Useful processes, that are independently repeatable, are utilised in all branches of science and traditional engineering disciplines but seldom in software engineering. This is particularly so with processes used for detection and correction of defects in software systems. Code inspection, as introduced by Michael Fagan at IBM in the mid 1970's is widely recognised as an effective technique for finding defects in software. Despite its reputation, code inspection, as it is currently practiced, is not a strictly repeatable process. This is due to the problems faced by inspectors when they attempt to paraphrase the complicated semantics of a unit of computer code. Verification based software inspection, as advocated by the cleanroom software engineering community, requires that arguments of correctness be formulated with the code and its specification. These arguments rely on the reader being able to extract the semantics from the code. This thesis addresses the requirement for an independently repeatable, scalable and substantially automated method for yielding semantics from computer code in a complete, unambiguous and consistent manner in order to facilitate, and make repeatable, verification based code inspection. Current literature regarding the use of code inspection for verification of software is surveyed. Empirical studies are referenced, comparing inspection to software testing and program proof. Current uses of formal methods in software engineering will be discussed, with particular reference to formal method applications in verification. Forming the basis of the presented method is a systematic, and hence repeatable, approach to the derivation of program semantics. The theories and techniques proposed for deriving semantics from program code extend current algorithmic and heuristic techniques for deriving invariants. Additionally, the techniques introduced yield weaker forms of invariant information which are also useful for verification, defect detection and correction. Methods for using these weaker invariant forms, and tools to support these methods, are introduced. Algorithmic and heuristic techniques for investigating loop progress and termination are also introduced. Some of these techniques have been automated in supporting tools, and hence, the resulting defects can be repeatably identified. Throughout this thesis a strong emphasis is placed on describing implementable algorithms to realise the derivation techniques discussed. A number of these algorithms are implemented in a tool to support the application of the verification methods presented. The techniques and tools presented in this thesis are well suited, but not limited to, supporting rigorous methods of defect detection as well as formal and semi-formal reasoning of correctness. The automation of these techniques in tools to support practical, formal code reading and correctness argument will assist in addressing the needs of trusted component technologies and the general requirement for quality in software.
Identifer | oai:union.ndltd.org:ADTP/195408 |
Date | January 2003 |
Creators | Powell, Daniel, n/a |
Publisher | Griffith University. School of Computing and Information Technology |
Source Sets | Australiasian Digital Theses Program |
Language | English |
Detected Language | English |
Rights | http://www.gu.edu.au/disclaimer.html), Copyright Daniel Powell |
Page generated in 0.0011 seconds