[LUCENE-5796] Scorer.getChildren() can throw or hide a subscorer for some boolean queries

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: 4.9
• Fix Version/s: 4.10, 6.0
• Component/s: core/search
• Lucene Fields: New, Patch Available
• Assignee: Unassigned
• Reporter: Terry Smith
• Created: 30/Jun/14 17:52
• Updated: 28/Aug/22 14:11
• Resolved: 07/Jul/14 13:24

Description

I've isolated two example boolean queries that don't behave with release 4.9 of Lucene.

1. A BooleanQuery with three SHOULD clauses and a minimumNumberShouldMatch of 2 will throw an ArrayIndexOutOfBoundsException.

   java.lang.ArrayIndexOutOfBoundsException: 2
   	at __randomizedtesting.SeedInfo.seed([2F79B3DF917D071B:2539E6DBC4DF793C]:0)
   	at org.apache.lucene.search.MinShouldMatchSumScorer.getChildren(MinShouldMatchSumScorer.java:119)
   	at org.apache.lucene.search.TestBooleanQueryVisitSubscorers$ScorerSummarizingCollector.summarizeScorer(TestBooleanQueryVisitSubscorers.java:261)
   	at org.apache.lucene.search.TestBooleanQueryVisitSubscorers$ScorerSummarizingCollector.setScorer(TestBooleanQueryVisitSubscorers.java:238)
   	at org.apache.lucene.search.Weight$DefaultBulkScorer.score(Weight.java:161)
   	at org.apache.lucene.search.AssertingBulkScorer.score(AssertingBulkScorer.java:64)
   	at org.apache.lucene.search.IndexSearcher.search(IndexSearcher.java:621)
   	at org.apache.lucene.search.AssertingIndexSearcher.search(AssertingIndexSearcher.java:94)
   	at org.apache.lucene.search.IndexSearcher.search(IndexSearcher.java:309)
   	at org.apache.lucene.search.TestBooleanQueryVisitSubscorers.testGetChildrenMinShouldMatchSumScorer(TestBooleanQueryVisitSubscorers.java:196)
   

2. A BooleanQuery with two should clauses, one of which is a miss for all documents in the current segment will accidentally mask the scorer that was a hit.

Unit tests and patch based on branch_4x are available and will be attached as soon as this ticket has a number.

They are immediately available on GitHub on branch shebiki/bqgetchildren as commit c64bb6f.

I took the liberty of naming the relationship in BoostingScorer.getChildren() BOOSTING. Suspect someone will offer a better name for this. Here is a summary of the various relationships in play for all Scorer.getChildren() implementations on branch_4x to help choose.

class	relationships
org.apache.lucene.search.AssertingScorer	SHOULD
org.apache.lucene.search.join.ToParentBlockJoinQuery.BlockJoinScorer	BLOCK_JOIN
org.apache.lucene.search.ConjunctionScorer	MUST
org.apache.lucene.search.ConstantScoreQuery.ConstantScorer	constant
org.apache.lucene.queries.function.BoostedQuery.CustomScorer	CUSTOM
org.apache.lucene.queries.CustomScoreQuery.CustomScorer	CUSTOM
org.apache.lucene.search.DisjunctionScorer	SHOULD
org.apache.lucene.facet.DrillSidewaysScorer.FakeScorer	MUST
org.apache.lucene.search.FilterScorer	calls in.getChildren()
org.apache.lucene.search.ScoreCachingWrappingScorer	CACHED
org.apache.lucene.search.FilteredQuery.LeapFrogScorer	FILTERED
org.apache.lucene.search.MinShouldMatchSumScorer	SHOULD
org.apache.lucene.search.FilteredQuery	FILTERED
org.apache.lucene.search.ReqExclScorer	MUST
org.apache.lucene.search.ReqOptSumScorer	MUST, SHOULD
org.apache.lucene.search.join.ToChildBlockJoinQuery	BLOCK_JOIN

I also removed FilterScorer.getChildren() to prevent mistakes and force subclasses to provide a correct implementation.