Feature-specific profiling

High-level language and library features provide significant readability and maintainability benefits. Their performance costs, however, are usually not predictable, at least not easily. Programmers may accidentally misuse features and produce underperforming programs as a result. This paper introduces feature-specific profiling, a profil-ing technique that reports performance costs in terms of lan-guage and library features. With a feature-specific profiler, a programmer can identify specific feature instances that are responsible for performance problems. In addition the paper presents the architecture of our feature-specific profiler. At a high level, the profiler consists of two pieces: a generic ker-nel and feature-specific plug-ins. The API between the two relies on a lightweight form of stack inspection