State Model Inference Through the GUI Using Run-Time Test Generation

Abstract

Software testing is an important part of engineering trustworthy information systems. End-to-end testing through Graphical User Interface (GUI) can be done manually, but it is a very time consuming and costly process. There are tools to capture or manually define scripts for automating regression testing through a GUI, but the main challenge is the high maintenance cost of the scripts when the GUI changes. In addition, GUIs tend to have a large state space, so creating scripts to cover all the possible paths and defining test oracles to check all the elements of all the states would be an enormous effort. This paper presents an approach to automatically explore a GUI while inferring state models that are used for action selection in run-time GUI test generation, implemented as an extension to the open source TESTAR tool. As an initial validation, we experiment on the impact of using various state abstraction mechanisms on the model inference and the performance of the implemented action selection algorithm based on the inferred model. Later, we analyse the challenges and provide future research directions on model inference and scriptless GUI testing.