Modelagem estatística de mudanças estruturais para simulação de evolução de software.

Abstract

The scenario of recent research in computer science is undergoing a significant transformation. Statistical approaches, typically present in scientific studies from various fields of human knowledge, used to be neglected by researchers of our area. In software engineering, there are still few results that have adequate statistical validation. This happens by several factors, among which we may evidence the lack of samples. However, this problem is not exclusive of us, and strategies to minimize them are well known, among which stands out modeling. Models, by definition, have limitations, but allow the exercise of simulated scenarios, allowing eventually statistical validation. Particularly in the field of software evolution, models of change are known. But when it comes to specific niches like fine-grained changes (involving smaller entities and relationships of the software), there is little or no theoretical basis of support in the academic literature. That is, there is no theoretical model that supports, for example, how to evolve the number of classes and methods of a Java software in a realistic way. So, returning to the context of Software Engineering research and its lack of samples, as well as the insufficient structural changes models, the results emerge in a series of papers that show such changes being performed rather indiscriminate (ad hoc), and with all the limitations it imposes. Facing with this problem, this work shows up as the first effort to devise formal statistical models of the software structural changes and its application through simulation. In particular, models derived from the statistical analysis of data from a range of real structural changes of open source Java software. The application and evaluation of models takes place through its use in a simulator of software version implemented for this research. The results of the formal model and experimentation via simulator bring up a series of new results, and show good adequacy of the approach to the problem.