In my experience this is a much better approach. Most modern compilers can inline across linker boundaries anyways, so it's not a real performance issue either.