Functional Programming Languages and the JVM : Comparing Functional Language Compilation Techniques for the JVM

Abstract: Because of its security, high availability, and automatic memory management, the JVM (Java Virtual Machine) is a desirable execution environment. However, since the JVM was originally made for Java, which is an objectoriented language, it is hard for languages with other paradigms to run on it. This thesis explores the challenges of compiling a functional language for the JVM. We implement a backend generating JVM bytecode as an extension to the compiler of MCore, a minimal functional language based on the lambda calculus. The backend features two different ways of compiling functions. One generates classes representing functions and their environment at compile-time, and one creates these function classes dynamically at runtime. The two different versions of compiling functions are compared in regards to execution speed of the outputted bytecode, compiler output size, and recursion depth. The results show that the two different ways of compiling functions perform well on different metrics.