BUG #19401: Inconsistent predicate evaluation with derived table vs direct query involving NULL

PG Bug reporting form <noreply@postgresql.org> writes:

I would like to report a behavior that appears to be incorrect and
inconsistent in PostgreSQL when the same predicate is evaluated in (1) a
derived table and (2) a direct query.

The two queries are logically equivalent, but they return different
cardinalities.

CREATE TABLE t0 (c0 TEXT);
INSERT INTO t0 (c0) VALUES ('4');
-- result: length 0;
SELECT ref0 FROM (SELECT (any_value(c0)) AS ref0, ((('j' = NULL) <= (NULL =
NULL))) AS ref1 FROM t0) AS s WHERE ref1;
-- result: length 1; (NULL)
SELECT (any_value(c0)) FROM t0 WHERE (('j' = NULL) <= (NULL = NULL));

These are not "logically equivalent". The first one applies the
WHERE filter above the aggregation, the second one applies it
before the aggregation. An aggregate will produce some value
(typically NULL) even if there are zero input rows, so the second
query gives a single NULL result as-expected. In the first query,
the subselect produces a row (4,NULL) but then the outer WHERE
filters that row away.

regards, tom lane