[JCR-3539] NotQuery#advance (and for older versions skipTo) violates Lucene advance contract in case a Filter is used

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 0.9, 2.4.4, 2.6.1, 2.7
• Component/s: jackrabbit-core
• Assignee: Ard Schrijvers
• Reporter: Ard Schrijvers
• Created: 22/Mar/13 14:25
• Updated: 26/Aug/19 08:47
• Resolved: 22/Mar/13 14:50

Description

Plain jackrabbit does not directly suffer from the issue in the NotQuery, but it surfaces when searching with a org.apache.lucene.search.Filterorg.apache.lucene.search.Filter.

The following happens when calling JackrabbitIndexSearcher#search(Query,Filter,int) (or any other search method containing a filter). In Lucene 3.6.0 (older Lucene version in older jackrabbit versions similar issue) at FilterQuery#getFilterScorer, there is

@Override
public void score(Collector collector) throws IOException {
int filterDoc = filterIter.nextDoc();
int scorerDoc = scorer.advance(filterDoc);

The scorer gets advanced with 'scorer.advance(filterDoc)' : If the scorer contains a Jackrabbit NotQuery, this results in NotQuery#advance being invoked. At this time, NotQuery#nextDoc() has not yet ever been invoked.

The NotQuery#advance results in that docNo becomes some positive number (0 or larger). After this, when NotQuery#nextDoc() gets invoked, the docNo is not -1 any more, and the contextScorer.nextDoc(); is never called any more, ever. Hence, when a FIlter gets applied, the NotQuery is completely skipped.

Fix is simply to replace the

if (docNo == -1) {

with a

if (firstTime) {
firstTime = false;

(in older lucene versions the problem results in false 'more' for IndexSearcher in this line: boolean more = filterDocIdIterator.next() && scorer.skipTo(filterDocIdIterator.doc()); )