From 06b73410561e77a3cc4e496e3be3f1ef78a721f5 Mon Sep 17 00:00:00 2001
From: lifulong <lifulong@zhihu.com>
Date: Thu, 26 Mar 2026 09:56:14 +0000
Subject: [PATCH] fix native union use column type as name lead to result error

---
 .../gluten/execution/MiscOperatorSuite.scala  | 49 +++++++++++++++++++
 cpp/velox/substrait/SubstraitToVeloxPlan.cc   |  3 +-
 2 files changed, 51 insertions(+), 1 deletion(-)

diff --git a/backends-velox/src/test/scala/org/apache/gluten/execution/MiscOperatorSuite.scala b/backends-velox/src/test/scala/org/apache/gluten/execution/MiscOperatorSuite.scala
index f5196cb8c0c2..9fbd99752ed4 100644
--- a/backends-velox/src/test/scala/org/apache/gluten/execution/MiscOperatorSuite.scala
+++ b/backends-velox/src/test/scala/org/apache/gluten/execution/MiscOperatorSuite.scala
@@ -582,6 +582,55 @@ class MiscOperatorSuite extends VeloxWholeStageTransformerSuite with AdaptiveSpa
     }
   }
 
+  test("native union_all with two level union keeps distinct output columns") {
+    withTempView("union_src_a", "union_src_b", "union_src_c") {
+      Seq(
+        ("valueA", "value1", "value11", "value111"),
+        ("valueA", "value2", "value22", "value222")
+      ).toDF("col1", "col2", "col3", "col4")
+        .createOrReplaceTempView("union_src_a")
+      Seq(
+        ("valueB", "value3", "value33", "value333"),
+        ("valueB", "value4", "value44", "value444")
+      ).toDF("col1", "col2", "col3", "col4")
+        .createOrReplaceTempView("union_src_b")
+
+      withSQLConf(GlutenConfig.NATIVE_UNION_ENABLED.key -> "true") {
+        compareDfResultsAgainstVanillaSpark(
+          () =>
+            spark.sql("""
+                        |with deduplicated_data as (
+                        |  select col1, col2, col3, col4
+                        |  from (
+                        |    select
+                        |      u.col1,
+                        |      u.col2,
+                        |      u.col3,
+                        |      u.col4,
+                        |      row_number() over (partition by u.col2 order by u.col5 desc) as rn
+                        |    from (
+                        |      select col1, col2, col3, col4, 98 as col5 from union_src_a
+                        |      union all
+                        |      select col1, col2, col3, col4, 100 as col5 from union_src_b
+                        |    ) u
+                        |  ) t
+                        |  where t.rn = 1
+                        |)
+                        |select col1, col2, col3, col4
+                        |from deduplicated_data
+                        |where col1 != 'valueC'
+                        |union all
+                        |select col1, col2, col3, col4
+                        |from deduplicated_data
+                        |where col1 = 'valueC'
+                        |""".stripMargin),
+          compareResult = true,
+          checkGlutenPlan[UnionExecTransformer]
+        )
+      }
+    }
+  }
+
   test("union two tables") {
     runQueryAndCompare("""
                          |select count(orderkey) from (
diff --git a/cpp/velox/substrait/SubstraitToVeloxPlan.cc b/cpp/velox/substrait/SubstraitToVeloxPlan.cc
index adb7fc5f45b6..b964649f01d2 100644
--- a/cpp/velox/substrait/SubstraitToVeloxPlan.cc
+++ b/cpp/velox/substrait/SubstraitToVeloxPlan.cc
@@ -1244,7 +1244,8 @@ core::PlanNodePtr SubstraitToVeloxPlanConverter::toVeloxPlan(const ::substrait::
       const RowTypePtr outRowType = asRowType(children[0]->outputType());
       std::vector<std::string> outNames;
       for (int32_t colIdx = 0; colIdx < outRowType->size(); ++colIdx) {
-        const auto name = outRowType->childAt(colIdx)->name();
+        // Using field names from the unified output row type instead child type names
+        const auto name = outRowType->nameOf(colIdx);
         outNames.push_back(name);
       }