[
https://issues.apache.org/jira/browse/LUCENE-2578?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12895213#action_12895213
]
Michael McCandless commented on LUCENE-2578:
--------------------------------------------
bq.The only bizzare thing is... somehow, the changeup in the index causes the
scores to be off by a bit (eg 1199.8008 vs 1199.8007),
OK I tracked this down: it's due to the ScorerDocQueue inside
DisjunctionSumScorer. The test that was failing was a BQ with 3 SHOULD
clauses. The ScorerDocQueue, which holds the current sub-scorers "ordered" by
their docIDs, does not break ties (it has no real reason to). When you simply
.nextDoc().score() the matching docs, these three subs' scores were summed in
one order. But then QueryUtil does a scorer.advance() to each doc, and that
.advance() was then pulling the matching docs in a different order from the
queue. Summing the same 3 floats in different orders can easily result in a
slightly different float due to different truncation.
I'll commit shortly w/ the threshold increased from 1e-5 to 1e-3.
> cutover FunctionQuery tests to use RandomIndexWriter, for better testing
> ------------------------------------------------------------------------
>
> Key: LUCENE-2578
> URL: https://issues.apache.org/jira/browse/LUCENE-2578
> Project: Lucene - Java
> Issue Type: Improvement
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Attachments: LUCENE-2578.patch
>
>
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
