Building Trust in a Model-Based Automatic Code Generator
S. Tucker Taft, Elie Richa, Andres Toom · ACM SIGAda Ada Letters · 2017
If we wish to use an automatic code generator for the modelbased development of a safety-critical system, how can we gain sufficient confidence in the correctness of the tool? For a tool like a code generator, which could insert an error into an airborne system, the US Federal Aviation Administration (FAA) requires the highest level of tool qualification [2], Tool Qualification Level 1 (TQL-1), if the tool is going to be used for a Level-A subsystem (one whose failure could be catastrophic). Achieving TQL-1 for such a code generator is analogous to achieving Level A certification for an embedded software component, but the lines of code in the tool can be substantially greater. In this paper we describe approaches to manage the complexity of specification and testing required for Level-1 qualification of a tool like an automatic code generator, a tool which includes multiple phases that transform an input model into optimized generated code.