[
https://issues.apache.org/jira/browse/LUCENE-6190?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Smiley updated LUCENE-6190:
---------------------------------
Attachment: LUCENE-6190__pointsOnly_RPT_predicate_bug.patch
This patch removes the predicate swap by RPT, and enhances existing randomized
testing for pointsOnly. And it refactored the "hasIndexedLeaves" flag
(generally !pointsOnly) to be passed through to WithinPrefixTreeFilter and
moved it from AbstractVisitingPrefixTreeFilter.VisitorTemplate to its parent
class.
In the process, I fixed a bug in which WithinPrefixTreeFilter was missing
equals/hashcode. It would have only been a problem if you were explicitly
creating WithinPrefixTreeFilter with custom buffered query distance args.
> spatial pointsOnly flag shouldn't force predicate to Intersects
> ---------------------------------------------------------------
>
> Key: LUCENE-6190
> URL: https://issues.apache.org/jira/browse/LUCENE-6190
> Project: Lucene - Core
> Issue Type: Improvement
> Components: modules/spatial
> Reporter: David Smiley
> Assignee: David Smiley
> Fix For: 5.1
>
> Attachments: LUCENE-6190__pointsOnly_RPT_predicate_bug.patch
>
>
> In the process of testing the pointsOnly flag, I realized RPT's optimization
> to force the predicate to Intersects from Within|Contains isn't sound. In
> the case of Within, this is only valid if there is one point per document but
> not multiple (since _all_ points on a doc need to intersect the query shape),
> and for Contains it was simply wrong.
> Note that the strategy has no multi-valued hint or some-such. If it did,
> then if !multiValued && pointsOnly, *then* Within could be changed to
> Intersects. Regardless, swapping the predicate can be done at a higher level
> (Solr/ES).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
