Towards a Semantics of Activity Diagrams with Semantic Variation Points

UML activity diagrams have become an established notation to model control and data ow on various levels of abstraction, ranging from fine-grained descriptions of algorithms to high-level workflow models in business applications. A formal semantics has to capture the flexibility of the interpretation of activity diagrams in real systems, which makes it inappropriate to define a fixed formal semantics. In this paper, we define a semantics with semantic variation points that allow for a customizable, application-specific interpretation of activity diagrams. We examine concrete variants of the activity diagram semantics which may also entail variants of the syntax re ecting the intended use at hand.

💡 Research Summary

The paper addresses a fundamental limitation in the formal semantics of UML Activity Diagrams (ADs): the tendency to prescribe a single, immutable interpretation that cannot accommodate the diverse ways ADs are used in practice. To overcome this, the authors introduce the concept of Semantic Variation Points (SVPs), which are explicit, configurable options embedded in the metamodel that allow the meaning of diagram elements to be altered according to the needs of a particular application domain.

The work is organized into four main sections. First, the authors survey the role of ADs in modeling both control flow and data flow across a spectrum of abstraction levels—from low‑level algorithmic descriptions to high‑level business workflows. They illustrate, with concrete examples, how the same diagram can be interpreted differently in a real‑time embedded system versus a business process management environment, thereby exposing the inadequacy of a one‑size‑fits‑all semantics.

Second, the paper defines SVPs formally. Each SVP is declared as an optional flag in the metamodel and is associated with a concrete “variation implementation” that specifies how the flag influences the semantics. The authors categorize SVPs into two broad families: (1) Execution‑model SVPs, which affect token propagation, concurrency control, and scheduling policies; and (2) Semantic‑level SVPs, which govern side‑effects of actions, exception handling, data transformation rules, and domain‑specific business constraints. By toggling these flags, a modeler can selectively enable or disable particular semantic aspects without altering the diagram’s visual structure.

Third, the authors present a rigorous formalization of SVP‑based semantics using a labeled transition system (LTS). A system state comprises a token distribution together with a variable environment that records data values. Transitions correspond to the execution of actions, the traversal of control nodes, or the movement of object flows. An SVP modifies the transition relation by either replacing a subset of transition rules or augmenting them with additional guards and actions. For example, a “time‑bound SVP” adds a temporal guard to every transition, while a “priority SVP” introduces a selection function that prefers higher‑priority actions when multiple transitions are enabled. This modular approach ensures that the core LTS remains unchanged; only the selected variation implementations are swapped in, preserving compositionality and enabling systematic reasoning about different semantics.

In addition to the semantic layer, the paper tackles the impact of SVPs on the diagram’s syntax. The authors propose a “variation syntax” mechanism whereby the presence of specific SVP flags dynamically alters the well‑formedness rules of the diagram. For instance, when a “concurrency‑disable SVP” is active, parallel edges are prohibited, whereas an “exception‑handling SVP” mandates the existence of exception handler nodes attached to actions that may raise exceptions. This dynamic syntax checking can be performed automatically during model validation, preventing inconsistent configurations before code generation or simulation.

The final section demonstrates the practicality of the approach through two case studies. The first case study targets a real‑time embedded system. By activating a “deadline SVP” together with a “priority‑based scheduling SVP,” the authors show how token propagation respects temporal constraints and how higher‑priority actions pre‑empt lower‑priority ones. The second case study focuses on a business process model. Here, a “business‑rule SVP” injects domain‑specific validation logic before and after certain actions, while an “exception‑handling SVP” automatically routes violations to designated compensation activities. In both scenarios, the same underlying activity diagram yields distinct execution traces and verification outcomes solely because different SVP configurations are selected.

In conclusion, the paper argues that SVP‑driven semantics provide a flexible, systematic, and formally grounded way to bridge the gap between high‑level modeling and concrete system implementation. By exposing variation points explicitly, modelers can tailor the meaning of activity diagrams to the precise requirements of their domain, while tool developers can support multiple semantics without duplicating the entire metamodel. The authors suggest future work on SVP management tooling, analysis of dependencies among SVPs, and integration of SVP concepts with other UML diagram types such as sequence diagrams and state machines.