Meshes can be easily "pierced though"

[kkkmail]

1. Create a flat static mesh.
2. Start shooting objects at it with increasing speed.
3. Observe that after some threshold the objects experience "quantum tunneling effect" and go through the mesh. This heavily depends at least on on velocityIterationCount, substepCount, and frequency.
4. The lower is the frequency in SpringSettings, the smaller velocity is needed for that effect to be observable. Solid static bodies (e.g. boxes, spheres, etc.) don't exhibit this effect. Even when the colliding bodies managed to get inside, then they eventually get pushed out.
5. While this is kind of cool in some cases (e.g. you can "model" a frame dragging and that is effect when a rotating black hole drags the spacetime around it).

Proposed solution 1: A mesh (or an object hitting the mesh) should keep track on which side of the mesh it is approaching, so that to push the object out when it pierces the mesh.

Proposed solution 2: Allow specifying mesh as a semi-infinite body. In this case one side of the mesh will be a free space and the other side of the mesh will contain the solid body thus always pushing the objects out of that region. This is likely much simpler to implement than the first variant.

[RossNordby]

Some background: there are two main circumstances where contact constraint softness will cause tunneling on thin surfaces like meshes.

1. For isolated bodies, if the spring compression that would be achieved prior to rebound is greater than the depth necessary to tunnel, then it'll tunnel.
2. For multiple interacting bodies, if the combined interaction just happens to overcome the spring and shoves a body through the ground. This happens a lot more frequently.

Increasing the contact frequency will indeed tend to help with both; sometimes increasing the maximum recovery velocity will too. Higher contact stiffness effectively reduces the maximum depth that the bodies will achieve.

(One demo I haven't gotten around to is dynamic contact stiffness. Since it's just a callback, you can use it to approximate depth-dependent or velocity-dependent compression stiffness. That can be handy both for pure gameplay reasons, but some real materials aren't well captured by a single fixed spring configuration.)

Solidity/thickness would also definitely make this far less visible even for very soft springs. Solid meshes are (and have been) on the to-do list for a very long time. In fact, v1 had them! Both the Terrain and MobileMesh supported thickness/volumetric versions, but they haven't been brought over to v2 yet because of the usual infinitetodolist problems (and I would like to implement a much better version of solidity than v1 had; it was slow).