[
https://issues.apache.org/jira/browse/OPTIQ-312?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14044408#comment-14044408
]
Vladimir Sitnikov commented on OPTIQ-312:
-----------------------------------------
It looks like there's a simple solution: ensure the first "CalcRelSplitter
level" is CalcRelType, that triggers projection optimization and results in the
desired plan without new rules.
See https://github.com/julianhyde/optiq/pull/307
> Need PushProjectThroughWindowRel
> --------------------------------
>
> Key: OPTIQ-312
> URL: https://issues.apache.org/jira/browse/OPTIQ-312
> Project: optiq
> Issue Type: Improvement
> Reporter: Vladimir Sitnikov
> Assignee: Julian Hyde
> Priority: Minor
>
> Currently optiq does not push projects through {{WindowRel}}, thus it feeds
> excessive data to {{WindowRel}}.
> {code:sql}
> select *
> from (
> select "empid", "deptno", count(*) over (partition by "deptno") c
> from "hr"."emps"
> )
> where "deptno"=10 and "empid"=100{code}
> Optiq should recognize it is required to track just fields {{empid}} and
> {{deptno}}.
> However the java code is as follows (note that empid, name, salary, and
> commission are projected):
> {code:java}
> /* 47 */ _list.add(new Object[] {
> /* 48 */ row.empid,
> /* 49 */ row.deptno,
> /* 50 */ row.name,
> /* 51 */ row.salary,
> /* 52 */ row.commission,
> /* 53 */ COUNTa0w0});
> {code}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
