Evaluation Overview
About
Here are some preliminary evaluation results of droidbot.
The best matrix of performance of automatic testing bots is test coverage.
However, because we don’t have a mature test coverage tool in Android,
(which I mean, a test coverage tool that does not need app repackaging.)
we use a alternative approach:
Comparing the number of droidbox logs generated when using different test bots.
I compare droidbot with droidbox default mode and adb monkey mode. Thus, there are five modes compared:
- default. (the droidbox default, which just start the app and do nothing)
- monkey. (adb monkey)
- random. (droidbot sends events randomly)
- static. (droidbot sends events according to static analysis)
- dynamic. (droidbot sends events according to dynamic device states)
In my evaluation, for each mode:
- droidbox keeps collecting logs for a duration
- the testing bot (monkey or droidbot) sends event at intervals during this time
- the number of logs is recorded at each time.
Comparisons are made between the log counts, and more logs mean higher coverage.
The script DroidboxEvaluator.py is what I used to generate the result data.