[SPARK-16648][SQL] Make ignoreNullsExpr a child expression of First and Last

## What changes were proposed in this pull request?

Default `TreeNode.withNewChildren` implementation doesn't work for `Last` and when both constructor arguments are the same, e.g.:

```sql
LAST_VALUE(FALSE) -- The 2nd argument defaults to FALSE
LAST_VALUE(FALSE, FALSE)
LAST_VALUE(TRUE, TRUE)
```

This is because although `Last` is a unary expression, both of its constructor arguments, `child` and `ignoreNullsExpr`, are `Expression`s. When they have the same value, `TreeNode.withNewChildren` treats both of them as child nodes by mistake. `First` is also affected by this issue in exactly the same way.

This PR fixes this issue by making `ignoreNullsExpr` a child expression of `First` and `Last`.

## How was this patch tested?

New test case added in `WindowQuerySuite`.

Author: Cheng Lian <lian@databricks.com>

Closes #14295 from liancheng/spark-16648-last-value.

1 files changed, 12 insertions, 0 deletions

diff --git a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/WindowQuerySuite.scala b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/WindowQuerySuite.scala
index c6b7eb6366..0ff3511c87 100644
--- a/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/WindowQuerySuite.scala
+++ b/sql/hive/src/test/scala/org/apache/spark/sql/hive/execution/WindowQuerySuite.scala
@@ -247,4 +247,16 @@ class WindowQuerySuite extends QueryTest with SQLTestUtils with TestHiveSingleto
         |from part
         """.stripMargin))
   }
+
+  test("SPARK-16646: LAST_VALUE(FALSE) OVER ()") {
+    checkAnswer(sql("SELECT LAST_VALUE(FALSE) OVER ()"), Row(false))
+    checkAnswer(sql("SELECT LAST_VALUE(FALSE, FALSE) OVER ()"), Row(false))
+    checkAnswer(sql("SELECT LAST_VALUE(TRUE, TRUE) OVER ()"), Row(true))
+  }
+
+  test("SPARK-16646: FIRST_VALUE(FALSE) OVER ()") {
+    checkAnswer(sql("SELECT FIRST_VALUE(FALSE) OVER ()"), Row(false))
+    checkAnswer(sql("SELECT FIRST_VALUE(FALSE, FALSE) OVER ()"), Row(false))
+    checkAnswer(sql("SELECT FIRST_VALUE(TRUE, TRUE) OVER ()"), Row(true))
+  }
 }