[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user asfgit closed the pull request at: https://github.com/apache/spark/pull/13155 --- 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66682369
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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, None)
+case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId,
None)
+case _ => (ne.exprId, 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)
+ }
+
+ /**
+ * Split the plan for a scalar subquery into the parts above the
innermost query block
+ * (first part of returned value), the HAVING clause of the innermost
query block
+ * (optional second part) and the parts below the HAVING CLAUSE (third
part).
+ */
+ private def splitSubquery(plan: LogicalPlan) : (Seq[LogicalPlan],
Option[Filter], Aggregate) = {
+val topPart = ArrayBuffer.empty[LogicalPlan]
+var bottomPart : LogicalPlan = plan
+while (true) {
+ bottomPart match {
+case havingPart@Filter(_, aggPart@Aggregate(_, _, _)) =>
+ return (topPart, Option(havingPart),
aggPart.asInstanceOf[Aggregate])
+
+case aggPart@Aggregate(_, _, _) =>
+ // No HAVING clause
+ return (topPart, None, aggPart)
+
+case p@Project(_, child) =>
+
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66682052
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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, None)
+case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId,
None)
+case _ => (ne.exprId, 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)
+ }
+
+ /**
+ * Split the plan for a scalar subquery into the parts above the
innermost query block
+ * (first part of returned value), the HAVING clause of the innermost
query block
+ * (optional second part) and the parts below the HAVING CLAUSE (third
part).
+ */
+ private def splitSubquery(plan: LogicalPlan) : (Seq[LogicalPlan],
Option[Filter], Aggregate) = {
+val topPart = ArrayBuffer.empty[LogicalPlan]
+var bottomPart : LogicalPlan = plan
+while (true) {
+ bottomPart match {
+case havingPart@Filter(_, aggPart@Aggregate(_, _, _)) =>
+ return (topPart, Option(havingPart),
aggPart.asInstanceOf[Aggregate])
+
+case aggPart@Aggregate(_, _, _) =>
+ // No HAVING clause
+ return (topPart, None, aggPart)
+
+case p@Project(_, child) =>
+
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66681873
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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, None)
+case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId,
None)
+case _ => (ne.exprId, 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)
+ }
+
+ /**
+ * Split the plan for a scalar subquery into the parts above the
innermost query block
+ * (first part of returned value), the HAVING clause of the innermost
query block
+ * (optional second part) and the parts below the HAVING CLAUSE (third
part).
+ */
+ private def splitSubquery(plan: LogicalPlan) : (Seq[LogicalPlan],
Option[Filter], Aggregate) = {
+val topPart = ArrayBuffer.empty[LogicalPlan]
+var bottomPart : LogicalPlan = plan
+while (true) {
+ bottomPart match {
+case havingPart@Filter(_, aggPart@Aggregate(_, _, _)) =>
+ return (topPart, Option(havingPart),
aggPart.asInstanceOf[Aggregate])
+
+case aggPart@Aggregate(_, _, _) =>
+ // No HAVING clause
+ return (topPart, None, aggPart)
+
+case p@Project(_, child) =>
+
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66681029
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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, None)
+case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId,
None)
+case _ => (ne.exprId, 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)
+ }
+
+ /**
+ * Split the plan for a scalar subquery into the parts above the
innermost query block
+ * (first part of returned value), the HAVING clause of the innermost
query block
+ * (optional second part) and the parts below the HAVING CLAUSE (third
part).
+ */
+ private def splitSubquery(plan: LogicalPlan) : (Seq[LogicalPlan],
Option[Filter], Aggregate) = {
+val topPart = ArrayBuffer.empty[LogicalPlan]
+var bottomPart : LogicalPlan = plan
+while (true) {
+ bottomPart match {
+case havingPart@Filter(_, aggPart@Aggregate(_, _, _)) =>
--- End diff --
style spaces between @ elements, i.e.: `havingPart @ Filter`
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66680913
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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, None)
+case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId,
None)
+case _ => (ne.exprId, 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)
+ }
+
+ /**
+ * Split the plan for a scalar subquery into the parts above the
innermost query block
+ * (first part of returned value), the HAVING clause of the innermost
query block
+ * (optional second part) and the parts below the HAVING CLAUSE (third
part).
