[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...

[frreiss <gi...@git.apache.org>]

Github user frreiss commented on a diff in the pull request:

    https://github.com/apache/spark/pull/13155#discussion_r65454461
  
    --- Diff: sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala ---
    @@ -1695,16 +1695,176 @@ object RewriteCorrelatedScalarSubquery extends Rule[LogicalPlan] {
       }
     
       /**
    +   * Statically evaluate an expression containing zero or more placeholders, given a set
    +   * of bindings for placeholder values.
    +   */
    +  private def evalExpr(expr : Expression, bindings : Map[Long, Option[Any]]) : Option[Any] = {
    +    val rewrittenExpr = expr transform {
    +      case r @ AttributeReference(_, dataType, _, _) =>
    +        bindings(r.exprId.id) match {
    +          case Some(v) => Literal.create(v, dataType)
    +          case None => Literal.default(NullType)
    +        }
    +    }
    +    Option(rewrittenExpr.eval())
    +  }
    +
    +  /**
    +   * Statically evaluate an expression containing one or more aggregates on an empty input.
    +   */
    +  private def evalAggOnZeroTups(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) =>
    +        aggFunc.defaultResult.getOrElse(Literal.default(NullType))
    +    }
    +    Option(rewrittenExpr.eval())
    +  }
    +
    +  /**
    +   * Statically evaluate a scalar subquery on an empty input.
    +   *
    +   * <b>WARNING:</b> This method only covers subqueries that pass the checks under
    +   * [[org.apache.spark.sql.catalyst.analysis.CheckAnalysis]]. If the checks in
    +   * CheckAnalysis become less restrictive, this method will need to change.
    +   */
    +  private def evalSubqueryOnZeroTups(plan: LogicalPlan) : Option[Any] = {
    +    // Inputs to this method will start with a chain of zero or more SubqueryAlias
    +    // and Project operators, followed by an optional Filter, followed by an
    +    // Aggregate. Traverse the operators recursively.
    +    def evalPlan(lp : LogicalPlan) : Map[Long, Option[Any]] = {
    +      lp match {
    +        case SubqueryAlias(_, child) => evalPlan(child)
    +        case Filter(condition, child) =>
    +          val bindings = evalPlan(child)
    +          if (bindings.size == 0) bindings
    +          else {
    +            val exprResult = evalExpr(condition, bindings).getOrElse(false)
    +              .asInstanceOf[Boolean]
    +            if (exprResult) bindings else Map()
    +          }
    +
    +        case Project(projectList, child) =>
    +          val bindings = evalPlan(child)
    +          if (bindings.size == 0) {
    +            bindings
    +          } else {
    +            projectList.map(ne => (ne.exprId.id, evalExpr(ne, bindings))).toMap
    +          }
    +
    +        case Aggregate(_, aggExprs, _) =>
    +          // Some of the expressions under the Aggregate node are the join columns
    +          // for joining with the outer query block. Fill those expressions in with
    +          // nulls and statically evaluate the remainder.
    +          aggExprs.map(ne => ne match {
    +            case AttributeReference(_, _, _, _) => (ne.exprId.id, None)
    +            case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId.id, None)
    +            case _ => (ne.exprId.id, evalAggOnZeroTups(ne))
    +          }).toMap
    +
    +        case _ => sys.error(s"Unexpected operator in scalar subquery: $lp")
    +      }
    +    }
    +
    +    val resultMap = evalPlan(plan)
    +
    +    // By convention, the scalar subquery result is the leftmost field.
    +    resultMap(plan.output.head.exprId.id)
    +  }
    +
    +  /**
    +   * Split the plan for a scalar subquery into the parts above the Aggregate node
    +   * (first part of returned value) and the parts below the Aggregate node, including
    +   * the Aggregate (second part of returned value)
    +   */
    +  private def splitSubquery(plan : LogicalPlan) : Tuple2[Seq[LogicalPlan], Aggregate] = {
    --- End diff --
    
    Fixed in my local copy.


---
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 infrastructure@apache.org or file a JIRA ticket
with INFRA.
---

---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscribe@spark.apache.org
For additional commands, e-mail: reviews-help@spark.apache.org