[
https://issues.apache.org/jira/browse/LUCENE-3121?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13189110#comment-13189110
]
Michael McCandless commented on LUCENE-3121:
--------------------------------------------
bq. Would the next step being applying this to the Field Cache?
That would be interesting! Or, maybe to DocValues instead...
Today neither DocValues nor FieldCache use an FST to hold the term data, but
you're right we could explore this now, since we can lookup-by-ord or
lookup-by-address (with this patch).
For example, the DocValues.BYTES_VAR/FIXED_SORTED_DEREF could hold the Term <->
ord/address map as an FST (doc -> ord/address would still be packed ints)...
same for FieldCache.DocTermsIndex/.DocTermOrds This should be a sizable
reduction in RAM required for the term data... but lookup time would get slower
too.
> FST should offer lookup-by-output API when output strictly increases
> --------------------------------------------------------------------
>
> Key: LUCENE-3121
> URL: https://issues.apache.org/jira/browse/LUCENE-3121
> Project: Lucene - Java
> Issue Type: Improvement
> Components: core/other
> Reporter: Michael McCandless
> Assignee: Michael McCandless
> Fix For: 4.0
>
> Attachments: LUCENE-3121.patch
>
>
> Spinoff from "FST and FieldCache" java-dev thread
> http://lucene.markmail.org/thread/swoawlv3fq4dntvl
> FST is able to associate arbitrary outputs with the sorted input keys, but in
> the special (and, common) case where the function is strictly monotonic (each
> output only "increases" vs prior outputs), such as mapping to term ords or
> mapping to file offsets in the terms dict, we should offer a lookup-by-output
> API that efficiently walks the FST and locates input key (exact or floor or
> ceil) matching that output.
--
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]
