Adding a cast around the null literal makes sense. I would use “rexBuilder.makeCast(type, constantNull())” rather than “rexBuilder.makeNullLiteral(type.getSqlTypeName())” because there’s a chance that the type needs precision & scale.
Can you please create a Calcite jira case and submit a pull request? And by the way, I think I see similar patterns in AggregateExpandDistinctAggregatesRule, AggregateValuesRule and even RexBuilder.makeOver. Julian > On Feb 2, 2017, at 12:57 AM, Anton Mushin <anton_mus...@epam.com> wrote: > > Hello community, > I have worked for support standard deviation/variance functions in Flink SQL > [1] and I got exception " Type NULL is not supported. Null values must have a > supported type.". > I started search root cause and found what problem come up from > AggregateReduceFunctionsRule[2]. @twalthr confirmed it after[3]. > > Now I have been used WA - the copy AggregateReduceFunctionsRule inside Flink > with changes. > Could we resolve the problem some way? > I look forward to your reply. > > [1] https://issues.apache.org/jira/browse/FLINK-4604# > [2] > https://issues.apache.org/jira/browse/FLINK-4604?focusedCommentId=15615280&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15615280 > [3] > https://issues.apache.org/jira/browse/FLINK-4604?focusedCommentId=15667697&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15667697 > > Regards, > Anton Mushin >
