On Faults and Faulty Programs

A fault is an attribute of a program that precludes it from satisfying its specification; while this definition may sound clear-cut, it leaves many details unspecified. An incorrect program may be corrected in many different ways, involving different numbers of modifications. Hence neither the location nor the number of of faults may be defined in a unique manner; this, in turn, sheds a cloud of uncertainty on such concepts as fault density, and fault forecasting. In this paper, we present a more precise definition of a program fault, that has the following properties: it recognizes that the same incorrect behavior may be remedied in more than one way; it recognizes that removing a fault does not necessarily make the program correct, but may make it less incorrect (in a sense to be defined); it characterizes fault removals that make the program less incorrect, as opposed to fault removals that may remedy one aspect of program behavior at the expense of others; it recognizes that isolating a fault in a program is based on implicit assumptions about the remaining program parts; it identifies instances when a fault may be localized in a program with absolute certainty.