+ */
+ private def splitSubquery(plan: LogicalPlan) : (Seq[LogicalPlan],
Option[Filter], Aggregate) = {
+val topPart = ArrayBuffer.empty[LogicalPlan]
+var bottomPart : LogicalPlan = plan
--- End diff --
style: no space before colon
---
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.
---
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66680885
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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, None)
+case Alias(AttributeReference(_, _, _, _), _) => (ne.exprId,
None)
--- End diff --
style: `case Alias(_: AttributeReference, _) =>`
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66680828
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
+ lp match {
+case SubqueryAlias(_, child) => evalPlan(child)
+case Filter(condition, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) bindings
+ else {
+val exprResult = evalExpr(condition, bindings).getOrElse(false)
+ .asInstanceOf[Boolean]
+if (exprResult) bindings else Map.empty
+ }
+
+case Project(projectList, child) =>
+ val bindings = evalPlan(child)
+ if (bindings.isEmpty) {
+bindings
+ } else {
+projectList.map(ne => (ne.exprId, 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 {
--- End diff --
style: do the pattern match directly, i.e.:
```scala
aggExpr.map {
case _: AttributeReference =>
case ...
}
```
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66680599
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
+}
+Option(rewrittenExpr.eval())
+ }
+
+ /**
+ * Statically evaluate a scalar subquery on an empty input.
+ *
+ * WARNING: 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[ExprId, Option[Any]] = {
--- End diff --
style: no space before colon
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66680557
--- Diff:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
---
@@ -1695,16 +1696,205 @@ 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[ExprId,
Option[Any]]) : Option[Any] = {
+val rewrittenExpr = expr transform {
+ case r @ AttributeReference(_, dataType, _, _) =>
+bindings(r.exprId) 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.
+// Also replace attribute refs (for example, for grouping columns)
with NULL.
+val rewrittenExpr = expr transform {
+ case a @ AggregateExpression(aggFunc, _, _, resultId) =>
+aggFunc.defaultResult.getOrElse(Literal.default(NullType))
+
+ case AttributeReference(_, _, _, _) => Literal.default(NullType)
--- End diff --
style: `case _: AttributeReference => ...`
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66564793
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
+var bottomPart : LogicalPlan = plan
+while (! bottomPart.isInstanceOf[Aggregate]) {
+ topPart = bottomPart :: topPart
+ bottomPart = bottomPart.children.head
+}
+(topPart, bottomPart.asInstanceOf[Aggregate])
+ }
+
+ /**
+ * Rewrite the nodes above the Aggregate in a subquery so that they
generate an
+ * auxiliary column "isFiltered"
+ * @param subqueryPlan plan before rewrite
+ * @param filteredId expression ID for the "isFiltered" column
+ */
+ private def addIsFiltered(subqueryPlan : LogicalPlan, filteredId :
ExprId) : LogicalPlan = {
+val
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66561815
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
+var bottomPart : LogicalPlan = plan
+while (! bottomPart.isInstanceOf[Aggregate]) {
+ topPart = bottomPart :: topPart
+ bottomPart = bottomPart.children.head
+}
+(topPart, bottomPart.asInstanceOf[Aggregate])
+ }
+
+ /**
+ * Rewrite the nodes above the Aggregate in a subquery so that they
generate an
+ * auxiliary column "isFiltered"
+ * @param subqueryPlan plan before rewrite
+ * @param filteredId expression ID for the "isFiltered" column
+ */
+ private def addIsFiltered(subqueryPlan : LogicalPlan, filteredId :
ExprId) : LogicalPlan = {
+val
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66561017
--- 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 {
--- End diff --
Yes, that is true. Added a test case and an additional clause in that case
statement 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 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66560947
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
+var bottomPart : LogicalPlan = plan
+while (! bottomPart.isInstanceOf[Aggregate]) {
+ topPart = bottomPart :: topPart
+ bottomPart = bottomPart.children.head
+}
+(topPart, bottomPart.asInstanceOf[Aggregate])
+ }
+
+ /**
+ * Rewrite the nodes above the Aggregate in a subquery so that they
generate an
+ * auxiliary column "isFiltered"
+ * @param subqueryPlan plan before rewrite
+ * @param filteredId expression ID for the "isFiltered" column
+ */
+ private def addIsFiltered(subqueryPlan : LogicalPlan, filteredId :
ExprId) : LogicalPlan = {
+val
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66560868
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
+var bottomPart : LogicalPlan = plan
+while (! bottomPart.isInstanceOf[Aggregate]) {
+ topPart = bottomPart :: topPart
+ bottomPart = bottomPart.children.head
+}
+(topPart, bottomPart.asInstanceOf[Aggregate])
+ }
+
+ /**
+ * Rewrite the nodes above the Aggregate in a subquery so that they
generate an
+ * auxiliary column "isFiltered"
+ * @param subqueryPlan plan before rewrite
+ * @param filteredId expression ID for the "isFiltered" column
+ */
+ private def addIsFiltered(subqueryPlan : LogicalPlan, filteredId :
ExprId) : LogicalPlan = {
+val
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66558119
--- 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.
