Although data flow analysis was first developed for use in compilers, its usefulness is now recognized in many software engineering tools. Because of the origin of global data flow analysis, that is, compilers, the computation of data flow for software tools is based on the traditional data flow framework. However, this framework for computing data flow is not producing the efficiency, the preciseness, or the type of data flow problem needed for software tools. These problems are due to the requirements of exhaustive data flow that information is computed for all possible executions, at all points, and all information is computed at each program point. This work investigates the definition of data flow problems when the properties of all executions, all program points and all information at each program point of exhaustive data flow are relaxed, producing partial data flow information computed on demand. The goal is to define a formal framework, in the vein of the traditional framework, where partial data flow problems can be defined and computed. Because the number of problems will grow with partial data flow, and the data flow problems will not be known in many cases until run time, this project investigates the implementation of partial data flow analysis by defining a specification technique to express the demand driven problem and automatically generating the algorithm that computes the information. In order to define partial data flow, information has to be provided that reduces the scope of data flow analysis. This information can either be static information, provided before execution, or dynamic information that is used to either further refine static data flow or to drive its computation. This work first focuses on debugging and testing tools and then explores other applications in software engineering.
