anchovYu commented on code in PR #39508: URL: https://github.com/apache/spark/pull/39508#discussion_r1068953886
########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveReferencesInAggregate.scala: ########## @@ -0,0 +1,202 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import org.apache.spark.sql.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{Alias, Attribute, Expression, GetStructField, NamedExpression} +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression +import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, AppendColumns, LogicalPlan} +import org.apache.spark.sql.catalyst.trees.TreePattern.UNRESOLVED_ATTRIBUTE + +/** + * A virtual rule to resolve [[UnresolvedAttribute]] in [[Aggregate]]. It's only used by the real + * rule `ResolveReferences`. The column resolution order for [[Aggregate]] is: + * 1. Resolves the column to [[AttributeReference]] with the output of the child plan. This + * includes metadata columns as well. + * 2. Resolves the column to a literal function which is allowed to be invoked without braces, e.g. + * `SELECT col, current_date FROM t`. + * 3. If `Aggregate.aggregateExpressions` are all resolved, resolve GROUP BY alias and GROUP BY ALL + * for `Aggregate.groupingExpressions`: + * 3.1. If the grouping expressions contain an unresolved column whose name matches an alias in the + * SELECT list, resolves that unresolved column to the alias. This is to support SQL pattern + * like `SELECT a + b AS c, max(col) FROM t GROUP BY c`. + * 3.2. If the grouping expressions only have one single unresolved column named 'ALL', expanded it + * to include all non-aggregate columns in the SELECT list. This is to support SQL pattern like + * `SELECT col1, col2, agg_expr(...) FROM t GROUP BY ALL`. + * 4. Resolves the column in `Aggregate.aggregateExpressions` to [[LateralColumnAliasReference]] if + * it references the alias defined previously in the SELECT list. The rule + * `ResolveLateralColumnAliasReference` will further resolve [[LateralColumnAliasReference]] and + * rewrite the plan. This is to support SQL pattern like + * `SELECT col1 + 1 AS x, x + 1 AS y, y + 1 AS z FROM t`. + * 5. Resolves the column to outer references with the outer plan if we are resolving subquery + * expressions. + */ +object ResolveReferencesInAggregate extends SQLConfHelper with ColumnResolutionHelper { + def apply(a: Aggregate): Aggregate = { + val planForResolve = a.child match { + // SPARK-25942: Resolves aggregate expressions with `AppendColumns`'s children, instead of + // `AppendColumns`, because `AppendColumns`'s serializer might produce conflict attribute + // names leading to ambiguous references exception. + case appendColumns: AppendColumns => appendColumns + case _ => a + } + + val resolvedAggExprsNoOuter = a.aggregateExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false) + .asInstanceOf[NamedExpression]) + + val resolvedGroupingExprs = a.groupingExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false)) + // SPARK-31670: Resolve Struct field in groupByExpressions and aggregateExpressions + // with CUBE/ROLLUP will be wrapped with alias like Alias(GetStructField, name) with + // different ExprId. This cause aggregateExpressions can't be replaced by expanded + // groupByExpressions in `ResolveGroupingAnalytics.constructAggregateExprs()`, we trim + // unnecessary alias of GetStructField here. + .map(trimTopLevelGetStructFieldAlias) + + // Only makes sense to do the rewrite once all the aggregate expressions have been resolved. + // Otherwise, we might incorrectly pull an actual aggregate expression over to the grouping + // expression list (because we don't know they would be aggregate expressions until resolved). + if (resolvedAggExprsNoOuter.forall(_.resolved)) { + val finalGroupExprs = resolveGroupByAll( + resolvedAggExprsNoOuter, + resolveGroupByAlias(resolvedAggExprsNoOuter, resolvedGroupingExprs) Review Comment: As we discussed, `select a + 1 as b, b + 1 as c .. group by c` will fail as LCA now doesn't handle grouping expr. Need a followup if we want to support above query (which I think we should). ########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveReferencesInAggregate.scala: ########## @@ -0,0 +1,202 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import org.apache.spark.sql.