Segment_2 has_on returns unexpected results

[barulicm]

Issue Details

Segment_2::has_on can return the wrong answer given segments that are slightly off axis. In some cases, the order of the segment points changes the answer given.

We are getting the points passed to has_on from a projection to the segment's supporting line. The order of the segment points does not appear to change the projected point.

Source Code

#include <iostream>
#include <CGAL/Simple_cartesian.h>

using Kernel = CGAL::Simple_cartesian<double>;
using Point = Kernel::Point_2;
using Segment = Kernel::Segment_2;

int main() {
  Point p{-3.0, 1.0};

  Segment s{
    Point{0.0, 0.0},
    Point{-7.0, 1e-4}
  };
  // Expected true, actually returns true
  std::cout << s.has_on(s.supporting_line().projection(p)) << "\n";

  Segment s2{
    Point{-7.0, 1e-4},
    Point{0.0, 0.0}
  };
  // Expected true, actually returns false
  std::cout << s2.has_on(s2.supporting_line().projection(p)) << "\n";

  Point p2{-1.0, 1.0};

  Segment s3{
    Point{-2.0, 1e-4},
    Point{0.0, 0.0}
  };
  // Expected true, actually returns false
  std::cout << s3.has_on(s3.supporting_line().projection(p2)) << "\n";

  Segment s4{
    Point{0.0, 0.0},
    Point{-2.0, 1e-4}
  };
  // Expected true, actually returns false
  std::cout << s4.has_on(s4.supporting_line().projection(p2)) << "\n";

  return 0;
}

Environment

• Operating system: Ubuntu 22.04
• Compiler: g++
• Release or debug mode: happens in both Debug and Release
• Specific flags used (if any): n/a
• CGAL version: 5.4 (git hash: c58ac97)
• Boost version: 1.74
• Other libraries versions if used (Eigen, TBB, etc.): n/a

[sloriot]

Replacing using Kernel = CGAL::Simple_cartesian<double>; by using Kernel = CGAL::Exact_predicates_exact_constructions_kernel; should fix the wrong results.

See all this FAQ entry

[barulicm]

Thanks for the info!