invalid intersection for CCW polygons

[vissarion]

The following program

#include <boost/geometry.hpp>

int main() {

    using Point = boost::geometry::model::d2::point_xy<double>;
    using MultiPoint = boost::geometry::model::multi_point<Point>;
    using Polygon = boost::geometry::model::polygon<Point, false>;
    using Linestring = boost::geometry::model::linestring<Point>;
    using MultiPolygon = boost::geometry::model::multi_polygon<Polygon>;
    using MultiLinestring = boost::geometry::model::multi_linestring<Linestring>;

    Polygon p1, p2, p3;
    boost::geometry::read_wkt("POLYGON((0 5,-6 -17,12 17,0 5),(4 6,5 5,0 1,4 6))", p1);
    boost::geometry::read_wkt("POLYGON((3 9,-15 -5,13 -11,3 9))", p2);

    MultiPolygon mp;

    std::cout << boost::geometry::is_valid(p1) << std::endl;
    std::cout << boost::geometry::is_valid(p2) << std::endl;


    std::tuple<MultiPoint, MultiLinestring, MultiPolygon> result;
    boost::geometry::intersection(p1, p2, result);

    std::cout << boost::geometry::wkt(std::get<0>(result)) << std::endl;
    std::cout << boost::geometry::wkt(std::get<1>(result)) << std::endl;
    std::cout << boost::geometry::wkt(std::get<2>(result)) << std::endl;
    std::cout << boost::geometry::area(std::get<2>(result)) << std::endl;

    std::cout << boost::geometry::is_valid(std::get<2>(result)) << std::endl;

    return 0;
}

returns the invalid geometry
MULTIPOLYGON(((3.33333 8.33333,0 5,-3.40491 -7.48466,-1.21132 -7.95472,5.31429 4.37143,5 5,0 1,4 6,5 5,3.33333 8.33333)))
while for CW polygons ie. if we change line
using Polygon = boost::geometry::model::polygon<Point, false>;
above with
using Polygon = boost::geometry::model::polygon<Point, true>;
it returns
MULTIPOLYGON(((3.33333 8.33333,5 5,5.31429 4.37143,-1.21132 -7.95472,-3.40491 -7.48466,0 5,3.33333 8.33333),(5 5,4 6,0 1,5 5)))
which is valid and correct.

This behavior is tested in 1.84 and also 1.78.
However, in 1.77 the above code gives the correct valid result in all cases.

The difference between the invalid and valid cases is that in the latter there is the creation of a hole (5 5,4 6,0 1,5 5) while in the former this is not true and the result is invalid. Notably the hole shares a point with the boundary of the output polygon.

[barendgehrels]

I can reproduce it.
Output (in slightly modified test program):

70: issue_1231 : MULTIPOLYGON(((3.33333 8.33333,5 5,5.31429 4.37143,-1.21132 -7.95472,-3.40491 -7.48466,0 5,3.33333 8.33333),(5 5,4 6,0 1,5 5)))
70: .../boost/libs/geometry/test/algorithms/set_operations/intersection/./test_intersection.hpp:120: error: in "test_main_caller( argc_ argv )": intersection: issue_1231 not valid: Geometry has invalid self-intersections. A self-intersection point was found at (5, 5); method: t; operations: i/i; segment IDs {source, multi, ring, segment}: {0, 0, -1, 4}/{0, 0, -1, 7} type: d ccw
70: issue_1231 ccw : MULTIPOLYGON(((3.33333 8.33333,0 5,-3.40491 -7.48466,-1.21132 -7.95472,5.31429 4.37143,5 5,0 1,4 6,5 5,3.33333 8.33333)))
70: 

which is the same as you have. So it fails to generate an interior ring, while it should. The generation itself is OK.

Also, we don't test ccw by default. If you turn it on, there are three errors in the intersection unit test, which might all be related. These are:

70: Test configuration:
70:   - Debug mode
70:   - No rescaling
70:   - Testing only one type
70:   - Default test type: d
70: 
70: .../boost/libs/geometry/test/algorithms/set_operations/intersection/./test_intersection.hpp:120: error: in "test_main_caller( argc_ argv )": intersection: issue_1231 not valid: Geometry has invalid self-intersections. A self-intersection point was found at (5, 5); method: t; operations: i/i; segment IDs {source, multi, ring, segment}: {0, 0, -1, 4}/{0, 0, -1, 7} type: d ccw
70: .../boost/libs/geometry/test/algorithms/set_operations/intersection/./test_intersection.hpp:120: error: in "test_main_caller( argc_ argv )": intersection: case_105 not valid: Geometry has invalid self-intersections. A self-intersection point was found at (5, 3); method: m; operations: u/u; segment IDs {source, multi, ring, segment}: {0, 0, 0, 0}/{0, 0, 1, 3} type: d ccw
70: .../boost/libs/geometry/test/algorithms/set_operations/intersection/./test_intersection.hpp:120: error: in "test_main_caller( argc_ argv )": intersection: mysql_23023665_10 not valid: Geometry has invalid self-intersections. A self-intersection point was found at (5, 5); method: t; operations: i/i; segment IDs {source, multi, ring, segment}: {0, 0, -1, 4}/{0, 0, -1, 7} type: d ccw
70: .../boost/libs/geometry/test/algorithms/set_operations/intersection/./test_intersection.hpp:120: error: in "test_main_caller( argc_ argv )": intersection: mysql_23023665_11 not valid: Geometry has invalid self-intersections. A self-intersection point was found at (3, -1); method: t; operations: i/i; segment IDs {source, multi, ring, segment}: {0, 0, -1, 0}/{0, 0, -1, 3} type: d ccw
70: 
70: *** 4 failures are detected in the test module "Test Program"
70: 
1/1 Test #70: boost_geometry_algorithms_intersection ...***Failed    0.71 sec

For the record, I added two lines, in overlay_cases.hpp

static std::string issue_1231[2] =
{
    "POLYGON((0 5,-6 -17,12 17,0 5),(4 6,5 5,0 1,4 6))",
    "POLYGON((3 9,-15 -5,13 -11,3 9))"
};

and in intersection.cpp

TEST_INTERSECTION(issue_1231, 1, -1, 54.701340543162516);

The output looks identical in both, and to your picture (here with TEST_WITH_SVG

[barendgehrels]

It's difference is also wrong, both clockwise and counterclockwise.
I'm making a PR for the testcase (not for the fix)

[vissarion]

Fixed by #1263