"Shareability" improves through splitting. This is first a matter of granularity: break into "shareable" components, thus keeping dependencies minimal.
Fine granularity also enforces a mathematical notion of
continuity.
Small increments mean no surprises.
They improve the feeling of control among members of the community.
This supports "programming by differences".
Differences make sense only on top of commonalities.
This gives birth to dependencies, which can be
structured,
thus offering a generic representation of meaning.
Modules and their interfaces: IRPs or others. But splitting is a strategy to increase visibility. Hiding is only methodological, and under local control: I decide what I want to ignore, and want the guarantee of doing it safely. Hiding should not be imposed on me.