[
https://issues.apache.org/jira/browse/LUCENE-9103?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17000174#comment-17000174
]
ASF subversion and git services commented on LUCENE-9103:
---------------------------------------------------------
Commit bc02ab906445fcf4e297f4ef00ab4a54fdd72ca2 in lucene-solr's branch
refs/heads/branch_8_4 from Adrien Grand
[ https://gitbox.apache.org/repos/asf?p=lucene-solr.git;h=bc02ab9 ]
LUCENE-9103: WANDScorer can miss some hits in some rare conditions.
> WANDScorer can miss some hits
> -----------------------------
>
> Key: LUCENE-9103
> URL: https://issues.apache.org/jira/browse/LUCENE-9103
> Project: Lucene - Core
> Issue Type: Bug
> Reporter: Adrien Grand
> Priority: Blocker
> Fix For: 8.4
>
>
> I found the issue while building the release candidate for 8.4. There is a
> rare bug in WANDScorer that occurs when a block is ignored because the sum of
> the maximum scores is less than the minimum competitive score. In that case
> WANDScorer advances until it finds a block where the sum of the maximum
> scores is greater than the minimum competitive score. Then it pops clauses
> from the tail until the maximum score of the tail gets smaller than the
> minimum competitive score, advances these clauses beyond the current target,
> and treats the smallest doc ID as the next candidate. This is where the bug
> lies: sometimes this candidate will be beyond the current block and
> WANDScorer won't update maximum scores of other clauses. In this case it
> might skip over hits thinking that they can't have a competitive score when
> in fact they could.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@lucene.apache.org
For additional commands, e-mail: issues-h...@lucene.apache.org
