Merge lp://staging/~toykeeper/ubuntu-power-tests/python-refactor-tk into lp://staging/ubuntu-power-tests

Fixed the power graph size / DPI.
Moved the time origin to the test begin tag.

Fixed incorrect argument order for a non-refactored function.
Added a wait_after test option, to keep measuring for a bit before rebooting.
Tried to calibrate display_on timing for both arale and krillin,
  but they seem to need different timings.
Added my krillin to the device list.

Refactor milestone: Uses TestResult object and keeps adt logs.
Moved a variety of things to TestResult class...
Keep entire result set in a dir, with a subdir per run.
Made partial() arguments explicit to avoid ordering issues.
Renamed run() to avoid name collisions.
  (still need to remove it entirely and merge the two sets of run()s)

Added logging to file(s) during test runs.
Fixed code which tried to directly access good_runs.
  (which is intended to be true/false, not a list of run objects)

Extended timeout so krillin tests will work. (setup on it is slow!)
Forced a visible axis at power=0 on graphs, in case it goes negative.
Made graph bounds smarter, based mostly on the samples where the
  device wasn't plugged in at the time.

Added adt log verification, so it requires adt to pass or it will ignore
  that test run and try again (up to the max tries).

Fixed the case where a graph has no "unplugged" data.
Made it easier to switch between devices in new-build test script.

Fixed graph layout.

Finally got adt-run --keep-screen-active to work.
display_on test is finally working.
Uses value from test file instead of test filename to control behavior.

Made grapher work again with legacy graphs.
(let it use autoscale if no USB tags were found)

power.py now collects crash dumps and screenshots and the date.
Adjusted test parameters to allow for more failures
  (since the tests now fail a lot more often).
Cleaned up some power test code.

Added graphs for bad runs too.
Might help reveal info for fixing the bugs which cause failures.

Added my flash script and minimal supporting files.
(includes recovery images)
Cleaned up the flash script a little, but it still has some options
  we don't actually use for power testing.

Added date sanity checking. Phone must be within 24 hours of host.
This catches the boots where the phone thinks it's January 1 2010.

Added on-device monitoring, at least the basics.
  (took a while to find a way to make it run in the background
   and not die as soon as USB gets unplugged)
Made power.py log full tracebacks instead of just the error string.
Now ensures the clock is correct before starting a test
  (though it might skew again on pre-test reboot, d'oh).
Moved adt verify check lower so it won't prevent collecting logs
  when adt fails.
Added an attempt to collect logs after failure, just in case it
  wasn't done before.
Added a reboot after failure to avoid cases where the error also
  breaks the next test run.
Removed the need to pass anything to usb_connect / usb_disconnect.
Fixed the case where a bad run has no log to graph.
Now collects syslog too.
Refactored a little to allow starting on-device monitoring scripts
  at the appropriate point in the code.
Replaced asserts with exceptions which provide more human-friendly errors.

Made test-new-builds accept a device name/list on the command line.
Also made it not wait after running tests, but instead poll again immediately.

Increased display_on test time to 30 seconds again
  (since the --keep-screen-active thing is working).

Er, stop trying to grab a file which never exists. :)
(I forgot I moved it)

Moved fix_clock() and date-pre to adt's prepare-device script.
Added pretty-printed stats to the 'stats' result file,
  since this is an important part of the results and should be visible.
Took all the detail out of graph filenames, since the parent dir has it.

Hah, make sure the DUT is plugged in before trying to collect logs post-fail.
Also log more detail about post-fail exceptions.
Grab dmesg after each test run.

Check the date later, because if it fails it prevents collecting other logs.

Oops, tell phablet-screenshot which device to use.
Also, show output from screenshot command for debugging.
Added -h / --help to usage docs.

Changed channel order so it'll test rc-proposed last.
This helps reduce how much time the power supply fan spends spinning
between tests, since rc-proposed behaves better and is less likely to
leave the device in a high-power state.

Updated the README quite a bit.
Removed outdated info and merged my notes into the main document.

Addressed remaining review comments.
Removed a variety of dead code.
Replaced bash with sh.
Pep8'd the graph script.
Renamed the good_runs variables to make the purpose more clear.
Removed good_count.
Switched run_cmd to use .communicate(); should be safer.
Ran a single test to make sure nothing huge broke.

Save a little extra network info before each test.