A literate process for the certification of scientific and engineering software

点击翻译按钮获取中文摘要

Our principal objective is to develop certifiable scientific and engineering software. One problem with current processes is that there is considerable duplication between software artifacts, as well as a definite lack of traceability between these artifacts. We believe that we can solve both of these problems by adapting the ideas behind Literate Programming (LP), in which programs are written first with the goal of being understandable to human readers, rather than being artificially constrained by the ordering and formatting required by conventional computer code.In classical LP, from literate source files two items are extracted: the user readable documentation and the computer readable source code. Our proposed literate process will extend this idea so that "source" files can be processed in multiple ways to extract not just the code and its explanation, but also the requirements specification, the Application Program Interface (API), the test cases, the test report and the build instructions. This will allow us to share the considerable amount of information in common between all these software artifacts, thus decreasing duplication. At the same time traceability is increased, since this shared information will be extracted into the correct artifacts.To develop, motivate and document the literate process we will use a case study of code generation for a family of ordinary differential equation (ODE) solvers. In the family approach code is generated taking into account the commonalities in the family, but distinguishing each member based on its variabilities. The code generation approach facilitates reuse of certified components, leading to quick certification. The family of ODE solvers will be put in the context of nuclear safety analysis. In particular, an example will be developed that solves a system of ODEs as part of the thermal analysis of fuel pins in a nuclear reactor.