[LUCENE-4300] BooleanQuery inconsistently applies coord() if it rewrites itself

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 4.0, 3.6.2, 6.0
• Lucene Fields: New
• Assignee: Unassigned
• Reporter: Robert Muir
• Created: 09/Aug/12 23:50
• Updated: 28/Aug/22 13:24
• Resolved: 10/Aug/12 11:10

Description

Tripped by the new random sim from LUCENE-4297:

The basics are this:

• BooleanQuery has the following rewrite():

    public Query rewrite(IndexReader reader) throws IOException {
      if (minNrShouldMatch == 0 && clauses.size() == 1) {                    // optimize 1-clause queries
  

• you have a coord() impl that doesnt return 1.0 if overlap == maxOverlap, particularly:

  return overlap / ((float)maxOverlap + 1);
  

• TestBooleanMinShouldMatch.testRandomQueries generates random boolean queries (Q1), then compares the scores of the random query to the same query but with minNrShouldmatch applied to its should clauses (Q2)

• in the case of a single term BQ, the rewrite applies to Q1, making it a term query, but not to Q2. so the coord() only gets called for Q2, not Q1. and with this crazy coord it means the scores are different.

I think the rewrite is wrong, we should also rewrite single-query BQs where minNrShouldMatch = 1 and there is a single optional clause.