Fix EvalPlanQual bug when query contains both locked and not-locked rels.

In commit afb9249d, we (probably I) made ExecLockRows assign
null test tuples to all relations of the query while setting up to do an
EvalPlanQual recheck for a newly-updated locked row.  This was sheerest
brain fade: we should only set test tuples for relations that are lockable
by the LockRows node, and in particular empty test tuples are only sensible
for inheritance child relations that weren't the source of the current
tuple from their inheritance tree.  Setting a null test tuple for an
unrelated table causes it to return NULLs when it should not, as exhibited
in bug #14034 from Bronislav Houdek.  To add insult to injury, doing it the
wrong way required two loops where one would suffice; so the corrected code
is even a bit shorter and faster.

Add a regression test case based on his example, and back-patch to 9.5
where the bug was introduced.

src/backend/executor/nodeLockRows.c

@@ -262,29 +262,15 @@ lnext:
 	 */
 	if (epq_needed)
 	{
-		int			i;
-
 		/* Initialize EPQ machinery */
 		EvalPlanQualBegin(&node->lr_epqstate, estate);
 
 		/*
-		 * Transfer already-fetched tuples into the EPQ state, and make sure
-		 * its test tuples for other tables are reset to NULL.
-		 */
-		for (i = 0; i < node->lr_ntables; i++)
-		{
-			EvalPlanQualSetTuple(&node->lr_epqstate,
-								 i + 1,
-								 node->lr_curtuples[i]);
-			/* freeing this tuple is now the responsibility of EPQ */
-			node->lr_curtuples[i] = NULL;
-		}
-
-		/*
-		 * Next, fetch a copy of any rows that were successfully locked
-		 * without any update having occurred.  (We do this in a separate pass
-		 * so as to avoid overhead in the common case where there are no
-		 * concurrent updates.)
+		 * Transfer any already-fetched tuples into the EPQ state, and fetch a
+		 * copy of any rows that were successfully locked without any update
+		 * having occurred.  (We do this in a separate pass so as to avoid
+		 * overhead in the common case where there are no concurrent updates.)
+		 * Make sure any inactive child rels have NULL test tuples in EPQ.
 		 */
 		foreach(lc, node->lr_arowMarks)
 		{
@@ -293,15 +279,25 @@ lnext:
 			HeapTupleData tuple;
 			Buffer		buffer;
 
-			/* ignore non-active child tables */
+			/* skip non-active child tables, but clear their test tuples */
 			if (!erm->ermActive)
 			{
 				Assert(erm->rti != erm->prti);	/* check it's child table */
+				EvalPlanQualSetTuple(&node->lr_epqstate, erm->rti, NULL);
 				continue;
 			}
 
-			if (EvalPlanQualGetTuple(&node->lr_epqstate, erm->rti) != NULL)
-				continue;		/* it was updated and fetched above */
+			/* was tuple updated and fetched above? */
+			if (node->lr_curtuples[erm->rti - 1] != NULL)
+			{
+				/* yes, so set it as the EPQ test tuple for this rel */
+				EvalPlanQualSetTuple(&node->lr_epqstate,
+									 erm->rti,
+									 node->lr_curtuples[erm->rti - 1]);
+				/* freeing this tuple is now the responsibility of EPQ */
+				node->lr_curtuples[erm->rti - 1] = NULL;
+				continue;
+			}
 
 			/* foreign tables should have been fetched above */
 			Assert(erm->relation->rd_rel->relkind != RELKIND_FOREIGN_TABLE);

src/test/isolation/expected/eval-plan-qual.out

@@ -144,3 +144,22 @@ accountid      balance
 
 checking       1050           
 savings        600            
+
+starting permutation: updateforss readforss c1 c2
+step updateforss: 
+	UPDATE table_a SET value = 'newTableAValue' WHERE id = 1;
+	UPDATE table_b SET value = 'newTableBValue' WHERE id = 1;
+
+step readforss: 
+	SELECT ta.id AS ta_id, ta.value AS ta_value,
+		(SELECT ROW(tb.id, tb.value)
+		 FROM table_b tb WHERE ta.id = tb.id) AS tb_row
+	FROM table_a ta
+	WHERE ta.id = 1 FOR UPDATE OF ta;
+ <waiting ...>
+step c1: COMMIT;
+step readforss: <... completed>
+ta_id          ta_value       tb_row         
+
+1              newTableAValue (1,tableBValue)
+step c2: COMMIT;

src/test/isolation/specs/eval-plan-qual.spec

@@ -16,12 +16,18 @@ setup
  INSERT INTO c1 SELECT 0, a / 3, a % 3 FROM generate_series(0, 9) a;
  INSERT INTO c2 SELECT 1, a / 3, a % 3 FROM generate_series(0, 9) a;
  INSERT INTO c3 SELECT 2, a / 3, a % 3 FROM generate_series(0, 9) a;
+
+ CREATE TABLE table_a (id integer, value text);
+ CREATE TABLE table_b (id integer, value text);
+ INSERT INTO table_a VALUES (1, 'tableAValue');
+ INSERT INTO table_b VALUES (1, 'tableBValue');
 }
 
 teardown
 {
  DROP TABLE accounts;
  DROP TABLE p CASCADE;
+ DROP TABLE table_a, table_b;
 }
 
 session "s1"
@@ -64,6 +70,15 @@ step "lockwithvalues"	{
 	  FOR UPDATE OF a1;
 }
 
+# these tests exercise EvalPlanQual with a SubLink sub-select (which should be
+# unaffected by any EPQ recheck behavior in the outer query); cf bug #14034
+
+step "updateforss"	{
+	UPDATE table_a SET value = 'newTableAValue' WHERE id = 1;
+	UPDATE table_b SET value = 'newTableBValue' WHERE id = 1;
+}
+
+
 session "s2"
 setup		{ BEGIN ISOLATION LEVEL READ COMMITTED; }
 step "wx2"	{ UPDATE accounts SET balance = balance + 450 WHERE accountid = 'checking'; }
@@ -81,6 +96,13 @@ step "returningp1" {
 	WITH u AS ( UPDATE p SET b = b WHERE a > 0 RETURNING * )
 	  SELECT * FROM u;
 }
+step "readforss"	{
+	SELECT ta.id AS ta_id, ta.value AS ta_value,
+		(SELECT ROW(tb.id, tb.value)
+		 FROM table_b tb WHERE ta.id = tb.id) AS tb_row
+	FROM table_a ta
+	WHERE ta.id = 1 FOR UPDATE OF ta;
+}
 step "c2"	{ COMMIT; }
 
 session "s3"
@@ -95,3 +117,4 @@ permutation "readp1" "writep1" "readp2" "c1" "c2"
 permutation "writep2" "returningp1" "c1" "c2"
 permutation "wx2" "partiallock" "c2" "c1" "read"
 permutation "wx2" "lockwithvalues" "c2" "c1" "read"
+permutation "updateforss" "readforss" "c1" "c2"