Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r63933593
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1648,16 +1648,56 @@ object RewriteCorrelatedScalarSubquery extends
Rule[LogicalPlan] {
}
/**
+ * Statically evaluate an expression containing one or more aggregates
on an empty input.
+ */
+ private def evalOnZeroTups(expr : Expression) : Option[Any] = {
+ // AggregateExpressions are Unevaluable, so we need to replace all
aggregates
+ // in the expression with the value they would return for zero input
tuples.
+ val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+ val resultLit = aggFunc.defaultResult match {
+ case Some(lit) => lit
+ case None => Literal.default(NullType)
+ }
+ Alias(resultLit, "aggVal") (exprId = resultId)
+ }
+ Option(rewrittenExpr.eval())
+ }
+
+ /**
* Construct a new child plan by left joining the given subqueries to a
base plan.
*/
private def constructLeftJoins(
child: LogicalPlan,
subqueries: ArrayBuffer[ScalarSubquery]): LogicalPlan = {
subqueries.foldLeft(child) {
case (currentChild, ScalarSubquery(query, conditions, _)) =>
+ val aggOutputExpr =
query.asInstanceOf[Aggregate].aggregateExpressions.head
--- End diff --
Upon further reflection, a Filter node above the Aggregate creates
additional issues. If you rewrite the subquery into an outer join above the
Filter, it may be impossible to distinguish between two cases:
- The original subquery returns null because an aggregate value did not
pass the HAVING clause
- The aggregate in the original subquery is evaluated over zero tuples
In both cases, the outer join will return a tuple containing nulls.
For example, the query:
```sql
select * from l where (select count(*) cnt from r where l.a = r.c having
cnt = 0) = 0
```
would turn into
```sql
select *
from l left outer join (select c, count(*) cnt from r group by c having cnt
= 0) sq
on l.a = sq.c
where sq.cnt = 0
```
Note how the rewritten subquery `select c, count(*) cnt from r group by c
having cnt = 0` always returns zero tuples.
I'm working through a few possible solutions, trying to see if one of them
is guaranteed to be correct.
---
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
