Constraint-Based Operational Consistency of Evolving Software Systems (COCoS)

• Applicants: Professor Dr. Reiner Hähnle; Professor Dr. Bernhard Steffen
• Subject Area: Software Engineering and Programming Languages ; Theoretical Computer Science
• Term: from 2017 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 351097374

In this project we will combine static analysis, runtime verification,automata learning and monitoring to automatically enforce behavioralconsistency between specification, implementation, and running system. Key to theenvisioned technology is the continuous synchronization of thetraceable HOW knowledge of the design and implementation and the learning-based and therefore self-adapting behavioral WHAT knowledgegained at runtime. The different knowledge domains of the system areencoded in a common constraint view expressible in first-order logicand the evolution is reflected by adaption of these constraints.Consistency is then established and preserved using automatedreasoning technology, which allows us to expose sources of conflict atthe level at which they are caused. Maintaining overall coherentsystem knowledge is thereby reduced to local conflict resolution, whichrequires manual interaction only at the specification level, in casedesign decisions are required. It bridges the inevitable semantic gapin software engineering with a minimum of manual interaction. This ispossible due to the looseness of the considered operationalconsistency, which does not reflect syntactic features but focuses onruntime behavior, the true primary concern. The impact of ourapproach will be evaluated by direct application in the developmentand evolution process of Springers Online Conference Service (OCS),an enterprise-level information system for multi-role online evaluationand production of conference proceedings. In particular, we will illustrate how changes at the HOW level, e.g. the software realization of a new feature like double blind reviewing, become visible at the WHAT level, i.e., in the newly learned behavioural model, which no longer prohibits reviewers to access authors information. Our approach allows one to monitor intended change, as well as to automatically reveal unwanted feature interaction.

• DFG Programme: Research Grants