Self-refining aspects for dynamic program analysis

Abstract

Accurate and efficient dynamic analysis tools are needed for understanding and improving performance of programs. However, such tools often produce perturbed and misleading results, since the inserted analysis code may interfere with runtime optimizations performed by the just-in-time compiler. In this work, I present an approach based on aspect-oriented programming (AOP) to dynamically refine the scope of an analysis, gradually reducing the number of intercepted join points at runtime, considering only those that are relevant for the specific analysis. The motivating example is an aspect-based hotspot locator based on execution time profiling.