[
https://issues.apache.org/jira/browse/LUCENE-914?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12629130#action_12629130
]
Paul Elschot commented on LUCENE-914:
-------------------------------------
I had another look at these lines in DisjunctionScorer.skipTo() from
LUCENE-413:
{code}
if (target <= currentDoc) { // CHECKME: skipTo() semantics?
return true;
}
{code}
I think these lines can simply be dropped. The bug at LUCENE-413 was probably
fixed
by the changes to SpanScorer then.
Dropping these lines also makes DisjunctionSumScorer advance (effectively use
next())
when skipTo(target) is called, even when target <= doc().
Also, I'm changing my opinion on what to do when target <= doc(). I think a
DocIdSetIterator
(and any Scorer) should always advance in that case, just like the javadocs of
TermDocs.
When Scorers/DocIdSetIterators are combined, for example in disjunctions or
conjunctions
or in the more complex cases of BooleanScorer2, the test for target <= doc()
can simply
be done before calling skipTo() on any of the combined
Scorers/DocIdSetIterators.
Now, if such a change surfaces some bugs like LUCENE-413, so be it. It's better
to have
clearly defined semantics than to have lingering bugs due to explicitly
undefined semantics
in the case of target <= doc().
Btw. such a change would leave a small bug in SpanScorer, the (freq != 0) test
there should
be replaced by a boolean check for actual matches. See
SpanScorer.setFreqCurrentDoc().
The bug would only surface when the Similarity used in SpanScorer returns 0 for
sloppyFreq().
> Scorer.skipTo(current) remains on current for some scorers
> ----------------------------------------------------------
>
> Key: LUCENE-914
> URL: https://issues.apache.org/jira/browse/LUCENE-914
> Project: Lucene - Java
> Issue Type: Bug
> Components: Search
> Reporter: Doron Cohen
> Priority: Minor
> Attachments: lucene-914.patch
>
>
> Background in http://www.nabble.com/scorer.skipTo%28%29-contr-tf3880986.html
> It appears that several scorers do not strictly follow the spec of
> Scorer.skipTo(n), and skip to current location remain in current location
> whereas the spec says: "beyond current".
> We should (probably) either relax the spec or fix the implementations.
--
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]
