[
https://issues.apache.org/jira/browse/LUCENE-2378?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Michael McCandless updated LUCENE-2378:
---------------------------------------
Attachment: LUCENE-2378.patch
I found the source of the massive slowdown when enum'ing docs -- attached patch
fixes it. I messed up the logic on whether the incoming DocsEnum can be
reused, thus causing a new DocsEnum to be created for each sub-segment on each
.docs() call!
With this I'm seeing trunk (6432 msec) slightly faster than 3x (6659 msec), for
term + docs enum.
But terms enum alone is still slightly slower (trunk is 5312; 3x is 5019 msec).
> Cutover remaining usage of pre-flex APIs
> ----------------------------------------
>
> Key: LUCENE-2378
> URL: https://issues.apache.org/jira/browse/LUCENE-2378
> Project: Lucene - Java
> Issue Type: Improvement
> Components: Index
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Fix For: 4.0
>
> Attachments: LUCENE-2378.patch, LUCENE-2378.patch, LUCENE-2378.patch,
> LUCENE-2378_FileFloatSource.patch, LUCENE-2378_PagedBytes.patch,
> LUCENE-2378_UnInvertedField.patch, TestEnumPerf.java, TestEnumPerf.java
>
>
> A number of places still use the pre-flex APIs.
> This is actually healthy, since it gives us ongoing testing of the back
> compat emulation layer.
> But we should at some point cut them all over to flex. Latest we can do this
> is 4.0, but I'm not sure we should do them all for 3.1... still marking this
> as 3.1 to "remind us" :)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
