Characterizing Performance Regression Introducing Code Changes

Performance regression testing is highly expensive as it delays system development when optimally conducted after each code change. As a result, performance regression testing should be devoted to code changes highly probably encountering regression. In this context, recent studies focus on the early identification of potentially problematic code changes through characterizing them using static and dynamic metrics. The aim of my research thesis is to support performance regression by better identifying and characterizing performance regression introducing code changes. Our first contribution has tackled the detection of these changes as an optimization problem. Our proposed approach used a combination of static and dynamic metrics and built using evolutionary computation, a detection rule, which was shown to outperform recent state-of-the-art studies. To extend our research, we are planning to increase metrics used, to better profile problematic code changes. We also plan on reducing the identification cost by searching for a traedeoff that reduces the use of dynamic metrics, while maintaining the detection performance. In addition, we would like to prioritize test case based on code changes characteristics to be conducted when regression predicted.