Extending Source Code Generators for Evidence-Based Software Certification

Automated code generation offers many advantages over manual software development but treating generators as trusted black boxes raise problems for certification. Traditional process-oriented approaches to certification thus require that the generator be verified to the same level of assurance as the generated code, but this is infeasible for realistic generators. However, generators can be extended to support an evidence-based approach to certification. By careful design of the trusted kernel, assurance of the generator itself is not required. In this paper, we describe several related extensions to two in-house code generators to provide two forms of evidence along with the code: safety proofs and safety explanations. We also describe how additionally provided links are used to trace between the code and the safety artifacts.