PhD Thesis - staffweb - University of Greenwich

PhD Thesis by John Ewer.nature of the data access mechanisms that would be imposed by mixed language developmentor the interface to the library routines.2.4.6 Reverse engineering of a legacy CFD system back to basic design andre-implement from this designThis process would take a considerable time and is by no means easy. The reverse-engineeringtechniques available [BYRNE91] [BRAND96] to "mine" an existing software system for theunderlying design are by no means infallible [BERGEY]. Furthermore it is possible and indeedquite likely that errors will be introduced in the re-design or re-implementation phases of reengineering.This technique does have the benefit of considerable flexibility in the choice ofimplementation language. As with a new development there is a long time before the newlydeveloped code can be tested against the legacy code behaviour. However the behaviour is moreassured than simply developing a new code from first principles because the design has beenextracted from a working and complete software system. Reverse engineering will generallypreserve all of the algorithms and empirical methods but may not necessarily preserve an audittrail back to the legacy software. This is particularly true if the structure and data architecturehave been drastically altered in the re-design process. It is likely that any updated methods andbug fixes would also have to go through a complete re-engineering and re-implementationprocess in order to be assured of functional consistency.2.4.7 Reverse engineering of a legacy CFD system and re-implementationusing an incremental approach and imposed data and controlarchitectureThis approach to re-engineering has to make concessions in the design to maintain consistencywith legacy system whilst incrementally re-engineering. One of the most important benefits isthat the re-engineered system (and the incremental stages) are never very far from a workingCFD code that can be validated as being algorithmically correct and consistent with solutionbehaviour from the legacy software. This gives considerable assurance that the algorithms andempirical methods are preserved during re-engineering. The effort involved is significantly lessthat would be needed for a complete re-engineering re-implementation from a reverse-2-26