Towards automation of checklist-based code-reviews

Different types of code-reviews (Fagan-style code-inspections, Parnas-like active design reviews and walkthroughs) have been found to be very useful in improving the quality of software. In many cases reviewers use checklists to guide their analysis during review sessions. However, valuable, checklist-based code-reviews have the principal shortcoming of their high costs due to lack of supporting tools enabling at least partial automation of typical multiple appearing rules. This paper describes an approach towards semi-automation of some steps of individual review processes based on checklists. The method proposed is interactive, i.e. reviewers will be enabled to actualize, extend, and check the consistency and redundancy of their checklists. The basic idea underlying the approach is the usage of a rule-based system, adapting concepts of the compiler theory and knowledge engineering, for acquisition and representation of knowledge about the program. Redundant and conflicting knowledge about the program under study is recognized and solved by means of an embedded truth maintenance system. As a result of fault diagnosis, rules for fault classification are used. Software reliability models are applied to validate the results of each review session. The approach has shown promising preliminary results in analyses of conventional C-programs developed in the automotive industry.