The automated analysis of object-oriented designs
This thesis concerns the use of software measures to assess the quality of object-oriented designs. It examines the ways in which design assessment can be assisted by measurement and the areas in which it can't. Other work in software measurement looks at defining and validating measures,or building prediction systems. This work is distinctive in that it examines the use of measures to help improve design quality during design time. To evaluate a design based on measurement results requires a means of relating measurement values to particular design problems or quality levels. Design heuristics were used to make this connection between measurement and quality. A survey was carried out to find suggestions for guidelines, rules and heuristics from the 00 design literature. This survey resulted in a catalogue of 288 suggestions for 00 design heuristics. The catalogue was structured around the 00 constructs to which the heuristics relate, and includes information on various heuristic attributes. This scheme is intended to allow suitable heuristics to be quickly located and correctly applied. Automation requires tool support. A tool was built which augmented the functionality available in existing sets, and taking input from multiple sources of design information (e.g., CASE tools and source code) and the described so far presents a potential method for automated design assessment provides the means of automation. An empirical study was then required to consider the efficacy of the method and evaluate the novel features of the tool. A case study was used to explore the approach taken by, and evaluate the effectiveness of, 15 subjects using measures and heuristics to assess the design of a small 00 system(IS classes). This study showed that semantic heuristics tended to highlight significant problems, but where attempts were made to automate these it often led to false problems being identified. This result, along with a previous finding that around half of quality criteria are not automatically assessable at design time, strongly suggeststhat people are still a necessary part of design assessment. The main result of the case study was that the subjects correctly identified 90% of the major design problems and were very positive about their experience of using measurement to support design assessment.