|
Original research ARCHITECTURE DESCRIPTION LANGUAGES AND THE "4+1" VIEW MODEL: A STRUCTURED CRITICAL REVIEW OF SOFTWARE ARCHITECTURE REPRESENTATION APPROACHESPages 397-422
Abstract
Software architecture description emerged as a formal subdiscipline of software engineering in the early-to-mid 1990s, producing two influential and complementary contributions: Kruchten's "4+1" view model, which organizes architectural description into logical, process, development, and physical views unified by representative scenarios, and Medvidovic and Taylor's classification and comparison framework for architecture description languages (ADLs), which evaluates formal modeling notations along component, connector, and configuration dimensions. Three decades later, mainstream software development has largely abandoned the formal ADL tooling that these frameworks were built to evaluate, while informally retaining much of the conceptual vocabulary they introduced. This review conducts a structured critical synthesis of the classical software architecture description literature and contemporary architecture documentation practice in order to establish what has transferred, what has not, and why. The review first reconstructs the 4+1 view model and the Medvidovic and Taylor framework in depth, together with the ten classical ADLs the framework was used to compare, namely Aesop, C2, Rapide, MetaH, UniCon, Weaves, Darwin, Wright, SADL, and ACME. It then surveys five contemporary developments: the ISO/IEC/IEEE 42010 standardization of architecture description; lightweight documentation approaches such as the C4 model, arc42, and architecture decision records; architecture description for microservice and cloud-native systems; model-driven and architecture-as-code approaches; and the emerging use of large language models for architecture documentation and recovery. The critical analysis argues that the separation-of-concerns logic underlying the 4+1 view model persists in modern notations, but that three assumptions embedded in classical ADLs break down under contemporary conditions: that a system possesses a single configuration rather than a population of request-scoped topologies; that reconfiguration is an act performed against a specification by an identifiable agent; and that the architecture description is the source from which an implementation is derived rather than an artifact recovered from one. Each of the ten classical ADLs is then revisited individually against contemporary practice, and the review proposes a five-factor contingency framework, comprising defect cost, architectural volatility, organizational scope, structural recoverability, and the relative half-lives of structure and rationale, that yields testable predictions about when heavyweight formal description remains warranted and which artifact should absorb documentation effort when it does not. Three summary tables consolidate the comparison, and the review closes with an agenda for empirical work on architecture representation adequacy.
Keywords:
Software architecture, architecture description language, 4+1 view model, ISO/IEC/IEEE 42010, C4 model, microservices, continuous architecture, architecture decision records, architecture as code, architecture erosion.
|