Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/3249#discussion_r22004962
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/Analyzer.scala
---
@@ -314,6 +318,113 @@ class Analyzer(catalog: Catalog, registry:
FunctionRegistry, caseSensitive: Bool
protected def containsStar(exprs: Seq[Expression]): Boolean =
exprs.collect { case _: Star => true }.nonEmpty
}
+
+ /**
+ * Transforms the query which has subquery expressions in where clause
to join queries.
+ * Case 1 Uncorelated queries
+ * -- original query
+ * select C from R1 where R1.A in (Select B from R2)
+ * -- rewritten query
+ * Select C from R1 left semi join (select B as sqc0 from R2) subquery
on R1.A = subquery.sqc0
+ *
+ * Case 2 Corelated queries
+ * -- original query
+ * select C from R1 where R1.A in (Select B from R2 where R1.X = R2.Y)
+ * -- rewritten query
+ * select C from R1 left semi join (select B as sqc0, R2.Y as sqc1 from
R2) subquery
+ * on R1.X = subquery.sqc1 and R1.A = subquery.sqc0
+ *
+ * Refer:
https://issues.apache.org/jira/secure/attachment/12614003/SubQuerySpec.pdf
+ */
+ object SubQueryExpressions extends Rule[LogicalPlan] {
+
+ def apply(plan: LogicalPlan): LogicalPlan = plan transform {
+ case filter @ Filter(conditions, child) =>
+ val subqueryExprs = new
scala.collection.mutable.ArrayBuffer[SubqueryExpression]()
+ val nonSubQueryConds = new
scala.collection.mutable.ArrayBuffer[Expression]()
+ val transformedConds = conditions.transform{
+ // Replace with dummy
+ case s @ SubqueryExpression(exp,subquery) =>
+ subqueryExprs += s
+ Literal(true)
+ }
+ if (subqueryExprs.size == 1) {
+ val subqueryExpr = subqueryExprs.remove(0)
+ createLeftSemiJoin(
+ child, subqueryExpr.value,
+ subqueryExpr.subquery, transformedConds)
+ } else if (subqueryExprs.size > 1) {
+ // Only one subquery expression is supported.
+ throw new TreeNodeException(filter, "Only 1 SubQuery expression
is supported.")
+ } else {
+ filter
+ }
+ }
+
+ /**
+ * Create LeftSemi join with parent query to the subquery which is
mentioned in 'IN' predicate
+ * And combine the subquery conditions and parent query conditions.
+ */
+ def createLeftSemiJoin(left: LogicalPlan,
+ value: Expression, subquery: LogicalPlan,
+ parentConds: Expression) : LogicalPlan = {
+ val (transformedPlan, subqueryConds) = transformAndGetConditions(
+ value, subquery)
+ // Unify the parent query conditions and subquery conditions and add
these as join conditions
+ val unifyConds = And(parentConds, subqueryConds)
+ Join(left, transformedPlan, LeftSemi, Some(unifyConds))
+ }
+
+ /**
+ * Transform the subquery LogicalPlan and add the expressions which
are used as filters to the
+ * projection. And also return filter conditions used in subquery
+ */
+ def transformAndGetConditions(value: Expression,
+ subquery: LogicalPlan): (LogicalPlan, Expression) = {
+ val expr = new scala.collection.mutable.ArrayBuffer[Expression]()
+ val transformedPlan = subquery transform {
+ case project @ Project(projectList, f @ Filter(condition, child))
=>
+ // Don't support more than 1 item in select list of subquery
+ if(projectList.size > 1) {
+ throw new TreeNodeException(project, "SubQuery can contain
only 1 item in Select List")
+ }
+ val resolvedChild = ResolveRelations(child)
+ // Add the expressions to the projections which are used as
filters in subquery
+ val toBeAddedExprs = f.references.filter(
+ a=>resolvedChild.resolve(a.name, resolver) != None &&
!projectList.contains(a))
--- End diff --
A few things here
- I'd add the guard above so we don't have to resolve here. Hand
resolution like you are doing here will work for simple cases but doesn't run
to fixed point.
- Never compare attributes by name, and don't inspect the project list to
see what the output will be. The reasons are as follows: names can be
ambiguous and the project list could have aliases and other things that confuse
the inspection. Instead use `outputSet` which returns an `AttributeSet` that
does comparisons correctly.
- if you need multi line lambas, prefer the following syntax:
```
list.filter { a =>
a....
}
```
All together I think you want something like:
```scala
f.references.filterNot(project.outputSet.contains)
```
---
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
