Static Analysis of an Actor-based Process Calculus by Abstract Interpretation

Overview & Synopsis

The Actor model, introduced by HEWITT and AGHA in the late 80s, describes a concurrent communicating system as a set of autonomous agents, with non uniform interfaces and communicating by the use of labeled messages. The CAP process calculus, proposed by COLAÇO, is based on this model and allows to describe non trivial realistic systems, without the need of complex encodings. CAP is a higher-order calculus: messages can carry actor behaviors. Multiple works address the analysis of CAP properties, mainly by the use of inference-based type systems using behavioral types and sub-typing. Otherwise, more recent works, by VENET and later FERET, propose the use of abstract interpretation to analyze process calculi. These approaches allow to compute non-uniform properties. For example, they are able to differentiate recursive instances of the same thread. This thesis is at the crossroad of these two approaches, applying abstract interpretation to the analysis of CAP. Following the framework of FERET, CAP is firstly expressed in a non standard form, easing its analysis. The set of reachable states is then over-approximated via a sound by construction representation within existing abstract domains. New general abstract domains are then introduced in order to improve the accuracy of existing analyses or to represent local properties. CAP specific properties such as the linearity of terms or the absence of orphan messages, are then considered in this framework. Specific abstract domains are defined and used to check these properties. The proposed framework is able to relax any existing restriction of previous analyses such as constraints on the shape of terms or limitation in the use of CAP behavior passing. The whole analyses have been implemented in a prototype.