From 0119573408f91775df7a969df0df0d3577ccc4e8 Mon Sep 17 00:00:00 2001
From: Norio <norio@apache.org>
Date: Sat, 25 Apr 2026 00:00:00 +0000
Subject: [PATCH] [GLUTEN-11916][VL][TEST] Enable
 subquery/exists-subquery/exists-orderby-limit.sql with SPARK-57125 workaround

GlutenSQLQueryTestSuite excludes ConvertToLocalRelation, ConstantFolding and
NullPropagation by default to force queries through Gluten's offload paths.
However, EXISTS+OFFSET queries in exists-orderby-limit.sql hit Spark's
LimitPushDown rule which rewrites

  LocalLimit(le, Offset(oe, child))

into

  Offset(oe, LocalLimit(Add(le, oe), child))

and relies on ConstantFolding to subsequently fold `Add(Literal(N), Literal(M))`
to `Literal(N + M)`. Without ConstantFolding the unfolded Add reaches physical
planning where BasicOperators only matches LocalLimit(IntegerLiteral, _),
producing

  AssertionError: No plan for LocalLimit (1 + 2)

wrapped as [INTERNAL_ERROR] during the planning phase.

This patch enables the test and re-enables ConstantFolding for just this SQL
file via a per-file `--SET spark.sql.optimizer.excludedRules=...` directive
that keeps only ConvertToLocalRelation excluded.

The upstream Spark fix is tracked as SPARK-57125 (Apache Spark PR #56180), which
makes LimitPushDown produce a literal sum directly so the rule no longer depends
on ConstantFolding. Once that lands and Gluten picks up the Spark version, the
`--SET` directive in this file can be removed.

Note: the test framework's `--SET` parser splits values by comma, so multiple
excluded rules cannot be specified in a single directive (recorded separately
for a future Spark/Gluten follow-up). NullPropagation getting re-enabled is
acceptable for this test.
---
 .../exists-subquery/exists-orderby-limit.sql         | 12 ++++++++++++
 .../utils/velox/VeloxSQLQueryTestSettings.scala      |  2 +-
 2 files changed, 13 insertions(+), 1 deletion(-)

diff --git a/gluten-ut/spark41/src/test/resources/backends-velox/sql-tests/inputs/subquery/exists-subquery/exists-orderby-limit.sql b/gluten-ut/spark41/src/test/resources/backends-velox/sql-tests/inputs/subquery/exists-subquery/exists-orderby-limit.sql
index 9ff3409b21a..1b2a884d158 100644
--- a/gluten-ut/spark41/src/test/resources/backends-velox/sql-tests/inputs/subquery/exists-subquery/exists-orderby-limit.sql
+++ b/gluten-ut/spark41/src/test/resources/backends-velox/sql-tests/inputs/subquery/exists-subquery/exists-orderby-limit.sql
@@ -5,6 +5,18 @@
 --CONFIG_DIM1 spark.sql.codegen.wholeStage=false,spark.sql.codegen.factoryMode=CODEGEN_ONLY
 --CONFIG_DIM1 spark.sql.codegen.wholeStage=false,spark.sql.codegen.factoryMode=NO_CODEGEN
 
+-- GlutenSQLQueryTestSuite excludes ConvertToLocalRelation, ConstantFolding and
+-- NullPropagation by default to force queries through Gluten's offload paths.
+-- However, EXISTS+OFFSET queries hit Spark's LimitPushDown rule which produces
+-- `LocalLimit(Add(limit, offset), ...)` and assumes ConstantFolding will fold
+-- the Add. Without ConstantFolding the planner fails with
+-- `AssertionError: No plan for LocalLimit (1 + 2)` because BasicOperators only
+-- matches LocalLimit(IntegerLiteral, _). Re-enable ConstantFolding (and
+-- NullPropagation, which the test framework's --SET parser would split on
+-- the comma) for this file by keeping only ConvertToLocalRelation excluded.
+-- Tracked upstream as SPARK-57125.
+--SET spark.sql.optimizer.excludedRules=org.apache.spark.sql.catalyst.optimizer.ConvertToLocalRelation
+
 --ONLY_IF spark
 CREATE TEMPORARY VIEW EMP AS SELECT * FROM VALUES
   (100, "emp 1", date "2005-01-01", 100.00D, 10),
diff --git a/gluten-ut/spark41/src/test/scala/org/apache/gluten/utils/velox/VeloxSQLQueryTestSettings.scala b/gluten-ut/spark41/src/test/scala/org/apache/gluten/utils/velox/VeloxSQLQueryTestSettings.scala
index ea58cc7a5cd..0202ecd24c0 100644
--- a/gluten-ut/spark41/src/test/scala/org/apache/gluten/utils/velox/VeloxSQLQueryTestSettings.scala
+++ b/gluten-ut/spark41/src/test/scala/org/apache/gluten/utils/velox/VeloxSQLQueryTestSettings.scala
@@ -127,7 +127,7 @@ object VeloxSQLQueryTestSettings extends SQLQueryTestSettings {
     "subquery/exists-subquery/exists-cte.sql",
     "subquery/exists-subquery/exists-having.sql",
     "subquery/exists-subquery/exists-joins-and-set-ops.sql",
-    // TODO: fix on Spark-4.1 "subquery/exists-subquery/exists-orderby-limit.sql",
+    "subquery/exists-subquery/exists-orderby-limit.sql",
     "subquery/exists-subquery/exists-outside-filter.sql",
     "subquery/exists-subquery/exists-within-and-or.sql",
     "subquery/in-subquery/in-basic.sql",