Taming the Software Development Complexity with Domain Specific Languages

Abstract

Modern computed tomography (CT) scanners are highly complex and flexible devices. This versatility is realized with a multitude of interconnected parameters and rules which are defined by domain experts in so-called scanner model specifications distributed over almost one hundred documents. The primarily used tools to write these documents (e.g. MS Word, MS Excel) are domain agnostic and they support only plain natural-language for the specification. Consequently, maintaining a valid scanner specification is a tedious, error-prone and therefore expensive process. To tackle the complexity of scanners parameters specifications, over the last two years we developed and deployed an eco-system of domain specific languages (DSLs) and associated tooling, covering a central portion of the scanner domain. The languages are developed using the Jetbrains’ MPS language workbench. In this paper, we present our experiences with developing our language eco-system. We briefly describe the language architecture, the design and development process that led us there, and discuss variation points of our approach and present in more detail a set of lessons learnt and best practices.