We present a technique for detecting semantically infeasible paths in programs using abstract interpretation. Our technique uses a sequence of path-insensitive forward and backward runs of an abstract interpreter to infer paths in the control flow graph that cannot be exercised in concrete executions of the program.

We then present a syntactic language refinement (SLR) technique that automatically excludes semantically infeasible paths from a program during static analysis. SLR allows us to iteratively prove more properties. Specifically, our technique simulates the effect of a path-sensitive analysis by performing syntactic language refinement over an underlying path- insensitive static analyzer. Finally, we present experimental results to quantify the impact of our technique on an abstract interpreter for C programs.