[SPARK-17863][SQL] should not add column into Distinct

## What changes were proposed in this pull request?

We are trying to resolve the attribute in sort by pulling up some column for grandchild into child, but that's wrong when the child is Distinct, because the added column will change the behavior of Distinct, we should not do that.

## How was this patch tested?

Added regression test.

Author: Davies Liu <davies@databricks.com>

Closes #15489 from davies/order_distinct.

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

@@ -838,6 +838,8 @@ class Analyzer(
           // attributes that its child might have or could have.
           val missing = missingAttrs -- g.child.outputSet
           g.copy(join = true, child = addMissingAttr(g.child, missing))
+        case d: Distinct =>
+          throw new AnalysisException(s"Can't add $missingAttrs to $d")
         case u: UnaryNode =>
           u.withNewChildren(addMissingAttr(u.child, missingAttrs) :: Nil)
         case other =>

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

@@ -1106,6 +1106,30 @@ class SQLQuerySuite extends QueryTest with SharedSQLContext {
     )
   }
 
+  test("SPARK-17863: SELECT distinct does not work correctly if order by missing attribute") {
+    checkAnswer(
+      sql("""select distinct struct.a, struct.b
+          |from (
+          |  select named_struct('a', 1, 'b', 2, 'c', 3) as struct
+          |  union all
+          |  select named_struct('a', 1, 'b', 2, 'c', 4) as struct) tmp
+          |order by a, b
+          |""".stripMargin),
+      Row(1, 2) :: Nil)
+
+    val error = intercept[AnalysisException] {
+      sql("""select distinct struct.a, struct.b
+            |from (
+            |  select named_struct('a', 1, 'b', 2, 'c', 3) as struct
+            |  union all
+            |  select named_struct('a', 1, 'b', 2, 'c', 4) as struct) tmp
+            |order by struct.a, struct.b
+            |""".stripMargin)
+    }
+    assert(error.message contains "cannot resolve '`struct.a`' given input columns: [a, b]")
+
+  }
+
   test("cast boolean to string") {
     // TODO Ensure true/false string letter casing is consistent with Hive in all cases.
     checkAnswer(