- Author
- Atif Memon, Ishan Banerjee and Adithya Nagarajan, University of Maryland
- Published
- Undated
- Kind
- Research paper describing a tool, with a small empirical study
- Original
- www.cs.umd.edu/~atif/pubs/MemonWCRE2003.pdf
- Subject
- Formal methods and proof
What it is
Atif Memon and two co-authors at the University of Maryland wrote the paper. They present the GUI Ripper tool, part of their GUITAR and DART testing systems. The aim is to cut the cost of building models for automated GUI testing.
What it says
- GUI ripping is a dynamic process run on a live GUI. It opens windows and extracts their widgets, properties and values.
- The extracted model has three parts: a GUI forest, event-flow graphs and an integration tree. Test cases are then generated from it.
- The ripper needs no source code. It reads information from the executable using low-level system calls.
- The extracted model needs a human tester to inspect and correct it. Windows the ripper misses can be added by hand.
- Ripping took under a minute for almost all tested applications. Java applications took longer than Windows ones.
- In a 1000-test run, ripping time was small beside oracle generation and replay. Faults were found in all three Java applications.
- The authors say capture/replay tools need 20-30 minutes for a test of 50 events. Ripping aims to avoid building the model by hand.
What evidence it gives
Table 2 gives rip time, windows ripped and missed, and manual effort for seven applications. Table 3 gives timings and fault counts for 1000 test cases on three Java applications. The faults found were 2, 47 and 12.
Interests and limits
The authors built the tool and ran its evaluation, and no other method is compared. Four of the Java applications were student-built. The tool covers only Windows and Java Swing GUIs. The paper does not say how many faults came from ripping itself.