You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
squared_distance(Segment_3, Ray_3) is wrong when the segment and the ray are parallel to the z-axis #9858
CGAL::squared_distance() between a Segment_3 and a Ray_3 gives a wrong (too large) value when the two are parallel to the z-axis and point in opposite directions. I ran into it while comparing the distance functions against brute force on small integer inputs.
In the example below the ray runs alongside the whole segment, so the squared distance is 5 (e.g. between (2 2 1) and (0 1 1)), but 6 is returned, which is the squared distance between the ray's source and the segment's target. If I just swap the endpoints of the segment I get 5.
I think the cause is in squared_distance_parallel() in Distance_3/include/CGAL/Distance_3/Segment_3_Ray_3.h. It decides whether the segment and the ray have the same direction by comparing the sign of hx() or of hy(), whichever is larger for the segment. When both directions are along z, hx() and hy() are 0 for both, the signs are equal, and opposite directions are taken to be the same. So it only goes wrong along the z-axis; along x or y the result is fine. There is already a same_direction() in Distance_3/internal/squared_distance_utils_3.h that also looks at z (Ray_3_Ray_3.h uses it), so that could be used here as well.
The only Segment_3 - Ray_3 case in test_distance_3.cpp is a non-parallel one, so this path isn't tested at the moment.
I have a fix with a test and can open a PR for it.
Source Code
#include<CGAL/Exact_predicates_exact_constructions_kernel.h>
#include<iostream>typedefCGAL::Exact_predicates_exact_constructions_kernel K;
typedef K::Point_3 P;
typedef K::Segment_3 S;
typedef K::Ray_3 R;
intmain()
{
// both parallel to the z-axis, pointing in opposite directions
S s(P(2, 2, 2), P(2, 2, 1));
R r(P(0, 1, 0), P(0, 1, 1));
std::cout << CGAL::squared_distance(s, r) << "\n"; // expected 5
std::cout << CGAL::squared_distance(s.opposite(), r) << "\n"; // 5
}
This prints 6 and 5. I get the same with Epick and Simple_cartesian.
Environment
Operating system (Windows/Mac/Linux, 32/64 bits): Linux (Ubuntu 24.04), 64 bits
I checked Distance_3/include/CGAL/Distance_3/Segment_3_Ray_3.h: the parallel branch compares only hx()/hy(), so when both directions are along z it compares sign(0) and treats opposite directions as matching. That is consistent with the reported 6-versus-5 result; same_direction() in squared_distance_utils_3.h looks like the natural shared helper.
A focused regression in Distance_3/test/Distance_3/test_distance_3.cpp could assert the reported result for both segment endpoint orders, then cover same-direction z and the reversed Ray_3/Segment_3 overload. Since you already have a fix in mind, could you assign me to add the test/follow-up if you are not already preparing that change?
Thanks for looking into it @victrvondoom The fix is already up in #9859. It switches the parallel case to same_direction(), and the test in test_distance_3.cpp covers both endpoint orders of the segment and rays pointing both ways along each of the three axes. check_squared_distance also checks the Ray_3/Segment_3 order, so that’s covered too. If you have time, a review or a test run of #9859 would be very welcome.
Activity
victrvondoom commentedon Nov 19, 2026
I checked
Distance_3/include/CGAL/Distance_3/Segment_3_Ray_3.h: the parallel branch compares onlyhx()/hy(), so when both directions are along z it comparessign(0)and treats opposite directions as matching. That is consistent with the reported 6-versus-5 result;same_direction()insquared_distance_utils_3.hlooks like the natural shared helper.A focused regression in
Distance_3/test/Distance_3/test_distance_3.cppcould assert the reported result for both segment endpoint orders, then cover same-direction z and the reversedRay_3/Segment_3overload. Since you already have a fix in mind, could you assign me to add the test/follow-up if you are not already preparing that change?Ishaan400 commentedon Nov 20, 2026
Thanks for looking into it @victrvondoom The fix is already up in #9859. It switches the parallel case to same_direction(), and the test in test_distance_3.cpp covers both endpoint orders of the segment and rays pointing both ways along each of the three axes. check_squared_distance also checks the Ray_3/Segment_3 order, so that’s covered too. If you have time, a review or a test run of #9859 would be very welcome.