[
https://issues.apache.org/jira/browse/CALCITE-7104?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18009468#comment-18009468
]
Silun Dong commented on CALCITE-7104:
-------------------------------------
I prefer the recursive check approach. If I understand correctly, as you said,
the Project with duplicate columns may not be a child of Sort, but a
descendant. Directly matching Sort-Project does not fundamentally solve this
problem.
Another point, you mentioned,
{code:java}
select d1 from (select deptno as d1, deptno + 1 as d2 from dept) as tmp order
by d1, d2{code}
This case can be simplified to order by d1 only. In other words, the
requirement for removing duplicate sort keys can be relaxed from "two keys are
essentially the same" to "two keys are generated from the same column
idempotent operation".
> Remove duplicate sort keys
> --------------------------
>
> Key: CALCITE-7104
> URL: https://issues.apache.org/jira/browse/CALCITE-7104
> Project: Calcite
> Issue Type: Improvement
> Reporter: weihua zhang
> Priority: Major
> Labels: pull-request-available
>
> {code:sql}
> select d1 from (select deptno as d1, deptno as d2 from dept ) as tmp order
> by d1, d2;
> {code}
> {noformat}
> LogicalProject(D1=[$0]), id = 2284
> LogicalSort(sort0=[$0], sort1=[$1], dir0=[ASC], dir1=[ASC]), id = 2283
> LogicalProject(D1=[$0], D2=[$0]), id = 2282
> LogicalTableScan(table=[[default, dept]]), id = 2278
> {noformat}
> LogicalSort can change to:
> {noformat}
> LogicalProject(D1=[$0]), id = 2284
> LogicalSort(sort0=[$0], dir0=[ASC]), id = 2283
> LogicalProject(D1=[$0], D2=[$0]), id = 2282
> LogicalTableScan(table=[[default, dept]]), id = 2278
> {noformat}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
