normally, the only pratical reason for such requirements (to get the same
results on all platforms) is that you are doing financial math with floating
point. Otherwise, if you do scientific math, it is more important that the
results are as exact as possible, but not the same on all platforms.
So, assumed that you are doing financial math, the error in the computations
using floating point are in the 10e-15 range (in relation to the result), so for
computing dollars and cents, this doesn't really matter. I always assume that
you are using double (8 bytes), not float (4 bytes).
The real problem is rounding.
Maybe your computation yields a result of 12.345 dollar, which should be rounded
But due to floating point number representation and small errors in the float
arithmetic, your result is (in decimal) 12.3449999996. So it get's rounded to
12.34, and you're in a real mess.
This problem needs to be fixed. So after some computations (and: before doing
output or converting into decimal representation or such things), you have to
add to the FP numbers some little epsilon, which corrects the error and allows
the subsequent rounding to take the correct value.
Our experience is: if you choose the epsilon carefully, you will get the same
results even on different platforms in MOST cases. But there is still no
guarantee that it will work in ALL cases. Some ten years ago I explained our
rounding function to a math scientist, and he said: "it's not bad, what you're
doing, but you can do anything you want: I'll always find an example where it
will not work".
Am 23.03.2011 13:08, schrieb M.S.:
> Thank you.
> Our problem is: which option and where to change to get the same result.
> Till now we have ignored this differences (in AFP documents, position
> difference sometimes in 1 pel),
> but some of our customers are saying, they would like to get the same
> result, in all platforms.