Automated Model-based Interface Test Generation for Mobile Applications

Mobile applications have become an integral part of modern society. The majority of the population uses a smartphone to some degree and Android has grown to be the most used operating system for mobile devices. As the uses for, and revenue gained by mobile applications grow, so too does the need for developers to validate the quality and integrity of their applications. Sudden loss of data, or an inconsistent user experience are examples of the impact defects can have, and given the wide range of applications available, they often result in a loss of users and therefore revenue. Software testing aims to highlight faults in a software system before release. While many aspects of mobile applications require robust testing, some of the most important defects impact or originate from a mishandled user interaction with the user interface (UI). As these defects are easily observed by the user, their detection is critical for user retention and satisfaction. Integration testing using test inputs on the UI is one method of detecting such defects and the fast pace of the mobile industry coupled with time consuming labour intensive testing methods, has made automated test input generation a key topic of research over the last decade. Automated test input generation has shown promising results with a variety of approaches and levels of success. Random input generation is commonly used and the easiest to maintain, but ultimately inefficient. Systematic and search-based approaches produce effective tests but require an excessive runtime. A model-based approach has the additional overhead of modelling the application under test (AUT) but results in a faster test generation. In this thesis, I focus on modelling Android applications and automatically generating user input tests from these models. I first provide a proof of concept traversal of a preliminary model to show the potential benefits of a more comprehensive model of an Android application in providing more effective and efficient test inputs. I then explore both static and dynamic approaches to modelling Android applications and their limitations. Finally, I present Precise AnDRoid Automated Input Generation (PADRAIG), a new model-based framework that relies on a detailed model of the AUT to generate tests that achieve higher coverage, with a lower test generation runtime.