Personally I think we should reuse the same tools as we did for winmo. We already have a sutagent for Android which would make these changes very little if not nothing. That was part of the reason for writing the sutagent and devicemanager so we can avoid rewriting code when we hit a new platform. let me know if there is an android specific reason why we should use adb vs sutagent.
Created attachment 435818 [details] [diff] [review] patch one drawback of running the tests over adb is that adb shell doesn't return the exit code of the test. So make check will continue to run after the first failure.
This is probably helpful for developers locally; I don't think we should do our testing infrastructure with this though, since it'll require each deivce to be tethered to a host. (However, unlike with winmo, adb can deal with multiple devices connected at the same time.)
I would agree that this is more useful for developers than QA/RelEng who would probably want to stay consistent with the telnet/test agent connection. I should note that adb supports "tethering" over tcp though.
Would having an android agent available now help? If we use the same tools for development as we do for the unittests, that will help make them more stable as time goes on.
So we fixed bug 811411, which I think fixes all the cases you cared about here. Maybe the usability needs some work, I haven't actually tried them, but dmose has used this successfully. I also just filed bug 858622 on making it possible to run the jit-tests on mobile, which are the other major component of "make check".