Compiling and distributing generic libraries with heterogeneous data and code representation

Generic programming has evolved at fast pace in recent years. There now exist numerous programming languages that support genericity, and generic libraries were developed that range from small utility libraries to huge general purpose or domain specific libraries. The importance of specification and verification was further emphasized in the context of generic algorithms and data structures, which became major issues in the research community. The largest generic code base is available in C++ which plays a central role for generic programming, because the Standard Template Library (STL) was implemented in C++. The STL ignited the application of its underlying principles to other domains, resulting in influential C++ libraries like the Matrix Template Library (MTL) or the Boost Graph Library (BGL). This thesis deals with a problem that became apparent for languages like C++, Ada, or Modula-3 that support genericity with heterogeneous data and code representation, especially with regard to large generic libraries. These libraries cannot be compiled and distributed without their source code. Application programs using generic libraries will have to compile the occurring instantiations of generic constructs from the libraries’ source code repeatedly during their development cycle. We first present a general approach to overcome the conceptual problem that necessitates this practice. It mandates a strict separation of compiler front-end and back-end which are connected by a special intermediate representation. Then an overview of the GILF system is given, our incarnation of the solution. Although the GILF system was developed for SUCHTHAT, a generic programming language devised by Sibylle Schupp, GILF is designed to support a wide range of generic programming languages with varying semantics for instantiation, specialization and overloading. We achieve this by separating declarations from definitions, as well as splitting the instantiation process into analysis and application. The information gathered by instantiation analysis is propagated from a compiler’s front-end to its back-end as explicit bindings of instantiation parameters to arguments in the mentioned intermediate language. GILF even allows binding multiple algorithms to one function symbol, which provides the basic mechanisms required for online or off-line algorithm selection. Algorithm selection can be exploited with instantiation at load-time or run-time. Thereafter, we introduce XGILF, the XML-based external GILF representation. An extensive specification is elaborated, covering all core features of GILF. Finally, the GILF prototype implementation is discussed in detail, showing the feasibility of our approach.