Automatic Transformation of UML to Decision Models and Assessment of Architectural Decisions

The design structure matrix (DSM) and its formal counterpart, the augmented constraint network (ACN), have been used to explicitly model software design decisions, design rules and the resulting modular structure. To address the difficulty of modeling software design in terms of decisions, we formally define the decisions and their assumption relations embedded within UML class diagrams and automatically transform a UML class diagram into an augmented constraint network, from which a DSM can be automatically derived. We also use a real-world software project to assess the difference between syntactical DSMs derived from source code with UML-derived DSMs in terms of their ability of assessing software architecture. The result shows that the UML-derived ACN and DSM models can more effectively help designers to isolate the impact of each individual architectural design rule and feature, assess the effectiveness of refactoring, and locate the maintenance activities that may cause modularity decay.