silundong commented on code in PR #4619: URL: https://github.com/apache/calcite/pull/4619#discussion_r2563400708
########## core/src/main/java/org/apache/calcite/sql2rel/TopDownGeneralDecorrelator.java: ########## @@ -0,0 +1,1009 @@ +/* + * 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.calcite.sql2rel; + +import org.apache.calcite.plan.RelOptUtil; +import org.apache.calcite.plan.Strong; +import org.apache.calcite.plan.hep.HepPlanner; +import org.apache.calcite.plan.hep.HepProgram; +import org.apache.calcite.rel.RelCollation; +import org.apache.calcite.rel.RelNode; +import org.apache.calcite.rel.core.Aggregate; +import org.apache.calcite.rel.core.AggregateCall; +import org.apache.calcite.rel.core.Correlate; +import org.apache.calcite.rel.core.CorrelationId; +import org.apache.calcite.rel.core.Filter; +import org.apache.calcite.rel.core.Join; +import org.apache.calcite.rel.core.JoinRelType; +import org.apache.calcite.rel.core.Project; +import org.apache.calcite.rel.core.SetOp; +import org.apache.calcite.rel.core.Sort; +import org.apache.calcite.rel.rules.CoreRules; +import org.apache.calcite.rel.type.RelDataType; +import org.apache.calcite.rex.RexCall; +import org.apache.calcite.rex.RexCorrelVariable; +import org.apache.calcite.rex.RexFieldAccess; +import org.apache.calcite.rex.RexInputRef; +import org.apache.calcite.rex.RexNode; +import org.apache.calcite.rex.RexShuttle; +import org.apache.calcite.rex.RexUtil; +import org.apache.calcite.rex.RexWindow; +import org.apache.calcite.sql.SqlAggFunction; +import org.apache.calcite.sql.SqlKind; +import org.apache.calcite.sql.fun.SqlCountAggFunction; +import org.apache.calcite.sql.fun.SqlStdOperatorTable; +import org.apache.calcite.sql2rel.RelDecorrelator.CorDef; +import org.apache.calcite.sql2rel.RelDecorrelator.Frame; +import org.apache.calcite.tools.RelBuilder; +import org.apache.calcite.util.ImmutableBitSet; +import org.apache.calcite.util.Litmus; +import org.apache.calcite.util.Pair; +import org.apache.calcite.util.ReflectUtil; +import org.apache.calcite.util.ReflectiveVisitor; +import org.apache.calcite.util.mapping.Mappings; + +import com.google.common.collect.ImmutableList; +import com.google.common.collect.ImmutableSet; + +import org.checkerframework.checker.nullness.qual.Nullable; + +import java.util.ArrayList; +import java.util.Collections; +import java.util.HashMap; +import java.util.List; +import java.util.Map; +import java.util.NavigableMap; +import java.util.NavigableSet; +import java.util.Set; +import java.util.TreeMap; +import java.util.TreeSet; +import java.util.stream.IntStream; + +import static java.util.Objects.requireNonNull; + +/** + * A top‑down, generic decorrelation algorithm that can handle deep nestings of correlated + * subqueries and that generalizes to complex query constructs. More details are in paper: + * <a href="https://dl.gi.de/items/b9df4765-d1b0-4267-a77c-4ce4ab0ee62d";> + * Improving Unnesting of Complex Queries</a> + */ +public class TopDownGeneralDecorrelator implements ReflectiveVisitor { + + private final RelBuilder builder; + + // record the CorDef in the current context (including those in the parent Correlate). + private final NavigableSet<CorDef> corDefs; + + // a map from RelNode to whether existing correlated expressions (according to corDefs). + private final Map<RelNode, Boolean> hasCorrelatedExpressions; + + // a map from RelNode to its UnnestQuery. + private final Map<RelNode, UnnestQuery> mapRelToUnnestQuery; + + private final boolean hasParent; + + // whether allow empty output resulting from decorrelate rewriting. + private boolean emptyOutputDisable; + + // the domain of the free variables (i.e. corDefs) D, it's duplicate free. + private DedupFreeVarsNode dedupFreeVarsNode; + + @SuppressWarnings("method.invocation.invalid") + private final ReflectUtil.MethodDispatcher<RelNode> dispatcher = + ReflectUtil.createMethodDispatcher( + RelNode.class, getVisitor(), "unnestInternal", RelNode.class); + + @SuppressWarnings("initialization.fields.uninitialized") + public TopDownGeneralDecorrelator( + RelBuilder builder, + boolean hasParent, + boolean emptyOutputDisable, + @Nullable Set<CorDef> parentCorDefs, + @Nullable Map<RelNode, Boolean> parentHasCorrelatedExpressions, + @Nullable Map<RelNode, UnnestQuery> parentMapRelToUnnestInfo) { + this.builder = builder; + this.hasParent = hasParent; + this.emptyOutputDisable = emptyOutputDisable; + this.corDefs = new TreeSet<>(); + if (parentCorDefs != null) { + this.corDefs.addAll(parentCorDefs); + } + this.hasCorrelatedExpressions = parentHasCorrelatedExpressions == null + ? new HashMap<>() + : parentHasCorrelatedExpressions; + this.mapRelToUnnestQuery = parentMapRelToUnnestInfo == null + ? new HashMap<>() + : parentMapRelToUnnestInfo; + } + + private static TopDownGeneralDecorrelator createEmptyDecorrelator(RelBuilder builder) { + return new TopDownGeneralDecorrelator(builder, false, false, null, null, null); + } + + private TopDownGeneralDecorrelator createSubDecorrelator() { + TopDownGeneralDecorrelator subDecorrelator = + new TopDownGeneralDecorrelator( + builder, + true, + emptyOutputDisable, + corDefs, + hasCorrelatedExpressions, + mapRelToUnnestQuery); + subDecorrelator.dedupFreeVarsNode = this.dedupFreeVarsNode; + return subDecorrelator; + } + + /** + * Decorrelates a query. This is the entry point for this class. + * + * @param rel Root node of the query + * @param builder RelBuilder + * @return Equivalent node without correlation + */ + public static RelNode decorrelateQuery(RelNode rel, RelBuilder builder) { + HepProgram preProgram = HepProgram.builder() + .addRuleCollection( + ImmutableList.of( + CoreRules.FILTER_PROJECT_TRANSPOSE, + CoreRules.FILTER_INTO_JOIN, + CoreRules.FILTER_CORRELATE)) + .build(); + HepPlanner prePlanner = new HepPlanner(preProgram); + prePlanner.setRoot(rel); + RelNode preparedRel = prePlanner.findBestExp(); + + // start decorrelating + TopDownGeneralDecorrelator decorrelator = createEmptyDecorrelator(builder); + RelNode decorrelateNode = rel; + try { + decorrelateNode = decorrelator.correlateElimination(preparedRel); + } catch (UnsupportedOperationException e) { + // if the correlation exists in an unsupported operator, retain the original plan. + } + + HepProgram postProgram = HepProgram.builder() + .addRuleCollection( + ImmutableList.of( + CoreRules.FILTER_PROJECT_TRANSPOSE, + CoreRules.FILTER_INTO_JOIN, + CoreRules.MARK_TO_SEMI_OR_ANTI_JOIN_RULE, + CoreRules.PROJECT_MERGE, + CoreRules.PROJECT_REMOVE)) + .build(); + HepPlanner postPlanner = new HepPlanner(postProgram); + postPlanner.setRoot(decorrelateNode); + return postPlanner.findBestExp(); + } + + /** + * Eliminates Correlate. + * + * @param rel RelNode + * @return Equivalent RelNode without Correlate + */ + private RelNode correlateElimination(RelNode rel) { + if (rel instanceof Correlate) { + final Correlate correlate = (Correlate) rel; + final RelNode newLeft; + if (hasParent) { + // if the current decorrelator has a parent, it means that the Correlate must have + // correlation from above. + assert hasCorrelatedExpressions.containsKey(correlate) + && hasCorrelatedExpressions.get(correlate); + newLeft = unnest(correlate.getLeft()); + } else { + // otherwise, start a new decorrelation for the left side. + newLeft = decorrelateQuery(correlate.getLeft(), builder); + } + + // create or update UnnestQuery of left side and corDefs of this decorrelator. + UnnestQuery leftInfo = mapRelToUnnestQuery.get(correlate.getLeft()); + TreeMap<CorDef, Integer> corDefOutputs = new TreeMap<>(); + Map<Integer, Integer> oldToNewOutputs = new HashMap<>(); + for (int i = 0; i < correlate.getLeft().getRowType().getFieldCount(); i++) { + int