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{Alias, Attribute, Expression, GetStructField, NamedExpression} +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression +import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, AppendColumns, LogicalPlan} +import org.apache.spark.sql.catalyst.trees.TreePattern.UNRESOLVED_ATTRIBUTE + +/** + * A virtual rule to resolve [[UnresolvedAttribute]] in [[Aggregate]]. It's only used by the real + * rule `ResolveReferences`. The column resolution order for [[Aggregate]] is: + * 1. Resolves the column to [[AttributeReference]] with the output of the child plan. This + * includes metadata columns as well. + * 2. Resolves the column to a literal function which is allowed to be invoked without braces, e.g. + * `SELECT col, current_date FROM t`. + * 3. If `Aggregate.aggregateExpressions` are all resolved, resolve GROUP BY alias and GROUP BY ALL + * for `Aggregate.groupingExpressions`: + * 3.1. If the grouping expressions contain an unresolved column whose name matches an alias in the + * SELECT list, resolves that unresolved column to the alias. This is to support SQL pattern + * like `SELECT a + b AS c, max(col) FROM t GROUP BY c`. + * 3.2. If the grouping expressions only have one single unresolved column named 'ALL', expanded it + * to include all non-aggregate columns in the SELECT list. This is to support SQL pattern like + * `SELECT col1, col2, agg_expr(...) FROM t GROUP BY ALL`. + * 4. Resolves the column in `Aggregate.aggregateExpressions` to [[LateralColumnAliasReference]] if + * it references the alias defined previously in the SELECT list. The rule + * `ResolveLateralColumnAliasReference` will further resolve [[LateralColumnAliasReference]] and + * rewrite the plan. This is to support SQL pattern like + * `SELECT col1 + 1 AS x, x + 1 AS y, y + 1 AS z FROM t`. + * 5. Resolves the column to outer references with the outer plan if we are resolving subquery + * expressions. + */ +object ResolveReferencesInAggregate extends SQLConfHelper with ColumnResolutionHelper { + def apply(a: Aggregate): Aggregate = { + val planForResolve = a.child match { + // SPARK-25942: Resolves aggregate expressions with `AppendColumns`'s children, instead of + // `AppendColumns`, because `AppendColumns`'s serializer might produce conflict attribute + // names leading to ambiguous references exception. + case appendColumns: AppendColumns => appendColumns + case _ => a + } + + val resolvedAggExprsNoOuter = a.aggregateExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false) + .asInstanceOf[NamedExpression]) + + val resolvedGroupingExprs = a.groupingExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false)) + // SPARK-31670: Resolve Struct field in groupByExpressions and aggregateExpressions + // with CUBE/ROLLUP will be wrapped with alias like Alias(GetStructField, name) with + // different ExprId. This cause aggregateExpressions can't be replaced by expanded + // groupByExpressions in `ResolveGroupingAnalytics.constructAggregateExprs()`, we trim + // unnecessary alias of GetStructField here. + .map(trimTopLevelGetStructFieldAlias) + + // Only makes sense to do the rewrite once all the aggregate expressions have been resolved. + // Otherwise, we might incorrectly pull an actual aggregate expression over to the grouping + // expression list (because we don't know they would be aggregate expressions until resolved). + if (resolvedAggExprsNoOuter.forall(_.resolved)) { + val finalGroupExprs = resolveGroupByAll( Review Comment: Add tests for this resolution order: 1) resolve group by alias 2) resolve group by all. E.g. `select .. as all .. group by all` ########## sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/analysis/ResolveReferencesInAggregate.scala: ########## @@ -0,0 +1,202 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one or more + * contributor license agreements. See the NOTICE file distributed with + * this work for additional information regarding copyright ownership. + * The ASF licenses this file to You under the Apache License, Version 2.0 + * (the "License"); you may not use this file except in compliance with + * the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +package org.apache.spark.sql.catalyst.analysis + +import org.apache.spark.sql.catalyst.SQLConfHelper +import org.apache.spark.sql.catalyst.expressions.{Alias, Attribute, Expression, GetStructField, NamedExpression} +import org.apache.spark.sql.catalyst.expressions.aggregate.AggregateExpression +import org.apache.spark.sql.catalyst.plans.logical.