In English

Automatic Regression Testing using Visual GUI Tools

Johan Sjöblom ; Caroline Strandberg
Göteborg : Chalmers tekniska högskola, 2015. 84 s.
[Examensarbete på avancerad nivå]

Present-day software companies are facing growing demands of quality of their products but shorter delivery times, which a ects the whole development process. The testing process, taking place at all levels of development, veri es that the product meets the requirements and expected quality. This process is often carried out manually, which is costly, time consuming and error-prone. Automated testing is proposed as a solution to this as it can raise the test frequency, giving faster feedback to the developers, and may run without human interaction. The advantages of automated testing is widely recognised, however the support for higher level { i.e. the graphical user interface (GUI) { system testing is limited. Automated testing becomes even more important when it comes to regression testing, since this will assure that the system ful ls the requirements after updates have been made. This thesis aims to investigate the deployability of automating regression tests for a Command, Control and Communication system at Saab AB, using visual GUI testing. Visual GUI testing uses image recognition to nd visible objects on the screen and has several advantages over previous techniques for GUI testing. It was invented in the early nineties, however due to the computational heaviness of the image recognition algorithms it was not until recently the hardware and software became powerful enough to make the technique usable in an industrial context. The investigation performed consists of an evaluation of two existing visual GUI testing tools { Sikuli and JAutomate { which is done with respect to empirically measured performance and perceived usability based on the tools' speci cations. Due to the complexity of the tested system, animated objects and fault tolerance also becomes important in the evaluation. The evaluation concludes with a comparison between the tools, as well as a recommendation to Saab AB of which is most suitable for the tested system. Sikuli was chosen due to the expressivity and powerfulness of the scripting language. The text recognition algorithms worked poorly, which was improved by implementing a training function for the text recognition in Sikuli. This improvement was then evaluated by comparing the results from running the original text recognition and the trained text recognition. The research on animated interfaces and visual GUI testing is limited, making our contribution in this aspect valuable. We can conclude that both of the tools can handle this; however the fault tolerance of the scripts becomes crucial. A topic for future work would be to investigate the long-term costs for development and maintenance. However, so far visual GUI testing have been found to be a very promising technique that work well and have a lot of potential.



Publikationen registrerades 2015-04-14.

CPL ID: 215187

Detta är en tjänst från Chalmers bibliotek