924060929 commented on code in PR #12130:
URL: https://github.com/apache/doris/pull/12130#discussion_r956860931
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/glue/translator/PhysicalPlanTranslator.java:
##########
@@ -363,32 +367,48 @@ public PlanFragment
visitPhysicalHashJoin(PhysicalHashJoin<Plan, Plan> hashJoin,
// NOTICE: We must visit from right to left, to ensure the last
fragment is root fragment
PlanFragment rightFragment = hashJoin.child(1).accept(this, context);
PlanFragment leftFragment = hashJoin.child(0).accept(this, context);
- PlanNode leftFragmentPlanRoot = leftFragment.getPlanRoot();
- PlanNode rightFragmentPlanRoot = rightFragment.getPlanRoot();
- JoinType joinType = hashJoin.getJoinType();
if (JoinUtils.shouldNestedLoopJoin(hashJoin)) {
throw new RuntimeException("Physical hash join could not execute
without equal join condition.");
}
- List<Expr> execEqConjunctList =
hashJoin.getHashJoinConjuncts().stream()
+ PlanNode leftPlanRoot = leftFragment.getPlanRoot();
+ PlanNode rightPlanRoot = rightFragment.getPlanRoot();
+ JoinType joinType = hashJoin.getJoinType();
+
+ List<Expr> execEqConjuncts = hashJoin.getHashJoinConjuncts().stream()
.map(EqualTo.class::cast)
.map(e -> swapEqualToForChildrenOrder(e,
hashJoin.left().getOutput()))
.map(e -> ExpressionTranslator.translate(e, context))
.collect(Collectors.toList());
+
+ // in Nereids, all node only has one TupleDescriptor, so we can use
the first one.
+ TupleDescriptor leftTuple =
context.getTupleDesc(leftPlanRoot.getTupleIds().get(0));
+ TupleDescriptor rightTuple =
context.getTupleDesc(rightPlanRoot.getTupleIds().get(0));
+
+ // Nereids does not care about output order of join,
+ // but BE need left child's output must be before right child's output.
+ // So we need to swap the output order of left and right child if
necessary.
+ // TODO: revert this after Nereids could ensure the output order is
correct.
TupleDescriptor outputDescriptor = context.generateTupleDesc();
- List<Expr> srcToOutput = hashJoin.getOutput().stream()
+ Map<ExprId, SlotReference> slotReferenceMap = Maps.newHashMap();
+ hashJoin.getOutput().stream()
.map(SlotReference.class::cast)
+ .forEach(s -> slotReferenceMap.put(s.getExprId(), s));
+ List<Expr> srcToOutput = Stream.concat(leftTuple.getSlots().stream(),
rightTuple.getSlots().stream())
Review Comment:
if there is only one TupleDescriptor, you no need to concat slots
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/glue/translator/PhysicalPlanTranslator.java:
##########
@@ -363,32 +367,48 @@ public PlanFragment
visitPhysicalHashJoin(PhysicalHashJoin<Plan, Plan> hashJoin,
// NOTICE: We must visit from right to left, to ensure the last
fragment is root fragment
PlanFragment rightFragment = hashJoin.child(1).accept(this, context);
PlanFragment leftFragment = hashJoin.child(0).accept(this, context);
- PlanNode leftFragmentPlanRoot = leftFragment.getPlanRoot();
- PlanNode rightFragmentPlanRoot = rightFragment.getPlanRoot();
- JoinType joinType = hashJoin.getJoinType();
if (JoinUtils.shouldNestedLoopJoin(hashJoin)) {
throw new RuntimeException("Physical hash join could not execute
without equal join condition.");
}
- List<Expr> execEqConjunctList =
hashJoin.getHashJoinConjuncts().stream()
+ PlanNode leftPlanRoot = leftFragment.getPlanRoot();
+ PlanNode rightPlanRoot = rightFragment.getPlanRoot();
+ JoinType joinType = hashJoin.getJoinType();
+
+ List<Expr> execEqConjuncts = hashJoin.getHashJoinConjuncts().stream()
.map(EqualTo.class::cast)
.map(e -> swapEqualToForChildrenOrder(e,
hashJoin.left().getOutput()))
.map(e -> ExpressionTranslator.translate(e, context))
.collect(Collectors.toList());
+
+ // in Nereids, all node only has one TupleDescriptor, so we can use
the first one.
+ TupleDescriptor leftTuple =
context.getTupleDesc(leftPlanRoot.getTupleIds().get(0));
+ TupleDescriptor rightTuple =
context.getTupleDesc(rightPlanRoot.getTupleIds().get(0));
Review Comment:
If Nereids only has one TupleDescriptor, we should hide the tupleId in the
context, or else people who add new translate method maybe misuse？
So I suggest
```java
TupleDescriptor joinTuple = context.getTupleDesc();
```
--
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: commits-unsubscr...@doris.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@doris.apache.org
For additional commands, e-mail: commits-h...@doris.apache.org
