Towards Pattern Based Architectural Conformance Checking

Abstract: Patterns are a source of knowledge when architecting software systems. They provide abstract and time-tested solutions that show how a system should be structured to achieve needed qualities. However, when developing software there is a chance that small mistakes are introduced in the source code. Over time, these mistakes can accumulate and break the structure of the pattern and its qualities are lost. There are methods that can help find such errors, but none of these provide a pattern abstraction. In this work, we describe a method that raises the level of abstraction from checking individual dependencies to checking key dependencies in the pattern. We implement our method, apply it to check the Model-View-Controller pattern. We show that the method can find architectural problems in real source code and examine how removal of detected erosions affects the source code. We conducted an experiment in a software project setting to determine if using the method affects the number of architectural problems. Some project teams were randomly assigned to use a software service that automated our method. It checked how well their implementation conformed to Model-View-Controller every time they updated the source code. The experiment showed that developers that used the tool had significantly fewer detected architectural problems during the course of the project. Our method makes conformance checking easier to use. This might help increase the adoption of conformance checking in industry.