Advances in Theory and Applications of Token Causality in Trace Analysis

Token causality refers to the causal relationship between system events on an individual system execution, as opposed to general causality, where relationships between types of events are studied. When a system fails at run-time, represented as the violations of certain system safety properties on an observed system execution trace, the investigation of token causality is helpful in understanding the root cause why such system safety properties are violated. In this report, three different techniques in defining token causality are surveyed. Though all based on counterfactual reasoning, the detailed formulations of the causality definitions, and hence the algorithms for deciding/approximating causalities, vary depending on the system definitions and on the intended use of the causalities. For each of the three surveyed approaches, the causality definitions are summarized and illustrated with examples, based on which the strength and limitations of each approach are discussed. The survey is concluded with comparisons across the three approaches and related work, as well as the challenges and potential extensions in the study of token causality.