Re: [boost] Fixed-Point Decimal Type

From: Ed Brey (brey_at_[hidden])
Date: 2001-05-09 16:04:02

From: "Fernando Cacciola" <fcacciola_at_[hidden]>
> > Yes, and you must be able to do binary operations with all numeric
> > types without having to do explicit type conversions; using a
> > float/double operand should give a float/double result,
>
> I don't think that mixed operations with floating point values should
give a
> floating point result.
> If the class has absent/explicit constructors from fp types, this will
> prevent many usual expressions;
> and if it has regular constructors from fp-types it will produce too
many
> temporaries and conversions.

Consider the sales tax use case:

double rate1 = 1./3.;
double rate2 = .051;

fixed<2> subtotal(1, 23);
fixed<2> tax1 = subtotal * rate1;
fixed<2> tax2 = subtotal * rate2;
fixed<2> total = subtotal + tax1 + tax 2;

Observations:

- The precision of the rate has no correlation to the precision of the
currency, and moreover it is perfectly reasonable for a rate to use all
the precision that the processor can handle, as in the 1/3 case.  This
leads to the temptation to perform the multiplication in floating point
and then round back to fixed, but then an evenly divisible rate like .5
of .1 might not come out right.  Converting rate1 to a fixed seems to be
the way to go, so long as it is a fixed<18> (or whatever the maximum
size is).

- To avoid the currencies-not-adding embarrassment, the calculation
should return a fixed-point number, properly rounded off, even if this
causes the total to not be equal to fixed<2>(1.23 * (1 + rate1 +
rate2)).  (Another approach is to figure out the final total by adding
the rates first, and then work backwards, fudging the taxes to make it
all add up.  I don't know whether this is common practice in the
business world, but in any case it seems like it would have to be hand
crafted beyond what the "fixed" class would provide.)

- Given guarantees like those implied above, hopefully it will not be
necessary to require explicitly conversion when multiplying/dividing
fixed and floating types.