Profiling For Laziness

While many programmers appreciate the benefits of lazy program-ming at an abstract level, determining which parts of a concrete program to evaluate lazily poses a significant challenge for most of them. Over the past thirty years, experts have published numer-ous papers on the problem, but developing this level of expertise requires a significant amount of experience. We present a profiling-based technique that captures and auto-mates this expertise for the insertion of laziness annotations into strict programs. To make this idea precise, we show how to equip a formal semantics with a metric that measures waste in an evalu-ation. Then we explain how to implement this metric as a dynamic profiling tool that suggests where to insert laziness into a program. Finally, we present evidence that our profiler’s suggestions either match or improve on an expert’s use of laziness in a range of real-world applications