Identification of atomic code blocks for model checking using deductive verification based abstraction

Sammanfattning

Model checking is a formal verification technique for verifying temporal properties in state-transition models. The main problem with using model checking is the state explosion problem, where the number of states in the model can grow exponentially, making verification infeasible. Previous work has tried to mitigate the state explosion problem by representing code blocks as Hoare-logic contracts in an abstract state-transition model using the temporal logic TLA. This is achieved by treating the block as atomic. In order to ensure that the abstract state-transition model is faithful with respect to the temporal properties you want to verify, only some code blocks can be considered atomic. This thesis aims to answer how atomic code blocks can be identified atomically and evaluate their potential for reducing state space during model checking. We give a theoretical foundation of what it means for a code block to be considered as atomic in TLA. Moreover, we introduce a property that characterizes these atomic code blocks and presents an algorithm to identify them for sequential programs written in a subset of C. Experimental results demonstrate that the identification of atomic code blocks using our algorithm can be used to significantly reduce the state space during model checking, with an average reduction factor of 62. The potential of this verification approach is promising, however, further case studies are necessary to better understand the extent of this reduction across different program types and properties.