PhD Thesis - staffweb - University of Greenwich

PhD Thesis by John Ewer.interface and dynamic visualisation indicate that Object Oriented design and the event drivenparadigm are necessary implementation strategies. These architectural changes are desirablefrom the point of view of the final delivery system but they have large implications for theapproach used and difficulties to overcome in order to re-engineer the legacy software into anappropriate form.In order to perform the software reverse engineering of the legacy software it was necessary toidentify the complete calling architecture, common procedural code and code duplication as wellas any "inline" methods. The full design and algorithmic methods used within the legacy softwarehad to be extracted for use in the new system in the most concise and self-consistent formpossible to support extended research.It was also necessary to determine if the legacy software contained any computational objectswhich could be created within the new system to encapsulate concepts, related data andmethods. The creation of abstract types provided significant flexibility and ease of maintenancewithin the target system. There was also an improvement in general code clarity when objectshad intuitive and self-consistent meanings. Often the use of alternative data structures couldsupport lateral research thinking because new ways of problem solving become apparent due tothe nature of the objects themselves. The potential problem of using objects was the potentiallysignificant overheads that have been noted in some Object Oriented (OO) implementations[ANGUS91] [DUBOIS-PELERIN93]. It has already been noted that performance is frequentlyan important consideration when choosing a CFD code and it is vital that the imposition ofObject Orientation should not drastically affect performance. Angus et al [ANGUS94] hadnoticed quite a large overhead for Object Orientation when applied to lower "processing" levelsof a "flutter analysis" simulation code. Their approach to overcoming these overheads was tokeep the lower levels using simple data structures but to group the simple structures andmethods at quite a high level of the calling architecture so that the performance hit of ObjectOrientation was minimal. Such an approach was considered for the re-engineering of the legacyCFD code but there are a number of potential Object Oriented exploitations that would not bepossible with this half-way house re-structuring. Care was taken to avoid the potential problems[WEBSTER] of using an Object Oriented development by investigating the possible problems3-39