Agreed! This is how I've always developed FFI code, for both Rust and Ruby. A thin wrapper that does little more than FFI and maybe mapping onto native memory-management techniques, and then a separate higher-level wrapper that provides as idiomatic an interface on top of the library as possible.