BUG #19835: view with GROUP BY ROLLUP over duplicate columns changes results after dump/restore

The following bug has been logged on the website:

Bug reference: 19835
Logged by: Atish Prakash
Email address: 125atishprak(at)gmail(dot)com
PostgreSQL version: 19rc1
Operating system: Linux
Description:

pg_get_viewdef() prints a grouping-set member as its expression. If an
earlier output column has
an equal expression, the parser binds the printed expression to that earlier
column. The two
columns are not interchangeable under grouping sets, so a re-created view
can silently return
different rows. pg_dump uses this output, so a dump and restore changes the
view with no error.

Reproducer

CREATE TABLE t (a int);
INSERT INTO t VALUES (1), (2);
CREATE VIEW v AS SELECT a AS x, a AS y, count(*) AS n FROM t GROUP BY
ROLLUP(1, 2);
SELECT * FROM v ORDER BY 1, 2;
x | y | n
---+---+---
1 | 1 | 1
1 | | 1
2 | 2 | 1
2 | | 1
| | 2

SELECT pg_get_viewdef('v'::regclass);
-- ... GROUP BY ROLLUP(a, a);

After pg_dump and restore into a new database:

SELECT * FROM v ORDER BY 1, 2;
x | y | n

1 | 1 | 1
1 | 1 | 1
2 | 2 | 1
2 | 2 | 1
| | 2

CUBE and GROUPING SETS deparse the same way.

Cause

get_rule_groupingset() calls get_rule_sortgroupclause() with force_colno =
false. The parser
(findTargetlistEntrySQL99) resolves an expression to the first equal
targetlist entry, so the
second member's sortgroupref moves to the first column.

Proposed fix

print the column number for a grouping-set member when an earlier non-junk
targetlist
entry has an equal expression. This gives GROUP BY ROLLUP(a, 2). All
regression tests pass with
the patch, and a new test is included in groupingsets.sql.