Controlled evolution of adaptive programs

摘要

Adaptive programming (AP) is a programming paradigm for expressing structure-shy computations over semi-structured data graphs. Structure-shyness means that adaptive programs hard code a minimal set of assumptions about the structure of their input. Because of this, adaptive programs are more susceptible to unsafe evolutions; evolutions that jeopardize the correctness of adaptive programs yet go uncaught. In this paper we study the evolution of adaptive programs and present two complementary approaches for controlling their unsafe evolution: a language for expressing application-specific constraints on the runtime behavior of adaptive programs, and a stricter notion of compatibility between the parts of an adaptive program, that does not sacrifice the expressiveness of the AP paradigm.