Scoping Constructs for Program Generators

Program generation is the process of generating code in a high-level language (e.g., C, C++, Java) to implement an abstract specification of a program. Generated programs are created by synthesizing and composing code fragments. Binding identifiers in generated code with their correct variable declarations has been the focus of a lot of research work in the context of macro-expansion (e.g., hygienic macro expansion and syntactic closures mechanisms). The common solutions include automatically maintaining identifier environments, which determine the legal bindings for an identifier. In this paper we present generation scoping: an adaptation of hygienic macro-expansion techniques to general-purpose program generation. The conceptual novelty of generation scoping lies in making identifier environments first-class objects and organizing them explicitly into directed graphs. This approach yields significant benefits: We are able to express scoping relationships independently of the structure of the generated program. This way we get a stronger tool for detecting errors in the specification of generated code. Additionally, we are able to simplify the specification by employing powerful implicit qualification. Thus, generation scoping becomes a useful layer of infrastructure for implementing software generators: it both solves binding problems and makes code specification more convenient.