[ 
https://issues.apache.org/jira/browse/CALCITE-5155?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17602583#comment-17602583
 ] 

Julian Hyde commented on CALCITE-5155:
--------------------------------------

Thanks for the feedback, [~tdsilva]. That all makes sense.

I realized that there's an interesting corner case, '7 hour', which cannot roll 
up to day (because 7 hours do not divide 24 hours) but can roll up to week 
(because 7 hours divides 168 hours).

And (an even rarer case) if a time unit is not aligned with its parent (i.e. 
start when its parent starts), then the question arises: what does it align 
with? You could imagine a 'year-aligned week' time unit where each week starts 
on a Thursday if the year starts on a Thursday. But of course the vast majority 
of time units are aligned on the epoch.

> Extensible time units
> ---------------------
>
>                 Key: CALCITE-5155
>                 URL: https://issues.apache.org/jira/browse/CALCITE-5155
>             Project: Calcite
>          Issue Type: Bug
>            Reporter: Julian Hyde
>            Priority: Major
>
> Allow a type system to define its own time units and how they are rolled up.
> Currently, time units are used in the {{EXTRACT}}, {{FLOOR}}, {{TRUNC}} 
> functions, and include {{YEAR}}, {{QUARTER}}, {{MONTH}}, {{HOUR}}, 
> {{MINUTE}}, {{NANOSECOND}}. For example {{FLOOR(t TO HOUR)}} is valid.
> A type system would be allowed to define extra time units. Once a time unit 
> is defined the {{EXTRACT}}, {{FLOOR}} and {{TRUNC}} functions should just 
> work.
> The definition of might consist of a base unit and multiplier. So 
> {{MINUTE15}} would be based on {{MINUTE}} with a multiplier of 15. 
> Various rules know that you can roll up {{FLOOR(t TO DAY)}} to {{FLOOR(t TO 
> MONTH)}} but you cannot roll {{FLOOR(t TO WEEK)}} to {{FLOOR(t TO MONTH)}}. 
> When you define a new time unit, the type system can deduce that full set of 
> time units that it can roll up to, and which can roll up to it.
> Should we support time units that do not evenly divide the next largest time 
> unit? For example the number of seconds since the top of the hour modulo 7. 
> 60 and 3,600 are not a multiples of 7, so {{SecondOfHourMod7}} would be 
> different from {{SecondOfMinuteMod7}}.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to