A test-driven development strategy for the construction of grammar-based software.

Grammar-based software is increasingly becoming a prominent and well defined subset of software engineering through the popularity of analysis tools, metrics evaluators and the increasing prevalence of software tools that rely upon extensible data such as XML. Software-testing plays a crucial role in the lifecycle of any modern software system, hence the correct and adequate testing of grammar-based software is essential. This thesis provides a review of the current research and practice in the field of grammar-based software. The motivation behind software-testing is examined and a summary of existing testing techniques is presented. This leads to a detailed empirical comparison of test-suites for grammar-based software. The effectiveness of test-suite generation via methods such as Purdom’s algorithm is contrasted with test-suite reduction via a novel elaboration of the notion of coverage for a grammar. The effectiveness of this strategy is discussed with regard to two testing analysis criteria: code coverage and fault detection. The second contribution of this thesis is the definition of a framework for developing grammar-based software within a strict test-driven environment. We describe the techniques used to develop a static analysis tool for the ISO C++ grammar and the verification of its correct operation via standardised test-suites. This work was completed in two distinct phases. The first phase involves porting an existing system from C++ to Java by combining eXtreme Porting with an order for porting derived from a novel use of Object Relation Diagrams. The second phase involves a large automated testing process including black-box testing, coverage comparisons and comparisons of reverse-engineered UML sequence diagrams from program traces. The third contribution of this thesis involves the utilisation of the generalised parsing algorithm, GLR, in the implementation of a parser generator. We describe the design and test-driven development of this parser generator, and its practical application as part of an analysis tool for ISO C++. Finally, we provide a study of the performance of a generated GLR parser when parsing ISO C++ programs, and a comparison with an existing system based on a generated backtracking parser.