Feature-based methodology for supporting architecture refactoring and maintenance of long-life software systems

摘要

The long-life software systems withstand many significant changes throughout their life-cycle in orderto follow the evolution of the problem domains. Usually, the software system architecture can notfollow the rapid evolution of a problem domain and with time, the diversion of the architecture inrespect to the domain features becomes prohibiting for software evolution. For avoiding this problem,periodical refactorings of the system architecture are required.Usually, architecture refactorings are postponed until the very last moment, because of the relativelyhigh risk involved and the lack of short-term profit. As a rule, the management is unwilling to acceptarchitecture refactorings unless they become absolutely necessary. The preferred way of working is toadd new system features in an ad-hoc manner and to keep the rule ”Never touch a running system!”.The final result is an architecture decay. The need of performing small refactoring activities turns intoneed for architecture reengineering. In contrast to refactoring, which is a normal evolutionary activity,reengineering is a kind of software ”revolution”. Reengineering projects are risky and expensive. Theeffectiveness of reengineering is also usually not as high as expected. When finally after reengineeringthe required architecture changes take place, it can be too late. Despite the enormous invested efforts,the results of the reengineering usually do not satisfy the expectations. It might happen that verysoon a new expensive reengineering is required.This thesis deals with the problem of software evolution and the decay of software architectures.It presents a method, which assists software evolution in its crucial part, the architecture refactoring.The assistance is performed for both technical and organizational aspects of the software evolution.The thesis provides new techniques for supporting reverse engineering, architecture recovery and redesigningactivities. It also proposes changes to the software engineering process, which can forcetimely architecture refactorings and thus avoid the need of performing architecture reengineering.For the work in this thesis feature modeling is utilized as a main asset. Features are used to fill theabstraction gap between domain requirements and system architecture. Feature models are also usedas an outline for recovering of lost system architectures. Through feature-based analyses a number ofuseful hints and clues for architecture redesign are produced. Finally, feature modeling is used as acommunication between different stakeholders of the software engineering process and on this basis anew requirements engineering process is proposed, which forces the needed architecture refactorings.