always equal bms_union(outer_relids, inner_rel->relids)
- * plus ojrelid if that's not zero.  We could simplify this function's API by
- * computing it internally, but most callers have the value at hand anyway.
+ * plus whatever add_outer_joins_to_relids() would add.  We could simplify
+ * this function's API by computing it internally, but most callers have the
+ * value at hand anyway.
  */
 List *
 generate_join_implied_equalities(PlannerInfo *root,
 								 Relids join_relids,
 								 Relids outer_relids,
 								 RelOptInfo *inner_rel,
-								 Index ojrelid)
+								 SpecialJoinInfo *sjinfo)
 {
 	List	   *result = NIL;
 	Relids		inner_relids = inner_rel->relids;
@@ -1396,8 +1397,10 @@ generate_join_implied_equalities(PlannerInfo *root,
 		nominal_inner_relids = inner_rel->top_parent_relids;
 		/* ECs will be marked with the parent's relid, not the child's */
 		nominal_join_relids = bms_union(outer_relids, nominal_inner_relids);
-		if (ojrelid != 0)
-			nominal_join_relids = bms_add_member(nominal_join_relids, ojrelid);
+		nominal_join_relids = add_outer_joins_to_relids(root,
+														nominal_join_relids,
+														sjinfo,
+														NULL);
 	}
 	else
 	{
@@ -1418,7 +1421,7 @@ generate_join_implied_equalities(PlannerInfo *root,
 	 * At inner joins, we can be smarter: only consider eclasses mentioning
 	 * both input rels.
 	 */
-	if (ojrelid != 0)
+	if (sjinfo && sjinfo->ojrelid != 0)
 		matching_ecs = get_eclass_indexes_for_relids(root, nominal_join_relids);
 	else
 		matching_ecs = get_common_eclass_indexes(root, nominal_inner_relids,
@@ -1467,7 +1470,7 @@ generate_join_implied_equalities(PlannerInfo *root,
  * generate_join_implied_equalities_for_ecs
  *	  As above, but consider only the listed ECs.
  *
- * For the sole current caller, we can assume ojrelid == 0, that is we are
+ * For the sole current caller, we can assume sjinfo == NULL, that is we are
  * not interested in outer-join filter clauses.  This might need to change
  * in future.
  */

src/backend/optimizer/path/indxpath.c

@@ -3364,7 +3364,7 @@ check_index_predicates(PlannerInfo *root, RelOptInfo *rel)
 																   otherrels),
 														 otherrels,
 														 rel,
-														 0));
+														 NULL));
 
 	/*
 	 * Normally we remove quals that are implied by a partial index's

src/backend/optimizer/path/joinrels.c

@@ -691,6 +691,7 @@ make_join_rel(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2)
 	Relids		joinrelids;
 	SpecialJoinInfo *sjinfo;
 	bool		reversed;
+	List	   *pushed_down_joins = NIL;
 	SpecialJoinInfo sjinfo_data;
 	RelOptInfo *joinrel;
 	List	   *restrictlist;
@@ -710,9 +711,12 @@ make_join_rel(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2)
 		return NULL;
 	}
 
-	/* If we have an outer join, add its RTI to form the canonical relids. */
-	if (sjinfo && sjinfo->ojrelid != 0)
-		joinrelids = bms_add_member(joinrelids, sjinfo->ojrelid);
+	/*
+	 * Add outer join relid(s) to form the canonical relids.  Any added outer
+	 * joins besides sjinfo itself are appended to pushed_down_joins.
+	 */
+	joinrelids = add_outer_joins_to_relids(root, joinrelids, sjinfo,
+										   &pushed_down_joins);
 
 	/* Swap rels if needed to match the join info. */
 	if (reversed)
@@ -740,7 +744,8 @@ make_join_rel(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2)
 		sjinfo->ojrelid = 0;
 		sjinfo->commute_above_l = NULL;
 		sjinfo->commute_above_r = NULL;
-		sjinfo->commute_below = NULL;
+		sjinfo->commute_below_l = NULL;
+		sjinfo->commute_below_r = NULL;
 		/* we don't bother trying to make the remaining fields valid */
 		sjinfo->lhs_strict = false;
 		sjinfo->semi_can_btree = false;
@@ -753,7 +758,8 @@ make_join_rel(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2)
 	 * Find or build the join RelOptInfo, and compute the restrictlist that
 	 * goes with this particular joining.
 	 */
-	joinrel = build_join_rel(root, joinrelids, rel1, rel2, sjinfo,
+	joinrel = build_join_rel(root, joinrelids, rel1, rel2,
+							 sjinfo, pushed_down_joins,
 							 &restrictlist);
 
 	/*
@@ -775,6 +781,108 @@ make_join_rel(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2)
 	return joinrel;
 }
 
