[
https://issues.apache.org/jira/browse/LUCENE-2090?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robert Muir updated LUCENE-2090:
--------------------------------
Attachment: LUCENE-2090_TermRef_flex.patch
Attached is a patch to TermRef to implement endsWith()
this is a huge win on flex, even though constant suffix gain is very minor on
trunk, because it avoids unicode conversion (char[]) for the worst cases that
must do lots of comparisons.
*N 1705.7ms avg -> 1195.4ms avg
*NNNNNN 1844.9ms avg -> 1192.3ms avg
it doesn't really matter if the suffix is short, if there is a way in
FilteredTermsEnum.accept() for a multitermquery to accept/reject a term without
unicode conversion, it helps a lot.
in my opinion, this is the cleanest way to improve these cases, other crazy
ideas i have tossed around out here like the iterative "reader-like" conversion
or even TermRef substring matching will probably not gain much more over this,
will be a lot more complex, and only apply to automatonquery.
Mike, if you get a chance to review, this, I'll commit it to flex branch (the
tests pass).
> convert automaton to char[] based processing and TermRef / TermsEnum api
> ------------------------------------------------------------------------
>
> Key: LUCENE-2090
> URL: https://issues.apache.org/jira/browse/LUCENE-2090
> Project: Lucene - Java
> Issue Type: Improvement
> Components: Search
> Reporter: Robert Muir
> Priority: Minor
> Fix For: 3.1
>
> Attachments: LUCENE-2090_TermRef_flex.patch
>
>
> The automaton processing is currently done with String, mostly because
> TermEnum is based on String.
> it is easy to change the processing to work with char[], since behind the
> scenes this is used anyway.
> in general I think we should make sure char[] based processing is exposed in
> the automaton pkg anyway, for things like pattern-based tokenizers and such.
--
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]
