Github user davies commented on a diff in the pull request:
https://github.com/apache/spark/pull/10678#discussion_r50152633
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
---
@@ -521,38 +524,138 @@ class Analyzer(
*/
object ResolveSortReferences extends Rule[LogicalPlan] {
def apply(plan: LogicalPlan): LogicalPlan = plan resolveOperators {
- case s @ Sort(ordering, global, p @ Project(projectList, child))
- if !s.resolved && p.resolved =>
- val (newOrdering, missing) = resolveAndFindMissing(ordering, p,
child)
+ case s @ Sort(_, _, child) if !s.resolved && child.resolved =>
+ val (newOrdering, missingResolvableAttrs) =
+ collectResolvableMissingAttrs(s.order, plans =
mutable.Queue(child))
- // If this rule was not a no-op, return the transformed plan,
otherwise return the original.
- if (missing.nonEmpty) {
- // Add missing attributes and then project them away after the
sort.
- Project(p.output,
- Sort(newOrdering, global,
- Project(projectList ++ missing, child)))
- } else {
- logDebug(s"Failed to find $missing in ${p.output.mkString(",
")}")
+ if (missingResolvableAttrs.isEmpty) {
+ val unresolvableAttrs = s.order.filterNot(_.resolved)
+ logDebug(s"Failed to find $unresolvableAttrs in
${child.output.mkString(", ")}")
s // Nothing we can do here. Return original plan.
}
+ else {
+ // Transform the whole tree in post-order. Add into the self's
outputSet
+ // all the children attributes that are part of
missingResolvableAttrs.
+ // Assumption: all the conflicting attributes between left and
right have been resolved
+ val newChild = child transformUp {
+ case p: Project =>
+ p.copy(projectList = p.projectList ++
+ findNotResolvedMissingAttrs(p.outputSet, p.inputSet,
missingResolvableAttrs))
+ case w: Window =>
+ w.copy(projectList = w.projectList ++
+ findNotResolvedMissingAttrs(w.outputSet, w.inputSet,
missingResolvableAttrs))
+ case a: Aggregate =>
+ val newAggregateExpressions = a.aggregateExpressions ++
+ findNotResolvedMissingAttrs(
+ a.aggregateExpressions, a.inputSet,
missingResolvableAttrs)
+ a.copy(aggregateExpressions = newAggregateExpressions)
+ case o => o
+ }
+
+ // Add missing attributes and then project them away after the
sort.
+ Project(child.output,
+ Sort(newOrdering, s.global, newChild))
+ }
+ }
+
+ private def findNotResolvedMissingAttrs(
+ outputSet: AttributeSet,
+ inputSet: AttributeSet,
+ missingAttrs: Seq[Attribute]): Seq[Attribute] = {
+ val resolvedAttrs =
+ missingAttrs.filter(attr =>
inputSet.exists(_.semanticEquals(attr)))
+ resolvedAttrs.filterNot(attr =>
outputSet.exists(_.semanticEquals(attr)))
+ }
+
+ private def findNotResolvedMissingAttrs(
+ outputSet: Seq[Expression],
+ inputSet: AttributeSet,
+ missingAttrs: Seq[Attribute]): Seq[Attribute] = {
+ val resolvedAttrs =
+ missingAttrs.filter(attr =>
inputSet.exists(_.semanticEquals(attr)))
+ resolvedAttrs.filterNot(attr =>
outputSet.exists(_.semanticEquals(attr)))
+ }
+
+ /**
+ * Traverse the tree until resolving the sorting attributes
+ * Return all the resolvable missing sorting attributes
+ */
+ @tailrec
+ private def collectResolvableMissingAttrs(
+ ordering: Seq[SortOrder],
+ plans: mutable.Queue[LogicalPlan]): (Seq[SortOrder],
Seq[Attribute]) = {
+ if (plans.isEmpty) (Seq.empty[SortOrder], Seq.empty[Attribute])
+ else {
+ plans.dequeue() match {
+ // Only Windows and Project have projectList-like attribute.
+ case un: UnaryNode if un.isInstanceOf[Project] ||
un.isInstanceOf[Window] =>
+ val (newOrdering, missingAttrs) =
resolveAndFindMissing(ordering, un, un.child)
+ // If missingAttrs is non empty, that means we got it and
return it;
+ // Otherwise, continue to traverse the tree.
+ if (missingAttrs.nonEmpty) (newOrdering, missingAttrs)
+ else {
+ plans.enqueue(un.child)
+ collectResolvableMissingAttrs(ordering, plans)
+ }
+ // Jump over the following UnaryNode types
+ // The output of these types is the same as their child's output
+ case un: UnaryNode
+ if un.isInstanceOf[Distinct] ||
+ un.isInstanceOf[Filter] ||
+ un.isInstanceOf[Limit] ||
+ un.isInstanceOf[RedistributeData] ||
+ un.isInstanceOf[Repartition] ||
+ un.isInstanceOf[RepartitionByExpression] ||
+ un.isInstanceOf[Sample] ||
+ un.isInstanceOf[Sort] ||
+ un.isInstanceOf[SortPartitions] ||
+ un.isInstanceOf[Subquery] ||
+ un.isInstanceOf[With] ||
+ un.isInstanceOf[WithWindowDefinition] =>
--- End diff --
It's correct to list all these here, but I can't image a case that those
except Filter are needed.
Subquery/With/WithWindowDefinition/Limit should be part of subquery, How to
use use Distinct/RedistributeData/Repartition/RepartitionByExpression in SQL
query? For DataFrame, we may not support adding missing attributes for Sort,
it's a little bit confusing.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
