[
https://issues.apache.org/jira/browse/LUCENE-4242?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13420117#comment-13420117
]
Yonik Seeley commented on LUCENE-4242:
--------------------------------------
bq. maxDoc doesn't reflect deletions.
Holding nDocs steady, as deletions rise so does maxDoc. But now that I think
about it, this depends on the semantics of "taking deletes into account". It
depends on the use-case and exactly what you're using DocTermOrds for (and why
you're skipping some terms). It may make sense for Solr to pro-rate for
deletions... I'm not sure.
bq. So then the question is what contract should we have (should caller be
expected to pro-rate by deletes themselves, or should DocTermOrds do so
I think the current behavior is easier to understand and explain - it's a
direct check of the term docFreq.
A pro-rate is just a guess since we wouldn't be guaranteeing that the "real"
docFreq would be below the max.
> UnInverted cache uses term freq to filter out terms (but deleted docs are
> included in the freq count)
> -----------------------------------------------------------------------------------------------------
>
> Key: LUCENE-4242
> URL: https://issues.apache.org/jira/browse/LUCENE-4242
> Project: Lucene - Java
> Issue Type: Bug
> Components: core/index
> Affects Versions: 4.0
> Reporter: roman
> Priority: Minor
> Attachments: LUCENE-4242.patch, LUCENE-4242.patch, LUCENE-4242.patch
>
>
> TermEnum.docFreq() count is used to compute uninverted index
> (DocTermOrds.uninvert()). The code goes like:
> final int df = te.docFreq();
> if (df <= maxTermDocFreq) {
> So, if there are deleted documents in the index and maxTermDocFreq is
> low, then the term will be excluded (even if the freq of the livedocs
> is OK). Most likely, the cache will be incomplete.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
