peter-toth commented on a change in pull request #28318:
URL: https://github.com/apache/spark/pull/28318#discussion_r414359577
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/CTESubstitution.scala
##########
@@ -31,53 +31,43 @@ object CTESubstitution extends Rule[LogicalPlan] {
def apply(plan: LogicalPlan): LogicalPlan = {
LegacyBehaviorPolicy.withName(SQLConf.get.getConf(LEGACY_CTE_PRECEDENCE_POLICY))
match {
case LegacyBehaviorPolicy.EXCEPTION =>
- assertNoNameConflictsInCTE(plan, inTraverse = false)
- traverseAndSubstituteCTE(plan, inTraverse = false)
+ assertNoNameConflictsInCTE(plan)
+ traverseAndSubstituteCTE(plan)
case LegacyBehaviorPolicy.LEGACY =>
legacyTraverseAndSubstituteCTE(plan)
case LegacyBehaviorPolicy.CORRECTED =>
- traverseAndSubstituteCTE(plan, inTraverse = false)
+ traverseAndSubstituteCTE(plan)
}
}
/**
* Check the plan to be traversed has naming conflicts in nested CTE or not,
traverse through
- * child, innerChildren and subquery for the current plan.
+ * child, innerChildren and subquery expressions for the current plan.
*/
private def assertNoNameConflictsInCTE(
plan: LogicalPlan,
- inTraverse: Boolean,
- cteNames: Set[String] = Set.empty): Unit = {
- plan.foreach {
+ namesInChildren: Set[String] = Set.empty,
+ namesInExpressions: Set[String] = Set.empty): Unit = {
Review comment:
IMHO this trick is necessary to find ambiguous situations in one
traversal pass. Ambiguous here doesn't only mean the the same CTE name is used
again in a nested CTE, but it is also important that we don't want to throw an
exception when the legacy and the corrected substitution returns the same
result. E.g.
```
WITH t(c) AS (SELECT 1)
SELECT max(c) FROM (
WITH t(c) AS (SELECT 2)
SELECT * FROM t
)
```
returns 2 in both modes so we don't throw the exception.
----------------------------------------------------------------
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]
