Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/15238#discussion_r80802568
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/planning/patterns.scala
---
@@ -191,25 +191,34 @@ object ExtractFiltersAndInnerJoins extends
PredicateHelper {
/**
* A pattern that collects all adjacent unions and returns their children
as a Seq.
+ * If the top union is wrapped in a [[Distinct]], then the [[Distinct]] in
the adjacent unions, if
+ * any, will be eliminated.
*/
object Unions {
- def unapply(plan: LogicalPlan): Option[Seq[LogicalPlan]] = plan match {
- case u: Union => Some(collectUnionChildren(mutable.Stack(u),
Seq.empty[LogicalPlan]))
+ def unapply(plan: LogicalPlan): Option[(Seq[LogicalPlan], Boolean)] =
plan match {
+ case u: Union =>
+ Some(collectUnionChildren(mutable.Stack(u), Seq.empty[LogicalPlan],
false), false)
+ case Distinct(u: Union) =>
+ Some(collectUnionChildren(mutable.Stack(u), Seq.empty[LogicalPlan],
true), true)
case _ => None
}
// Doing a depth-first tree traversal to combine all the union children.
@tailrec
private def collectUnionChildren(
plans: mutable.Stack[LogicalPlan],
- children: Seq[LogicalPlan]): Seq[LogicalPlan] = {
+ children: Seq[LogicalPlan],
+ dedupDistinct: Boolean): Seq[LogicalPlan] = {
--- End diff --
Ok, so I took a longer look. The current approach uses tail recursion with
a bunch of state. I tried rewriting this as a tail recursive function without
that much state, but that proved to be very painful. So it seems simpler to
just use a purely imperative function:
```scala
private def flattenUnion(union: Union, flattenDistinct: Boolean): Union = {
val stack = mutable.Stack[LogicalPlan](union)
val flattened = mutable.ArrayBuffer.empty[LogicalPlan]
while (stack.nonEmpty) {
stack.pop() match {
case Distinct(Union(children)) if flattenDistinct =>
stack.pushAll(children)
case Union(children) =>
stack.pushAll(children)
case child =>
flattened += child
}
}
Union(flattened)
}
```
@gatorsmile why did you use `reverseMap`?
---
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 infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
