A Verification-Driven Approach to Traceability and Documentation for Auto-Generated Mathematical Software

Abstract

Automated code generators are increasingly used in safety-critical applications, but since they are typically not qualified, the generated code must still be fully tested, reviewed, and certified. For mathematical and engineering software this requires reviewers to trace subtle details of textbook formulas and algorithms to the code, and to match requirements (e.g., physical units or coordinate frames) not represented explicitly in models or code. We support these tasks by using the AutoCert verification system to identify and verify mathematical concepts in the code, recovering verified traceability links between concepts, code, and verification conditions. We then exploit these links to construct a natural language report that provides a high-level structured argument explaining where the code uses specified assumptions and why and how it complies with the requirements. We have applied our approach to generate review documents for several sub-systems of NASA's Project Constellation.