Testing without Scripts: An Approach to Smart GUI Exploration

Software testing through graphical user interfaces (GUIs) remains a critical challenge in quality assurance, particularly as software systems grow in complexity
and evolve rapidly. Traditional script-based testing approaches, which rely on
predefined test cases, are widely used in industry but often struggle with high
maintenance costs, limited adaptability to GUI changes, and restricted coverage
of unforeseen user behaviours.
Scriptless GUI testing has emerged as a powerful alternative, dynamically
exploring applications without the need for predefined test scripts. This ap proach introduces randomness, allowing the execution of unexpected sequences
of actions and the discovery of faults that scripted tests often miss. This thesis
investigates the effectiveness of scriptless testing, examining how its exploratory
nature complements existing testing practices and reduces manual efforts.
To establish a strong foundation, this research first analyses thirty years
of GUI testing literature, tracing the evolution of the field. The findings reveal a
growing transition from manual and script-based testing to scriptless approaches.
With this motivation, this thesis investigates the effectiveness of scriptless GUI
testing through the lens of testar, an open-source tool that serves as this study’s
primary research vehicle. A generalisation study of the tool allowed the introduction of an architectural analogy for scriptless testing deployment, built upon
the existing industrial case studies with testar.
This thesis examines the role of state models in guiding scriptless testing by
evaluating how different levels of state abstraction can influence model inference
and test coverage. The results provide guidelines for balancing model complexity
with exploration effectiveness. Additionally, this thesis explores the impact of
reinforcement learning-driven reward mechanisms in balancing pure randomness
with targeted exploration to enhance test effectiveness.
The thesis further evaluates the industrial applicability of scriptless testing
through empirical studies in collaboration with companies participating in the
European IVVES project (Industrial-grade Verification and Validation of Evolving
Systems). The research aims to bridge the gap between traditional testing ade quacy criteria and quality-oriented metrics by investigating whether code smell
could serve as a complementary adequacy criterion when evaluating scriptless
testing effectiveness. Findings reveal that while increasing traditional code cov erage leads to broader exploration, it does not necessarily translate into covering
code with deeper structural or maintainability issues.
This research extends testar’s scriptless testing into the mobile domain, adapt ing it for mobile platforms and introducing MINTestar, a specialised Android test ing tool. Developed as part of the industry collaboration within the IVVES project,
these efforts explore the feasibility of scriptless testing in real-world mobile envi ronments, and the integration of mobile-specific oracles and probabilistic explo ration strategies. The results highlight the adaptability of scriptless approaches
across platforms and their potential for adoption in industrial mobile testing work flows.
This thesis integrates insights from literature reviews, empirical evaluations,
and industrial case studies to provide both theoretical and practical contributions to the field of scriptless GUI testing. By improving state models, leveraging
reward-based exploration, refining test adequacy metrics, and extending automation to mobile platforms, this thesis lays the foundation for future advancements
in smart testing, domain-specific oracles, and distributed testing architectures