Boost.Geometry intersection() returns empty result in 1.84/85/86 but not in 1.72 for valid overlapping polygons

[SowmyaKoppuravuri]

We are observing a behavioral difference in boost::geometry::intersection() between Boost 1.72 and Boost 1.86.

For the same valid input polygons:

• bg::is_valid() returns true for both polygons
• bg::intersects() returns true
• Boost ≤ 1.72 returns a non-empty intersection polygon
• Boost ≥ 1.84 (tested with 1.86) returns an empty result
  We are trying to determine whether this is:
• an intended robustness change in overlay behavior,
• degenerate result filtering, or a regression.

Environment
Working version:
Boost 1.72
MSVC 19.xx
C++17

Failing version:
Boost 1.84/85/86
MSVC 19.xx
C++17

Minimal Reproducible Example

#include <boost/geometry.hpp>
#include <vector>
#include <iostream>

namespace bg = boost::geometry;

using Point = bg::model::d2::point_xy<double>;
using Polygon = bg::model::polygon<Point>;

int main()
{
    Polygon a;
    bg::append(a, Point(-2.7262688429326545, 16.826043180279360));
    bg::append(a, Point(-2.7262688429326545, 17.976043180279358));
    bg::append(a, Point(0.59999999999990905, 17.976043180279358));
    bg::append(a, Point(0.59999999999990905, 16.826043180279360));
    bg::append(a, Point(-2.7262688429326545, 16.826043180279360));

    Polygon b;
    bg::append(b, Point(2.5905051825378118, 20.985713718593683));
    bg::append(b, Point(2.5905051825378118, 19.985713718593683));
    bg::append(b, Point(0.60479136394412836, 18.000000000000000));
    bg::append(b, Point(0.59999999999999998, 17.976043180279358));
    bg::append(b, Point(0.57000000000000006, 17.826043180279360));
    bg::append(b, Point(-2.6962688429326507, 17.826043180279360));
    bg::append(b, Point(-2.7262688429326509, 17.976043180279358));
    bg::append(b, Point(-2.7310602068767795, 18.000000000000000));
    bg::append(b, Point(-4.6257992677318862, 19.894738960855108));
    bg::append(b, Point(-4.6257992677318862, 20.985713718593683));
    bg::append(b, Point(2.5905051825378118, 20.985713718593683));

    bg::correct(a);
    bg::correct(b);

    std::cout << "is_valid(a): " << bg::is_valid(a) << "\n";
    std::cout << "is_valid(b): " << bg::is_valid(b) << "\n";
    std::cout << "intersects: " << bg::intersects(a, b) << "\n";

    std::vector<Polygon> result;
    bg::intersection(a, b, result);

    std::cout << "intersection count: " << result.size() << "\n";

    if (!result.empty())
        std::cout << "area: " << bg::area(result[0]) << "\n";

    return 0;
}

Observed Behavior :
Boost 1.72:
intersects: 1
intersection count: 1
area: (non-zero)

Boost 1.86:
intersects: 1
intersection count: 0

Question

Is this behavior difference expected due to overlay robustness/degenerate filtering changes introduced in later Boost versions, or is this considered a regression? The polygons are valid and clearly overlap geometrically.

[barendgehrels]

Could you maybe also check Boost 1.88 ? Because the algorithm considerably changed.

[SowmyaKoppuravuri]

@barendgehrels It seems to be working with 1.90 version, Thank you

[barendgehrels-tomtom]

Top, thanks for testing itt!