Classdesc and Graphcode: support for scientific programming in C++

Abstract

Object-oriented programming languages such as Java and Objective C have become popular for implementing agent-based and other object-based simulations since objects in those languages can {\em reflect} (i.e. make runtime queries of an object's structure). This allows, for example, a fairly trivial {\em serialisation} routine (conversion of an object into a binary representation that can be stored or passed over a network) to be written. However C++ does not offer this ability, as type information is thrown away at compile time. Yet C++ is often a preferred development environment, whether for performance reasons or for its expressive features such as operator overloading. In scientific coding, changes to a model's codes takes place constantly, as the model is refined, and different phenomena are studied. Yet traditionally, facilities such as checkpointing, routines for initialising model parameters and analysis of model output depend on the underlying model remaining static, otherwise each time a model is modified, a whole slew of supporting routines needs to be changed to reflect the new data structures. Reflection offers the advantage of the simulation framework adapting to the underlying model without programmer intervention, reducing the effort of modifying the model. In this paper, we present the {\em Classdesc} system which brings many of the benefits of object reflection to C++, {\em ClassdescMP} which dramatically simplifies coding of MPI based parallel programs and {\em Graphcode} a general purpose data parallel programming environment.