[
https://issues.apache.org/jira/browse/CALCITE-4018?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17113608#comment-17113608
]
Haisheng Yuan commented on CALCITE-4018:
----------------------------------------
Thanks, I have updated. Even though Values may not be worth optimizing in
practice, it will be a good example to demonstrate how useful it is to do trait
propagation on leaf nodes. The problem in CALCITE-2624 can be easily solved by
overriding EnumerableTableScan.passThrough(), return an ElasticScan or
IndexScan if there is an index that can satisfy the required collation.
> EnumerableValues should provide requested traits
> ------------------------------------------------
>
> Key: CALCITE-4018
> URL: https://issues.apache.org/jira/browse/CALCITE-4018
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Reporter: Haisheng Yuan
> Priority: Major
>
> Only passThrough is needed.
> Currently, when Values is created, it will enumerate all the possible
> collations no matter parent operator requires it or not, it will be a
> disaster if the Values has thousands of columns, and the parent operator may
> be just a hash aggregate or hashjoin, which doesn't care about its collation.
> The collation should be created on demand by calling passThrough.
> e.g.
> {code:java}
> SELECT * from (values
> (1, 1),
> (2, 1),
> (1, 2),
> (2, 2)
> ) as t(a, b)
> order by b, a
> {code}
> Currently Calcite will generate plan:
> {code:java}
> EnumerableSort(sort0=[$1], sort1=[$0], dir0=[ASC], dir1=[ASC])
> EnumerableValues(tuples=[[{ 1, 1 }, { 2, 1 }, { 1, 2 }, { 2, 2 }]])
> {code}
> But after this JIRA, I am expecting a plan without Sort.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
