Programming faults and errors: Towards a theory of software incorrectness

Abstract

Programming faults are defined in the framework of the program verification schema (proof outline). Component S in program P is faulty if P cannot be proved correct with the current implementation of S but it can be proved using the implementation specification for S. A programming error is a state that violates that specification. Conditions for error propagation and masking are expressed in terms of the relationships between the implementation and design specification of S, which defines the role of S in the overall design of P. Errors propagate due to the dependencies between program entities. It is shown that “classical” static dependencies, developed for the purpose of code optimization, are inadequate for the analysis of error propagation since they do not capture events that occur on individual paths through the program. A novel path analysis method is proposed to identify variables potentially corrupted on a same path due the existence of the fault. The method is based upon error propagation axioms. The axioms are used to define path relations for structured programming constructs. The relations provide a conservative structural approximation to the semantical theory of error creation and propagation and are shown useful in testing, debugging and static analysis.