[
https://issues.apache.org/jira/browse/CALCITE-4661?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17376024#comment-17376024
]
Julian Hyde commented on CALCITE-4661:
--------------------------------------
The PR looks very good. I've marked 'fix in 1.28' so that we don't drop this. A
few things:
* Does MODE need to reserved? Does it even need to be a keyword? I think this
change could be accomplished without any changes to the parser.
* I don't see why {{RelDataTypeSystem.deriveModeType}} is necessary. The type
is always the type of the expression, right?
* Add a comment to each test in agg.iq, with the intent of the test. E.g. one
test seems to be deliberately testing by a NULL group key. Another is
deliberately applying MODE to the group key.
* Is there a test of MODE applied to not-null column in an empty table (or
WHERE FALSE); if not, add one.
* Add a test (maybe in {{agg.iq}}) for {{MODE}} with {{FILTER (WHERE ...)}}
* I see that you allow {{DISTINCT}}. {{DISTINCT}} makes {{MODE}} trivial, so I
think you should omit ALL and DISTINCT from the syntax in reference.md.
(PostgreSQL throws if you specify DISTINCT. I think we should allow it.) Also
can you add a test case with DISTINCT, stating what is the expected behavior.
* Do you expect MODE to work as a windowed aggregate function? If so, add a
test to winagg.iq.
* Add a test case for MODE in a {{GROUPING SETS}} query.
> Add MODE aggregate function
> ---------------------------
>
> Key: CALCITE-4661
> URL: https://issues.apache.org/jira/browse/CALCITE-4661
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Reporter: duan xiong
> Assignee: duan xiong
> Priority: Major
> Labels: pull-request-available
> Time Spent: 2h 10m
> Remaining Estimate: 0h
>
> Add MODE functions in the operator table which returns the most frequent
> input value (arbitrarily choosing the first one if there are multiple
> equally-frequent results)
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
