Abstract

It has been pointed out that inheritance and synchronization constraints in concurrent object systems often conflict with each other. Several proposals have been made for resolving the conflicts. We show that, however, for many of the proposals, one cannot avoid anomaly in inheritance where re-definitions of all relevant parent methods are necessary. Our prime contribution is that we formally prove the occurrence of such an anomaly is not a minor problem in the language design, but is an inherent difficulty in the general scheme for specifying the behavior of objects with respect to synchronization. This is more serious than the violation of class encapsulation that has been pointed out by Snyder, for NONE of the parent methods can be inherited except for trivial cases. We then propose an alternative scheme in which the anomaly does not occur, while efficiency is retained using program transformation. Since this transformation is invisible to the programmer, the full benefit of inherit...

Citations