At 11:15 AM 24/01/2003 -0300, you wrote:

>MT> I guess that depends a little on what you want. Are the FLOAT fields in
>MT> Paradox _real_ floats?
>
>That's my doubt too. I do not know how paradox (or even the BDE)
>handle this. I would like to use the firebird datatype that has the
>most similar "handling" as Paradox-Float to avoid customers complaints
>about diferent results after the transition from Pdox to FB.

Carlos,
Paradox floats are single-precision and suffer from the same problems as IB
floats. If you don't see any evidence in the code that any rounding rules
have been applied to the values to correct for precision errors, I suppose
you can assume that your customer doesn't care about them.

If *you* care about them, you will probably apply the types that will
achieve the precision you need and will explain to your customer that this
is one more advantage of moving over to Firebird...that their calculation
results will possibly be different (and more accurate).

I like Ann's rule of thumb: use numeric() for things you count, double for
things you measure. Just make sure your numerics have enough scale and
precision to accommodate the results of calculations.

heLen