A method for re-modularising legacy code

Abstract

This thesis proposes a method for the re-modularisation of legacy COBOL. Legacy code often performs a number of functions that if split, would improve software maintainability. For instance, program comprehension would benefit from a reduction in the size of the code modules. The method aims to identify potential reuse candidates from the functions re-modularised, and to ensure clear interfaces are present between the new modules. Furthermore, functionality is often replicated across applications and so the re-modularisation process can also seek to reduce commonality and hence the overall amount of a company's code requiring maintenance. A 10 step method is devised which assembles a number of new and existing techniques into an approach suitable for use by staff not having significant reengineering experience. Three main approaches are used throughout the method; that is the analysis of the PERFORM structure, the analysis of the data, and the use of graphical representations. Both top-down and bottom-up strategies to program comprehension are incorporated within the method as are automatable, and user controlled processes to reuse candidate selection. Three industrial case studies are used to demonstrate and evaluate the method. The case studies range in size to gain an indication of the scalability of the method. The case studies are used to evaluate the method on a step by step basis; both strong points and deficiencies are identified, as well as potential solutions to the deficiencies. A review is also presented to assesses the three main approaches of the methods; the analysis of the PERFORM and data structures, and the use of graphical representations. The review uses the process of software evolution for its evaluation using successive versions of COBOL software. The method is retrospectively applied to the earliest version and the known changes identified from the following versions are used to evaluate the re-modularisations. Within the evaluation chapters a new link within the dominance tree is proposed as is an approach for dealing with multiple dominance trees. The results show that «ach approach provides an important contribution to the method as well as giving a useful insight (in the form of graphical representations) of the process of software evolution.