[ 
https://issues.apache.org/jira/browse/DRILL-1450?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14157507#comment-14157507
 ] 

Julian Hyde commented on DRILL-1450:
------------------------------------

Maybe previous Calcite (formerly known as Optiq) versions had a problem with 
missing ORDER BY, but it's OK now:

{code:sql}
select *, count(*) over (partition by deptno) as c from emp;
 ENAME | DEPTNO | GENDER | C
-------+--------+--------+---
 Adam  |     50 | M      | 2
 Alice |     30 | F      | 2
 Bob   |     10 | M      | 2
 Eric  |     20 | M      | 1
 Eve   |     50 | F      | 2
 Grace |     60 | F      | 1
 Jane  |     10 | F      | 2
 Susan |     30 | F      | 2
 Wilma |        | F      | 1
(9 rows)
{code}

If there are other issues with Calcite please log them.

> Window function when window definition does not have order by clause would 
> hit assertion error. 
> ------------------------------------------------------------------------------------------------
>
>                 Key: DRILL-1450
>                 URL: https://issues.apache.org/jira/browse/DRILL-1450
>             Project: Apache Drill
>          Issue Type: Bug
>            Reporter: Jinfeng Ni
>
> This query will hit as
> Q1:
> select sum(position_id) over w from cp.`employee.json` window w as ( 
> partition by position_id)
> Stack trace:
> java.lang.AssertionError
>      at 
> org.eigenbase.sql.validate.SqlValidatorUtil.lookup(SqlValidatorUtil.java:242) 
> ~[optiq-core-0.9-drill-r3.jar:na]
>      at 
> org.eigenbase.sql.SqlIdentifier.getMonotonicity(SqlIdentifier.java:261) 
> ~[optiq-core-0.9-drill-r3.jar:na]
>      at 
> org.eigenbase.sql.validate.SelectScope.getMonotonicity(SelectScope.java:149) 
> ~[optiq-core-0.9-drill-r3.jar:na]
>      at org.eigenbase.sql.SqlWindow.isTableSorted(SqlWindow.java:340) 
> ~[optiq-core-0.9-drill-r3.jar:na]
>      at org.eigenbase.sql.SqlWindow.validate(SqlWindow.java:561) 
> ~[optiq-core-0.9-drill-r3.jar:na]
> The cause of this issue is the star column in schema-less system. Drill 
> currently will by default add "*" to a schema-less table.  In optiq, * would 
> be expanded to a list of regular columns if table is schema-aware, but does 
> not expand for schema-less table (That's the behavior starting from optiq 
> 0.9-drill-r2).  That would cause issue in some optiq code logic, since it 
> does not expect * ( should already be expanded).
> For Q1, we could remove the logic of adding * by default to schema-less 
> table, and it would work fine. However, if the query explicitly ask for *, 
> then still, run into similar problem:
> Q2:
> select *, sum(position_id) over w from cp.`employee.json` window w as ( 
> partition by position_id)
> Therefore, the fix had better to be in Optiq. 
>  



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to