Compilation of Heterogeneous Models: Motivations and Challenges

International audienceThe widespread use of model driven engineering in the development of software-intensive systems, including high-integrity embedded systems, gave rise to a "Tower of Babel" of modeling languages. System architects may use languages such as OMG SysML and MARTE, SAE AADL or EAST-ADL; control and command engineers tend to use graphical tools such as MathWorks Simulink/Stateflow or Esterel Technologies SCADE, or textual languages such as MathWorks Embedded Matlab; software engineers usually rely on OMG UML; and, of course, many in-house domain specific languages are equally used at any step of the development process. This heterogeneity of modeling formalisms raises several questions on the verification and code generation for systems described using heterogeneous models: How can we ensure consistency across multiple modeling views? How can we generate code, which is optimized with respect to multiple modeling views? How can we ensure model-level verification is consistent with the run-time behavior of the generated executable application?In this position paper we describe the motivations and challenges of analysis and code generation from heterogeneous models when intra-view consistency, optimization and safety are major concerns. We will then introduce Project P 2 and Hi-MoCo 3-respectively FUI and Eurostars-funded collaborative projects tackling the challenges above. This work continues and extends, in a wider context, the work carried out by the Gene-Auto 4 project [1], [2]. Hereby we will present the key elements of Project P and Hi-MoCo, in particular: (i) the philosophy for the identification of safe and minimal practical subsets of input modeling languages; (ii) the overall architecture of the toolsets, the supported analysis techniques and the target languages for code generation; and finally, (iii) the approach to cross-domain qualification for an open-source, community-driven toolset