fix: TopK operator should return correct results on dictionary column…

… with nulls
apache · Oct 23, 2024 · 297c242 · 297c242
1 parent cb3e977
commit 297c242
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 1 deletion.
diff --git a/native/core/src/execution/operators/copy.rs b/native/core/src/execution/operators/copy.rs
@@ -262,7 +262,9 @@ fn copy_or_unpack_array(array: &Arc<dyn Array>, mode: &CopyMode) -> Result<Array
     match array.data_type() {
         DataType::Dictionary(_, value_type) => {
             let options = CastOptions::default();
-            cast_with_options(array, value_type.as_ref(), &options)
+            // We need to copy the array after `cast` because arrow-rs `take` kernel which is used
+            // to unpack dictionary array might reuse the input array's null buffer.
+            Ok(copy_array(&cast_with_options(array, value_type.as_ref(), &options)?))
         }
         _ => {
             if mode == &CopyMode::UnpackOrDeepCopy {

diff --git a/spark/src/test/scala/org/apache/comet/exec/CometExecSuite.scala b/spark/src/test/scala/org/apache/comet/exec/CometExecSuite.scala
@@ -64,6 +64,25 @@ class CometExecSuite extends CometTestBase {
     }
   }
 
+  test("TopK operator should return correct results on dictionary column with nulls") {
+    withSQLConf(SQLConf.USE_V1_SOURCE_LIST.key -> "") {
+      withTable("test_data") {
+        val tableDF = spark.sparkContext
+          .parallelize(Seq((1, null, "A"), (2, "BBB", "B"), (3, "BBB", "B"), (4, "BBB", "B")), 3)
+          .toDF("c1", "c2", "c3")
+        tableDF
+          .coalesce(1)
+          .sortWithinPartitions("c1")
+          .writeTo("test_data")
+          .using("parquet")
+          .create()
+
+        val df = sql("SELECT * FROM test_data ORDER BY c1 LIMIT 3")
+        checkSparkAnswer(df)
+      }
+    }
+  }
+
   test("DPP fallback") {
     withTempDir { path =>
       // create test data