[SPARK-14000][SQL] case class with a tuple field can't work in Dataset

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala

@@ -596,7 +596,10 @@ class Analyzer(
       exprs.exists(_.collect { case _: Star => true }.nonEmpty)
   }
 
-  private def resolveExpression(expr: Expression, plan: LogicalPlan, throws: Boolean = false) = {
+  protected[sql] def resolveExpression(
+      expr: Expression,
+      plan: LogicalPlan,
+      throws: Boolean = false) = {
     // Resolve expression in one round.
     // If throws == false or the desired attribute doesn't exist
     // (like try to resolve `a.b` but `a` doesn't exist), fail and return the origin one.

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/encoders/ExpressionEncoder.scala

@@ -282,8 +282,14 @@ case class ExpressionEncoder[T](
     // If we have nested tuple, the `fromRowExpression` will contains `GetStructField` instead of
     // `UnresolvedExtractValue`, so we need to check if their ordinals are all valid.
     // Note that, `BoundReference` contains the expected type, but here we need the actual type, so
-    // we unbound it by the given `schema` and propagate the actual type to `GetStructField`.
-    val unbound = fromRowExpression transform {
+    // we unbound it by the given `schema` and propagate the actual type to `GetStructField`, after
+    // we resolve the `fromRowExpression`.
+    val resolved = SimpleAnalyzer.resolveExpression(
+      fromRowExpression,
+      LocalRelation(schema),
+      throws = true)
+
+    val unbound = resolved transform {
       case b: BoundReference => schema(b.ordinal)
     }
 

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/expressions/complexTypeExtractors.scala

@@ -110,7 +110,12 @@ case class GetStructField(child: Expression, ordinal: Int, name: Option[String]
 
   override def dataType: DataType = childSchema(ordinal).dataType
   override def nullable: Boolean = child.nullable || childSchema(ordinal).nullable
-  override def toString: String = s"$child.${name.getOrElse(childSchema(ordinal).name)}"
+
+  override def toString: String = {
+    val fieldName = if (resolved) childSchema(ordinal).name else s"_$ordinal"
+    s"$child.${name.getOrElse(fieldName)}"
+  }
+
   override def sql: String =
     child.sql + s".${quoteIdentifier(name.getOrElse(childSchema(ordinal).name))}"
 

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

@@ -27,8 +27,6 @@ import org.apache.spark.sql.functions._
 import org.apache.spark.sql.test.SharedSQLContext
 import org.apache.spark.sql.types.{IntegerType, StringType, StructField, StructType}
 
-case class OtherTuple(_1: String, _2: Int)
-
 class DatasetSuite extends QueryTest with SharedSQLContext {
   import testImplicits._
 
@@ -636,8 +634,19 @@ class DatasetSuite extends QueryTest with SharedSQLContext {
       Seq(OuterObject.InnerClass("foo")).toDS(),
       OuterObject.InnerClass("foo"))
   }
+
+  test("SPARK-14000: case class with tuple type field") {
+    checkDataset(
+      Seq(TupleClass((1, "a"))).toDS(),
+      TupleClass(1, "a")
+    )
+  }
 }
 
+case class OtherTuple(_1: String, _2: Int)
+
+case class TupleClass(data: (Int, String))
+
 class OuterClass extends Serializable {
   case class InnerClass(a: String)
 }