Concurrent Self-Explaining Computation

Self-explaining computation is an approach to program execution in which every value comes with an explanation of how it was computed. The explanation can be used to reverse the computation and to slice the original program relative to any part of the output of interest. As a result, self-explaining computation is a suitable foundation for o ine dynamic program analyses such as taint analysis and algorithmic debugging. Building on prior work in the functional setting, we develop the foundations of concurrent self-explanation for a higher-order process calculus. We represent explanations as traces, which record inter-process synchronisation as well as intra-process functional evaluation. We show that any part of the state of a concurrent computation can be accounted for by a unique minimal slice of the initial con guration. This result is established using a Galois connection describing forward and backward executions of a concurrent program that share a synchronisation structure. As our main practical result, we provide a reverse operational semantics for processes which computes the lower adjoint of the Galois connection.