Many autopilot tests will fail if the orientation of the app is landscape on startup

I don’t think this issue should be worked around in the browser app. It’s an issue in the shell, and it should be fixed there. It has existed since the very beginning, and it affects all applications that react to orientation changes and expose some sort of functionality through the bottom edge.

Note that webbrowser-app uses the standard MainView component from the UITK, which handles orientation changes under the hood, so I don’t think we can hack it to force portrait mode. Besides, even if we could, it wouldn’t make sense to break the UX (starting in portrait mode while in landscape orientation is unexpected) to make the tests more reliable.

What about the orientation sensor’s sensitivity? It looks like this is a relatively recent regression, couldn’t it be fixed/improved there?

Finally, have we considered a setup where test devices are standing upright, to prevent this from happening at all?

The short term workaround would definitely be to elevate the devices during testing. Note that even a small gradient (like resting an object underneath the top of the device) prevents it launching in landscape mode, so the devices do not need to be fully standing. In the long term, probably there is a bug to fix somewhere in the shell/sensors, but it's worth noting that when I launch the web browser from the shell it never launches in landscape mode, even when lying flat - so the test is doing something a bit different (not sure what)

Note that this bug is mainly for tracking the issue and not an indictment on webbrowser-app as being at fault. It will hopefully be moved to a more appropriate place in due course.

I can reproduce this like this:

/usr/bin/webbrowser-app -testability --desktop_file_hint=/usr/share/applications/webbrowser-app.desktop http://people.canonical.com/~brendan-donegan/browse.html

This is how autopilot launches the web browser from the test.

If I remove -testability:

/usr/bin/webbrowser-app --desktop_file_hint=/usr/share/applications/webbrowser-app.desktop http://people.canonical.com/~brendan-donegan/browse.html

Then it stays in portrait mode.

I can’t reproduce this here on N7. My device is lying on a (mostly) flat surface, and the -testability switch doesn’t seem to make any difference (I’m having a hard time imagining how it possibly could).

I haven't tried on flo yet - I'm currently focused on mako as I don't have time to try things on both. Even on mako this doesn't happen every time, with the -testability switch. I know it's hard to imagine - it's baffling me as well, but this does seem to be the case.

A couple of findings on this over the weekend while debugging in the qtubuntu-sensors code.

1.) Adding to many qDebug statements in the code causes the unit tests to fail. This suggests the code is timing sensitive.
2.) Maybe related - this section of code is found in the core_orientation_sensor.cpp code:

int nearestRotation = 0;

void core::OrientationSensor::onAccelerometerReadingChanged(QSharedPointer<QAccelerometerReading> reading)
{

i.e. nearestRotation is defined outside any function. I'm not sure if this is intended/necessary, but it certainly looks wrong.

This bug was fixed in the package qtubuntu-sensors - 0.6+14.10.20140707-0ubuntu1

---------------
qtubuntu-sensors (0.6+14.10.20140707-0ubuntu1) utopic; urgency=low

  [ Brendan Donegan ]
  * Initialize m_last{X,Y,Z} to avoid faulty values (LP: #1337284)
 -- Ubuntu daily release <email address hidden> Mon, 07 Jul 2014 21:49:14 +0000