[GitHub] [spark] Ngone51 commented on a change in pull request #29166: [SPARK-32280][SPARK-32372][SQL] ResolveReferences.dedupRight should only rewrite attributes for ancestor nodes of the conflict plan

[GitBox <gi...@apache.org>]

Ngone51 commented on a change in pull request #29166:
URL: https://github.com/apache/spark/pull/29166#discussion_r459179732



##########
File path: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
##########
@@ -1237,20 +1250,79 @@ class Analyzer(
       if (conflictPlans.isEmpty) {
         right
       } else {
-        val attributeRewrites = AttributeMap(conflictPlans.flatMap {
-          case (oldRelation, newRelation) => oldRelation.output.zip(newRelation.output)})
-        val conflictPlanMap = conflictPlans.toMap
-        // transformDown so that we can replace all the old Relations in one turn due to
-        // the reason that `conflictPlans` are also collected in pre-order.
-        right transformDown {
-          case r => conflictPlanMap.getOrElse(r, r)
-        } transformUp {
-          case other => other transformExpressions {
+        rewritePlan(right, conflictPlans.toMap)._1
+      }
+    }
+
+    private def rewritePlan(plan: LogicalPlan, conflictPlanMap: Map[LogicalPlan, LogicalPlan])
+      : (LogicalPlan, Seq[(Attribute, Attribute)]) = {
+      if (conflictPlanMap.contains(plan)) {
+        // If the plan is the one that conflict the with left one, we'd
+        // just replace it with the new plan and collect the rewrite
+        // attributes for the parent node.
+        val newRelation = conflictPlanMap(plan)
+        newRelation -> plan.output.zip(newRelation.output)
+      } else {
+        val attrMapping = new mutable.ArrayBuffer[(Attribute, Attribute)]()
+        val newPlan = plan.mapChildren { child =>
+          // If not, we'd rewrite child plan recursively until we find the
+          // conflict node or reach the leaf node.
+          val (newChild, childAttrMapping) = rewritePlan(child, conflictPlanMap)
+          // Only return rewrite attributes which could be used by the parent node.
+          // Otherwise, it could introduce duplicate rewrite attributes. For example,
+          // for the following plan, if we don't do filter for the `childAttrMapping`,
+          // the node `SubqueryAlias b` will return rewrite attribute of [kind#220 -> kind#228]
+          // (which is from the conflict plan `Project [id#218, foo AS kind#228]`), and the node
+          // `SubqueryAlias c` will return rewrite attribute of [kind#220 -> kind#229] (which
+          // is from the conflict plan `Project [id#227, foo AS kind#229]`). As a result, the top
+          // Join will have duplicated rewrite attribute.
+          //
+          // The problem is, the plan `Join Inner, (kind#229 = kind#223)` shouldn't keep returning
+          // rewrite attribute of [kind#220 -> kind#229] to its parent node `Project [id#227]` as
+          // it doesn't really need it.
+          //
+          // Join Inner, (id#218 = id#227)
+          // :- SubqueryAlias b
+          // :  +- Project [id#218, foo AS kind#228]
+          // :     +- SubqueryAlias a
+          // :        +- Project [1 AS id#218]
+          // :           +- OneRowRelation
+          // +- SubqueryAlias c
+          //    +- Project [id#227]
+          //       +- Join Inner, (kind#229 = kind#223)
+          //          :- SubqueryAlias l
+          //          :  +- SubqueryAlias b
+          //          :     +- Project [id#227, foo AS kind#229]
+          //          :        +- SubqueryAlias a
+          //          :           +- Project [1 AS id#227]
+          //          :              +- OneRowRelation
+          //          +- SubqueryAlias r
+          //             +- SubqueryAlias b
+          //                +- Project [id#224, foo AS kind#223]
+          //                   +- SubqueryAlias a
+          //                      +- Project [1 AS id#224]
+          //                         +- OneRowRelation
+          attrMapping ++= childAttrMapping.filter { case (oldAttr, _) =>
+            (plan.references ++ plan.outputSet ++ plan.producedAttributes).contains(oldAttr)

Review comment:
       `plan.references` is also necessary? For example,  Join's condition references the attributes but its output doesn't contain the attribute.




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org



---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org