Bug : CanonicalBox3-Cylinder3 intersection has bugs in my test case!

[ottoGithub]

Version Is The Newest.
File Version: 8.0.2025.05.10

TestCase:

void IntersectBoxCylinderWindow3::myTestIntersection1()
{
    Cylinder3<double> cylinder;
    cylinder.axis.origin = { 1434.5667763484134, 389.35568672554501, 92.150001483748767 };
    cylinder.axis.direction = { 0.0f, 0.0f, 1.0f };
    cylinder.radius = 150.0f;
    cylinder.height = 10.0f;

    OrientedBox3<double> box;
    box.center = { 503.48061758921818, 510.45977283453658, 70.752710265075905 };
    box.axis[0] = { 1.0f, 0.0f, 0.0f };
    box.axis[1] = { 0.0f, 1.0f, 0.0f };
    box.axis[2] = { 0.0f, 0.0f, 1.0f };
    box.extent = { 10.7810440, 10.7810440, 25.5310440 };

    // roll, pitch, yaw
    Quaternion<double> incr = Rotation<3, double>(EulerAngles<double>(0, 1, 2, 0.20377350194043725 * GTE_C_DEG_TO_RAD, -4.3779816887311274 * GTE_C_DEG_TO_RAD, 78.185201704856738 * GTE_C_DEG_TO_RAD));
    for (int32_t i = 0; i < 3; ++i)
    {
        box.axis[i] = HProject(
            gte::Rotate(incr, HLift(box.axis[i], 0.0)));
    }


    TIQuery<double, OrientedBox3<double>, Cylinder3<double>> bcQuery;
    bool result = bcQuery(box, cylinder).intersect;
    printf("myTestIntersection: %d\n", result);  // result is true!
}

=====================
The intersect result is TRUE! It's obviously not right.

The distance cylinder2box is: 939.1728
the bounding sphere of cylinder is less than 160
the bounding sphere of box is less than 45
cylinder is to far away from box that they can not be intersect.

But the result is true.

========================
Debuging :

result.intersect = DoQueryNoZeros(C, D, r, hDiv2, E); // return true!
function: DoQueryNoZeros
    // (U2, -D)
    lower = (s5p >= zero ? -E[1] : -E[1] - s5p / D[2]);
    upper = (s6p <= zero ? +E[1] : +E[1] - s6p / D[2]);
    if (lower <= upper)
    {
        if (s5p >= zero)
        {
            P0 = { +E[0] - s5p / D[0], -E[1], +E[2] };
        }
        else
        {
            P0 = { +E[0], -E[1] - s5p / D[1], +E[2] };
        }

        if (s6p <= zero)
        {
            P1 = { -E[0] - s6p / D[0], +E[1], +E[2] };
        }
        else
        {
            P1 = { -E[0], E[1] - s6p / D[1], +E[2] };
        }
// HERE return true:

        sqrDistance = ComputeSqrDistance(P0, P1, C, W0, W1);
        if (sqrDistance <= sqrRadius)
        {
            return true;
        }
    }

sqrDistance: 22083.197885594218
sqrRadius: 22500

How to fix it?
mybe construct line segments using 'projection' is not correct?

[davideberly]

I am in the process of clearing out my queue of issues and bug reports. I am currently working on your issue and hopefully will get back to you soon regarding what I discover.

[davideberly]

In the case where your example leads to failure (U2,-D), the mathematics in the PDF is incorrect. I implemented the code from the PDF, so the code is incorrect. I am fixing this case and analyzing the other cases involving directions D or -D.

[OttoX]

Got it, thanks

[davideberly]

@OttoX

I worked through every equation in the document for test-intersection query of canonical box and cylinder, verifying each with Mathematica. I do not see errors (although I fixed some typographical errors). I believe the problem is your perception of the rotated box. You rotate the box (using Euler angles). The problem is that the center of the box is NOT rotated, For a properly rotate box, the 8 vertices obtained by calling

std::array<Vector3<double, 8> vertices{};
box GetVertices(vertices));

should be the corners of the rotated box, but in fact they are not. The image is from Mathematica. The colored objects at the top of image are the canonical box (not rotated) and the cylinder. The points at the bottom of the image are NOT vertices of the rotated box. So in fact you have a malformed box and the query does not apply.

[ottoGithub]

I made a new testcase on intersecting CanonicalBox3 with Cylinder3.
in this case there is no rotation for box.
the distance of cylinder to box is ||cylinder.axis.origin|| = 558.79.
it's obviously not intersect,but result is true.

  void myTestIntersection()
  {
      CanonicalBox3<double> box;
      box.extent = { 9.5, 30.0, 17.0 };
  
      Cylinder3<double> cylinder;
      cylinder.axis.origin = { 537.4, -150.3, -29.4 };
      cylinder.axis.direction = { 0.025314638187811905, -0.20477058295580472, 0.97848258924181175 };
      cylinder.radius = 150.f;
      cylinder.height = 10.f;
  
      TIQuery<double, CanonicalBox3<double>, Cylinder3<double>> bcQuery;
      bool result = bcQuery(box, cylinder).intersect;
      printf("myTestIntersection: %d\n", result);  //  result is TRUE!!!
  }
  
  Failed on Test:
  // (D, -U2)
  sqrDistance = ComputeSqrDistance(P0, P1, C, W0, W1);
  if (sqrDistance <= sqrRadius) 
      return true;
  sqrDistance = 1323.9711150069654
  sqrRadius = 22500.000000000000

[davideberly]

I had posted my PDF modifications to the Geometric Tools website. It would have helped if I had also posted the modified C++ header file. The lower and upper calculations for the case (U2,-D) had typographical errors. See line 637 of the header file. Once this change is made, the query returns false.

Thank you for helping to make the code base better by providing me with diagnostic test information.

[ottoGithub]

#111 (comment)

It's still failed on this test case.
Maybe you should also fix (D, -U2) calculation?

[davideberly]

I ran a test with your dataset. Or so I thought I did. I acquired a Logitech Alto K86m keyboard and have been struggling with its layout and nonstandard number pad ever since. It turns out that I removed the 'f' suffixes on the double-precision numbers for cylinder radius and height. After your report, I saw that the height had become 100.0 instead of 10.0. The test with 100.0 returned the correct Boolean.

I believe that now I have matched the Mathematica code with the C++ code for the (U2, -D) (D, -U2) cases. Your previous example returns the correct Boolean.

I apologize for screwing this up. Once in a while I work on a bug report and turn it into a comedy of errors (on my part).

[ottoGithub]

All my test cases have passed, and I didn’t find any bugs now. great job