[
https://issues.apache.org/jira/browse/SOLR-8962?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15234076#comment-15234076
]
Dennis Gove commented on SOLR-8962:
-----------------------------------
Been giving this a bit of thought and I think I'm gonna go with the following
approach.
By default this will use Java's standard Collections sort method which provides
nlogn performance. If we also take into account the time to read all the tuples
from the incoming sort into a List (currently thinking LinkedList because
insertion is O(1) but I do want to confirm that sorting will still be done in
nlogn with a linked list) then that cost is just n. Total time cost for the
sort would be 2nlogn.
That said, there may be cases where a more performant sort, like counting sort,
could be used. For those cases the stream will also accept an algorithm
parameter asking for the use of a different sorting algorithm.
> Add sort Streaming Expression
> -----------------------------
>
> Key: SOLR-8962
> URL: https://issues.apache.org/jira/browse/SOLR-8962
> Project: Solr
> Issue Type: New Feature
> Reporter: Joel Bernstein
> Priority: Critical
> Fix For: 6.1
>
>
> The sort Streaming Expression does an in memory sort of the Tuples returned
> by it's underlying stream. This is intended to be used for sorting sets
> gathered during local graph traversals. This will make it easy to gather sets
> during a traversal and use all of the sort based set operations (merge,
> innerJoin, outerJoin, reduce, complement, intersect).
> This will be particularly useful with the gatherNodes expression (SOLR-8925).
> Sample syntax:
> {code}
> intersect(
> sort(gatherNodes(...), "fieldA asc"),
> sort(gatherNodes(...), "fieldA asc"),
> on)
> {code}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org
For additional commands, e-mail: dev-h...@lucene.apache.org
