A Foundation for Development of Programming Languages for Real-Time Systems

Applications of real-time systems have grown considerably in both diversity and popularity, and the demand for real-time software has never been higher. In contrast, the choice of programming languages used to develop these systems has mostly remained limited to decades-old languages, namely Ada and C/C++, and more recently real-time Java. We postulate that the main reason for this mono-culture is the difficulty of developing new programming languages for real-time systems, due to their strict correctness requirements. Wang et al. argue that implementing even general-purpose languages is not easy, and is the source of many problems in today's languages. They propose the Micro Virtual Machine as a minimal abstraction layer to relieve the challenges of implementing a managed language, and design a micro VM specification named Mu. Compared to conventional language VMs, a micro VM is minimal and low-level. This makes micro VM an appealing idea towards creating a platform for development of programming languages for real-time systems, as it allows supporting a wide range of languages, aligned with the huge diversity of real-time systems. It also makes correct implementation and formal verification of the platform easier, which is vital for many real-time systems. Prior to this thesis, the only concrete micro VM specification was [Mu, 2018]. However,Mu is not designed for real-time systems and lacks some of the essential features. My thesis is that a real-time-enabled micro virtual machine is capable of providing an efficient and usable foundation for development of program-ming languages, suitable for building real-time software. The first high-level contribution of this thesis is the design of RTMu, a micro VM instance aiming at programming languages for real-time systems. We build on the Mu specification and propose a set of modifications to its abstractions over concurrency and memory management to make it suitable for real-time systems. The second contribution is the affirmation of the implementability of the RTMu's abstractions. For this purpose, we build a high-performance implementation of the RTMu specification, based-on a high-performance implementation of Mu. The third contribution is the design of a real-time extension to RPython, to make it a fully-featured language for real-time systems, named RT-RPython. We implement RT-RPython on top of RTMu and evaluate its performance through the CollisionDetection benchmark suite. This thesis is a proof of concept for establishing the use of micro VMs to build new high-quality programming languages for real-time systems. It also provides empirical demonstration of performance and predictability for micro VMs in the real-time domain. We believe that RTMu can help in tackling the current lack of diversity in programming languages for real-time systems