slow-J commented on code in PR #16271:
URL: https://github.com/apache/lucene/pull/16271#discussion_r3444948182
##########
lucene/core/src/java/org/apache/lucene/search/AbstractKnnVectorQuery.java:
##########
@@ -149,9 +149,64 @@ public Query rewrite(IndexSearcher indexSearcher) throws
IOException {
topK = runSearchTasks(tasks, taskExecutor, perLeafResults,
leafReaderContexts);
}
if (topK.scoreDocs.length == 0) {
- return MatchNoDocsQuery.INSTANCE;
+ return new MatchNoDocsQuery("No documents matched the nearest-neighbor
search");
}
- return DocAndScoreQuery.createDocAndScoreQuery(reader, topK, reentryCount);
+ return DocAndScoreQuery.createDocAndScoreQuery(
+ reader, topK, reentryCount, noMatchExplainer(topK, filterWeight));
+ }
+
+ /** Builds the explainer for documents this query did not collect, capturing
minTopKScore. */
+ private DocAndScoreQuery.NoMatchExplainer noMatchExplainer(TopDocs topK,
Weight filterWeight) {
+ // topK is score-descending, so the lowest collected score is the last
entry.
+ final float minTopKScore = topK.scoreDocs[topK.scoreDocs.length - 1].score;
+ return (context, doc, topN) ->
+ explainNotCollected(context, doc, topN, filterWeight, minTopKScore);
Review Comment:
Thanks, good catch!
Looking into this a bit, I don't think it would always throw an exception
when called with a different reader, but it would evaluate the filter against
the other reader's segment, giving the wrong answer.
Instead of the try/catch, I think we should gate the explain in
DocAndScoreQuery#createWeight on contextIdentity, like so:
```
if (noMatchExplainer != null
&& ReaderUtil.getTopLevelContext(context).id() ==
contextIdentity)
```
This is similar to the existing check in createWeight.
What do you think?
--
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]
