github-actions[bot] commented on code in PR #69292:
URL: https://github.com/apache/doris/pull/69292#discussion_r4321752024
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/rules/rewrite/CheckScoreUsage.java:
##########
@@ -54,18 +60,22 @@ public List<Rule> buildRules() {
return hasScore && !isOptimized;
})
.then(project -> {
- throw new AnalysisException(
- "score() function requires WHERE clause with MATCH
function, "
- + "ORDER BY and LIMIT for optimization");
+ throw new AnalysisException(SCORE_CONTEXT_ERROR);
+ }).toRule(RuleType.CHECK_SCORE_USAGE),
+
+ // A score() filter on a scan was bound by
PushDownScoreTopNIntoOlapScan. One left here
+ // filters another operator, e.g. an inner join, and would
evaluate the scalar placeholder
+ // as zero.
+ logicalFilter(any())
Review Comment:
[P2] Reject score filters before window predicate pushdown
An outer `WHERE score() > 0.5` on a windowed MATCH subquery produces this
reduced input:
```text
Filter(score() > 0.5)
Window(row_number() OVER (ORDER BY id) AS rn)
Filter(c1 MATCH 'apple')
OlapScan
```
This guard runs after `PUSH_DOWN_FILTERS`.
`PushDownFilterThroughWindow.canPushDown` accepts the predicate's empty
input-slot set: Score's `isDeterministic=false` does not make it a
`VolatileExpression` or `NoneMovableFunction`. `MergeFilters` then joins it
with MATCH, and the new standalone scan rule binds/removes it before this check
sees it. The query is therefore accepted with filtering below the window,
rather than rejected as an unsupported score context. With matching ids 1 and 2
where only id 2 exceeds the threshold (as in the ANN regression fixture), the
surviving row gets `rn=1` instead of retaining its post-window `rn=2`. Prevent
raw Score from crossing window/partition-TopN operators, or validate its
context before predicate pushdown, and cover an outer score filter on a
windowed subquery.
--
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]
