[
https://issues.apache.org/jira/browse/CALCITE-7828?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aleksandr Efimov updated CALCITE-7828:
--------------------------------------
Description:
ROLLUP(a) is GROUPING SETS ((a), ()), and the () set yields one row even when
the input is empty, as a query with no GROUP BY does (CALCITE-1059). Calcite
returns no rows instead:
{code:sql}
SELECT a, COUNT(*) AS c FROM (VALUES (1)) AS t(a) WHERE FALSE GROUP BY
ROLLUP(a);
{code}
The expected result is one row, (NULL, 0). CUBE(a, b) and GROUPING SETS ((a),
()) lose the row for () in the same way.
Two places drop the row. {{PruneEmptyRules.AGGREGATE_INSTANCE}} replaces the
aggregate with an empty {{Values}}, because its guard,
{{Aggregate.isNotGrandTotal}}, only checks that the group count is greater than
0. With that rule removed, the plan keeps an {{EnumerableAggregate}} over the
empty {{Values}}, but the query still returns no rows:
{noformat}
EnumerableAggregate(group=[{0}], groups=[[{0}, {}]], C=[COUNT()])
EnumerableValues(tuples=[[]])
{noformat}
Reproduced on 1.42.0 and on main (4e9460600c).
was:
ROLLUP(a) is GROUPING SETS ((a), ()), and the () set yields one row even when
the input is empty, as a query with no GROUP BY does (CALCITE-1059). Calcite
returns no rows instead:
{code:sql}
SELECT a, COUNT(*) AS c FROM (VALUES (1)) AS t(a) WHERE FALSE GROUP BY
ROLLUP(a);
{code}
The expected result is one row, (NULL, 0). CUBE(a, b) and GROUPING SETS ((a),
()) lose the row for () in the same way.
Two places drop the row. {{PruneEmptyRules.AGGREGATE_INSTANCE}} replaces the
aggregate with an empty {{Values}}, because its guard,
{{Aggregate.isNotGrandTotal}}, only checks that the group count is greater than
0. With that rule removed, the plan keeps {{EnumerableAggregate(group=[{0}],
groups=[[{0}, {}]])}} over the empty {{Values}}, but the query still returns no
rows.
Reproduced on 1.42.0 and on main (4e9460600c).
> Aggregate whose grouping sets include () returns no rows on empty input
> -----------------------------------------------------------------------
>
> Key: CALCITE-7828
> URL: https://issues.apache.org/jira/browse/CALCITE-7828
> Project: Calcite
> Issue Type: Bug
> Components: core
> Affects Versions: 1.42.0
> Reporter: Aleksandr Efimov
> Priority: Major
>
> ROLLUP(a) is GROUPING SETS ((a), ()), and the () set yields one row even when
> the input is empty, as a query with no GROUP BY does (CALCITE-1059). Calcite
> returns no rows instead:
> {code:sql}
> SELECT a, COUNT(*) AS c FROM (VALUES (1)) AS t(a) WHERE FALSE GROUP BY
> ROLLUP(a);
> {code}
> The expected result is one row, (NULL, 0). CUBE(a, b) and GROUPING SETS ((a),
> ()) lose the row for () in the same way.
> Two places drop the row. {{PruneEmptyRules.AGGREGATE_INSTANCE}} replaces the
> aggregate with an empty {{Values}}, because its guard,
> {{Aggregate.isNotGrandTotal}}, only checks that the group count is greater
> than 0. With that rule removed, the plan keeps an {{EnumerableAggregate}}
> over the empty {{Values}}, but the query still returns no rows:
> {noformat}
> EnumerableAggregate(group=[{0}], groups=[[{0}, {}]], C=[COUNT()])
> EnumerableValues(tuples=[[]])
> {noformat}
> Reproduced on 1.42.0 and on main (4e9460600c).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
