Re: Floating point implementation - was something else...

I finally understand the c-weenies tagline of "1+1=4 for sufficiently 
large values of 1".  But the comment made earlier was about three 
significant places.   Hundredths in your example is three significant 
digits. 

In my humble opinion, for most business reporting, accuracy is better 
served with three significant digits than it is with  careful attention to 
pennies.  Of course audit trails and balancing entries need to balance to 
pennies, but for sales, marketing, trend-analysis, and most reporting 
requirements anything beyond three significant digits is confusing and 
clouds the relationships of the numbers for most managers.

_______________________
Booth Martin
Booth@MartinVT.com
http://www.MartinVT.com
_______________________




Don <dr2@cssas400.com>
Sent by: owner-midrange-l@midrange.com
07/04/2000 10:55 PM
Please respond to MIDRANGE-L

 
        To:     Douglas Handy <dhandy1@bellsouth.net>
        cc:     MIDRANGE-L@midrange.com
        Subject:        Floating point implementation - was something else...



Doug, Hans, et al..

If I'm misunderstanding this, my apologies and would appreciate the
correctings.....:) 

I expect to see FP errors by definition.  That's not what my concern
was...my concern was with the point of significant figures that the error
would be expected to occur....HUNDRETHS!.  And for a small number! 

What I'm hearing is that if I multiply 1.1 * 2.2 and if I get 2.41 or 2.43
(2.42 is correct) I should be happy.  I contend that this is a compiler or
internal error...   OR, if I send a function a constant 2.42, I should be
content that if picks some number between 2.40 and 2.50...

That's a problem. 

Yeah, if I'm converting from Punts (0.8271/USD) to Turkish
Lira(621550.00/USD), I may have a natural floating point concern I would
expect to see in the ten-thousanths position...of the
RESULT(product)...but if you're going to sell me 1/10000 of a
Punt........:)  Which is to say, that in most non-cascading calculations,
this kind of error is normative and ignored due to it's insignificance...

Interestingly enough, I wrote some currency arbit code in Borland Pascal
years ago, using "real" fields that we never had any problems with...and
it worked....:) 

...just back from the fireworks...:)

Don in DC
 

On Tue, 4 Jul 2000, Douglas Handy wrote:

> Don,
> 
> >I've done many years of eco and financials sims on many languages and 
to
> >have roundoff in the HUNDRETHS position is just unforgivable. 
> 
> Which is why one normally wants to avoid floating-point for currency
> stuff (i.e., most business applications).  And when you do need to use
> it, make sure you use half-adjust.  Also it seems best to isolate just
> the part of the formula which uses float to just a single expression
> on the EVAL, and put the result back into a packed number (with
> half-adjust).  Then continue the formula in another expression.  Or
> use %DECH() in the formula where needed.
> 
> I think many languages just perform the half-adjust implicitly or as a
> by-product of a print formatting expression, whereas RPG makes you do
> the half-adjust explicitly.
> 
> Doug
> +---
> | This is the Midrange System Mailing List!
> | To submit a new message, send your mail to MIDRANGE-L@midrange.com.
> | To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
> | To unsubscribe from this list send email to 
MIDRANGE-L-UNSUB@midrange.com.
> | Questions should be directed to the list owner/operator: 
david@midrange.com
> +---
> 

+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to 
MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: 
david@midrange.com
+---




+---
| This is the Midrange System Mailing List!
| To submit a new message, send your mail to MIDRANGE-L@midrange.com.
| To subscribe to this list send email to MIDRANGE-L-SUB@midrange.com.
| To unsubscribe from this list send email to MIDRANGE-L-UNSUB@midrange.com.
| Questions should be directed to the list owner/operator: david@midrange.com
+---