API shaping (lightweight)

Aim

• •Small public surface
• •Clear data ownership (views vs owned)
• •Fast internals without leaking complexity

Prefer

• •Detector structs that own scratch buffers → avoid allocations per call.
• •Config structs with safe defaults, but don’t hide “magic” thresholds.
• •ImageView<T> / ImageViewMut<T> in APIs; keep crates buffer-agnostic.
• •Separate “core algorithm” from “pipeline convenience wrapper”.

Avoid

• •Generic abstractions that obscure hot loops.
• •Exposing internal scratch buffers in public API.
• •“One mega function” that does everything.

Patterns that work here

• •detect_*(&mut self, img: &ImageView<_>, cfg: &Config) -> Output
• •Output types that can be iterated cheaply (Vec<Edgel>, Vec<LaserSample>)
• •Optional features for parallelism/SIMD later, not in baseline.