Ulrich Thomas Gabor, Daniel Siegert, Olaf Spinczyk (TU Dortmund / Universität Osnabrück)
We present a novel method to inject software faults into source code, which is capable of operating on unpreprocessed C/C++ source code and can inject faults into source files, with as little modifications as possible. Tools that inject faults on lower levels, i.e. assembly or intermediate representation, suffer from limited accuracy regarding the emulated fault model. In contrast, our approach operates on source code and utilizes Clang AST Matchers to emulate faults as close to the model as possible. Additionally, we show how macros and other compile-time constructs, additional image modifications by other tools, and compiler optimization settings can have a negative impact on the validity of performed injection campaigns. Furthermore, we highlight some weaknesses in the description of the currently used fault models, which are based on the different injection levels and progress in the software development processes.