+ *
+ * WARNING: 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
--- 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 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66558125
--- 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.
+ *
+ * WARNING: 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()
--- 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 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user hvanhovell commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66540105
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
+var bottomPart : LogicalPlan = plan
+while (! bottomPart.isInstanceOf[Aggregate]) {
+ topPart = bottomPart :: topPart
+ bottomPart = bottomPart.children.head
+}
+(topPart, bottomPart.asInstanceOf[Aggregate])
+ }
+
+ /**
+ * Rewrite the nodes above the Aggregate in a subquery so that they
generate an
+ * auxiliary column "isFiltered"
+ * @param subqueryPlan plan before rewrite
+ * @param filteredId expression ID for the "isFiltered" column
+ */
+ private def addIsFiltered(subqueryPlan : LogicalPlan, filteredId :
ExprId) : LogicalPlan = {
+val
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66539170
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
+var bottomPart : LogicalPlan = plan
+while (! bottomPart.isInstanceOf[Aggregate]) {
+ topPart = bottomPart :: topPart
+ bottomPart = bottomPart.children.head
+}
+(topPart, bottomPart.asInstanceOf[Aggregate])
+ }
+
+ /**
+ * Rewrite the nodes above the Aggregate in a subquery so that they
generate an
+ * auxiliary column "isFiltered"
+ * @param subqueryPlan plan before rewrite
+ * @param filteredId expression ID for the "isFiltered" column
+ */
+ private def addIsFiltered(subqueryPlan : LogicalPlan, filteredId :
ExprId) : LogicalPlan = {
+val
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66509510
--- 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.
+ *
+ * WARNING: 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] = {
+var topPart = List[LogicalPlan]()
--- 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 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66509444
--- 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.
+ *
+ * WARNING: 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()
--- 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 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66509405
--- 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.
+ *
+ * WARNING: 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
--- 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 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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r66478261
--- 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] = {
--- End diff --
No particular reason for using `ExprId.id`. I'll change to using the entire
class as a key.
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r65584546
--- 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.
+ *
+ * WARNING: 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)
--- End diff --
The join attributes that the Analyzer adds to the Aggregate node are
AttributeReference nodes, and the `evalAggOnZeroTups` method as currently
written can't process them. A more comprehensive facility for statically
evaluating expressions would be nice to have; but I'm hesitant to add such a
mechanism as part of a bug fix. Perhaps a follow-on JIRA is in order?
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
Github user frreiss commented on a diff in the pull request:
https://github.com/apache/spark/pull/13155#discussion_r65583548
--- 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.
+ *
+ * WARNING: 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]
--- End diff --
The test on the next line doesn't compile without the cast. The condition
in a Filter node is of type Expression, and `Expresssion.eval()` returns `Any`.
---
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
[GitHub] spark pull request #13155: [SPARK-15370] [SQL] Update RewriteCorrelatedScala...
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.
+ *
+ * WARNING: 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 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
