GROUP BY ... WITH ROLLUP: duplicate grouping-column reference not NULL in super-aggregate row

Summary

In GROUP BY ... WITH ROLLUP, grouping columns must be NULL in the super-aggregate (grand total) row. MariaDB nullifies only the first select-list slot tied to the GROUP BY column. Any additional bare reference to the same column — duplicate column, alias (a AS c2), or extra slot from t.* expansion — keeps the last processed group's value in the super-aggregate row.

Wrapping the duplicate in an expression (a + 0) yields the correct NULL. Reproduced on MariaDB 12.3.1-MariaDB-asan-log. MySQL 9.6 returns NULL for both columns in the super-aggregate row.

Minimal Reproducer

DROP TABLE IF EXISTS t1;

CREATE TABLE t1 (a INT) ENGINE=InnoDB;

INSERT INTO t1 VALUES (10), (20), (30);

SELECT a, a AS c2

FROM t1

GROUP BY a WITH ROLLUP;

Actual Result (MariaDB 12.3.1-MariaDB-asan-log)

+------+------+ 

| a    | c2   | 

+------+------+ 

| 10   | 10   | 

| 20   | 20   | 

| 30   | 30   | 

| NULL | 30   |  <-- wrong: c2 should be NULL

+------+------+

Expected Result

Super-aggregate row: NULL | NULL (both grouping-column projections NULL), consistent with MySQL 9.6 and SQL ROLLUP semantics.

Root Cause (Hypothesis)

ROLLUP nullification applies to one select-list item per GROUP BY expression (bound during resolution). Additional bare Item_field nodes referencing the same physical column are not rewritten and still read the row buffer populated with the last group's value. Expression wrappers (a+0) force evaluation on the already-nullified value.

Likely fix: ensure all select-list references mapping to grouping items are nullified, not only the first per group expression.

Environment

Suggested Regression Test

CREATE TABLE t1 (a INT);

INSERT INTO t1 VALUES (10),(20),(30);

SELECT a, a AS c2 FROM t1 GROUP BY a WITH ROLLUP;

-- last row must be: NULL NULL

DROP TABLE t1;