On the Generation of Initial Contexts for Effective Deadlock Detection

It has been recently proposed that testing based on symbolic execution can be used in conjunction with static deadlock analysis to define a deadlock detection framework that: (i) can show deadlock presence, in that case a concrete test-case and trace are obtained, and (ii) can also prove deadlock freedom. Such symbolic execution starts from an initial distributed context, i.e., a set of locations and their initial tasks. Considering all possibilities results in a combinatorial explosion on the different distributed contexts that must be considered. This paper proposes a technique to effectively generate initial contexts that can lead to deadlock, using the possible conflicting task interactions identified by static analysis, discarding other distributed contexts that cannot lead to deadlock. The proposed technique has been integrated in the abovementioned deadlock detection framework hence enabling it to analyze systems without the need of any user supplied initial context.