Support `SUM` `AVG`, `MIN`, `MAX` on `Time` columns.

[alamb]

Is your feature request related to a problem or challenge? Please describe what you are trying to do.

DataFusion does not support SUM, AVG, MIN or MAX on Time type columns a standard SQL feature that could be used to sum how much cumulative time was spent doing some operation, for example

Reproducer is as follows:

select 
  sum(t), 
  avg(t)
from 
  (select cast ('11:00:00' as TIME) as t) as sq;

Doing so today will result in

❯ select 
  sum(t), 
  avg(t)
from 
  (select cast ('11:00:00' as TIME) as t) as sq;
Plan("The function Sum does not support inputs of type Time64(Nanosecond).")
❯ 

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
N/A

Additional context
As described in #200

[alamb]

I also think this is a relatively good first issue as it requires following the model for existing aggregates (e.g. SUM(int64)) so marking it as such.

[waitingkuo]

@alamb postgrseql cast time to interval and then do SUM and AVG

[alamb]

@waitingkuo notes that postgres does:

willy=# select pg_typeof(SUM(time '00:00:00')),  pg_typeof(AVg(time '00:00:00'));
 pg_typeof | pg_typeof 
-----------+-----------
 interval  | interval
(1 row)

this make sense otherwise SUM(Time) is just quite strange

[alamb]

java.lang.ConcurrentModificationException 😆 (our updates got crossed)

[waitingkuo]

@alamb i think we could add more expressions like MAX and MIN here (which is already implemented for timestamp

[waitingkuo]

do we have interval for now?

select '00:00:00'::interval;
NotImplemented("Unsupported SQL type Interval")

but this works

❯ select interval '1 second';
+------------------------------------------------+
| IntervalDayTime("1000")                        |
+------------------------------------------------+
| 0 years 0 mons 0 days 0 hours 0 mins 1.00 secs |
+------------------------------------------------+

[waitingkuo]

@alamb should we use IntervalMonthDayNano insead?

this is what we have for now

    /// Number of elapsed whole months
    IntervalYearMonth(Option<i32>),

    /// Number of elapsed days and milliseconds (no leap seconds)
    /// stored as 2 contiguous 32-bit signed integers
    IntervalDayTime(Option<i64>),

    /// A triple of the number of elapsed months, days, and nanoseconds.
    /// Months and days are encoded as 32-bit signed integers.
    /// Nanoseconds is encoded as a 64-bit signed integer (no leap seconds).
    IntervalMonthDayNano(Option<i128>),

[alamb]

Since we are treating Time literals as Nanosecond precision, I agree @waitingkuo that the output of SUM() and AVG() of TIME should be Interval(MonthDayNano) 👍

❯ select arrow_typeof(cast('10:00:00' as time));
+-----------------------------------------------------------+
| arrowtypeof(CAST(Utf8("10:00:00") AS Time64(Nanosecond))) |
+-----------------------------------------------------------+
| Time64(Nanosecond)                                        |
+-----------------------------------------------------------+
1 row in set. Query took 0.038 seconds.

[alamb]

do we have interval for now?

🤔 looks like another feature gap #3167 (comment)