Details
-
Type:
Bug
-
Status: Closed
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: 4.3.1
-
Fix Version/s: 4.4
-
Component/s: modules/join
-
Labels:None
-
Lucene Fields:New, Patch Available
Description
TermsIncludingScoreQuery has an inner class SVInnerScorer used when the "to" side of a join is single-valued. This has a nextDocOutOfOrder() method that is faulty when there are deleted documents, and a document that is deleted is matched by the join. It'll terminate with NO_MORE_DOCS prematurely. Interestingly, it appears MVInnerScorer (multi-valued) was coded properly to not have this problem.
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
This patch fixes the bug.
It also refactors MVInnerScorer so that it overrides a new method that has the essence of the delta needed over SVInnerScorer – its superclass.
Commit 1502784 from David Smiley
[ https://svn.apache.org/r1502784 ]
LUCENE-5103: A join on A single-valued field with deleted docs scored too few docs. Did a little refactoring of the inner scorers too.
Commit 1502785 from David Smiley
[ https://svn.apache.org/r1502785 ]
LUCENE-5103: A join on A single-valued field with deleted docs scored too few docs. Did a little refactoring of the inner scorers too.
This wasn't actually fixed on the 4.4 release branch.
Oh; I didn't notice the release branch. I'll merge later today when I have time; same with SOLR-5034.
Commit 1502832 from David Smiley
[ https://svn.apache.org/r1502832 ]
LUCENE-5103: A join on A single-valued field with deleted docs scored too few docs. Did a little refactoring of the inner scorers too.
@David Thanks for fixing this!
A working implementation of the method should looks like this:
int nextDocOutOfOrder() throws IOException { while (true) { if (docsEnum != null) { int docId = docsEnum.nextDoc(); if (docId == DocIdSetIterator.NO_MORE_DOCS) { docsEnum = null; } else { return doc = docId; } } if (upto == terms.size()) { return doc = DocIdSetIterator.NO_MORE_DOCS; } scoreUpto = upto; if (termsEnum.seekExact(terms.get(ords[upto++], spare), true)) { docsEnum = reuse = termsEnum.docs(acceptDocs, reuse, DocsEnum.FLAG_NONE); } } }I'll code a proper patch another day as it's very late right now.