Derivation of Architecture Decision Records for RCEnvironment and their Validation using jQAssistant

Abstract

Software projects are defined to a high degree by architecture decisions about the organization of code into components and how these components work together. These decisions are gradually made during the evolution of a product. Even though it would be beneficial for development, documentation of this process does not exist for most project. If architecture decisions are documented, a common problem is their correctness since documentation often does not get as much attention as the development of new features. Hence, architecture documentation easily gets outdated having low value for developers. We present a workflow that aids a software architect solving both problems. The first part of our workflow consists of using large language models in order to retrospectively document architecture decisions. The second part ensures that architecture documentation is up-to-date. To achieve this, we formulate architecture decisions as statements using logic-inspired human language and subsequently translate them to rules that can be validated with the quality assurance tool jQAssistant. We evaluate the whole workflow with the open-source project RCE that is developed at German Aerospace Center.