A factorial experimental evaluation of automated test input generation – Java platform testing in embedded devices

In English Background. When delivering an embedded product, such as a mobile phone, third party products, like games, are often bundled with it in the form of Java MIDlets. To verify the compatibility be- tween the runtime platform and the MIDlet is a labour-intensive task, if input data should be manually generated for thousands of MIDlets. Aim. In order to make the verification more efficient, we investigate four different automated input generation methods which do not require extensive modeling; random, feedback based, with and without a constant startup sequence. Method. We evaluate the methods in a factorial design experiment with manual input generation as a reference. One original experiment is run, and a partial replication. Result. The results show that the startup sequence gives good code coverage values for the selected MIDlets. The feedback method gives somewhat better code coverage than the random method, but requires real-time code coverage measurements, which decreases the run speed of the tests. Conclusion The random method with startup sequence is the best trade-off in the current setting.