Triangulating fails on a single invalid triangle.

[baardaap-blender]

Tested versions

Reproducible in: versions 4.0+ and probably below but didn't really test

System information

All platforms

Issue description

When triangulating a 2d polygon, triangulate_polygon() fails and returns an empty array while the underlaying triangulate() function generates a mostly usable (though maybe incomplete) triangularisation.

Longer description of the situation where this problem arises:
I'm creating a small drawing app where the user can draw filled polygons with the mouse. This easily generates some invalid points which causes the complete fill to disappear. This is because in geometry_2d.h the triangulate_polygon() function discards the generated triangularisation if the triangulate() returns false.

a possible fix is to just ignore the error status returned by the triangulate function and always return the result:

diff --git a/core/math/geometry_2d.h b/core/math/geometry_2d.h
index abd395d8df..946a57ab42 100644
--- a/core/math/geometry_2d.h
+++ b/core/math/geometry_2d.h
@@ -345,7 +345,8 @@ public:
        static Vector<int> triangulate_polygon(const Vector<Vector2> &p_polygon) {
                Vector<int> triangles;
                if (!Triangulate::triangulate(p_polygon, triangles)) {
-                       return Vector<int>(); //fail
+                       //return Vector<int>(); //fail
+                       // maybe somehow report a warning?
                }
                return triangles;
        }

This leads to much more stable triangularisations, but I'm not sure if we'd want to maybe generate a warning that some invalid triangles/points were detected? And how to do that.

Steps to reproduce

In the supplied project start it and draw with the mouse. As soon as the resulting polygon is invalid it completely disappears.

With the above patch it behaves much more like other drawing apps. I'm not sure if this is the 'correct' way to solve this, but it seems a shame to just throw away the (mostly valid) triangularisation.

Minimal reproduction project (MRP)

drawpolygon.zip

[AThousandShips]

This change would break compatibility as it is expected to always return a valid triangulation of the input, but we could add an argument (defaulting to false) to allow partial results

Note however that failing on "just" one triangle is no guarantee of the degree of success, it could fail to triangulate most of the input in this case, resulting in very limited output, so I'm not sure how reliable and useful this would be

[baardaap-blender]

That would be very useful. My experimentation (with the supplied test program) shows me that the partial triangulations are much better to have than nothing at all.

I admit it's maybe a bit of a niche application (user drawn polygons) but if adding such an 'allow_partial' parameter would be easy to do it would be very helpful.

[FireCatMagic]

I also use it for user drawn polygons, this would be very nice to have.