Draft patch. Added ScoreMode as parameter to JoinUtil#createJoinQuery(...).

Maybe ScoreMode should be a public enum inside the join package.

Updated patch.

• Started adding randomizing score mode in TestJoinUtil test class.
• Made ScoreMode a public enum in join package.

Updated patch.

• Fixed random tests.
• Added support for explain.
• Added ScoreMode support for documents that relate to more than one document.

I think it is ready to be committed.

Patch looks great!

You don't need to use your own growFactor ... just call ArrayUtil.grow
directly (it already oversizes under the hood for you).

Maybe remove @throws IAE from createJoinQuery's javadocs? (But, still
throw it... in case we add a new ScoreMode and forget to fix this
code, in the future). Because today all ScoreMode enum values work...

Fix omitted to emitted in the comment on top of "class MVInnerScorer".

Probably javadocs should somewhere explain about the "first time doc
is emitted it gets that score"?

Maybe explain added RAM requirements when scores are tracked in the
javadocs?

Maybe rename TermsWithScoreCollector.MV.Avg.ordScores -> .scoreCounts
(and .scores -> .scoreSums?).

Can we put back the non-wildcard imports?

Thanks for reviewing Mike! I've updated the patch.

You don't need to use your own growFactor ... just call ArrayUtil.grow directly (it already oversizes under the hood for you).

Sure. (I didn't release that the ArrayUtil#oversize() was doing this)

Maybe remove @throws IAE from createJoinQuery's javadocs? (But, still throw it... in case we add a new ScoreMode and forget to fix this code, in the future). Because today all ScoreMode enum values work...

Makes sense. We design it not to throw the exception. If the exception is thrown, then it is a bug.

Fix omitted to emitted in the comment on top of "class MVInnerScorer".

Done.

Probably javadocs should somewhere explain about the "first time doc is emitted it gets that score"?

Done.

Maybe explain added RAM requirements when scores are tracked in the javadocs?

Done

Maybe rename TermsWithScoreCollector.MV.Avg.ordScores -> .scoreCounts (and .scores -> .scoreSums?).

Done.

Can we put back the non-wildcard imports?

Done. (IDE was trying to be smart... I'll change my settings...)

I still see one omitted Otherwise this looks great: +1 to commit!

Oops... I see. I'll commit soon!

Committed to trunk and branch4x.

Hi, Is this possible to use in solr I tried setting

, but it doesn't seem to have any effect.

Solr uses a different joining implementation. Which doesn't support mapping the scores from the `from` side to the `to` side. If you want to use the Lucene joining implementation you could wrap this in a Solr QParserPlugin extension.

Could it be that you're JoinUtil doesn't clear a cache or something?(Could also be my code) My plugin always works once when I start Solr. It joins and gives good scores. The second query it always returns the same as the first one? After restarting Solr it works 1 time and then does the same.

public class TestParserPlugin extends ExtendedDismaxQParserPlugin {

@Override
public QParser createParser(String string, SolrParams sp, SolrParams sp1, SolrQueryRequest sqr)

@Override
public void init(NamedList nl) {
}

public class TestParser extends QParser {

public TestParser(String qstr, SolrParams localParams, SolrParams params, SolrQueryRequest req)

@Override
public org.apache.lucene.search.Query parse() throws org.apache.lucene.queryparser.classic.ParseException {
//IndexReader reader;
try

catch (IOException ex)

return null;
}

@Override
protected void finalize() throws Throwable

}

}

I think this happens b/c the Query that the JoinUtil returns doesn't override the equals and hashcode method (See TermsIncludingScoreQuery). This should be fixed, otherwise this query can never be cached (this is what the SolrIndexSearcher does). Can you check if the following works:

IndexSearcher searcher = new IndexSearcher(req.getSearcher().getIndexReader());
return JoinUtil.createJoinQuery("pageId", true, "fileId", q, searcher, ScoreMode.Max); 

Same problem if I use:
IndexSearcher searcher = new IndexSearcher(req.getSearcher().getIndexReader());

I can't override the class since the constructor is private. This is probably to only use the static methods.

I've used an output to see the hashcode and it stays the same. Can I change this behaviour somehow?

I got suggested to extend the Query class and return a Hash myself.

My query class contains:

@Override
public int hashCode()

It seems to work now.

Hi David, I just committed LUCENE-4704. Query instances returned from JoinUtil will implement equals and hashcode in future versions.

Thanks alot!

I'll try the patch when 4.2 get's released.