[
https://issues.apache.org/jira/browse/CALCITE-4644?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17369681#comment-17369681
]
Julian Hyde commented on CALCITE-4644:
--------------------------------------
I agree with [~liyafan]'s remark (in PR review) that {{isPercentile}} seems too
narrow. However I can't think of a better name right now. I propose that you
mark the {{isPercentile}} method experimental. This will allow us to refactor
it later.
{{PERCENTILE_DISC}} should allow any type in the {{ORDER BY}} clause, not just
numeric. (PostgreSQL and MSSQL doc agree with this.) I wonder whether we should
allow an {{ORDER BY}} with any number of keys (not just one). Because we don't
need to find the difference between keys.
You are checking that the {{ORDER BY}} of {{PERCENTILE_CONT}} is numeric, and
that agrees with PostgreSQL and MSSQL. However, I claim that we should also
allow datetime and interval values. Why? Because the difference between two of
these values is numeric. (An interval is basically a number.)
(Note that in {{OVER}}, you can specify {{RANGE}} if and only if the type of
the {{ORDER BY}} key is numeric, interval or datetime, for the same reasons.
Maybe you could re-use the validation logic and error message of {{RANGE}}.)
> Add PERCENTILE_CONT and PERCENTILE_DISC aggregate functions
> -----------------------------------------------------------
>
> Key: CALCITE-4644
> URL: https://issues.apache.org/jira/browse/CALCITE-4644
> Project: Calcite
> Issue Type: Improvement
> Components: core
> Reporter: Rafay A
> Priority: Major
> Labels: pull-request-available
> Time Spent: 2h 40m
> Remaining Estimate: 0h
>
> Add percentile functions in the operator table.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
