A Factor Graph Model for Software Bug Finding
Ted Kremenek, Andrew Y. Ng, Dawson Engler
Automatic tools for finding software errors require knowledge of the rules a program must obey, or "specifications," before they can identify bugs. We present a method that combines factor graphs and static program analysis to automatically infer specifications directly from programs. We illustrate the approach on inferring functions in C programs that allocate and release resources, and evaluate the approach on three codebases: SDL, OpenSSH, and the OS kernel for Mac OS X (XNU). The inferred specifications are highly accurate and with them we have discovered numerous bugs.