GUI ripping: reverse engineering of graphical user interfaces for testing

TLDR

Graphical user interfaces are critical to software correctness, yet manual testing is slow and building models for automated test case generation is difficult. This work proposes a method to reverse engineer a GUI model—comprising a forest, event‑flow graphs, and an integration tree—directly from the executable interface. The authors introduce GUI Ripping, a dynamic traversal that opens all windows, extracts widgets, properties, and values, verifies the data, and automatically generates test cases, implemented in a tool suite for Java and Windows GUIs. Case studies demonstrate that the approach requires minimal human intervention, excels at regression testing of frequently modified software, and has been successfully deployed in large experiments, with the tool available for download.

Abstract

Graphical user interfaces (GUIs) are important parts oftoday's software and their correct execution is required toensure the correctness of the overall software. A populartechnique to detect defects in GUIs is to test them by executingtest cases and checking the execution results. Testcases may either be created manually or generated automaticallyfrom a model of the GUI. While manual testingis unacceptably slow for many applications, our experiencewith testing has shown that creating a model that canbe used for automated test case generation is difficult.We describe a new approach to reverse engineer a modelrepresented as structures called a forest, event-flowgraphs and an integration tree directly from the executableGUI. We describe GUI Ripping, a dynamic process inwhich the software's is automatically traversed byopening all its windows and extracting all their widgets(GUI objects), properties, and values. The extracted informationis then verified by the test designer and used to automaticallygenerate test cases. We present algorithms for theripping process and describe their implementation in a toolsuite that operates on Java and Microsoft Windows' GUIs.We present results of case studies which show that ourapproach requires very little human intervention and is especiallyuseful for regression testing of software that is modifiedfrequently. We have successfully used the GUI Ripperin several large experiments and have made it availableas a downloadable tool.