Model driven architecture as a facilitator for automatic code generation

As the amount of software on platforms continues to increase, our capability to rigorously assess this software reduces. Like assembly code before it, software written in high level programming languages is becoming difficult to reason about due to its volume and complexity. For example, some aerospace platforms now have more than one million lines of code, much of it safety related. One approach would be to deploy tools to automatically generate the software code, yet this may compromise our ability to reason about the software's behaviour rigorously. This is especially the case where there is little evidence about how the code was constructed by the tool. The OMG's "model driven architectures" proposal may provide one approach to resolving this by separating analysis models from implementation models and defining clear mappings between them. This paper assesses the need for autocode tools, the various approaches to implementing them and the options for assessing correctness of their operation.