cloud-fan commented on a change in pull request #29166:
URL: https://github.com/apache/spark/pull/29166#discussion_r459198562
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
##########
@@ -1237,20 +1250,48 @@ 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)
+ attrMapping ++= childAttrMapping.filter { case (oldAttr, _) =>
+ // If an attribute from the child is not output or referenced by
the parent plan,
+ // it's useless to propagate this attribute to the parent plan, as
they are not
+ // going to use this attribute.
+ (plan.outputSet ++ plan.references).contains(oldAttr)
Review comment:
ah sorry, you are right. `attrMapping` is not only propagated to
parents, but is also used to rewrite attributes of the currennt `plan`, so
`plan.references` is necessary. Can we make the comment clearer about it?
----------------------------------------------------------------
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:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
