On the Evolutionary Nature of Architectural Violations

Architectural conformance checking is the process of verifying whether a given software implementation conforms to the rules and decisions in an architectural model. Different conformance checking techniques have been proposed, and both academic and commercial tools use them for architecture evaluation. In this paper, we present an exploratory and longitudinal study on architectural violations of software systems. As an exploratory study, our aim has been to reveal patterns, raise hypotheses and form an initial body of knowledge on architectural violations, rather than test specific hypotheses. We have studied the evolution of four widely known open source systems for which we have trusted architectural models. The study encompasses the analysis of 19 bi-weekly versions of each system. In total, we analyzed more than 3,000 violations. From our observations, we have derived a series of facts. Four of them, deserve attention: 1) development teams of all studied projects seem to be aware of the presence of architectural violations in the code and all of them do perform perfective maintenance aimed at eliminating such violations, 2) despite all effort, the number of architectural violations, in the long term, is continuously growing, 3) in all studied systems there is a critical core, i.e., just a few design entities are responsible for the majority of violations, and 4) some violations seem to be "respawning", i.e., they are eliminated, but are likely to be back in future versions of the system.