Coupling in Software Engineering
Yannick Loth · Zenodo (CERN European Organization for Nuclear Research) · 2026
Coupling has been studied for over five decades, since Stevens, Myers, and Constantine introduced the term in 1974. The literature has classified coupling into an expanding set of dimensions — structural, logical, semantic, temporal, organizational, deployment, and runtime — formalized it through metrics and dependency models, and attempted unification at the level of measurement. This paper surveys that literature. We catalog the origin taxonomy and its severity ordering, the dimensions that have been added beyond the structural, and the formal and metric frameworks that have been proposed: property-based measurement, information-theoretic coupling, dependency-structure matrices, and change-propagation models. We show that every surveyed unification attempt operates at one of three levels: a severity scale, a measurement framework, or a single numeric index. None of the surveyed work unifies all coupling dimensions under a common ontological structure: none answers what coupling is across all its manifestations. The survey identifies this unification gap as the open problem the field has not addressed. The fragmentation has a discernible explanation: each dimension has been studied by a different community with a different method, and the literature has refined each fragment without asking whether the fragments are instances of one phenomenon. We conclude by stating the requirement a unifying account must satisfy and the shape such an account would take. This paper is descriptive and historical. It makes no normative claim about how coupling should be minimised or managed; it establishes what the literature has said about what coupling is, and where that account stops.