Re: Strange result with exponents...

[Hans Boldt <boldt@xxxxxxxxxx>]

jt wrote:
> Cheers, Hans...  I couldn't figure out why anyone would WANT to use FP,
> until you mentioned speed.
>
> But I can't possibly disagree More with this statement: "My point is just
> that float format was primarily intended for scientific calculations, and an
> understanding of the concept of precision is important."  It's important to
> You, because you write Turing machines (compilers).
>
> And FP (afaik) IS covered in even 2-year programs and so yes, I agree,
> programmers Should be familiar.  HOWEVER, you have almost no understanding
> of what it is to do biz programming, as you've mentioned before.  And if you
> think biz programmers Should be as familiar as you are with FP, then you are
> mistaken.  When I do arithmetic I want a correct result, not an estimate via
> FP.  Because the arithmetic is only a small SUBSET of the issues I hafta
> deal with in writing a biz app.
>
> I should add, in my experience anyway...  But I'd be surprised if there were
> a whole LOTta people who disagree with this aspect of the problem.

A couple of points:

1) With respect to business programming, when you do arithmetic on 
monetary values, you want to, no, you NEED to follow the rules and 
regulations governing monetary computations in your jurisdiction. That 
may or may not require "exact" results. For example, as I understand it, 
the EU requires Euro computations to be exact to the whole Euro cent, 
with a certain style of rounding. That is, no fractional cents. Other 
jurisdictions may have different rules. And in cases where the 
jurisdiction does not lay down precise rules, companies may have 
specific accounting policies you have to follow.

My point here is that "correct result" is often largely a matter of 
accounting policy, and it's the job of the application programmer to 
match the capabilities of the chosen tool with the rules. As you've 
said, compiler writers are not generally well versed in accounting 
practice. ;-)

2) As I've said before, IEEE float will not give you an exact result 
anyways, so of course it should be avoided for monetary calculations.

3) "Infinite precision" integer operations are certainly available in 
some languages, like Python. But you might not like the performance of 
such interpreted languages compared to compiled languages like RPG. ;-)

Cheers! Hans