[
https://issues.apache.org/jira/browse/LUCENE-8806?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16848902#comment-16848902
]
Adrien Grand commented on LUCENE-8806:
--------------------------------------
The change to correct. Regarding tests, could you have a separate
testRandomTwoPhase from testRandom so that we get an idea whether a problem is
related to two-phase iteration or not when we get a failure? It'd also be nice
to have something similar to testBasics, but for two-phase iterators, maybe we
can somehow rewrite testBasics with phrase queries?
I'm wondering whether we could make this more efficient (possibly as a
follow-up, the way changes to WANDScorer are designed in your patch is already
an improvement, and the ideas I am suggesting may be difficult to implement):
- Right now your patch checks all clauses that are positioned on the current
doc. Maybe we could improve this logic to exit early if the fact that one
clause doesn't match makes leadMaxScore less than the min competitive score.
- As a follow-up of the above point, we could then look into advancing disis
from the tail lazily instead of calling advanceAllTail like the current patch
does. In some cases, leadMaxScore+tailMaxScore might become less than the min
competitive score before all disis from the tail have advanced.
- We could look into making TwoPhaseIterator#matches return true as soon as
one clause returns true instead of checking them all? This would help in the
case that score() isn't called.
> WANDScorer should support two-phase iterator
> --------------------------------------------
>
> Key: LUCENE-8806
> URL: https://issues.apache.org/jira/browse/LUCENE-8806
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Jim Ferenczi
> Priority: Major
> Attachments: LUCENE-8806.patch
>
>
> Following https://issues.apache.org/jira/browse/LUCENE-8770 the WANDScorer
> should leverage two-phase iterators in order to be faster when used in
> conjunctions.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
