Union with PolyPaths64 treats 45° collinear lines differently from 0°

[sethhillbrand]

Let's say that we start with the following two polygons:

Both are single-outline polygons. The one on the left has collinear lines at 0° on the bottom of the structure. The one on the right has collinear lines at 45° on the bottom right of the structure.

We run the following code:

#include <clipper2/clipper.h>
#include <Utils/clipper.svg.h>
#include <Utils/clipper.svg.utils.h>

using namespace Clipper2Lib;

int main(int argc, char * argv[])
{
  Paths64 subject, solution;
  SvgWriter svg;
  Clipper64 c;
  PolyTree64 pt_solution;

  Clipper64 c2;
  Paths64 solution2;
  PolyTree64 pt_solution2;
  SvgWriter svg2;

  subject.push_back(MakePath({7600000, 9000000, 6600000, 9000000, 6600000, 8750000, 7000000, 8750000, 7000000, 9000000, 7250000, 9000000, 7250000, 8500000, 7000000, 8500000, 7000000, 8750000, 6600000, 8750000, 6600000, 8000000, 7600000, 8000000, 7600000, 9000000}));
  c.AddSubject(subject);
  c.Execute(ClipType::Union, FillRule::NonZero, solution);
  c.Execute(ClipType::Union, FillRule::NonZero, pt_solution);

  Paths64 subject2;
  subject2.push_back(MakePath({9912310, 8325057, 9288816, 8948550, 8000000, 8567586, 9428364, 8547698, 9585009, 8652365, 9613140, 8624234, 9471719, 8482813, 9428364, 8547698, 8000000, 8567586, 9912310, 8325057}));
  c2.AddSubject(subject2);
  c2.Execute(ClipType::Union, FillRule::NonZero, solution2);
  c2.Execute(ClipType::Union, FillRule::NonZero, pt_solution2);

  std::cout << pt_solution << std::endl;
  std::cout << pt_solution2 << std::endl;

  SvgAddSolution( svg, solution, FillRule::NonZero, false );
  SvgSaveToFile( svg, "good_union.svg", 8000, 8000 );

  SvgAddSolution( svg2, solution2, FillRule::NonZero, false );
  SvgSaveToFile( svg2, "bad_union2.svg", 8000, 8000 );

}

The expected output here is that the PolyTree64 will say "Polytree with 1 polygon". Instead, we get:

Polytree with 1 polygon.


Polytree with 2 polygons.

Only the one on the left correctly merges the interior and exterior. The output looks like this:

The one on the right has output with two overlapping polygons:

I think that both polygons should behave the same way. Ideally, I think that the one on the left is the correct output but I'd love to have them synchronized either way

edit: I should have noted that this is testing using current master f1d4916

[AngusJohnson]

Hi Seth. I agree that it would certainly be better if the second clipping operation also returned a single path (instead of 2 paths).
But unfortunately reliably returning "canonical" solutions is extremely difficult to achieve (and time costly).
I'm currently disinclined to making further tweaks where these results are valid (ie correctly describe the filled regions), but I'll move this thread to the Discussion Forum, and to the "wish list" for whenever I'm not tied up with other things.

[sethhillbrand]

Understood. Please let me know if this is a work package that we can fund the research and development for at any point in the future