Re: [bug-gawk] Error Gawk Manual FP Precision

[arnold]

Hi.

> > Dear bug-gawk:
> >
> > The section on comparing two floating point values correctly recommends
> > making avoiding simple equality comparisons in favor of ensuring the
> > values are close enough.
> >
> > The code given in 15.4.1.2 "Be Careful Comparing Values" is clearly
> > wrong, though:
> > 
> >     The general wisdom when comparing floating-point values is to see
> >     if they are within some small range of each other (called a delta,
> >     or tolerance).  You have to decide how small a delta is important
> >     to you.  Code to do this looks something like the following:
> >             
> >             delta = 0.00001                 # for example
> >             difference = abs(a) - abs(b)    # subtract the two values
> >             if (difference < delta)
> >                 # all ok
> >             else
> >                 # not ok
> >
> > If the desired value, a is -1000, and b is +1000, the answer is off by
> > 2000. But difference will be zero.
>

How does this look to you? I'm not sure it's right.

Thanks,

Arnold
------------------------------------------------------------
function abs(f)
{
        return f < 0 ? -f : f
}

function within_delta(a, b, delta)
{
        if ((a > 0 && b > 0) || (a < 0 && b < 0))
                return abs(a - b) <= abs(delta)
        else
                return (a - b) <= abs(delta)
}

function test_delta(a, b, delta,        result)
{
        result = within_delta(a, b, delta)
        printf("%.12g, %.12g within delta %.12g -> %s\n", a, b, delta,
                        result ? "TRUE" : "FALSE")
}

BEGIN {
        delta = 0.001

        n = split("100.0001 100.0002 100 -100 100.0001 -100.0001", pairs)

        for (i = 1; i <= n; i++)
                pairs[i] += 0      # force numeric

        for (i = 1; i <= n; i += 2)
                test_delta(pairs[i], pairs[i+1], delta)
}