Odd crash on sanitized input (c++) about poly2tri HOT 8 CLOSED

crash data

I get this when I run it in my latest codebase for Java.

Skipping edge: Intersecting Constraints : 
[-3882000.0,7191000.0],[-3875026.0,7153693.0] intersects 
[-3883360.0,7168352.0],[-3876543.0,7172961.0]
Skipping edge: Intersecting Constraints : 
[-3846561.0,7193229.0],[-3840166.0,7189498.0] intersects 
[-3882000.0,7191000.0],[3776000.0,7191000.0]

So my guess is that one constraint(edge) from the hole intersects with the 
polygon hull.

This is a dataset that has been processed by clipper?

To detect these issues in the c++ code add a test here:

void Sweep::FlipEdgeEvent(SweepContext& tcx, Point& ep, Point& eq, Triangle* t, 
Point& p)
{
  Triangle& ot = t->NeighborAcross(p);
  Point& op = *ot.OppositePoint(*t, p);

// My java code
        if( t.getConstrainedEdgeAcross(p) )
        {
            int index = t.index( p );
            throw new TriangulationException( "Intersecting Constraints : "+ 
                                              p +","+ op
                                              +" intersects "+
                                              t.points[(index+1)%3] 
                                              +","+ 
                                              t.points[(index+2)%3]);            
        }

and you will have to add

    public boolean getConstrainedEdgeAcross( final TriangulationPoint p )
    {
        if( p == points[0] )
        {
            return cEdge[0];
        }
        else if( p == points[1] )
        {
            return cEdge[1];
        }
        return cEdge[2];
    }

to Triangle

Just looking at the holes point values I can see that some are outside the 
rectangular polygon hull. So something fishy is going on with the clipping.

I make an assumption the you wiggled the intersection points? That could 
explain why they end up outside the polygon.

Clipper v6 enforces strictly simple polygons:

http://sourceforge.net/p/polyclipping/discussion/1148419/thread/813a62c8/?limit=
50

Maybe upgrading to it can help?

You are absolutely correct.  We had rare-case in the game where a hole could 
have a point outside of the outer hull.  

This was our code problem.  Thank you very much for that code that detected the 
bad points.  Is that standard in the Java version?  Maybe I should use that 
port for debugging...

Thanks again!

I realized that this simple test could be made sometime after the inital 
release of Poly2Tri so don't really know if it has been ported to other 
version. It just handle some invalid input a bit more gentle.

But apart from this there shouldn't be many differences in the Java and other 
code bases in the core triangulation classes.

I always try to help if someone get stuck on some invalid data :)

I added this additional Java check in the JavaScript version