Indicateur de qualite pour les systemes orientes objet: Vers un modele unifiant plusieurs metriques.

摘要

Current industrial systems are increasingly complex. The management (insurance and control) of their quality answers important technical and economic stakes. The quality of software is a complex concept. It has several characteristics. Metrics, in this context, constitute genuine "probes" making it possible to evaluate several attributes of quality.;To probe the quality of a software, metrology, as in all the fields of engineering, remains the most objective means and most reliable. The field of metrics, in particular in the object oriented paradigm, was the subject of a multitude of proposals.;Several of these metrics were largely tested and discussed in the literature. On the other hand, some of them were not the subject of great validations. In addition, their number, rather significant, especially in the case of object-oriented systems, raises several problems, in particular, the quantity of information which they provide which is difficult to assimilate and manage (and even to interpret), on the one hand, and the overlapping which exist between several of them, on the other hand.;The idea to unify some metrics to capture high level attributes is the principal motivation of this work. The model that we propose (Quality Indicator Qi), in this work, is based on control graphs, probabilities of methods invocation, their cyclomatic complexity and their testing coverage. Its experimentation, on several large-scale software projects, enabled us to show his effectiveness, on several levels, relatively to several existing object-oriented metrics. The results obtained show that our model is a good indicator of several quality attributes such as the changeability and testability without counting certain attributes captured in an indirect way. We also developed a tool supporting this model, allowing to determine the indicators of quality of each component of a software system written in Java.