[ 
https://issues.apache.org/jira/browse/CALCITE-1493?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15726883#comment-15726883
 ] 

Vineet Garg commented on CALCITE-1493:
--------------------------------------

Following is one example of such test case:

{code}
create table table1(i int, j int);
create table table2(i int, j int);

insert into table1 values(1, 2);
insert into table1 values(1, 3);

insert into table2 values(NULL, 1);
insert into table2 values(2, 1);

select i, j from table1 where table1.j NOT IN (select i from table2 where 
table1.i=table2.j);
{code}

{noformat}
expected result: no rows
{noformat}

More could be found at 
https://github.com/apache/hive/blob/master/ql/src/test/queries/clientpositive/subquery_notin.q
Schema used by queries in subquery_notin.q is setup in 
https://github.com/apache/hive/blob/master/data/scripts/q_test_init.sql

> Wrong plan for NOT IN correlated queries
> ----------------------------------------
>
>                 Key: CALCITE-1493
>                 URL: https://issues.apache.org/jira/browse/CALCITE-1493
>             Project: Calcite
>          Issue Type: Bug
>          Components: core
>            Reporter: Vineet Garg
>            Assignee: Julian Hyde
>              Labels: sub-query
>         Attachments: CALCITE-1493.test.patch
>
>
> Plan generated by calcite with SubqueryRemoveRule followed by de-correlation 
> for the following query:
> {noformat} select sal from emp where empno NOT IN (select deptno from dept 
> where emp.job = dept.name) {noformat}
> is
> {noformat}
> LogicalProject(SAL=[$5])
>   LogicalProject(EMPNO=[$0], ENAME=[$1], JOB=[$2], MGR=[$3], HIREDATE=[$4], 
> SAL=[$5], COMM=[$6], DEPTNO=[$7], SLACKER=[$8])
>     LogicalFilter(condition=[IS NULL($11)])
>       LogicalProject(EMPNO=[$0], ENAME=[$1], JOB=[$2], MGR=[$3], 
> HIREDATE=[$4], SAL=[$5], COMM=[$6], DEPTNO=[$7], SLACKER=[$8], 
> DEPTNO0=[CAST($9):INTEGER], JOB0=[CAST($10):VARCHAR(10) CHARACTER SET 
> "ISO-8859-1" COLLATE "ISO-8859-1$en_US$primary"], $f2=[CAST($11):BOOLEAN])
>         LogicalJoin(condition=[AND(=($2, $10), =($0, $9))], joinType=[inner])
>           LogicalTableScan(table=[[CATALOG, SALES, EMP]])
>           LogicalProject(DEPTNO=[$0], JOB=[$1], $f2=[true])
>             LogicalAggregate(group=[{0, 1}])
>               LogicalProject(DEPTNO=[$0], JOB=[$2], i=[$1])
>                 LogicalProject(DEPTNO=[$0], i=[true], JOB=[$1])
>                   LogicalProject(DEPTNO=[$0], JOB=[$2])
>                     LogicalJoin(condition=[=($2, $1)], joinType=[inner])
>                       LogicalTableScan(table=[[CATALOG, SALES, DEPT]])
>                       LogicalAggregate(group=[{0}])
>                         LogicalProject(JOB=[$2])
>                           LogicalTableScan(table=[[CATALOG, SALES, EMP]])
> {noformat}
> As you can notice instead of doing Left Outer Join Calcite is doing Inner 
> Join. This will produce wrong results.
> Plan for same query just before SubqueryRemove Rule is:
> {noformat}
> LogicalProject(SAL=[$5])
>   LogicalFilter(condition=[NOT(IN($0, {
> LogicalProject(DEPTNO=[$0])
>   LogicalFilter(condition=[=($cor0.JOB, $1)])
>     LogicalTableScan(table=[[CATALOG, SALES, DEPT]])
> }))], variablesSet=[[$cor0]])
>     LogicalTableScan(table=[[CATALOG, SALES, EMP]])
> {noformat}
> Plan just after SubqueryRemove Rule:
> {noformat}
> LogicalProject(SAL=[$5])
>   LogicalProject(EMPNO=[$0], ENAME=[$1], JOB=[$2], MGR=[$3], HIREDATE=[$4], 
> SAL=[$5], COMM=[$6], DEPTNO=[$7], SLACKER=[$8])
>     LogicalFilter(condition=[IS NULL($10)])
>       LogicalFilter(condition=[=($0, $9)])
>         LogicalCorrelate(correlation=[$cor0], joinType=[LEFT], 
> requiredColumns=[{2}])
>           LogicalTableScan(table=[[CATALOG, SALES, EMP]])
>           LogicalAggregate(group=[{0, 1}])
>             LogicalProject(DEPTNO=[$0], i=[true])
>               LogicalProject(DEPTNO=[$0])
>                 LogicalFilter(condition=[=($cor0.JOB, $1)])
>                   LogicalTableScan(table=[[CATALOG, SALES, DEPT]])
> {noformat}
> Looking at above it seems RelDecorrelator have some issue where it is coming 
> up with Inner Join.



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

Reply via email to