Implementing RUSSELL

We have completed an implementation of the Russell programming language [Don 85]. This effort has been very helpful in the evaluation of the original language design. It has also served to pinpoint the difficulties in implementing languages with type systems as general as that of Russell. Russell treats both functions and types as data objects which can be freely manipulated by the program. Most operators present in conventional programming languages are viewed as function calls. In spite of this, our compiler produces surprisingly efficient machine code, even with minimal effort invested in the code generator. The generality of the language served to simplify some aspects of the compiler. We focus on the separate compilation mechanism. The most difficult implementation problem is that of inferring typing information omitted by the programmer. We argue that this is an essential part of type checking a language such as Russell. Our current solution is only partially satisfactory.