Re: BUG #5484: sum() bug

Robert Haas <robertmhaas@gmail.com> wrote:
> On Tue, Jun 1, 2010 at 9:24 AM, viras <viras@yandex.ru> wrote:

>> What type of the data is better for using? Numbers up to 100000
>> and accuracy of 2 fractional signs.
>
> numeric is a good choice to avoid loss of precision, but can be a
> bit slower.
>
> You could also try float8.

Yeah, as long as you remember that this is an *approximate* data
type.  If you really mean that you're satisfied with an *accuracy*
of two fractional digits for a number up to 100000, you are OK.  But
realize that means that 1.01 would actually be
1.0100000000000000088817841970012523233890533447265625 and that
100000.01 would actually be
100000.009999999994761310517787933349609375 -- accurate to far more
than two decimal digits, but not *exact*.

If you want exact values based on decimal fractions, you should use
numeric.

-Kevin