[
https://issues.apache.org/jira/browse/CALCITE-5655?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Runkang He updated CALCITE-5655:
--------------------------------
Description:
When the query contains multiple IN/SOME sub-queries connected with OR
predicate in WHERE clause, the result is wrong. The minimal reproducer is below:
SQL:
{code:SQL}
select empno from sales.empnullables
where deptno in (
select deptno from sales.deptnullables where name = 'dept1')
or deptno in (
select deptno from sales.deptnullables where name = 'dept2')
{code}
The Plan generated by calcite master branch: (Notice the bold part in the
downstream LogicalFilter)
{code:SQL}
LogicalProject(EMPNO=[$0])
LogicalProject(EMPNO=[$0], DEPTNO=[$1])
LogicalFilter(condition=[OR(AND(<>($2, 0), IS NOT NULL($5), IS NOT
NULL($1)), AND(*<>($2, 0)*, IS NOT NULL($9), IS NOT NULL($1)))])
LogicalJoin(condition=[=($1, $8)], joinType=[left])
LogicalJoin(condition=[true], joinType=[inner])
LogicalJoin(condition=[=($1, $4)], joinType=[left])
LogicalJoin(condition=[true], joinType=[inner])
LogicalProject(EMPNO=[$0], DEPTNO=[$7])
LogicalTableScan(table=[[CATALOG, SALES, EMPNULLABLES]])
LogicalAggregate(group=[{}], c=[COUNT()], ck=[COUNT($0)])
LogicalProject(DEPTNO=[$0])
LogicalFilter(condition=[=($1, 'dept1')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
LogicalProject(DEPTNO=[$0], i=[true])
LogicalFilter(condition=[=($1, 'dept1')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
LogicalAggregate(group=[{}], c=[COUNT()], ck=[COUNT($0)])
LogicalProject(DEPTNO=[$0])
LogicalFilter(condition=[=($1, 'dept2')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
LogicalProject(DEPTNO=[$0], i=[true])
LogicalFilter(condition=[=($1, 'dept2')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
{code}
The wrong part is that when build the downstream LogicalFilter for the two
sub-queries, the filter for the second sub-query is AND(<>($2, 0), IS NOT
NULL($9), IS NOT NULL($1)), notice that $2 should be the second sub-query's
intermediate table field ct.c(which field index is $6), but now the actual
reference is the first sub-query's, this leads to wrong plan, and wrong result.
The root cause is that intermediate table alias is the same as the previous
sub-query's, but when lookup intermediate table field, it always returns the
previous one which is not belong to the current subquery.
was:
When the query contains multiple IN/SOME sub-queries connected with OR
predicate in WHERE clause, the result is wrong. The minimal reproducer is below:
SQL:
{{select empno from sales.empnullables
where deptno in (
select deptno from sales.deptnullables where name = 'dept1')
or deptno in (
select deptno from sales.deptnullables where name = 'dept2')}}
The Plan generated by calcite master branch: (Notice the bold part in the
downstream LogicalFilter)
LogicalProject(EMPNO=[$0])
LogicalProject(EMPNO=[$0], DEPTNO=[$1])
LogicalFilter(condition=[OR(AND(<>($2, 0), IS NOT NULL($5), IS NOT
NULL($1)), AND(<>($2, 0), IS NOT NULL($9), IS NOT NULL($1)))])
LogicalJoin(condition=[=($1, $8)], joinType=[left])
LogicalJoin(condition=[true], joinType=[inner])
LogicalJoin(condition=[=($1, $4)], joinType=[left])
LogicalJoin(condition=[true], joinType=[inner])
LogicalProject(EMPNO=[$0], DEPTNO=[$7])
LogicalTableScan(table=[[CATALOG, SALES, EMPNULLABLES]])
LogicalAggregate(group=[{}], c=[COUNT()], ck=[COUNT($0)])
LogicalProject(DEPTNO=[$0])
LogicalFilter(condition=[=($1, 'dept1')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
LogicalProject(DEPTNO=[$0], i=[true])
LogicalFilter(condition=[=($1, 'dept1')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
LogicalAggregate(group=[{}], c=[COUNT()], ck=[COUNT($0)])
LogicalProject(DEPTNO=[$0])
LogicalFilter(condition=[=($1, 'dept2')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
LogicalProject(DEPTNO=[$0], i=[true])
LogicalFilter(condition=[=($1, 'dept2')])
LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
The wrong part is that when build the downstream LogicalFilter for the two
sub-queries, the filter for the second sub-query is AND(<>($2, 0), IS NOT
NULL($9), IS NOT NULL($1)), notice that $2 should be the second sub-query's
intermediate table field ct.c(which field index is $6), but now the actual
reference is the first sub-query's, this leads to wrong plan, and wrong result.
