[
https://issues.apache.org/jira/browse/JENA-111?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13098905#comment-13098905
]
Andy Seaborne commented on JENA-111:
------------------------------------
(project) does change the number of rows and (slice) does not depend on the
columns/variables, so the optimization which is the connection of slice and
order by, is unaffected by pushing projection around. In ARQ project is a
mask, it does not matter is it is inside or outside the slice but it's better
to keep it in the right place.
4 looks right.
3 is harder because DISTINCT interacts with projection. The transformation
above isn't right because distinct changes the number rows, so topN may not
generate enough unique rows for the slice. It sin't possible statically,
without reference to the data to know the right value for N in X+N. This one
needs some further thinking.
> Improving TopN optimization in case of an intermediate OpModifier
> -----------------------------------------------------------------
>
> Key: JENA-111
> URL: https://issues.apache.org/jira/browse/JENA-111
> Project: Jena
> Issue Type: Improvement
> Components: ARQ
> Reporter: Sara Magliacane
> Assignee: Paolo Castagna
> Priority: Minor
> Labels: arq, optimization
> Attachments: topk_project.patch
>
>
> In the TopN optimization (Jena-89) it would be useful to handle also the
> case in which there are some other OpModifiers (I think they are the only
> category of Ops that can be in that position in the tree) between Slice and
> Order By, for example OpProject:
> (slice _1
> (project ?s ...
> (order by <condition>
> ->
> (project ?s ...
> (top 1 <condition>
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
