lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Muir (JIRA)" <>
Subject [jira] [Commented] (LUCENE-3849) position increments should be implemented by TokenStream.end()
Date Mon, 20 Aug 2012 21:44:38 GMT


Robert Muir commented on LUCENE-3849:

where does the "positionIncrementGap" come in? because my naive impression is that the end()
method is precisely where something like posIncAtt.setPositionIncrement(getPositionIncrementaGap())
should be called.

PositionIncrementGap is not related to this bug, because its something done separately by
indexwriter. Its always factored in correctly, its not buggy.

The "bug" is if a field instance ends with a stopword, that accumulated 'hole' is lost.
doc.add(new TextField("body", "just a", Field.Store.NO));
doc.add(new TextField("body", "test of gaps", Field.Store.NO));
PhraseQuery pq = new PhraseQuery();
pq.add(new Term("body", "just"), 0);
pq.add(new Term("body", "test"), 2);
// body:"just ? test"
assertEquals(1,, 5).totalHits); // FAIL!

So the problem is the first instance of the field loses its hole for "a" (a stopword that
was removed),
only because it ended on a stopword, so incrementToken() returned false and there was no subsequent
to apply the 'hole' to.

This is the same problem as an instance of a field ending with say a space character and all
the offsets being wrong,
this was why end() was added, so the indexer can pull 'end of stream state'.

> position increments should be implemented by TokenStream.end()
> --------------------------------------------------------------
>                 Key: LUCENE-3849
>                 URL:
>             Project: Lucene - Core
>          Issue Type: Bug
>    Affects Versions: 3.6, 4.0-ALPHA
>            Reporter: Robert Muir
>         Attachments: LUCENE-3849.patch
> if you have pages of a book as multivalued fields, with the default position increment
> of (0), phrase queries won't work across pages if one ends with stopword(s).
> This is because the 'trailing holes' are not taken into account in end(). So I think
> TokenStream.end(), subclasses of FilteringTokenFilter (e.g. stopfilter) should do:
> {code}
> super.end();
> posIncAtt += skippedPositions;
> {code}
> One problem is that these filters need to 'add' to the posinc, but currently nothing
> the attributes for end() [they are dirty, except offset which is set by the tokenizer].
> Also the indexer should be changed to pull posIncAtt from end().

This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:!default.jspa
For more information on JIRA, see:


To unsubscribe, e-mail:
For additional commands, e-mail:

View raw message