Previously, the code calculated the volume of a component, and if that volume was too small calculated its moments of inertia based on the surface instead of the volume. The decision wasn't based on the thickness of the shell (which might have made sense to me), it was the actual volume. The result is any really small component had its moments of inertia calculated using this surface method, no matter how "solid" the object was.
The cause of #2403 was that this method was used to calculate the moments of inertia in the .ork that failed, but the actual CG was used in the parallel axis theorem. This put the CG in a different place than a "surface CG" would have, so the longitudinal moment of inertia ended up less than 0 triggering the crash.
Now, if a component is so small that's volume is 0 it is assumed its contribution to moment of inertia must also be negligible, so the moments of inertia are set to 0 in this case.
4b1c6a4f4b