{Aggregate, AppendColumns, LogicalPlan} +import org.apache.spark.sql.catalyst.trees.TreePattern.UNRESOLVED_ATTRIBUTE + +/** + * A virtual rule to resolve [[UnresolvedAttribute]] in [[Aggregate]]. It's only used by the real + * rule `ResolveReferences`. The column resolution order for [[Aggregate]] is: + * 1. Resolves the column to [[AttributeReference]] with the output of the child plan. This + * includes metadata columns as well. + * 2. Resolves the column to a literal function which is allowed to be invoked without braces, e.g. + * `SELECT col, current_date FROM t`. + * 3. If `Aggregate.aggregateExpressions` are all resolved, resolve GROUP BY alias and GROUP BY ALL + * for `Aggregate.groupingExpressions`: + * 3.1. If the grouping expressions contain an unresolved column whose name matches an alias in the + * SELECT list, resolves that unresolved column to the alias. This is to support SQL pattern + * like `SELECT a + b AS c, max(col) FROM t GROUP BY c`. + * 3.2. If the grouping expressions only have one single unresolved column named 'ALL', expanded it + * to include all non-aggregate columns in the SELECT list. This is to support SQL pattern like + * `SELECT col1, col2, agg_expr(...) FROM t GROUP BY ALL`. + * 4. Resolves the column in `Aggregate.aggregateExpressions` to [[LateralColumnAliasReference]] if + * it references the alias defined previously in the SELECT list. The rule + * `ResolveLateralColumnAliasReference` will further resolve [[LateralColumnAliasReference]] and + * rewrite the plan. This is to support SQL pattern like + * `SELECT col1 + 1 AS x, x + 1 AS y, y + 1 AS z FROM t`. + * 5. Resolves the column to outer references with the outer plan if we are resolving subquery + * expressions. + */ +object ResolveReferencesInAggregate extends SQLConfHelper with ColumnResolutionHelper { + def apply(a: Aggregate): Aggregate = { + val planForResolve = a.child match { + // SPARK-25942: Resolves aggregate expressions with `AppendColumns`'s children, instead of + // `AppendColumns`, because `AppendColumns`'s serializer might produce conflict attribute + // names leading to ambiguous references exception. + case appendColumns: AppendColumns => appendColumns + case _ => a + } + + val resolvedAggExprsNoOuter = a.aggregateExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false) + .asInstanceOf[NamedExpression]) + + val resolvedGroupingExprs = a.groupingExpressions + .map(resolveExpressionByPlanChildren(_, planForResolve, allowOuter = false)) + // SPARK-31670: Resolve Struct field in groupByExpressions and aggregateExpressions + // with CUBE/ROLLUP will be wrapped with alias like Alias(GetStructField, name) with + // different ExprId. This cause aggregateExpressions can't be replaced by expanded + // groupByExpressions in `ResolveGroupingAnalytics.constructAggregateExprs()`, we trim + // unnecessary alias of GetStructField here. + .map(trimTopLevelGetStructFieldAlias) + + // Only makes sense to do the rewrite once all the aggregate expressions have been resolved. + // Otherwise, we might incorrectly pull an actual aggregate expression over to the grouping + // expression list (because we don't know they would be aggregate expressions until resolved). + if (resolvedAggExprsNoOuter.forall(_.resolved)) { + val finalGroupExprs = resolveGroupByAll( + resolvedAggExprsNoOuter, + resolveGroupByAlias(resolvedAggExprsNoOuter, resolvedGroupingExprs) + ).map(resolveOuterRef) + a.copy(finalGroupExprs, resolvedAggExprsNoOuter, a.child) + } else { + // If the SELECT list is not full resolved at this point, we need to apply lateral column + // alias and outer reference resolution, which are not supported in GROUP BY. We can't + // resolve group by alias and group by all here. + // Aggregate supports Lateral column alias, which has higher priority than outer reference. + val resolvedAggExprsWithLCA = resolveLateralColumnAlias(resolvedAggExprsNoOuter) + val resolvedAggExprsWithOuter = resolvedAggExprsWithLCA.map(resolveOuterRef) + .map(_.asInstanceOf[NamedExpression]) + a.copy(resolvedGroupingExprs.map(resolveOuterRef), resolvedAggExprsWithOuter, a.child) Review Comment: It can be problematic, try with this query: ``` sql(s"select * from values(1) as t(something) " + s"where exists (select salary * 1.5 as something from $testTable group by something)") ``` where testTable has a salary field. (I just wrote in LCA suite.) -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
