BUG: findSpan may be stuck in an infinite loop in this case.

[萧萧]

Hi ~ Tinynurbs is a nice library ~
But I encountered a problem when I was using the following code.
findSpan will be stuck in an infinite loop.

vector<float> knots = {0, 1, 1.3, 2.1, 3.6, 4.0};
int x = tinynurbs::findSpan(2, knots, 2.1f);

[Henrik Jakob]

Hi,
I ran into the same bug. Apperently it gets stuck on the last knot.
I fixed it by uncommenting some lines in the upper part in the findSpan function.

// For u that is equal to last knot value
if (util::close(u, knots[n + 1])) {
    return n;
}

This at least cured my problems

[alektron]

I stumbled upon the same issue and I'm relatively sure I know what's up.
Uncommenting the code that @henrij22 mentioned can indeed work. However the issue lies in the lines of code directly below.

// For values of u that lies outside the domain
if (u > (knots[n + 1] - std::numeric_limits<T>::epsilon()))
{
    return n;
}
if (u < (knots[degree] + std::numeric_limits<T>::epsilon()))
{
    return degree;
}

std::numeric_limits::epsilon() is defined to be FLT_EPSILON which is defined as
difference between 1.0 and the next representable value for float.

That means if knots[n + 1] is indeed 1.0f, this works. By the nature of floating point representation however, the delta between a value and closest possible next value increases for bigger values. In my case where knots[n +1] was 3.0f, this delta is 2.384185791015625​e-7 which is bigger than FLT_EPSILON (1.192092896e-07F).
Therefore 3.0f - FLT_EPSILON is equal to 3.0f so the check fails and we enter the loop below which gets stuck in an infinite loop.

The question now is why this specific approach was chosen instead of just using <= and >= instead of < and > respectively.
As far as I can tell, this would resolve the issue.

[Manuel Antúnez]

Hi,
I've submitted a PR to fix the issue with the findSpan method entering an infinite loop. The fix replaces the custom binary search with std::lower_bound, which improves robustness and prevents the loop. Please review the PR here.