On Fri, Jun 25, 2010 at 6:53 PM, Lee Hachadoorian
<lee.hachadoorian@gmail.com> wrote:
> Is it documented anywhere that floating-point numbers round
> "scientifically", that is 0.5 rounds to the nearest even number? Compare:
>
> SELECT round(2.5::real), round(2.5::numeric), round(3.5::real),
> round(3.5::numeric);
>
> generates
>
> 2 | 3 | 4 | 4
>
> I stumbled across this when I was trying to use round(a::real/b::real)
> to generate a rounded result to dividing integers, and noticed sometimes
> 0.5 was truncated and sometimes it was rounded up. Couldn't find
> anything about this in the archives or the data type documentation. Is
> there something obvious that I'm I missing?
That all floating point representations are approximate?
