Optimizing runtime performance of hybrid dynamically and statically typed languages for the .NET platform

Abstract

Dynamically typed languages have become popular in scenarios where high flexibility and adaptability are important issues. On the other hand, statically typed languages provide important benefits such as earlier type error detection and, usually, better runtime performance. The main objective of hybrid statically and dynamically typed languages is to provide the benefits of both approaches, combining the adaptability of dynamic typing and the robustness and performance of static typing. The dynamically typed code of hybrid languages for the .Net platform typically use the introspection services provided by the platform, incurring a significant performance penalty. We propose a set of transformation rules to replace the use of introspection with optimized code that uses the services of the Dynamic Language Runtime. These rules have been implemented as a binary optimization tool, and included as part of an existing open source compiler. Our system has been used to optimize 37 programs in 5 different languages, obtaining significant runtime performance improvements. The additional memory resources consumed by optimized programs have always been lower than the corresponding performance gains obtained.