[
https://issues.apache.org/jira/browse/CALCITE-4726?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17399950#comment-17399950
]
Will Noble edited comment on CALCITE-4726 at 8/16/21, 7:02 PM:
---------------------------------------------------------------
I think it's a better idea to just count group cardinalities whenever there's a
filter. The unique semantics of `COUNT(*)` makes `ignoreNulls` un-intuitive;
you would probably want to say that `COUNT` always ignores nulls, it's just
that `COUNT(*)` would never encounter a null, but since we reduce `COUNT(a)` to
`COUNT(*)` in a number of places (including `AggregateReduceFunctionsRule`), we
can't always depend on the argument to reference it's true "intentional"
column. Doing it this way also makes the rule more robust against regressions
where we forget to set `ignoreNulls` correctly.
Other than that, feedback incorporated as suggested.
was (Author: wnoble):
I think it's a better idea to just count agg functions whenever there's a
filter. The unique semantics of `COUNT(*)` makes `ignoreNulls` un-intuitive;
you would probably want to say that `COUNT` always ignores nulls, it's just
that `COUNT(*)` would never encounter a null, but since we reduce `COUNT(a)` to
`COUNT(*)` in a number of places (including `AggregateReduceFunctionsRule`), we
can't always depend on the argument to reference it's true "intentional"
column. Doing it this way also makes the rule more robust against regressions
where we forget to set `ignoreNulls` correctly.
Other than that, feedback incorporated as suggested.
> Add support for filters in AggregateExpandWithinDistinctRule
> ------------------------------------------------------------
>
> Key: CALCITE-4726
> URL: https://issues.apache.org/jira/browse/CALCITE-4726
> Project: Calcite
> Issue Type: Improvement
> Reporter: Will Noble
> Assignee: Will Noble
> Priority: Blocker
>
> Currently, {{AggregateExpandWithinDistinctRule}} does not fire if any of the
> aggregate calls in the aggregate have a filter. We should support filters.
> For example, the following query will not be handled by the current rule due
> to the {{FILTER}} clause:
> {code:java}
> SELECT deptno,
> SUM(sal),
> SUM(sal) WITHIN DISTINCT (job) FILTER (WHERE ename LIKE 'A%')
> FROM emp
> GROUP BY deptno
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
