Instead of emitting one giant crate containing everything, we tweaked our SQL-to-Rust compiler to split the output into many smaller crates. Each one encapsulating just a portion of the logic, neatly depending on each other, with a single top-level main crate pulling them all in.