From 7f0c2d25a4bc4902380012811eae66b19480b7e2 Mon Sep 17 00:00:00 2001
From: Vladimir Petrzhikovskii <hello@0xdeafbeef.dev>
Date: Thu, 29 Oct 2026 20:13:00 +0100
Subject: [PATCH] test: cover grouping set equivalences and constants

On DataFusion 53.1.0 and 55.2.0 with target_partitions = 1, a
GROUPING SETS ((a, b), (a)) query over rows filtered by a = b returned
rows in the wrong ORDER BY order. The grouping-set AggregateExec
projected the input equivalence a = b to its output, but the (a) set
outputs b as NULL, so a = b is false on those rows. EnforceSorting
used the equivalence and rewrote the sort
[a ASC NULLS LAST, b ASC NULLS FIRST] to [b ASC NULLS LAST]. A filter
constant such as b = 1 fails the same way.

#24422 fixed this on main by building the properties of a grouping-set
aggregate from empty EquivalenceProperties. Its tests check that
grouping sets do not pick ordered aggregation. Add cases to
grouping.slt for the equivalence, the constant, a row_number() variant
and the query from #16965, in the target_partitions = 1 block.

Closes #16965.
---
 .../sqllogictest/test_files/grouping.slt      | 64 +++++++++++++++++++
 1 file changed, 64 insertions(+)

diff --git a/datafusion/sqllogictest/test_files/grouping.slt b/datafusion/sqllogictest/test_files/grouping.slt
index ce6ea17d01d4a..850f73f56b1cb 100644
--- a/datafusion/sqllogictest/test_files/grouping.slt
+++ b/datafusion/sqllogictest/test_files/grouping.slt
@@ -123,6 +123,70 @@ LIMIT 10;
 statement ok
 RESET datafusion.execution.enable_migration_aggregate;
 
+# a = b holds before the aggregate, but not after it: the (a) grouping set
+# outputs b as NULL.
+#   (a, b) set: (1, 1), (2, 2)
+#   (a) set:    (1, NULL), (2, NULL)
+# So the sort must keep both a and b. Queries from PostgreSQL's groupingsets.sql.
+query TT
+EXPLAIN SELECT a, b FROM (VALUES (1, 1), (2, 2)) AS t (a, b) WHERE a = b
+GROUP BY GROUPING SETS ((a, b), (a)) ORDER BY a, b NULLS FIRST;
+----
+logical_plan
+01)Sort: t.a ASC NULLS LAST, t.b ASC NULLS FIRST
+02)--Projection: t.a, t.b
+03)----Aggregate: groupBy=[[GROUPING SETS ((t.a, t.b), (t.a))]], aggr=[[]]
+04)------SubqueryAlias: t
+05)--------Projection: column1 AS a, column2 AS b
+06)----------Filter: column2 = column1
+07)------------Values: (Int64(1), Int64(1)), (Int64(2), Int64(2))
+physical_plan
+01)SortExec: expr=[a@0 ASC NULLS LAST, b@1 ASC], preserve_partitioning=[false]
+02)--ProjectionExec: expr=[a@0 as a, b@1 as b]
+03)----AggregateExec: mode=Final, gby=[a@0 as a, b@1 as b, __grouping_id@2 as __grouping_id], aggr=[]
+04)------AggregateExec: mode=Partial, gby=[(a@0 as a, b@1 as b), (a@0 as a, NULL as b)], aggr=[]
+05)--------ProjectionExec: expr=[column1@0 as a, column2@1 as b]
+06)----------FilterExec: column2@1 = column1@0
+07)------------DataSourceExec: partitions=1, partition_sizes=[1]
+
+query II
+SELECT a, b FROM (VALUES (1, 1), (2, 2)) AS t (a, b) WHERE a = b
+GROUP BY GROUPING SETS ((a, b), (a)) ORDER BY a, b NULLS FIRST;
+----
+1 NULL
+1 1
+2 NULL
+2 2
+
+query III
+SELECT a, b, row_number() OVER (ORDER BY a, b NULLS FIRST)
+FROM (VALUES (1, 1), (2, 2)) AS t (a, b) WHERE a = b
+GROUP BY GROUPING SETS ((a, b), (a)) ORDER BY a, b NULLS FIRST;
+----
+1 NULL 1
+1 1 2
+2 NULL 3
+2 2 4
+
+# Same for b = 1: the (a) set outputs (1, NULL), (2, NULL), so b stays in the sort.
+query II
+SELECT a, b FROM (VALUES (1, 1), (2, 1)) AS t (a, b) WHERE b = 1
+GROUP BY GROUPING SETS ((a, b), (a)) ORDER BY a, b NULLS FIRST;
+----
+1 NULL
+1 1
+2 NULL
+2 1
+
+# https://github.com/apache/datafusion/issues/16965: a constant, sorted input
+# must still produce one row per grouping set.
+query T rowsort
+SELECT id FROM (SELECT 'id' AS id UNION ALL SELECT 'id' AS id ORDER BY id)
+GROUP BY GROUPING SETS ((id), ());
+----
+NULL
+id
+
 statement ok
 SET datafusion.execution.target_partitions = 4;