github-actions[bot] commented on code in PR #62589:
URL: https://github.com/apache/doris/pull/62589#discussion_r3220461861
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/glue/translator/RuntimeFilterTranslator.java:
##########
@@ -322,6 +336,18 @@ public void createLegacyRuntimeFilter(RuntimeFilter
filter, PlanNode node, PlanT
Expr targetExpr = targetExprList.get(i);
origFilter.addTarget(new RuntimeFilterTarget(
scanNode, targetExpr, true, isLocalTarget));
+ RuntimeFilterPartitionPruneClassifier.Classification
classification =
+ RuntimeFilterPartitionPruneClassifier.classify(
+ targetExpr,
filter.getTargetExpressions().get(i), scanNode);
Review Comment:
The classifier is proving monotonicity from
`filter.getTargetExpressions().get(i)`, but the BE will project partition
boundaries through the final legacy `targetExpr` stored in the RF, including
any `CastExpr` injected by the type-adjustment block above. That means a target
expression that is locally monotonic before type adjustment can still be marked
safe even though the actual expression evaluated in
`ParsedPartitionBoundaries::get_or_compute_projection()` has one more cast on
top that was not checked against the partition range.
Please either classify an equivalent Nereids expression that includes the
inserted cast, or conservatively skip RF partition pruning whenever the
adjustment at line 307 wraps a non-identity target. The current tests cover the
bare-slot implicit-cast case, but not a locally monotonic non-slot target with
an added outer cast.
--
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]
