[
https://issues.apache.org/jira/browse/LUCENE-7957?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16153853#comment-16153853
]
Adrien Grand commented on LUCENE-7957:
--------------------------------------
Thanks for the details.
bq. I think I hope we don't remove this API.
I see how it could be useful, like in your case. But at the same time I see
multiple issues with this API:
- With some scorers, being able to track the matching sub scorers would be
additional overhead, eg. BooleanScorer today can't tell you which clauses
matched a given document.
- How do you check "did clause X or Y match?", you have to iterate over all
scorers and see whether the one you are interested in is there?
- Is it ok to perform heavy rewrites that make the scorer tree look very
different from the query tree, or even make the clause you are interested in
impossible to find? eg. by inlining nested disjunctions/conjunctions, rewriting
a TermInSetQuery to a BooleanQuery, splitting a bbox query that crosses the
dateline into 2 non-crossing bbox queries, rewriting a TermQuery to a
MatchAllDocsQuery because scores are not needed and docFreq == maxDoc, etc.
I tried to think about how to address these issues but I don't have a good
solution, especially about the last point: I think it would be a pity that an
improvement to query execution be seen as a regression because it makes it
harder to identify a matching clause.
I'd rather like this use-case to be addressed by consuming queries twice, once
by Lucene so that it can build an efficient iterator, and once from a
FilterScorer so that the score can be customized depending on whether a
particular query matches.
I haven't removed it because there was disagreement every time I suggested that
it should be removed but I don't see any way that we could support it as-is
realistically. To me it's also interesting that it is the first time this bug
is reported even though it has existed for almost two entire major versions
(since 5.1) and affects one of our main scorers.
However we still expose this API, so +1 to fixing.
> ConjunctionScorer.getChildren does not return all children
> ----------------------------------------------------------
>
> Key: LUCENE-7957
> URL: https://issues.apache.org/jira/browse/LUCENE-7957
> Project: Lucene - Core
> Issue Type: Bug
> Components: core/search
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Fix For: master (8.0), 7.1
>
>
> Today it returns all scoring children and misses the `FILTER` clauses; I
> think we just need to save the incoming `required` parameter to the ctor and
> iterate over those in `getChildren` since `scorers` is a subset of `required`?
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
