olson-sean-k / theon Goto Github PK

Theon should probably provide abstractions of foundational geometric primitives like translations, rotations, isometries, etc. that are built from the scalars, points, vectors, and matrices that are already abstracted. Such types would be analogous to similar types provided by some mathematics crates like nalgebra, but generalized to work with various implementations. This could lower the barrier to remaining agnostic to specific crates in library code. See the pictor support crate, for example.

Tangent: Plexus is using Theon to provide computational geometry APIs. I'm still not sure if this would be a good idea (see this issue), but it may be worth considering if the polygonal primitives in Plexus should still migrate into Theon. 🤷

See failures in this CI run. It looks like doctests are failing on the beta toolchain, because the Intel MKL dynamic library cannot be found. Interestingly, no such error occurs when running libtests. MKL is only used on Linux, so that's the only OS on which this failure occurs.

I first encountered this in Plexus' CI. It's not yet clear to me what the root cause is. In that repository, I tried various CI configurations and have found that the beta toolchain seems to be the most important variable. Using different versions of Ubuntu didn't avoid this error, and I was unable to install additional MKL packages in the workflow (and I doubt that is necessary).

The array feature uses ndarray and supporting crates to implement operations that require solving linear systems of arbitrary size. This allows Theon to optionally support certain useful spatial queries, such as computing a best-fit plane given a sampling of points.

This depends on ndarray-linalg, which builds and links BLAS/LAPACK libraries. These require building source packages and don't offer consistent support for different OSes. In the worst case, the output on different OSes differs (which is very likely a bug). 😬 Currently, this feature is gated behind cfg(target_os = "linux").

Only supporting certain OSes means that downstream code must always use conditional compilation. If possible, find some combination of packages that build and run on Linux, MacOS, and Windows.