The kernel boundary

crates/kernel is SolidCraft’s B-rep kernel boundary. The modelling kernel behind it is solidcore, the vendored Rust port of NVIDIA’s SMLib (solidcore/): crates/smlib wraps it in a safe API, and crates/kernel turns that into the operations the app uses. Callers get SolidCraft types (Body, face and edge indices, meshes, measures) and a KernelError instead of kernel types and panics.

guard

The vendored kernel can panic on geometry it does not handle. Every call into it goes through solidcraft_kernel::guard, which catches the panic and turns it into KernelError::Internal, leaving the model unchanged:

#![allow(unused)]
fn main() {
pub fn guard<T>(what: &str, f: impl FnOnce() -> Result<T>) -> Result<T>
}

A failing feature shows an error on the timeline; it never takes the app down.

What is our own

The B-rep operations are solidcore’s. SolidCraft’s own code in crates/kernel is the boundary around it: the public API, display meshes, measures, persistent names and STEP exchange. A failing kernel operation stays failing until the kernel is fixed; there is no workaround on SolidCraft’s side (tests/kernel-cases/README.md).

The vendored kernel

solidcore is vendored in the solidcore/ directory with git subtree and is built from there as the path dependencies solidcore and solidcore-record. Develop it there, in a normal commit; .github/workflows/solidcore-mirror.yml mirrors changes under it back to tangency-labs/solidcore. solidcore/ and vendor/ are excluded from our fmt and clippy rules because they are not our code.