Hypothesis-based concept assignment to support software maintenance

Abstract

Software comprehension is one of the most expensive activities in software maintenance and many tools have been developed to help the maintainer reduce the time and cost of the task. Of the numerous tools and methods available, one group has received relatively little attention: those using plausible reasoning to address the concept assignment problem. This problem is defined as the process of assigning descriptive terms to their implementation in source code, the terms being nominated by a user and usually relating to computational intent. It has two major research issues: Segmentation: finding the location and extent of concepts in the source code.Concept Binding', determining which concepts are implemented at these locations. This thesis presents a new concept assignment method: Hypothesis-Based Concept Assignment (HB-CA). A framework for the activity of software comprehension is defined using elements of psychological theory and software tools. In this context, HB-CA is presented as a successful concept assignment method for COBOL II, employing a simple knowledge base (the library) to model concepts, source code indicators, and inter-concept relationships. The library and source code are used to generate hypotheses on which segmentation and concept binding are performed. A two-part evaluation is presented using a prototype implementation of HB-CA. The first part shows that HB-CA has linear computational growth in the length of program under analysis. Other characteristics addressed include HB-CA's scalability, its applicability to other languages, the contribution made by different information sources, domain independence, representational power, and guidelines for the content of the library. The first part concludes by comparing the method and implementation to cognitive requirements for software comprehension tools. The second part considers applications of HB-CA in software maintenance. Five areas for potential cost reduction are identified: business-rule ripple analysis, code ripple analysis, module selection, software reuse, and software module comprehension.