[
https://issues.apache.org/jira/browse/CALCITE-5930?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17754933#comment-17754933
]
libopeng commented on CALCITE-5930:
-----------------------------------
I found that in sqlToRel, if the data type of the join condition is
inconsistent, the data type will be forcibly changed.
{code:java}
select d.*
from dept d
inner join (
select count(*) c
from emp
where comm is null) e on e.c=d.deptno;
---------------------------------------------------------------------------------------------
LogicalProject(DEPTNO=[$0], DNAME=[$1], LOC=[$2])
LogicalJoin(condition=[=($4, $3)], joinType=[inner])
LogicalProject(DEPTNO=[$0], DNAME=[$1], LOC=[$2], DEPTNO0=[CAST($0):BIGINT
NOT NULL]) <---here has cast
LogicalTableScan(table=[[scott, DEPT]])
LogicalAggregate(group=[{}], C=[COUNT()])
LogicalFilter(condition=[IS NULL($6)])
LogicalTableScan(table=[[scott, EMP]]){code}
If the join is converted from in, the condition does not perform a type change.
{code:java}
select *
from dept
where deptno in (
select count(*)
from emp
where comm is null);
---------------------------------------------------------------------------------------------
LogicalProject(DEPTNO=[$0], DNAME=[$1], LOC=[$2])
LogicalJoin(condition=[=($0, $3)], joinType=[inner])
LogicalTableScan(table=[[scott, DEPT]]) <----not
cast
LogicalAggregate(group=[{0}])
LogicalAggregate(group=[{}], EXPR$0=[COUNT()])
LogicalFilter(condition=[IS NULL($6)])
LogicalTableScan(table=[[scott, EMP]])
{code}
Therefore, I think adding 'cast' when converting in to join can solve this
problem.
If anyone has any other suggestions, please let me know.
> When data types do not match, calcite may make a calculation error
> ------------------------------------------------------------------
>
> Key: CALCITE-5930
> URL: https://issues.apache.org/jira/browse/CALCITE-5930
> Project: Calcite
> Issue Type: Bug
> Reporter: libopeng
> Priority: Major
>
> Wrong calculation result
> {code:java}
> select *
> from dept
> where deptno in (
> select count(*)
> from emp
> where comm is null);
> +--------+------------+----------+
> | DEPTNO | DNAME | LOC |
> +--------+------------+----------+
> +--------+------------+----------+ {code}
> Correct calculation results
> {code:java}
> select *
> from dept
> where cast(deptno as bigint) in (
> select count(*)
> from emp
> where comm is null);
> +--------+------------+----------+
> | DEPTNO | DNAME | LOC |
> +--------+------------+----------+
> | 10 | ACCOUNTING | NEW YORK |
> +--------+------------+----------+{code}
> The above two cases can be reproduced in agg.iq.
> Or look at this submission
> [https://github.com/apache/calcite/commit/6cef259f29884fbda3f165d579232285998d1b46].
>
>
>
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
