Negative Zero

[galagast]

I'm ran into an issue where:

1. I have a default resampled Line.
2. Calculated Normals using PolyFrame set to First Edge.
   @N:  { 0,-1, 0 }
3. Reversed the Normals using Facet.
   @N: {-0, 1,-0 }
4. Scattered points, inheriting Normals.
   @N: { 0, 1, 0 }
5. Finally used a wrangle to check if Normals == {0,1,0}

The results were a bit random. Sometime it is true, sometimes it is false.

I did find a couple of solutions to work around this for what I was doing, but I believe something in this simple sequence of nodes is not doing its job properly. I just wanted to hear your thoughts about this before (or even if I needed to) send this as a bug.

Thanks!

negative_zero.hiplc - H16.0.600 Indie

[Skybar]

If you test like this it seems its the -1 it fails on:

int test1 = @N.x == 0;
int test2 = @N.y == -1;
int test3 = @N.z == 0;

if(test2)
    i@check = 1;

Might be a rounding issue somewhere, because if you then change it to int test2 = @N.y >= -1; it works fine. Also if you normalize N in your example it works.

[f1480187]

It's not a negative zero, it's a very small value (-0.0000000596046) shown as a negative zero by Spreadsheet. To understand problem, compute:

@N = {0, -1, 0} - @N;

 

Borrow any good one "almost equal" function example from the web. Here is one of simpler versions:

// Simplified port of CPython's math.isclose() function.
int isclose(float a, b, rtol, atol)
{
    if (a == b)
    {
        return true;
    }

    float d = abs(b - a);
    return ((d<=abs(rtol*b))||
            (d<=abs(rtol*a))) ||
           (d <= atol);
}

 

It will make your comparisons more robust: almost_equal_example.hipnc

Edited June 12, 2017 by f1480187