PolygonCollider2D incorrectly calculates shape

I'm assuming it's the same thing but I was asked by a QA the general question of the warning that's produced which indicates some primitive shapes couldn't be created because they didn't pass validation. In the end, the libtess2 library we use sometimes produces *very* thin triangles when the outline its being asked to produce causes large fine fan-shapes. This makes it very susceptible to minor floating-point shifts (any transform or offset change). There's absolutely nothing we can do about that. If we pass it to the physics system then it causes very bad collision responses so we filter them out. In nearly all cases this means removing a fractional triangle or super thin polygon wedged inbetwen others but these are fairly rare.

If you are seeing significant visible polygons missing then that is indeed a bug but if the problem is that you don't like the warning because it's unexpected for you then that isn't a bug.

We intend to expose the ability to allow Delaunay Triangulation which deals with these horrible fan-shaped things much better and also produces less primitive shapes but in the meantime, there's no bug that can be fixed unfortunately. If you are seeing fine fan-shaped primitives then this is likely the root cause of that sensitivity to changes.

If this is for a Static collider then you can add a Rigidbody2D set to Static (this is the same as not adding a Rigidbody2D component btw). This means that when you move the body via the Transform in the editor or the body itself, it doesn't cause the collider to be recalculated so is a much more efficient way to offsetting a collider. It will act exactly the same.

 

Because at zero it changes nothing but non-zero slightly affects the points. Floating point changes mean the library sees a slightly different set of points. A small change of points can produce a larger change of decomposed convex polygons. The same happens with Transform changes although they're more evident if you were to rotate around the Z axis.

If this were rendering it'd be fine but for physics it can be a disaster. Indeed, there's asserts all over the place checking these polygons.

This is the library we use: https://github.com/memononen/libtess2

Anyway, add a Rigidbody2D set to Static should work perfectly fine without any hacks.

 

Sounds like you misunderstood. Don't add offset, set the Transform local position with a Static Rigidbody2D. You already said your shape is fine with no offset.

 

If I set the offset to be (0.54,0) on my machine then some shapes are removed.

Nevertheless, I'll reactivate the bug and assign to myself so I can take a deeper look at what's going on during the creation of that shape. Thanks for raising it as my only involvement so far was to answer QA regarding the question about offsets affecting primitive shapes being removed due to validation. That said QA were more than correct to close it upon my response so really pleased you raised it again as the shape size being removed is huge. Something is off there but it would seem that the slight floating-point difference is exposing the real bug.

Please bear with me, kind of a crazy busy time right now. I'll post this forum link in the case as a reminder to get back to you here. If there's a bug I'll find it.

 

No problem, sorry for the issues it's causing you. Will get back to you ASAP.

 

So I can confirm that the issue you've reported here isn't related to the normal exclusion of shapes due to them being collinear or having coincidental/bad vertex. The bug has been assigned and I'll get it fixed ASAP.

https://issuetracker.unity3d.com/is...n-the-polygoncollider2ds-points-are-too-close

Thanks for the report!