github-actions[bot] commented on code in PR #63763:
URL: https://github.com/apache/doris/pull/63763#discussion_r3433910962
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/rules/rewrite/AggScalarSubQueryToWindowFunction.java:
##########
@@ -278,6 +289,19 @@ private boolean checkRelation(List<Slot> correlatedSlots) {
.filter(node -> outerIds.contains(node.getTable().getId()))
.map(LogicalRelation.class::cast)
.map(LogicalRelation::getOutputExprIdSet).flatMap(Collection::stream).collect(Collectors.toSet());
+ partitionBySlots.addAll(apply.left().getOutput().stream()
+ .filter(slot ->
correlatedRelationOutput.contains(slot.getExprId()))
+ .collect(Collectors.toList()));
+ if (partitionBySlots.isEmpty()) {
+ return false;
Review Comment:
This still is not enough to preserve row identity for duplicate-preserving
tables. `DUPLICATE KEY` tables can contain two rows with identical values for
every slot in `correlatedRelationOutput`, so this equality check passes even
though the window partition cannot distinguish those two outer rows. For
example, if the new regression adds another identical `dim` row `(10, 1, 1)`,
the original scalar subquery evaluates `SUM(f2.v)` once per `dim` row and keeps
two `did=10`/`f.id=2` joined rows. After this rewrite, both identical `dim`
rows share one `(did, k, tag)` partition, each joined `fact` row appears twice
in that partition, and `SUM(v)` becomes `24` instead of `12`, so those rows are
filtered out. This is distinct from the previous pruned-output thread because
here `partitionBySlots` already equals `correlatedRelationOutput`; the missing
requirement is a proven unique row identity for the outer-only relation, or the
rule should return false for duplicate-preserving relations.
--
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]
