Re: BUG #18348: Inconsistency with EXTRACT([field] from INTERVAL);

On Mon, Jul  8, 2024 at 01:03:28PM -0400, Tom Lane wrote:
> I took another look at this issue and got annoyed by the fact that the
> proposed coding for "quarter" still doesn't satisfy the rule that
> the output for a negative interval should be the negative of the
> output for the sign-reversed interval.  Specifically, if the month
> field is zero, the v2 patch always emits 1:
> 
> regression=# select extract(quarter from interval '1 day');
>  extract 
> ---------
>        1
> (1 row)
> 
> regression=# select extract(quarter from interval '-1 day');
>  extract 
> ---------
>        1
> (1 row)
> 
> We could fix that by examining the sign of the lower-order fields
> when month is zero, as in the v3 patch attached.  However, I'm not
> at all sure this is really better than v2.  Notably, it makes the
> documentation's statement that the result is "the month field
> divided by 3 plus 1" even more incomplete.  I still don't really
> want to go into details about the behavior for negative intervals.
> OTOH if we did do that, I'd rather write a blanket statement
> about the result being the negative of the result for a positive
> interval.
> 
> Thoughts?

I tested master, patch version 2 and patch version 3 with some sample
extract() queires, attached.  I like patch version 2.  Patch version 3
bothers me because "-600 days" is ignored if months is non-zero, and
used for its sign for zero month values, which seems odd to me;  better
to ignore it.

-- 
  Bruce Momjian  <bruce@momjian.us>        https://momjian.us
  EDB                                      https://enterprisedb.com

  Only you can decide what is important to you.