Abstract

The goal of program analysis is to determine automatically properties of the run-time behavior of a program. Tools of software development, such as compilers, program-verification systems, and program-comprehension systems, are in large part based on program analyses. Most semantics-based program analyses model the runtime behavior of a program as a trace of execution states and compute a property of these states. Typically, this property is drawn from a predetermined language of semantic information, such as aliasing descriptions or types of values. The standard methodology of program analysis is to construct the property as a fixed point, a single execution step at a time. We explain that these ubiquitous methodological choices---the a priori choice of the describable program properties and the use of a fixed-point computation---have some fundamental limitations and can result in poor precision. In this dissertation, we present a different approach to semantics-based program analysis...

Citations