amcheck: Support for different header sizes of short varlena datum

In the heap, tuples may contain short varlena datum with both 1B header and 4B
headers.  But the corresponding index tuple should always have such varlena's
with 1B headers.  So, for fingerprinting, we need to convert.

Backpatch to all supported versions.

Discussion: https://postgr.es/m/flat/7bdbe559-d61a-4ae4-a6e1-48abdf3024cc%40postgrespro.ru
Author: Michael Zhilin
Reviewed-by: Alexander Lakhin, Andrey Borodin, Jian He, Alexander Korotkov
Backpatch-through: 12

Author: akorotkov

contrib/amcheck/expected/check_btree.out

@@ -199,6 +199,18 @@ SELECT bt_index_check('bttest_a_expr_idx', true);
 
 (1 row)
 
+-- Check support of both 1B and 4B header sizes of short varlena datum
+CREATE TABLE varlena_bug (v text);
+ALTER TABLE varlena_bug ALTER column v SET storage plain;
+INSERT INTO varlena_bug VALUES ('x');
+COPY varlena_bug from stdin;
+CREATE INDEX varlena_bug_idx on varlena_bug(v);
+SELECT bt_index_check('varlena_bug_idx', true);
+ bt_index_check 
+----------------
+ 
+(1 row)
+
 -- cleanup
 DROP TABLE bttest_a;
 DROP TABLE bttest_b;
@@ -208,3 +220,4 @@ DROP TABLE toast_bug;
 DROP FUNCTION ifun(int8);
 DROP OWNED BY regress_bttest_role; -- permissions
 DROP ROLE regress_bttest_role;
+DROP TABLE varlena_bug;

contrib/amcheck/sql/check_btree.sql

@@ -135,6 +135,16 @@ CREATE INDEX bttest_a_expr_idx ON bttest_a ((ifun(id) + ifun(0)))
 
 SELECT bt_index_check('bttest_a_expr_idx', true);
 
+-- Check support of both 1B and 4B header sizes of short varlena datum
+CREATE TABLE varlena_bug (v text);
+ALTER TABLE varlena_bug ALTER column v SET storage plain;
+INSERT INTO varlena_bug VALUES ('x');
+COPY varlena_bug from stdin;
+x
+\.
+CREATE INDEX varlena_bug_idx on varlena_bug(v);
+SELECT bt_index_check('varlena_bug_idx', true);
+
 -- cleanup
 DROP TABLE bttest_a;
 DROP TABLE bttest_b;
@@ -144,3 +154,4 @@ DROP TABLE toast_bug;
 DROP FUNCTION ifun(int8);
 DROP OWNED BY regress_bttest_role; -- permissions
 DROP ROLE regress_bttest_role;
+DROP TABLE varlena_bug;

contrib/amcheck/verify_nbtree.c

@@ -2642,7 +2642,7 @@ bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
 	TupleDesc	tupleDescriptor = RelationGetDescr(state->rel);
 	Datum		normalized[INDEX_MAX_KEYS];
 	bool		isnull[INDEX_MAX_KEYS];
-	bool		toast_free[INDEX_MAX_KEYS];
+	bool		need_free[INDEX_MAX_KEYS];
 	bool		formnewtup = false;
 	IndexTuple	reformed;
 	int			i;
@@ -2661,7 +2661,7 @@ bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
 		att = TupleDescAttr(tupleDescriptor, i);
 
 		/* Assume untoasted/already normalized datum initially */
-		toast_free[i] = false;
+		need_free[i] = false;
 		normalized[i] = index_getattr(itup, att->attnum,
 									  tupleDescriptor,
 									  &isnull[i]);
@@ -2684,11 +2684,32 @@ bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
 		{
 			formnewtup = true;
 			normalized[i] = PointerGetDatum(PG_DETOAST_DATUM(normalized[i]));
-			toast_free[i] = true;
+			need_free[i] = true;
+		}
+
+		/*
+		 * Short tuples may have 1B or 4B header. Convert 4B header of short
+		 * tuples to 1B
+		 */
+		else if (VARATT_CAN_MAKE_SHORT(DatumGetPointer(normalized[i])))
+		{
+			/* convert to short varlena */
+			Size		len = VARATT_CONVERTED_SHORT_SIZE(DatumGetPointer(normalized[i]));
+			char	   *data = palloc(len);
+
+			SET_VARSIZE_SHORT(data, len);
+			memcpy(data + 1, VARDATA(DatumGetPointer(normalized[i])), len - 1);
+
+			formnewtup = true;
+			normalized[i] = PointerGetDatum(data);
+			need_free[i] = true;
 		}
 	}
 
-	/* Easier case: Tuple has varlena datums, none of which are compressed */
+	/*
+	 * Easier case: Tuple has varlena datums, none of which are compressed or
+	 * short with 4B header
+	 */
 	if (!formnewtup)
 		return itup;
 
@@ -2698,6 +2719,11 @@ bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
 	 * (normalized input datums).  This is rather naive, but shouldn't be
 	 * necessary too often.
 	 *
+	 * In the heap, tuples may contain short varlena datums with both 1B
+	 * header and 4B headers.  But the corresponding index tuple should always
+	 * have such varlena's with 1B headers.  So, if there is a short varlena
+	 * with 4B header, we need to convert it for for fingerprinting.
+	 *
 	 * Note that we rely on deterministic index_form_tuple() TOAST compression
 	 * of normalized input.
 	 */
@@ -2706,7 +2732,7 @@ bt_normalize_tuple(BtreeCheckState *state, IndexTuple itup)
 
 	/* Cannot leak memory here */
 	for (i = 0; i < tupleDescriptor->natts; i++)
-		if (toast_free[i])
+		if (need_free[i])
 			pfree(DatumGetPointer(normalized[i]));
 
 	return reformed;