Re: BUG #1993: Adding/subtracting negative time intervals

Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > Tom Lane wrote:
> >> Urgh.  I think this is a serious thinko in Michael Glaesemann's rewrite
> >> of interval_mul.
>
> > The reason interval_justify_hours is called by interval multiplication
> > is so multipling an interval '2 days, 4 hours' by 10 doesn't return
> > values like 20 days, 40 hours, etc, but instead something like '21 days,
> > 16 hours', which seems more reasonable.
>
> That's utterly WRONG, though.  The entire *point* of the 8.1 change is
> that days and hours are incommensurable.  We are forced to down-convert
> in some cases --- for example, we can't compute a useful result for
> "0.5 * '1 day'" without imputing "12 hours" as the equivalent of 0.5 day
> --- but we never have to and never should up-convert, except by explicit
> user command ... which is what the justify_hours function is for.

OK, what about 1.5 * '1 day'.  By my logic multiplication and division
were by definition imprecise.  Is the logic that we spill down only for
non-integral values?

> > One solution would be
> > to suggest the use of interval_justify_hours() in the documentation for
> > interval multiplication, and prevent the justification from happening
> > automatically.
>
> Exactly.  Forcing the justification to happen is broken, because there's
> no way to get the other behavior.

If that's what people want, it is fine by me.

--
  Bruce Momjian                        |  http://candle.pha.pa.us
  pgman@candle.pha.pa.us               |  (610) 359-1001
  +  If your life is a hard drive,     |  13 Roberts Road
  +  Christ can be your backup.        |  Newtown Square, Pennsylvania 19073