Unique identification of elements in evolving models : towards fine-grained traceability in model-driven engineering

Model-driven engineering (MDE) is a widely accepted methodology in software engineering. At the same time, the ability to retrace the engineering process is an important success factor for software projects. In MDE, however, such traceability is often impeded by the inadequate management of model evolution. Although models have a very fine-grained structure, their different revisions and variants are prevalently managed as monoliths in a file-based software configuration management (SCM). This causes the identification problem: if the fine-grained elements are not assigned with globally unique identifiers, we cannot identify them over time. If such identifiers would be given, they can be misleading. As a consequence, we cannot comprehend the evolution of elements and traceability relationships among the elements cannot be managed sufficiently.

This thesis presents a novel solution to the identification problem. It establishes a representation to describe the history of a model and its fine-grained elements inside. The key feature of the representation is a new kind of traceability relationship, called identification links. They allow us to identify elements of a given revision in other revisions or variants of the model. The identification is even applicable to anonymous elements and model fragments. It provides us with a broad spectrum of opportunities: e.g. management of fine-grained traceability links, evolution analysis, merging of development branches. Due to the expression of model evolution in the history representation, we are further able to capture the changes that have been applied to the traced elements.
This thesis further presents an algorithm to infer the identification links automatically. The approach does not rely on persistent identifiers, but it utilizes a similarity-based model comparison technique to locate the model elements in
other revisions.

The algorithm and the history representation have been implemented in a prototype. It is metamodel and tool independent and can work with an arbitrary SCM. Existing modeling environments do not have to be modified. Traceability information and evolution information is accessible through a service interface and can thus be integrated in arbitrary tools. The evaluation of our approach by means of controlled experiments with data from real models attested excellent precision and recall values for the identification of model elements over time. Furthermore, different evolution analysis tools have already been built on our approach, which documents the practical applicability of our solution.