newColumnIndex = leftInfo == null ? i : requireNonNull(leftInfo.oldToNewOutputs.get(i)); + oldToNewOutputs.put(i, newColumnIndex); + if (correlate.getRequiredColumns().get(i)) { + CorDef corDef = new CorDef(correlate.getCorrelationId(), i); + corDefs.add(corDef); + corDefOutputs.put(corDef, newColumnIndex); + } + } + if (leftInfo != null) { + corDefOutputs.putAll(leftInfo.corDefOutputs); + } + leftInfo = new UnnestQuery(correlate.getLeft(), newLeft, corDefOutputs, oldToNewOutputs); + dedupFreeVarsNode = generateDedupFreeVarsNode(newLeft, leftInfo); + + // decorrelate right side + detectCorrelatedExpressions(correlate.getRight()); + emptyOutputDisable |= correlate.getJoinType() == JoinRelType.MARK; + RelNode newRight = unnest(correlate.getRight()); + UnnestQuery rightInfo = requireNonNull(mapRelToUnnestQuery.get(correlate.getRight())); + + // rewrite condition, adding the natural join condition between the left side and + // the domain D that in right side. + // + // Correlate(condition=[p]) + // / \ + // L ... with correlation + // \ + // R + // => + // Join(condition=[p AND (L is not distinct from D)]) + // / \ + // L ... without correlation + // \ + // x + // / \ + // D R + builder.push(newLeft).push(newRight); + RexNode unnestedJoinCondition = + createUnnestedJoinCondition(correlate.getCondition(), leftInfo, rightInfo, true); + RelNode unnestedRel = builder.join(correlate.getJoinType(), unnestedJoinCondition).build(); + + if (!hasParent) { + // ensure that the fields are in the same order as in the original plan. + builder.push(unnestedRel); + UnnestQuery unnestQuery = + createJoinUnnestInfo( + leftInfo, + rightInfo, + correlate, + unnestedRel, + correlate.getJoinType()); + List<RexNode> projects + = builder.fields(new ArrayList<>(unnestQuery.oldToNewOutputs.values())); + unnestedRel = builder.project(projects).build(); + } + return unnestedRel; + } else { + for (int i = 0; i < rel.getInputs().size(); i++) { + rel.replaceInput(i, correlateElimination(rel.getInput(i))); + } + } + return rel; + } + + /** + * Generate the domain of the free variables D. + * + * @param newLeft the left side (without correlation) of Correlate + * @param leftInfo the UnnestQuery of the left side of Correlate + * @return the domain of the free variables D + */ + private DedupFreeVarsNode generateDedupFreeVarsNode(RelNode newLeft, UnnestQuery leftInfo) { + List<Integer> columnIndexes = new ArrayList<>(); + for (CorDef corDef : corDefs) { + int fieldIndex = requireNonNull(leftInfo.corDefOutputs.get(corDef)); + columnIndexes.add(fieldIndex); + } + List<RexNode> inputRefs = builder.push(newLeft) + .fields(columnIndexes); + RelNode rel = builder.project(inputRefs).distinct().build(); + return new DedupFreeVarsNode(rel); + } + + /** + * Detects whether existing correlated expressions in the RelNode (according to corDefs). + * + * @param rel RelNode + * @return true when there are correlated expressions + */ + private boolean detectCorrelatedExpressions(RelNode rel) { + boolean hasCorrelation = false; + for (RelNode input : rel.getInputs()) { + hasCorrelation |= detectCorrelatedExpressions(input); + } + if (!hasCorrelation) { + RelOptUtil.VariableUsedVisitor variableUsedVisitor = + new RelOptUtil.VariableUsedVisitor(null); + rel.accept(variableUsedVisitor); + Set<CorrelationId> corrIdSet + = corDefs.stream() + .map(corDef -> corDef.corr) + .collect(ImmutableSet.toImmutableSet()); + hasCorrelation = + !variableUsedVisitor.variables.isEmpty() + && !Collections.disjoint(corrIdSet, variableUsedVisitor.variables); + } + hasCorrelatedExpressions.put(rel, hasCorrelation); + return hasCorrelation; + } + + /** + * Create the new join condition after decorrelating. + * + * @param oriCondition original Correlate/Join condition + * @param leftInfo UnnestQuery of the left side + * @param rightInfo UnnestQuery of