Dynamic Object Replacement and Implementation-Only Classes

GILGUL is an extension of the Java programming language that allows for dynamic object replacement without consistency problems. This is possible in a semantically clean way, because its model strictly separates the notions of reference and comparison that usually are subsumed in the concept of object identity. Since GILGUL’s new operations respect Java’s type system, objects still can be replaced only by instances of the same class or any subclasses thereof, but not by instances of any of the former object’s superclasses. The concept of implementation-only classes is an extension of the type system that allows classes to be declared that never must be used as types. Consequently, instances of implementation-only classes always can be replaced by instances of their superclasses. This effectively widens the range of both anticipated and unanticipated adaptations. 1 The TAILOR Project 1.1 Unanticipated Adaptation Software requirements are in a constant flux. Some changes in requirements can be anticipated by software developers, so that the necessary adaptations can be prepared for, for example by suitable parameterization or by application of dedicated design patterns. Within the scope of these anticipated options for adaptation, software can be used already in a flexible way, albeit in restricted limits und with a corresponding increase in development effort. However, unanticipated changes of requirements occur repeatedly in practice, and the above suggested techniques cannot tackle them by definition. Furthermore, the manual integration of hooks for any conceivable eventuality is not a feasible option, since this dramatically decreases reliability, efficiency and maintainability of systems. Alternatively, programming languages and runtime systems should be equipped with features that allow for far reaching manipulations of program internals without destructively modifying its source code. This leads to an increase of options for unanticipated adaptations as well as a decrease of effort to prepare for anticipated adaptations. Therefore the structure of programs can be kept much simpler from the outset and they can be focussed on solving their primary tasks. In order to provide for this significant simplification of software development, the goal of the TAILOR Project at the Institute of Computer Science III of the University of Bonn [21] is to conceive and implement enhancements of programming languages and runtime systems to allow for unanticipated adaptability of software. In doing so, special attention is payed to the following issues. 1.2 Component-Oriented Software The focus on Component-Oriented Software pushes the limits even further, since it generally is not required that components can be modified in unanticipated ways when being integrated into a system. On the contrary, components usually are deployed using a compiled format, and their source code is not available for modifications. Even if the source code can be accessed in some cases, destructive modifications still are not feasible, since they would lapse in the presence of new versions of a component. However, it is still highly desirable to be able to use components in unanticipated contexts and with unanticipated adaptations. How can such unanticipated adaptations be carried out on a software component when only its interfaces are guaranteed to be known and when the need to update to new versions in the future have to be taken into account? 1.3 Reduction of Downtime Essentially, unanticipated adaptations of software can take place at two points in time. They can be performed before a program is being linked into its final form, or they can happen during runtime. Adaptations before linktime can be tackled by systems that allow for transformations of source code or