[ https://issues.apache.org/jira/browse/LUCENE-7517?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15604773#comment-15604773 ]
Adrien Grand commented on LUCENE-7517: -------------------------------------- I started exploring making this change and would appreciate if somebody could cross-check these assumptions: - the Explanation API needs to be switched to doubles, so that it can combine scores from sub queries the same way as the Scorer API - TopDocsCollector and TopFieldCollector with SortField.Type.SCORE need to cast to a float _before_ comparing the bottom score and adding to the priority queue. Otherwise the final top docs could appear as being out of order. - score-based value sources still need to expose the score as a float rather than a double, so that sorting by a score value source yields the same result as sorting by score > Explore making Scorer.score() return a double > --------------------------------------------- > > Key: LUCENE-7517 > URL: https://issues.apache.org/jira/browse/LUCENE-7517 > Project: Lucene - Core > Issue Type: Wish > Reporter: Adrien Grand > Priority: Minor > > Follow-up to > http://search-lucene.com/m/l6pAi1BoyPJ1vr2382&subj=Re+JENKINS+EA+Lucene+Solr+master+Linux+64bit+jdk+9+ea+140+Build+18103+Unstable+. > We could make Scorer.score() return a double in order to lose less accuracy > when combining scores together, while still using floats on TopDocs and more > generally all parts of the code that need to store scores. -- This message was sent by Atlassian JIRA (v6.3.4#6332) --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org