Bug found in thors ConcaveShape

[Doodlemeat]

Hello.
I found a bug, maybe 2 of them with thor ConcaveShape.

First bug:
A ConcaveShape is not filled with any color. This happens randomly from what I have seen. And yes, I am 100% sure I fill it with color since. It fills with color more often than it does'nt


Second bug:
Sometimes when I call

rt.draw(shape);

where rt is rendertarget and shape is a thor::ConcaveShape, the program crashes with this message:

here is the callstack:


Here lies my code. In Soundripple.cpp.
There is not other file that manipulates the ConcaveShape.
https://github.com/doodlemeat/ProjectTerminal/blob/master/Stealth/SoundRipple.cpp

[thomas9459]

Can you provide a SSCCE so that others can test the code in question? Otherwise, it ends up just being guesswork...

[Doodlemeat]

So you want the .sln file ? I think you will need all other dependencies like boost and such?

[Nexus]

No, we want a complete and minimal example that reproduces the problem. Please read the link carefully.

[Doodlemeat]

Hello!
Sorry for taking so long time.

Below is a .rar file of everything that is included. That is

• SFML 2.1
• Thor
• Aurora (thor depends on this)

Built with visual studio 2013 at Windows 8.
Only configured for debug mode.
http://www.sendspace.com/file/ysni67

[Doodlemeat]

UPDATE!
I just want to tell you that the second bug is not related to the topic. Its a mistake by me, but the first bug is still there

[Nexus]

I could reduce the problem to the following list of points:

(344, 248)
(120, 248)
(184, 248)
(184, 181)

The problem here is that three points of adjacent triangles are on a line (y=248). This is a limitation of the Delaunay triangulation and stated in the documentation:

Note that the triangulation may meet problems at 4 co-circular points or at 3 or more collinear points.

This led to a removal of too many triangles in the cleanup step during thor::triangulatePolygon() called from thor::ConcaveShape, so that the convex shapes composing the concave shape were left empty.

I agree that this limitation is annoying, I'll investigate alternatives. For the concave shape it might suffice to simply offset points randomly by a tiny amount, but I actually prefer a clean solution that also works for other triangulations.

[Nexus]

Should be fixed now

I improved the triangulation so that this specific case is recognized. It won't be possible to handle all situations of 3 collinear points (e.g. only 3 vertices in total), but most reasonable scenarios should be handled correctly now.