Deriving with Derivatives: Efficient Implementations of Abstract Abstract Machines
Ben Quiring
Abstract
Abstract interpretation often involves computing an "abstract state space" that captures all behaviors of a program. Workset algorithms are almost always at the heart of this computation, and in simple cases these resolve to a simple graph traversal. This work inspects more complicated cases (coarser abstractions), and takes advantage of this to derive more efficient workset algorithms by looking at lattice-theoretic derivatives. We apply this to an example abstract machine to show how performance can be improved.
This talk is organized by Sankha Narayan Guria