I believer that PERCENTILE_DISC was implemented in https://issues.apache.org/jira/browse/CALCITE-4644 <https://issues.apache.org/jira/browse/CALCITE-4644>. There are open issues to support any sortable type [ https://issues.apache.org/jira/browse/CALCITE-4670 <https://issues.apache.org/jira/browse/CALCITE-4670> ] and also to change the implementation strategy [ https://issues.apache.org/jira/browse/CALCITE-4666 <https://issues.apache.org/jira/browse/CALCITE-4666> ].
Does your case map onto any of those? Is PERCENTILE_DISC able to deduce return types from its arguments today? If so, how does it do it? Julian > On Aug 3, 2022, at 1:40 PM, Itiel Sadeh <iti...@sqreamtech.com.INVALID> wrote: > > Hello, > > First of all I just wanted to thank all of you for the work you are doing > on Calcite. > > We want to add support for the percentile_disc aggregate function, but we > encounter a problem. > Our percentile_disc return type is dependent on the column of the "order > by" clause (just like postgresql > <https://www.postgresql.org/docs/current/functions-aggregate.html>). > However, I don't see how to achieve that on calcite. The issue is that the > sort column argument is not passed to the percentile_disc function, Rather, > it is stored outside of it as a collation. Therefore, I cannot use the > SqlReturnTypeInference mechanism. > > If someone has an idea on how to achieve this it would be very much > appreciated. > Thank you, > > Itiel