[
https://issues.apache.org/jira/browse/LUCENE-5798?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14048085#comment-14048085
]
Robert Muir commented on LUCENE-5798:
-------------------------------------
In my benchmarking this seems to give a fairly substantial improvement when
merging terms.
> minor optimizations to MultiDocs(AndPositions)Enum.reset()
> ----------------------------------------------------------
>
> Key: LUCENE-5798
> URL: https://issues.apache.org/jira/browse/LUCENE-5798
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Robert Muir
> Attachments: LUCENE-5798.patch
>
>
> This method is called by merging for each term, potentially many times, but
> only returning a few docs for each invocation (e.g. imagine high cardinality
> fields, unique id fields, normal zipf distribution on full text).
> Today we create a new EnumWithSlice[] array and new EnumWithSlice entry for
> each term, but this creates a fair amount of unnecessary garbage: instead we
> can just make this array up-front as size subReaderCount and reuse it.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
