[SPARK-11379][SQL] ExpressionEncoder can't handle top level primitive type correctly

For inner primitive type(e.g. inside `Product`), we use `schemaFor` to get the catalyst type for it, https://github.com/apache/spark/blob/master/sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala#L403.

However, for top level primitive type, we use `dataTypeFor`, which is wrong.

Author: Wenchen Fan <[email protected]>

Closes #9337 from cloud-fan/encoder.

Author: cloud-fan

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/ScalaReflection.scala

@@ -170,7 +170,7 @@ trait ScalaReflection {
         .getOrElse(BoundReference(ordinal, dataType, false))
 
     /** Returns the current path or throws an error. */
-    def getPath = path.getOrElse(BoundReference(0, dataTypeFor(tpe), true))
+    def getPath = path.getOrElse(BoundReference(0, schemaFor(tpe).dataType, true))
 
     tpe match {
       case t if !dataTypeFor(t).isInstanceOf[ObjectType] =>

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/encoders/ExpressionEncoderSuite.scala

@@ -57,6 +57,7 @@ class ExpressionEncoderSuite extends SparkFunSuite {
   encodeDecodeTest(false)
   encodeDecodeTest(1.toShort)
   encodeDecodeTest(1.toByte)
+  encodeDecodeTest("hello")
 
   encodeDecodeTest(PrimitiveData(1, 1, 1, 1, 1, 1, true))