Interesting post. As a non-expert, I wondered whether dependent types fit here: macros extend what you can write, and dependent types could let the compiler check what those new constructs mean. Lean 4 does both, with user-definable syntax on a dependently typed core. Curious whether that influenced Sweetener at all.

score 1 · canonical

Thanks for the detailed answer; that makes sense. I hadn’t considered that CPU↔GPU sync would amplify insertion/removal costs enough to rule out SoA, and that treating runtime-sized arrays as resources sidesteps the storage-buffer limit.

score 1 · canonical

Nice milestone. Sizing buffers per WGSL §14.4.1 instead of size_of seems like exactly the kind of Rust/WGSL mismatch that’s easy to get wrong.

Question about the ECS that runs on the GPU: are you planning to store components as struct-of-arrays (e.g., one storage buffer per component type), or keep everything in a single crabslab-style slab with array-of-structs records? I’m curious how wgsl-rs would handle each case, e.g., whether SoA means many bindings per shader, and whether the WGSL layout rules stay transparent when a component is a runtime-sized array.

score 1 · canonical