Re: The 85/0.0085 mistery ?

"Frederick W. Reimer" wrote:
>
> Nothing is wrong, that's just how computers work. Fractional numbers are
> stored in a variety of formats on different platforms. Most support the
> IEEE formats, but some use their own formats. In all cases, that I'm aware
> of, the numbers are stored as a mantissa and exponent. The numbers are
> "base 2", so certain base 10 numbers don't have an exact representation in
> base 2, no matter how many bits are used in the mantissa. Other base 10
> numbers have problems with the "short" mantissa formats, but are represented
> exactly with the "long" mantissa formats.

thank you for the lesson,
I probably have done too much database this time :o

what I was really affraid was the BUT clause ...

> > jctest=> select * from eigthyfive where (cs/csbytenthousand=10000);
> > cs | csbytenthousand | csbyhundred
> > ----+-----------------+-------------
> > 88 | 0.0088 | 0.88
> > 86 | 0.0086 | 0.86
> > 84 | 0.0084 | 0.84
> > (3 rows)
> >
> > *** oh oh 85 disappeared ! ***
> >
> > BUT
> >
> > jctest=> select (85/0.0085=10000);
> > ?column?
> > ----------
> > t
> > (1 row)

... It seems that this is due to the single precision
(float4) instead of double precision (float)
(and even with a test done in base 3 :P)
BTW there is no round(float4) function

thanks

jc