lucene-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Robert Muir (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (LUCENE-4694) Add back IndexReader.fields() -> Multi*, or discourage term vectors in some better way
Date Sun, 10 Mar 2013 08:31:13 GMT

    [ https://issues.apache.org/jira/browse/LUCENE-4694?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13598189#comment-13598189
] 

Robert Muir commented on LUCENE-4694:
-------------------------------------

{quote}
I personally think it's ok if IndexReader lets you get docsValues(doc), document(doc), getTV(doc)
and termDocsEnum(term). There's nothing inefficient about supporting them, as far as I can
see.
{quote}

this is not correct at all. 

for the sorted types we need to iterate through all of the values and create a datastructure
mapping per-segment ordinals to global ones, and also cache this somewhere. 

additionally, all docvalues types and norms on a composite reader would pay the cost of binary-search
for *each* docid access: and due to the way they are used, typically many docids are accessed.

stored fields are used for summary results, so on a 100 million doc index who cares if you
do 10 or 20 binary searches: who cares.

term vectors are used for highlighting summary results, MoreLikeThis, etc: both of which are
small top-N just like the stored fields case. so its also fine.

but docvalues is used in scoring and sorting, so this would be 100 million binary searches.
its a big damn difference.

the postings is pretty much just an additional check per document, so its a little more up
in the air what to do. but as mentioned in the description, users look at IndexReader.java
and the only postings api they see is term vectors.

                
> Add back IndexReader.fields() -> Multi*, or discourage term vectors in some better
way
> --------------------------------------------------------------------------------------
>
>                 Key: LUCENE-4694
>                 URL: https://issues.apache.org/jira/browse/LUCENE-4694
>             Project: Lucene - Core
>          Issue Type: Bug
>            Reporter: Robert Muir
>         Attachments: LUCENE-4694.patch
>
>
> Users can easily get term vectors from any indexreader, but not postings lists. this
encourages them to do really slow things: like pulling term vectors for every single document.
> this is really really so much worse than going through multifields or whatever. 

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org


Mime
View raw message