I've committed a patch that will make the problems go away and shouldn't
cause any issues when the dtype already matches the expectation, i.e.
when the platform's int type is equal or greater than np.int64.
There is a slight risk with the type conversion though: When the input
is very large and uses indices > 2³¹-1. This will cause data loss on
32-bit architectures without any warning. I did some simple tests and
found that this will cause failures in other places (memory allocation,
for example), but that's not very reliable.
Unfortunately, I couldn't find a way to do an efficient "unsafe"
conversion with a warning in numpy - a lot of boilerplate and
performance loss would be needed to do this manually.
The Fedora developers already reported the issue upstream, by the way:
https://github.com/mikedh/trimesh/issues/690