+/*
+ * add_outer_joins_to_relids
+ *	  Add relids to input_relids to represent any outer joins that will be
+ *	  calculated at this join.
+ *
+ * input_relids is the union of the relid sets of the two input relations.
+ * Note that we modify this in-place and return it; caller must bms_copy()
+ * it first, if a separate value is desired.
+ *
+ * sjinfo represents the join being performed.
+ *
+ * If the current join completes the calculation of any outer joins that
+ * have been pushed down per outer-join identity 3, those relids will be
+ * added to the result along with sjinfo's own relid.  If pushed_down_joins
+ * is not NULL, then also the SpecialJoinInfos for such added outer joins will
+ * be appended to *pushed_down_joins (so caller must initialize it to NIL).
+ */
+Relids
+add_outer_joins_to_relids(PlannerInfo *root, Relids input_relids,
+						  SpecialJoinInfo *sjinfo,
+						  List **pushed_down_joins)
+{
+	/* Nothing to do if this isn't an outer join with an assigned relid. */
+	if (sjinfo == NULL || sjinfo->ojrelid == 0)
+		return input_relids;
+
+	/*
+	 * If it's not a left join, we have no rules that would permit executing
+	 * it in non-syntactic order, so just form the syntactic relid set.  (This
+	 * is just a quick-exit test; we'd come to the same conclusion anyway,
+	 * since its commute_below_l and commute_above_l sets must be empty.)
+	 */
+	if (sjinfo->jointype != JOIN_LEFT)
+		return bms_add_member(input_relids, sjinfo->ojrelid);
+
+	/*
+	 * We cannot add the OJ relid if this join has been pushed into the RHS of
+	 * a syntactically-lower left join per OJ identity 3.  (If it has, then we
+	 * cannot claim that its outputs represent the final state of its RHS.)
+	 * There will not be any other OJs that can be added either, so we're
+	 * done.
+	 */
+	if (!bms_is_subset(sjinfo->commute_below_l, input_relids))
+		return input_relids;
+
+	/* OK to add OJ's own relid */
+	input_relids = bms_add_member(input_relids, sjinfo->ojrelid);
+
+	/*
+	 * Contrariwise, if we are now forming the final result of such a commuted
+	 * pair of OJs, it's time to add the relid(s) of the pushed-down join(s).
+	 * We can skip this if this join was never a candidate to be pushed up.
+	 */
+	if (sjinfo->commute_above_l)
+	{
+		Relids		commute_above_rels = bms_copy(sjinfo->commute_above_l);
+		ListCell   *lc;
+
+		/*
+		 * The current join could complete the nulling of more than one
+		 * pushed-down join, so we have to examine all the SpecialJoinInfos.
+		 * Because join_info_list was built in bottom-up order, it's
+		 * sufficient to traverse it once: an ojrelid we add in one loop
+		 * iteration would not have affected decisions of earlier iterations.
+		 */
+		foreach(lc, root->join_info_list)
+		{
+			SpecialJoinInfo *othersj = (SpecialJoinInfo *) lfirst(lc);
+
+			if (othersj == sjinfo ||
+				othersj->ojrelid == 0 || othersj->jointype != JOIN_LEFT)
+				continue;		/* definitely not interesting */
+
+			if (!bms_is_member(othersj->ojrelid, commute_above_rels))
+				continue;
+
+			/* Add it if not already present but conditions now satisfied */
+			if (!bms_is_member(othersj->ojrelid, input_relids) &&
+				bms_is_subset(othersj->min_lefthand, input_relids) &&
+				bms_is_subset(othersj->min_righthand, input_relids) &&
+				bms_is_subset(othersj->commute_below_l, input_relids))
+			{
+				input_relids = bms_add_member(input_relids, othersj->ojrelid);
+				/* report such pushed down outer joins, if asked */
+				if (pushed_down_joins != NULL)
+					*pushed_down_joins = lappend(*pushed_down_joins, othersj);
+
+				/*
+				 * We must also check any joins that othersj potentially
+				 * commutes with.  They likewise must appear later in
+				 * join_info_list than othersj itself, so we can visit them
+				 * later in this loop.
+				 */
+				commute_above_rels = bms_add_members(commute_above_rels,
+													 othersj->commute_above_l);
+			}
+		}
+	}
+
+	return input_relids;
+}
+
 /*
  * populate_joinrel_with_paths
  *	  Add paths to the given joinrel for given pair of joining relations. The
@@ -1534,9 +1642,8 @@ try_partitionwise_join(PlannerInfo *root, RelOptInfo *rel1, RelOptInfo *rel2,
 
 		/* Build correct join relids for child join */
 		child_joinrelids = bms_union(child_rel1->relids, child_rel2->relids);
-		if (child_sjinfo->ojrelid != 0)
-			child_joinrelids = bms_add_member(child_joinrelids,
-											  child_sjinfo->ojrelid);
+		child_joinrelids = add_outer_joins_to_relids(root, child_joinrelids,
+													 child_sjinfo, NULL);
 
 		/* Find the AppendRelInfo structures */
 		appinfos = find_appinfos_by_relids(root, child_joinrelids, &nappinfos);