Using Squeeziness to test component-based systems defined as Finite State Machines

Context: Testing is the main validation technique used to increase the reliability of software systems. The effectiveness of testing can be strongly reduced by Failed Error Propagation. This situation happens when the System Under Test executes a faulty statement, the state of the system is affected by this fault, but the expected output is observed. Squeeziness is an information theoretic measure designed to quantify the likelihood of Failed Error Propagation and previous work has shown that Squeeziness correlates strongly with Failed Error Propagation in white-box scenarios. Despite its usefulness, this measure, in its current formulation, cannot be used in a black-box scenario where we do not have access to the source code of the components. Objective: The main goal of this paper is to adapt Squeeziness to a black-box scenario and evaluate whether it can be used to estimate the likelihood that a component of a software system introduces Failed Error Propagation. Method: First, we defined our black-box scenario. Specifically, we considered the Failed Error Propagation that a component introduces when it receives its input from another component. We were interested in this since such fault masking makes it more difficult to find faults in the previous component when testing. Second, we defined our notion of Squeeziness in this framework. Finally, we carried out experiments in order to evaluate our measure. Results: Our experiments showed a strong correlation between the likelihood of Failed Error Propagation and Squeeziness. Conclusion: We can conclude that our new notion of Squeeziness can be used as a measure that estimates the probability of Failed Error Propagation being introduced by a component. As a result, it has the potential to be used as a measure of testability, allowing testers to assess how easy it is to test either the whole system or a single component. We considered a simple model (Finite State Machines) but the notions and results can be extended/adapted to deal with more complex state-based models, in particular, those containing data.