viirya commented on a change in pull request #28898:
URL: https://github.com/apache/spark/pull/28898#discussion_r448753272
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/NestedColumnAliasing.scala
##########
@@ -172,13 +188,23 @@ object NestedColumnAliasing {
(f, Alias(f, s"_gen_alias_${exprId.id}")(exprId, Seq.empty, None))
}
+
+ // Do deduplication based on semanticEquals, and then sum.
+ val nestedFieldNum = nestedFieldToAlias
+ .foldLeft(Seq[ExtractValue]()) {
+ (unique, curr) => if (!unique.exists(curr._1.semanticEquals(_))) {
+ curr._1 +: unique
+ } else {
+ unique
+ }
+ }
+ .map { t => totalFieldNum(t.dataType) }
+ .sum
Review comment:
The problem is your approach is, which one is selected from the
`ExtractValue`s with different qualifier, is non-deterministic.
Later when we query the `nestedFieldToAlias` map, you might fail to find the
corresponding item from the map due to qualifier difference.
I think the safer approach is to clean up all qualifier.
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/NestedColumnAliasing.scala
##########
@@ -172,13 +188,23 @@ object NestedColumnAliasing {
(f, Alias(f, s"_gen_alias_${exprId.id}")(exprId, Seq.empty, None))
}
+
+ // Do deduplication based on semanticEquals, and then sum.
+ val nestedFieldNum = nestedFieldToAlias
+ .foldLeft(Seq[ExtractValue]()) {
+ (unique, curr) => if (!unique.exists(curr._1.semanticEquals(_))) {
+ curr._1 +: unique
+ } else {
+ unique
+ }
+ }
+ .map { t => totalFieldNum(t.dataType) }
+ .sum
Review comment:
oh, I think it is also due different `qualifier` in the `ExtractValue`s.
As above, I think we can just clean up `qualifier`.
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/NestedColumnAliasing.scala
##########
@@ -172,13 +188,23 @@ object NestedColumnAliasing {
(f, Alias(f, s"_gen_alias_${exprId.id}")(exprId, Seq.empty, None))
}
+
+ // Do deduplication based on semanticEquals, and then sum.
+ val nestedFieldNum = nestedFieldToAlias
+ .foldLeft(Seq[ExtractValue]()) {
+ (unique, curr) => if (!unique.exists(curr._1.semanticEquals(_))) {
+ curr._1 +: unique
+ } else {
+ unique
+ }
+ }
+ .map { t => totalFieldNum(t.dataType) }
+ .sum
Review comment:
Like:
```scala
def removeQualifier(f: ExtractValue): ExtractValue = {
f.transform {
case a: AttributeReference => a.withQualifier(Seq.empty)
}.asInstanceOf[ExtractValue]
}
```
```scala
val dedupNestedFields = nestedFields.filter {
case e @ (_: GetStructField | _: GetArrayStructFields) =>
val child = e.children.head
nestedFields.forall(f => child.find(_.semanticEquals(f)).isEmpty)
case _ => true
}.map(removeQualifier)
```
And when we need to query the map, we do:
```scala
nestedFieldToAlias.contains(removeQualifier(f))
```
or
```scala
nestedFieldToAlias(removeQualifier(f)).toAttribute
```
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/NestedColumnAliasing.scala
##########
@@ -152,7 +168,7 @@ object NestedColumnAliasing {
val exclusiveAttrSet = AttributeSet(exclusiveAttrs ++ otherRootReferences)
val aliasSub = nestedFieldReferences.asInstanceOf[Seq[ExtractValue]]
.filter(!_.references.subsetOf(exclusiveAttrSet))
- .groupBy(_.references.head)
+ .groupBy(t => (t.references.head.exprId, t.references.head.dataType))
Review comment:
We don't need to change `collectRootReferenceAndExtractValue`.
```scala
val aliasSub = nestedFieldReferences.asInstanceOf[Seq[ExtractValue]]
.filter(!_.references.subsetOf(exclusiveAttrSet))
.groupBy(_.references.head.withQualifier(Seq.empty))
...
```
----------------------------------------------------------------
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:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