the right side + * @param needNatureJoinCond whether need to add the natural join condition for domain D + * @return the new join condition + */ + private RexNode createUnnestedJoinCondition( + RexNode oriCondition, + UnnestQuery leftInfo, + UnnestQuery rightInfo, + boolean needNatureJoinCond) { + // create a virtual inner join and its UnnestQuery to help rewrite the + // original condition by CorrelatedExprRewriter + Map<Integer, Integer> virtualOldToNewOutputs = new HashMap<>(); + int oriLeftFieldCount = leftInfo.oldRel.getRowType().getFieldCount(); + int newLeftFieldCount = leftInfo.r.getRowType().getFieldCount(); + virtualOldToNewOutputs.putAll(leftInfo.oldToNewOutputs); + rightInfo.oldToNewOutputs.forEach((oriIndex, newIndex) -> + virtualOldToNewOutputs.put( + requireNonNull(oriIndex, "oriIndex") + oriLeftFieldCount, + requireNonNull(newIndex, "newIndex") + newLeftFieldCount)); + + TreeMap<CorDef, Integer> virtualCorDefOutputs = new TreeMap<>(); + if (!leftInfo.corDefOutputs.isEmpty()) { + virtualCorDefOutputs.putAll(leftInfo.corDefOutputs); + } else if (!rightInfo.corDefOutputs.isEmpty()) { + rightInfo.corDefOutputs.forEach((corDef, index) -> + virtualCorDefOutputs.put(corDef, index + newLeftFieldCount)); + } else { + throw new IllegalArgumentException("The UnnestInfo for both sides of Join/Correlate that has " + + "correlation should not all be empty."); + } + RelNode virtualOldRel = builder.push(leftInfo.oldRel).push(rightInfo.oldRel) + .join(JoinRelType.INNER) + .build(); + RelNode virtualNewRel = builder.push(leftInfo.r).push(rightInfo.r) + .join(JoinRelType.INNER) + .build(); + UnnestQuery virtualInfo = + new UnnestQuery(virtualOldRel, virtualNewRel, virtualCorDefOutputs, virtualOldToNewOutputs); + RexNode rewriteOriCondition = CorrelatedExprRewriter.rewrite(oriCondition, virtualInfo); + List<RexNode> unnestedJoinConditions = new ArrayList<>(); + unnestedJoinConditions.add(rewriteOriCondition); + + if (needNatureJoinCond) { + for (CorDef corDef : corDefs) { + int leftIndex = requireNonNull(leftInfo.corDefOutputs.get(corDef)); + RelDataType leftColumnType + = leftInfo.r.getRowType().getFieldList().get(leftIndex).getType(); + int rightIndex = requireNonNull(rightInfo.corDefOutputs.get(corDef)); + RelDataType rightColumnType + = rightInfo.r.getRowType().getFieldList().get(rightIndex).getType(); + RexNode notDistinctFrom = + builder.isNotDistinctFrom( + new RexInputRef(leftIndex, leftColumnType), + new RexInputRef(rightIndex + newLeftFieldCount, rightColumnType)); + unnestedJoinConditions.add(notDistinctFrom); + } + } + return RexUtil.composeConjunction(builder.getRexBuilder(), unnestedJoinConditions); + } + + /** + * Create UnnestQuery for Join/Correlate after decorrelating. + * + * @param leftInfo UnnestQuery of the left side + * @param rightInfo UnnestQuery of the right side + * @param oriJoinNode original Join/Correlate node + * @param unnestedJoinNode new node after decorrelating + * @param joinRelType join type of original Join/Correlate + * @return UnnestQuery + */ + private UnnestQuery createJoinUnnestInfo( + UnnestQuery leftInfo, + UnnestQuery rightInfo, + RelNode oriJoinNode, + RelNode unnestedJoinNode, + JoinRelType joinRelType) { + Map<Integer, Integer> oldToNewOutputs = new HashMap<>(); + oldToNewOutputs.putAll(leftInfo.oldToNewOutputs); + int oriLeftFieldCount = leftInfo.oldRel.getRowType().getFieldCount(); + int newLeftFieldCount = leftInfo.r.getRowType().getFieldCount(); + switch (joinRelType) { + case SEMI: + case ANTI: + break; + case MARK: + oldToNewOutputs.put(oriLeftFieldCount, newLeftFieldCount); + break; + default: + rightInfo.oldToNewOutputs.forEach((oriIndex, newIndex) -> + oldToNewOutputs.put( + requireNonNull(oriIndex, "oriIndex") + oriLeftFieldCount, + requireNonNull(newIndex, "newIndex") + newLeftFieldCount)); + break; + } + + TreeMap<CorDef, Integer> corDefOutputs = new TreeMap<>(); + if (!leftInfo.corDefOutputs.isEmpty()) { + corDefOutputs.putAll(leftInfo.corDefOutputs); + } else if (!rightInfo.corDefOutputs.isEmpty()) { + Litmus.THROW.check(joinRelType.projectsRight(), + "If the joinType doesn't project right, its left side must have UnnestInfo."); + rightInfo.corDefOutputs.forEach((corDef, index) -> + corDefOutputs.put(corDef, index + newLeftFieldCount)); + } else { + throw new IllegalArgumentException("The UnnestInfo for both sides of Join/Correlate that has " + + "correlation should not all be empty."); + } + return new UnnestQuery(oriJoinNode, unnestedJoinNode, corDefOutputs, oldToNewOutputs); + } + + /** + * Unnests a RelNode. If there is no correlation in the node, create the cross product + * with domain D; otherwise, dispatch to specific method to push down D. + * + * @param rel RelNode + * @return new node (contains domain D) without correlation + */ + private RelNode unnest(RelNode rel) { + if (!requireNonNull(hasCorrelatedExpressions.get(rel))) { + RelNode newRel + = builder.push(decorrelateQuery(rel, builder)) + .push(dedupFreeVarsNode.r) + .join(JoinRelType.INNER) + .build(); + Map<Integer, Integer> oldToNewOutputs = new HashMap<>(); + IntStream.range(0, rel.getRowType().getFieldCount()) + .forEach(i -> oldToNewOutputs.put(i, i)); + + int offset = rel.getRowType().getFieldCount(); + TreeMap<CorDef, Integer> corDefOutputs = new TreeMap<>(); + for (CorDef corDef : corDefs) { + corDefOutputs.put(corDef, offset++); + } + + UnnestQuery unnestQuery + = new UnnestQuery(rel, newRel, corDefOutputs, oldToNewOutputs); + mapRelToUnnestQuery.put(rel, unnestQuery); + return newRel; + } + return dispatcher.invoke(rel); + } + + public RelNode unnestInternal(Filter filter) { + Map<Integer, Integer> oldToNewOutputs = new HashMap<>(); + TreeMap<CorDef, Integer> corDefOutputs = new TreeMap<>(); + List<RexNode> newConditions = new ArrayList<>(); + // try to replace all free variables to input refs according to equi-conditions + if (tryReplaceFreeVarsToInputRef(filter, corDefOutputs, newConditions)) { + // all free variables can be replaced, no need to push down D, that is, D is eliminated. + builder.push(filter.getInput()).filter(newConditions); + for (int i = 0; i < filter.getRowType().getFieldCount(); i++) { + oldToNewOutputs.put(i, i); + } + } else { + // push down D + RelNode newInput = unnest(filter.getInput()); + UnnestQuery inputInfo = requireNonNull(mapRelToUnnestQuery.get(filter.getInput())); + RexNode newCondition = + CorrelatedExprRewriter.rewrite(filter.getCondition(), inputInfo); + builder.push(newInput).filter(newCondition); + oldToNewOutputs = inputInfo.oldToNewOutputs; + corDefOutputs.putAll(inputInfo.corDefOutputs); + } + RelNode newFilter = builder.build(); + UnnestQuery unnestQuery = new UnnestQuery(filter, newFilter, corDefOutputs, oldToNewOutputs); + mapRelToUnnestQuery.put(filter, unnestQuery); + return newFilter; + } + + public RelNode unnestInternal(Project project) { + for (RexNode expr : project.getProjects()) { + if (!Strong.isStrong(expr)) { + emptyOutputDisable = true; + } + } + RelNode newInput = unnest(project.getInput()); + UnnestQuery inputInfo = requireNonNull(mapRelToUnnestQuery.get(project.getInput())); + List<RexNode> newProjects + = CorrelatedExprRewriter.rewrite(project.getProjects(), inputInfo); + + int oriFieldCount = newProjects.size(); + Map<Integer, Integer> oldToNewOutputs = new HashMap<>(); + IntStream.range(0, oriFieldCount).forEach(i -> oldToNewOutputs.put(i, i)); + + builder.push(newInput); + TreeMap<CorDef, Integer> corDefOutputs = new TreeMap<>(); + for (CorDef corDef : corDefs) { + newProjects.add(builder.field(requireNonNull(inputInfo.corDefOutputs.get(corDef)))); + corDefOutputs.put(corDef, oriFieldCount++); + } + RelNode newProject = builder.project(newProjects, ImmutableList.of(), true).build(); + UnnestQuery unnestQuery + = new UnnestQuery(project, newProject, corDefOutputs, oldToNewOutputs); + mapRelToUnnestQuery.put(project, unnestQuery); + return newProject; + } + + public RelNode unnestInternal(Aggregate aggregate) { + RelNode newInput = unnest(aggregate.getInput()); + UnnestQuery inputUnnestQuery = + requireNonNull(mapRelToUnnestQuery.get(aggregate.getInput())); + builder.push(newInput); + + // create new groupSet and groupSets, adding the fields in D to group keys + ImmutableBitSet.Builder corKeyBuilder = ImmutableBitSet.builder(); + for (CorDef corDef : corDefs) { + int corKeyIndex = requireNonNull(inputUnnestQuery.corDefOutputs.get(corDef)); + corKeyBuilder.set(corKeyIndex); + } + ImmutableBitSet corKeyBitSet = corKeyBuilder.build(); + ImmutableBitSet newGroupSet + = aggregate.getGroupSet().permute(inputUnnestQuery.oldToNewOutputs) + .union(corKeyBitSet); + List<ImmutableBitSet> newGroupSets = new ArrayList<>(); + for (ImmutableBitSet bitSet : aggregate.getGroupSets()) { + ImmutableBitSet newBitSet + = bitSet.permute(inputUnnestQuery.oldToNewOutputs).union(corKeyBitSet); + newGroupSets.add(newBitSet); + } + + // create new aggregate functions + boolean hasCountFunction = false; + List<AggregateCall> permutedAggCalls = new ArrayList<>(); + Mappings.TargetMapping targetMapping = + Mappings.target( + inputUnnestQuery.oldToNewOutputs, + inputUnnestQuery.oldRel.getRowType().getFieldCount(), + inputUnnestQuery.r.getRowType().getFieldCount()); + for (AggregateCall aggCall : aggregate.getAggCallList()) { + hasCountFunction |= aggCall.getAggregation() instanceof SqlCountAggFunction; + permutedAggCalls.add(aggCall.transform(targetMapping)); + } + // create new Aggregate node + RelNode newAggregate + = builder.aggregate(builder.groupKey(newGroupSet, newGroupSets), permutedAggCalls).build(); + + // create UnnestQuery + Map<Integer, Integer> oldToNewOutputs = new HashMap<>(); + for (int groupKey : aggregate.getGroupSet()) { + int oriIndex = aggregate.getGroupSet().indexOf(groupKey); + int newIndex = newGroupSet.indexOf(groupKey); + oldToNewOutputs.put(oriIndex, newIndex); + } + for (int i = 0; i < aggregate.getAggCallList().size(); i++) { + oldToNewOutputs.put( + aggregate.getGroupCount() + i, + newGroupSet.cardinality() + i); + } + TreeMap<CorDef, Integer> corDefOutputs = new TreeMap<>(); + for (CorDef corDef : corDefs) { + int index = requireNonNull(inputUnnestQuery.corDefOutputs.get(corDef)); + corDefOutputs.put(corDef, newGroupSet.indexOf(index)); + } + + if (aggregate.hasEmptyGroup() + && (emptyOutputDisable || hasCountFunction)) { + // create a left join with D to avoid rewriting from non-empty to empty output + builder.push(dedupFreeVarsNode.r).push(newAggregate); + List<RexNode> leftJoinConditions = new ArrayList<>(); + int freeVarsIndex = 0; + for (CorDef corDef : corDefs) { + RexNode notDistinctFrom = + builder.isNotDistinctFrom( + builder.field(2, 0, freeVarsIndex), + builder.field(2, 1, requireNonNull(corDefOutputs.get(corDef)))); + leftJoinConditions.add(notDistinctFrom); + + corDefOutputs.put(corDef, freeVarsIndex++); + } + builder.join(JoinRelType.LEFT, leftJoinConditions); + + // rewrite COUNT to case when + List<RexNode> aggCallProjects = new ArrayList<>(); + final int aggCallStartIndex = + dedupFreeVarsNode.r.getRowType().getFieldCount() + newGroupSet.cardinality(); + for (int i = 0; i < permutedAggCalls.size(); i++) { + int index = aggCallStartIndex + i; + SqlAggFunction aggregation = permutedAggCalls.get(i).getAggregation(); + if (aggregation instanceof SqlCountAggFunction) { Review Comment: Yes. When the input is empty set, if the Aggregate has empty group, both COUNT(*) and COUNT(x) return 0; if the Aggregate doesn't have empty group, they return an empty set. -- 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]
