cloud-fan commented on code in PR #41347:
URL: https://github.com/apache/spark/pull/41347#discussion_r1266553718
##########
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/DeduplicateRelations.scala:
##########
@@ -95,59 +102,118 @@ object DeduplicateRelations extends Rule[LogicalPlan] {
case p: LogicalPlan if p.isStreaming => (plan, false)
case m: MultiInstanceRelation =>
- val planWrapper = RelationWrapper(m.getClass, m.output.map(_.exprId.id))
- if (existingRelations.contains(planWrapper)) {
- val newNode = m.newInstance()
- newNode.copyTagsFrom(m)
- (newNode, true)
- } else {
- existingRelations.add(planWrapper)
- (m, false)
- }
+ deduplicate(existingRelations, m)(Some(_.output.map(_.exprId.id)))(node
=> node.newInstance()
+ .asInstanceOf[LogicalPlan with MultiInstanceRelation])
+
+ case p: Project =>
+ deduplicate(existingRelations, p)(Some(newProject =>
findAliases(newProject.projectList)
+ .map(_.exprId.id).toSeq))(newProject =>
newProject.copy(newAliases(newProject.projectList)))
+
+ case s: SerializeFromObject =>
+ deduplicate(existingRelations,
s)(Some(_.serializer.map(_.exprId.id)))(newSer =>
+ newSer.copy(newSer.serializer.map(_.newInstance())))
+
+ case f: FlatMapGroupsInPandas =>
+ deduplicate(existingRelations,
f)(Some(_.output.map(_.exprId.id)))(newFlatMap =>
+ newFlatMap.copy(output = newFlatMap.output.map(_.newInstance())))
+
+ case f: FlatMapCoGroupsInPandas =>
+ deduplicate(existingRelations,
f)(Some(_.output.map(_.exprId.id)))(newFlatMap =>
+ newFlatMap.copy(output = newFlatMap.output.map(_.newInstance())))
+
+ case m: MapInPandas =>
+ deduplicate(existingRelations,
m)(Some(_.output.map(_.exprId.id)))(newMap =>
+ newMap.copy(output = newMap.output.map(_.newInstance())))
+
+ case p: PythonMapInArrow =>
+ deduplicate(existingRelations,
p)(Some(_.output.map(_.exprId.id)))(newMap =>
+ newMap.copy(output = newMap.output.map(_.newInstance())))
+
+ case a: AttachDistributedSequence =>
+ deduplicate(existingRelations,
a)(Some(_.producedAttributes.map(_.exprId.id).toSeq))(
+ newAttach => newAttach.copy(sequenceAttr = newAttach.producedAttributes
+ .map(_.newInstance()).head))
+
+ case g: Generate =>
+ deduplicate(existingRelations,
g)(Some(_.generatorOutput.map(_.exprId.id)))(newGenerate =>
+ newGenerate.copy(generatorOutput =
newGenerate.generatorOutput.map(_.newInstance())))
+
+ case e: Expand =>
+ deduplicate(existingRelations,
e)(Some(_.producedAttributes.map(_.exprId.id).toSeq))(
+ newExpand => newExpand.copy(output =
newExpand.output.map(_.newInstance())))
+
+ case w: Window =>
+ deduplicate(existingRelations, w)(Some(_.windowExpressions
+ .map(_.exprId.id)))(newWindow => newWindow.copy(windowExpressions =
+ newWindow.windowExpressions.map(_.newInstance())))
+
+ case s: ScriptTransformation =>
+ deduplicate(existingRelations, s)(Some(_.output.map(_.exprId.id)))(
+ newScript => newScript.copy(output =
newScript.output.map(_.newInstance())))
case plan: LogicalPlan =>
- var planChanged = false
- val newPlan = if (plan.children.nonEmpty) {
- val newChildren = mutable.ArrayBuffer.empty[LogicalPlan]
- for (c <- plan.children) {
- val (renewed, changed) = renewDuplicatedRelations(existingRelations,
c)
- newChildren += renewed
- if (changed) {
- planChanged = true
- }
- }
+ deduplicate(existingRelations, plan)()(p => p)
+ }
- val planWithNewSubquery = plan.transformExpressions {
- case subquery: SubqueryExpression =>
- val (renewed, changed) =
renewDuplicatedRelations(existingRelations, subquery.plan)
- if (changed) planChanged = true
- subquery.withNewPlan(renewed)
+ private def deduplicate[T <: LogicalPlan](
+ existingRelations: mutable.HashSet[RelationWrapper], plan: T)
+ (getExprIds: Option[T => Seq[Long]] = Option.empty)
Review Comment:
I find the code a bit hard to read with too many lambda parameters. How
about the following code structure?
```
def deduplicate(...) // deduplicate relations in children and update
attriubte references
def deduplicateAndRenew(...) // invokes `deduplicate` and renew attributes
in itself
```
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
