Change root bounds in RenderWidgetHostViewAuraTest.ParentMovementUpdatesScreenRect

Change root bounds in RenderWidgetHostViewAuraTest.ParentMovementUpdatesScreenRect
test to avoid initializing portrait mode static primary rotation and cause test
failures on memory waterfall.

It turns out that on normal run each test is run separately, but on
memory waterfall they are run in such a way  (in one process in bulk?)
that if previous test sets up some static values they are preserved
for all consecutive tests, creating hidden dependency between them.
In this case values of primary portrait and landscape angles that are in
RenderWidgetHostViewBase::GetOrientationTypeForDesktop are preserved
between tests and primary_portrait_angle was set up to unexpected value (0)
by test I've added, which broke RenderWidgetHostViewBaseTest.OrientationTypeForDesktop
which expected to set those up by itself and test it.

BUG=499914, 500011
Committed: https://crrev.com/a7f0ee1973ca033d3e5f4544cb3d3e6f5b582eec
Cr-Commit-Position: refs/heads/master@{#334792}

[mharanczyk, 2015-06-15 12:36:35]

mharanczyk@opera.com changed reviewers:
+ sadrul@chromium.org, sky@chromium.org

[mharanczyk, 2015-06-15 12:45:28]

Please take a look, that is just simple test update to fix hidden dependency
between tests on memory waterfalls.

On the side note: such hidden dependencies between test in not very intuitive
and cool, is this known and accepted/respected by you on memory waterfall trys?
Basically this passes just because most tests init landscape priority angle to 0
and 
RenderWidgetHostViewBaseTest.OrientationTypeForDesktop also follows this rule
(that is why it passes), so if you would reverse the logic of test you would get
it failing all the time. This also means there is hidden dependency on test
execution order.

[sadrul, 2015-06-17 03:00:53]

This ... seems unfortunate. I assume the issue is the statics in
RenderWidgetHostViewBase::GetOrientationTypeForDesktop() that get set for the
first test that calls it, and are re-used for subsequent tests? It would be nice
to fix that properly. (maybe introduce a ResetForTest or something like that;
that is a pattern used in a few places). Would you mind filing a bug for this?
(and maybe submit a patch too! :) )

For now, this LGTM

[oshima, 2015-06-17 05:03:48]

On 2015/06/17 03:00:53, sadrul wrote:
> This ... seems unfortunate. I assume the issue is the statics in
> RenderWidgetHostViewBase::GetOrientationTypeForDesktop() that get set for the
> first test that calls it, and are re-used for subsequent tests? It would be
nice
> to fix that properly. (maybe introduce a ResetForTest or something like that;
> that is a pattern used in a few places). Would you mind filing a bug for this?
> (and maybe submit a patch too! :) )
> 
> For now, this LGTM

Many unittests runs in one process (not sure about content_unittests though).
My guess is that the tests pass on normal bot thanks to sharding, and they
just happen to run on different shards.

Just my guess, but I've seen similar issue in the past.

[mharanczyk, 2015-06-17 08:18:47]

The CQ bit was checked by mharanczyk@opera.com

[commit-bot: I haz the power, 2015-06-17 08:19:24]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1186843002/1

[commit-bot: I haz the power, 2015-06-17 08:45:34]

Committed patchset #1 (id:1)

[commit-bot: I haz the power, 2015-06-17 08:46:52]

Patchset 1 (id:??) landed as
https://crrev.com/a7f0ee1973ca033d3e5f4544cb3d3e6f5b582eec
Cr-Commit-Position: refs/heads/master@{#334792}

[mharanczyk, 2015-06-17 09:07:32]

On 2015/06/17 05:03:48, oshima wrote:
> On 2015/06/17 03:00:53, sadrul wrote:
> > This ... seems unfortunate. I assume the issue is the statics in
> > RenderWidgetHostViewBase::GetOrientationTypeForDesktop() that get set for
the
> > first test that calls it, and are re-used for subsequent tests? It would be
> nice
> > to fix that properly. (maybe introduce a ResetForTest or something like
that;
> > that is a pattern used in a few places). Would you mind filing a bug for
this?
> > (and maybe submit a patch too! :) )
> > 
> > For now, this LGTM
> 
> Many unittests runs in one process (not sure about content_unittests though).
> My guess is that the tests pass on normal bot thanks to sharding, and they
> just happen to run on different shards.
> 
> Just my guess, but I've seen similar issue in the past.

Reported crbug.com/501219 and committed this change as there are some actions
happening in reported bugs that I am afraid of :). I will submit a patch for it
today a bit later hopefully. I am still not certain if it is enough to reset
those values for RenderWidgetHostViewBaseTest.OrientationTypeForDesktop test
only, as it seems that most (if not all) of RWHVA tests are setting that up
indirectly, but lets discuss it when patch is reported.