morrySnow commented on code in PR #10659:
URL: https://github.com/apache/doris/pull/10659#discussion_r916497641
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/glue/translator/PhysicalPlanTranslator.java:
##########
@@ -112,62 +131,92 @@ public PlanFragment visit(Plan plan,
PlanTranslatorContext context) {
/**
* Translate Agg.
+ * todo: support DISTINCT
*/
@Override
- public PlanFragment visitPhysicalAggregation(
- PhysicalUnaryPlan<PhysicalAggregation, Plan> agg,
PlanTranslatorContext context) {
-
- PlanFragment inputPlanFragment = visit(agg.child(0), context);
-
- AggregationNode aggregationNode;
- List<Slot> slotList = new ArrayList<>();
- PhysicalAggregation physicalAggregation = agg.getOperator();
- AggregateInfo.AggPhase phase =
physicalAggregation.getAggPhase().toExec();
-
- List<Expression> groupByExpressionList =
physicalAggregation.getGroupByExprList();
+ public PlanFragment visitPhysicalAggregate(
+ PhysicalUnaryPlan<PhysicalAggregate, Plan> aggregate,
PlanTranslatorContext context) {
+ PlanFragment inputPlanFragment = visit(aggregate.child(0), context);
+ PhysicalAggregate physicalAggregate = aggregate.getOperator();
+
+ // TODO: stale planner generate aggregate tuple in a special way.
tuple include 2 parts:
+ // 1. group by expressions: removing duplicate expressions add to
tuple
+ // 2. agg functions: only removing duplicate agg functions in
output expression should appear in tuple.
+ // e.g. select sum(v1) + 1, sum(v1) + 2 from t1 should only
generate one sum(v1) in tuple
+ // We need:
+ // 1. add a project after agg, if agg function is not the top
output expression.
+ // 2. introduce canonicalized, semanticEquals and deterministic in
Expression
+ // for removing duplicate.
+ List<Expression> groupByExpressionList =
physicalAggregate.getGroupByExprList();
+ List<NamedExpression> outputExpressionList =
physicalAggregate.getOutputExpressionList();
+
+ // 1. generate slot reference for each group expression
+ List<SlotReference> groupSlotList = Lists.newArrayList();
+ for (Expression e : groupByExpressionList) {
+ if (e instanceof SlotReference &&
outputExpressionList.stream().anyMatch(o -> o.anyMatch(e::equals))) {
+ groupSlotList.add((SlotReference) e);
+ } else {
+ groupSlotList.add(new SlotReference(e.sql(), e.getDataType(),
e.nullable(), Collections.emptyList()));
+ }
+ }
ArrayList<Expr> execGroupingExpressions =
groupByExpressionList.stream()
- // Since output of plan doesn't contain the slots of groupBy,
which is actually needed by
- // the BE execution, so we have to collect them and add to the
slotList to generate corresponding
- // TupleDesc.
- .peek(x ->
slotList.addAll(x.collect(SlotReference.class::isInstance)))
.map(e -> ExpressionTranslator.translate(e,
context)).collect(Collectors.toCollection(ArrayList::new));
- slotList.addAll(agg.getOutput());
- TupleDescriptor outputTupleDesc = generateTupleDesc(slotList, context,
null);
-
- List<NamedExpression> outputExpressionList =
physicalAggregation.getOutputExpressionList();
- ArrayList<FunctionCallExpr> execAggExpressions =
outputExpressionList.stream()
- .map(e ->
e.<List<AggregateFunction>>collect(AggregateFunction.class::isInstance))
+ // 2. collect agg functions and generate agg function to slot
reference map
Review Comment:
in translator, this could not happen at all
--
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]
