[SPARK-12828][SQL] Natural join follow-up

This is a small addendum to #10762 to make the code more robust again future changes.

Author: Reynold Xin <[email protected]>

Closes #11070 from rxin/SPARK-12828-natural-join.

Author: rxin

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

@@ -1239,35 +1239,38 @@ class Analyzer(
    */
   object ResolveNaturalJoin extends Rule[LogicalPlan] {
     override def apply(plan: LogicalPlan): LogicalPlan = plan resolveOperators {
-      // Should not skip unresolved nodes because natural join is always unresolved.
       case j @ Join(left, right, NaturalJoin(joinType), condition) if j.resolvedExceptNatural =>
-        // find common column names from both sides, should be treated like usingColumns
+        // find common column names from both sides
         val joinNames = left.output.map(_.name).intersect(right.output.map(_.name))
         val leftKeys = joinNames.map(keyName => left.output.find(_.name == keyName).get)
         val rightKeys = joinNames.map(keyName => right.output.find(_.name == keyName).get)
         val joinPairs = leftKeys.zip(rightKeys)
+
         // Add joinPairs to joinConditions
         val newCondition = (condition ++ joinPairs.map {
           case (l, r) => EqualTo(l, r)
-        }).reduceLeftOption(And)
+        }).reduceOption(And)
+
         // columns not in joinPairs
         val lUniqueOutput = left.output.filterNot(att => leftKeys.contains(att))
         val rUniqueOutput = right.output.filterNot(att => rightKeys.contains(att))
-        // we should only keep unique columns(depends on joinType) for joinCols
+
+        // the output list looks like: join keys, columns from left, columns from right
         val projectList = joinType match {
           case LeftOuter =>
             leftKeys ++ lUniqueOutput ++ rUniqueOutput.map(_.withNullability(true))
           case RightOuter =>
             rightKeys ++ lUniqueOutput.map(_.withNullability(true)) ++ rUniqueOutput
           case FullOuter =>
             // in full outer join, joinCols should be non-null if there is.
-            val joinedCols = joinPairs.map {
-              case (l, r) => Alias(Coalesce(Seq(l, r)), l.name)()
-            }
-            joinedCols ++ lUniqueOutput.map(_.withNullability(true)) ++
+            val joinedCols = joinPairs.map { case (l, r) => Alias(Coalesce(Seq(l, r)), l.name)() }
+            joinedCols ++
+              lUniqueOutput.map(_.withNullability(true)) ++
               rUniqueOutput.map(_.withNullability(true))
-          case _ =>
+          case Inner =>
             rightKeys ++ lUniqueOutput ++ rUniqueOutput
+          case _ =>
+            sys.error("Unsupported natural join type " + joinType)
         }
         // use Project to trim unnecessary fields
         Project(projectList, Join(left, right, joinType, newCondition))

sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/plans/joinTypes.scala

@@ -62,5 +62,7 @@ case object LeftSemi extends JoinType {
 }
 
 case class NaturalJoin(tpe: JoinType) extends JoinType {
+  require(Seq(Inner, LeftOuter, RightOuter, FullOuter).contains(tpe),
+    "Unsupported natural join type " + tpe)
   override def sql: String = "NATURAL " + tpe.sql
 }

sql/catalyst/src/test/scala/org/apache/spark/sql/catalyst/analysis/ResolveNaturalJoinSuite.scala

@@ -30,10 +30,10 @@ class ResolveNaturalJoinSuite extends AnalysisTest {
   lazy val aNotNull = a.notNull
   lazy val bNotNull = b.notNull
   lazy val cNotNull = c.notNull
-  lazy val r1 = LocalRelation(a, b)
-  lazy val r2 = LocalRelation(a, c)
+  lazy val r1 = LocalRelation(b, a)
+  lazy val r2 = LocalRelation(c, a)
   lazy val r3 = LocalRelation(aNotNull, bNotNull)
-  lazy val r4 = LocalRelation(bNotNull, cNotNull)
+  lazy val r4 = LocalRelation(cNotNull, bNotNull)
 
   test("natural inner join") {
     val plan = r1.join(r2, NaturalJoin(Inner), None)