Date of this Version
Our previous studies of DUTs addressed research questions of effectiveness, efficiency, and robustness with respect to one software artifact, Siena, and we believe the findings generalize to similar artifacts. Still, we realize that existing studies suffer from threats to validity. Specifically, the selected artifact provided limited exposure to CR in the presence of deeper heap structures, extensive software changes, and high number of methods invocations. We start to address those threats to the validity of our findings by investigating the performance of CR in the presence of such settings.
The findings in this report reveal that the performance of the different carving strategies can vary significantly in programs with complex heap structures, that the ReplayAnomalyHandler can enhance DUTs reuse and potential for fault detection with affordable replay costs, and that the clustering projection can be very effective to reduce the number of DUTs on high-frequency methods.