Procedurally Generated Meshes Sometimes Not Colliding

I'm creating a voxel terrain system that loads the land in chunks, and I create MeshColliders on the fly. The issue is maybe one out of every twenty chunks or so doesn't collide with anything or respond to raycasts. The mesh and edges are visible when using PhysicsDebugDisplayData. Re-running the physics mesh creation system on the relevant terrain chunk fixes the issue, as does creating new terrain chunks in unrelated areas.
My hunch is that the broadphase isn't capturing them properly and there's some internal optimization that only runs when there's a change in a memory chunk, and somehow that's not getting triggered, but I don't know how to tell or what to do about it. Is there anything else that could cause a collision mesh to be visible with Debug but not interact with anything?

 

Hey, if you're up for some deeper debugging, you could go into BuildPhysicsWorld.cs and look a t haveStatucBodiesChanged array of size 1. haveStaticBodiesChanges[0] will be set to 1 if we detected a change in the colliders of any kind (I'm assuming the stuff you are adding is static). If it doesn't detect the change, that's most likely the reason you are not seeing collisions with it. This is just a guess of course, so let's check and continue from there...

 

Hey thanks for the response. Yeah they're static, or at least just don't have a PhysicsVelocity component added. I added a debug statement into the body count changed section, and that being triggered sometimes caused the mesh to be fixed, but not all the time. In fact, when body count is triggered, it sometimes also seems to break formerly working colliders. However, it doesn't seem like whether they work or not ever changes without the debug message being triggered. Also worth noting I've tried both including the PhysicsCollider in the archetype and then setting its value, as well as adding it in a command buffer, to no avail. Also the issue occurs even when disabling Use Job Threads, so it's not a race condition.

 

Just to clarify: When you create these on the fly, are you going through MeshCollider.Create() each time and writing back to PhysicsCollider.Value, or are you using unsafe code to modify some mesh collider that was already created?

 

I'm using MeshCollider.Create in an Entities.ForEach, and have tried both using a CommandBuffer to add and set a new PhysicsCollider with the Value from it, as well as just setting the value of a PhysicsCollider passed by ref. Both behave identically as described earlier.

 

I think this might be hard for us to diagnose further here. Are you able to submit a bug report with a minimal repro that we can take a look at?

 

I'll see if I can put one together in the next few days. Thanks in advance.

 

@craftsmanbeck, in case you might not have seen it, check out this explanation of @petarmHavok about the order of systems: https://forum.unity.com/threads/ray...-finaljobhandle-complete.897467/#post-5904341

The tl;dr: It might be that your collider was not yet seen by the BuildPhysicsWorldSystem.

 

If I'm understanding it correctly, that would only cause a delay of a frame, not permanent collider not working. Further, that would cause the problem to be uniform across all meshes, not just happening on some. Additionally, it would not cause meshes that formerly worked to break. Correct me if I'm wrong.

 

You are right.