Habilitando a checagem estática de conformidade arquitetural de software em evolução.

Abstract

The refíexion model technique is a static conformance checking technique to keep architecture module views and implementation conformant. It can either prevent or remedy software aging by combating architecture deterioration. However, the amount of manual effort to apply the technique may prevent its use in practice, especially in the context of software evolution in lightweight development methods. More specifically, it can be time-consuming and costly to: i) produce a high-level model and the mapping between source code entities and this model; ii) keep both model and mapping up-to-date as software evolves; and Ui) analyze the usual large number of architectural violations in the source code reported by the technique. This work tries to enable static conformance checking of evolving software by partially automating the manual effort to apply the refíexion model technique. To do so, the potential of clustering techniques to generate high-level models and keep them up-to-date is evaluated. It is also proposed and evaluated an incremental mapping approach between source code entities and high-level models based on the combination of information retrieval of software vocabulary and structural dependencies. Last, a prioritizing technique based on software history to recommend architectural violations in the source code most likely to be relevant to software developers is reported and evaluated. Clustering techniques are evaluated by measures of accuracy and stability, and results for four different clustering algorithms show that none of them performs best for ali measures, and that they are limited to provide fully automated generation of high-level models. On the other hand, evaluation suggests that the mapping step in the refíexion model technique can be enabled by the proposed incremental automated mapping technique that combines structure and vocabulary. In two case studies, the combined technique showed the highest F-measure values for both singleton, small and large source code changes. Finally, evaluation for the prioritizing technique shows that, from five studied factors, violation duration and violation co-location correlate well with violation relevance. Results suggest that these factors can be used to rank the violations most likely to be relevant, with an improvement of at least 57% against a baseline of randomly selected violations. Analysis of the results suggests that the producton of high-level models for static conformance checking can be aided by a semi-automated architecture recovery process, and, as software evolves, by incremental clustering/mapping techniques. On the other hand, analysis of the results for the prioritizing technique suggests the effectiveness of an automated approach to recommend architectural violations to bí analyzed by software developers.