Backport "Use type-widened encoder for DataFrame rather than existing encoder to allow type-widening from set operations"

Author: HyukjinKwon

sql/core/src/main/scala/org/apache/spark/sql/Dataset.scala

@@ -1456,7 +1456,7 @@ class Dataset[T] private[sql](
    * @group typedrel
    * @since 2.0.0
    */
-  def union(other: Dataset[T]): Dataset[T] = withTypedPlan {
+  def union(other: Dataset[T]): Dataset[T] = withSetOperator {
     // This breaks caching, but it's usually ok because it addresses a very specific use case:
     // using union to union many files or partitions.
     CombineUnions(Union(logicalPlan, other.logicalPlan))
@@ -1472,7 +1472,7 @@ class Dataset[T] private[sql](
    * @group typedrel
    * @since 1.6.0
    */
-  def intersect(other: Dataset[T]): Dataset[T] = withTypedPlan {
+  def intersect(other: Dataset[T]): Dataset[T] = withSetOperator {
     Intersect(logicalPlan, other.logicalPlan)
   }
 
@@ -1486,7 +1486,7 @@ class Dataset[T] private[sql](
    * @group typedrel
    * @since 2.0.0
    */
-  def except(other: Dataset[T]): Dataset[T] = withTypedPlan {
+  def except(other: Dataset[T]): Dataset[T] = withSetOperator {
     Except(logicalPlan, other.logicalPlan)
   }
 
@@ -2607,4 +2607,14 @@ class Dataset[T] private[sql](
   @inline private def withTypedPlan[U : Encoder](logicalPlan: => LogicalPlan): Dataset[U] = {
     Dataset(sparkSession, logicalPlan)
   }
+
+  /** A convenient function to wrap a set based logical plan and produce a Dataset. */
+  @inline private def withSetOperator[U : Encoder](logicalPlan: => LogicalPlan): Dataset[U] = {
+    if (classTag.runtimeClass.isAssignableFrom(classOf[Row])) {
+      // Set operators widen types (change the schema), so we cannot reuse the row encoder.
+      Dataset.ofRows(sparkSession, logicalPlan).asInstanceOf[Dataset[U]]
+    } else {
+      Dataset(sparkSession, logicalPlan)
+    }
+  }
 }

sql/core/src/test/scala/org/apache/spark/sql/DataFrameSuite.scala

@@ -19,6 +19,7 @@ package org.apache.spark.sql
 
 import java.io.File
 import java.nio.charset.StandardCharsets
+import java.sql.{Date, Timestamp}
 import java.util.UUID
 
 import scala.language.postfixOps
@@ -1585,4 +1586,19 @@ class DataFrameSuite extends QueryTest with SharedSQLContext {
       }
     }
   }
+
+  test("SPARK-17123: Performing set operations that combine non-scala native types") {
+    val dates = Seq(
+      (BigDecimal.valueOf(1), new Timestamp(2)),
+      (BigDecimal.valueOf(4), new Timestamp(5))
+    ).toDF("timestamp", "decimal")
+
+    val widenTypedRows = Seq(
+      (10.5D, "string")
+    ).toDF("timestamp", "decimal")
+
+    dates.union(widenTypedRows).collect()
+    dates.except(widenTypedRows).collect()
+    dates.intersect(widenTypedRows).collect()
+  }
 }