[
https://issues.apache.org/jira/browse/LUCENE-5979?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14154811#comment-14154811
]
Adrien Grand commented on LUCENE-5979:
--------------------------------------
I don't think the documentation of the cost API prevents from interpreting the
cost as an absolute value? Even if it is not something that we do today, I
think all our current DocIdSetIterator (but FixedBitSet) return cost values
that are ok to intepret as an absolute value? Even if this estimation is very
vague it would be much better than what we have today to decide on the filter
strategy to use. Moreover I think this could also prove valuable in
CachingWrapperFilter to decide on the cacheable impl to use: some impls would
likely better at high-cardinality and vice-versa?
> Use the cost API instead of a heuristic on the first document in
> FilteredQuery to decide on whether to use random access
> ------------------------------------------------------------------------------------------------------------------------
>
> Key: LUCENE-5979
> URL: https://issues.apache.org/jira/browse/LUCENE-5979
> Project: Lucene - Core
> Issue Type: Improvement
> Reporter: Adrien Grand
> Assignee: Adrien Grand
> Priority: Minor
> Fix For: 5.0
>
> Attachments: LUCENE-5979.patch
>
>
> Now that some major filters such as TermsFilter and
> MultiTermQueryWrapperFilter return DocIdSets that have a better cost, we
> should switch to the cost API.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
