A Theory of Implementation-Dependent Low-Level Software

We present a theory describing implementation-dependent assumptions that a C program might make, such as the size and alignment of data. We define a static analysis to encode such assumptions in a constraint that describes language implementations (i.e., compilers and architectures) on which a program is memory-safe. The constraint produced by the analysis is a formula in a theory of first-order logic for which implementations are models. By defining an abstract machine that is parameterized by an implementation, we can prove the analysis sound. This foundation explains some common coding practices and the poorly understood assumptions they are implicitly making.