The root cause is that intermediate table alias is the same as the previous
sub-query's, but when lookup intermediate table field, it always returns the
previous one which is not belong to the current subquery.
> Wrong plan for multiple IN/SOME sub-queries with OR predicate
> -------------------------------------------------------------
>
> Key: CALCITE-5655
> URL: https://issues.apache.org/jira/browse/CALCITE-5655
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.34.0
> Reporter: Runkang He
> Priority: Major
>
> When the query contains multiple IN/SOME sub-queries connected with OR
> predicate in WHERE clause, the result is wrong. The minimal reproducer is
> below:
> SQL:
> {code:SQL}
> select empno from sales.empnullables
> where deptno in (
> select deptno from sales.deptnullables where name = 'dept1')
> or deptno in (
> select deptno from sales.deptnullables where name = 'dept2')
> {code}
> The Plan generated by calcite master branch: (Notice the bold part in the
> downstream LogicalFilter)
> {code:SQL}
> LogicalProject(EMPNO=[$0])
> LogicalProject(EMPNO=[$0], DEPTNO=[$1])
> LogicalFilter(condition=[OR(AND(<>($2, 0), IS NOT NULL($5), IS NOT
> NULL($1)), AND(*<>($2, 0)*, IS NOT NULL($9), IS NOT NULL($1)))])
> LogicalJoin(condition=[=($1, $8)], joinType=[left])
> LogicalJoin(condition=[true], joinType=[inner])
> LogicalJoin(condition=[=($1, $4)], joinType=[left])
> LogicalJoin(condition=[true], joinType=[inner])
> LogicalProject(EMPNO=[$0], DEPTNO=[$7])
> LogicalTableScan(table=[[CATALOG, SALES, EMPNULLABLES]])
> LogicalAggregate(group=[{}], c=[COUNT()], ck=[COUNT($0)])
> LogicalProject(DEPTNO=[$0])
> LogicalFilter(condition=[=($1, 'dept1')])
> LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
> LogicalProject(DEPTNO=[$0], i=[true])
> LogicalFilter(condition=[=($1, 'dept1')])
> LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
> LogicalAggregate(group=[{}], c=[COUNT()], ck=[COUNT($0)])
> LogicalProject(DEPTNO=[$0])
> LogicalFilter(condition=[=($1, 'dept2')])
> LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
> LogicalProject(DEPTNO=[$0], i=[true])
> LogicalFilter(condition=[=($1, 'dept2')])
> LogicalTableScan(table=[[CATALOG, SALES, DEPTNULLABLES]])
> {code}
> The wrong part is that when build the downstream LogicalFilter for the two
> sub-queries, the filter for the second sub-query is AND(<>($2, 0), IS NOT
> NULL($9), IS NOT NULL($1)), notice that $2 should be the second sub-query's
> intermediate table field ct.c(which field index is $6), but now the actual
> reference is the first sub-query's, this leads to wrong plan, and wrong
> result.
> The root cause is that intermediate table alias is the same as the previous
> sub-query's, but when lookup intermediate table field, it always returns the
> previous one which is not belong to the current subquery.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
