[Development] [Mac] tst_qquickwindow::testWindowVisibilityOrder() regression

[Alan Alpert]

On Thu, Mar 19, 2015 at 1:53 AM, Albert Astals Cid
<albert.astals at canonical.com> wrote:
>
> It seems tst_qquickwindow::testWindowVisibilityOrder() has regressed
> in the 5.4 branch (either because of qtdeclarative or qtbase changes)
> and is not letting changes to the branch integrate.
>
> I've run it locally here (linux) and get no failure, so it may be Mac
> specific since it always seem to fail in the Mac builder.
>
> Can anyone with a Mac please have a look?
>
> http://testresults.qt.io/ci/QtDeclarative_5.4_Integration/build_00605/macx-clang_developer-build_qtnamespace_OSX_10.7/log.txt.gz

For me the test consistently failed (both 5.4 and on dev) on one
machine and consistently passed (both 5.4 and dev) on another,
virtually identical machine (both running Fedora with KDE).

I finally tracked it down to the configuration of kwin, if the focus
policy is not set strictly to "Click to Focus" then you'll get
different behavior when a window is hidden and the test will fail.
This could explain how it "regressed", if the configuration of a WM
was changed on a CI machine or Apple altered their WM edge case
behavior in a patch.

The same behavior is found if you extend the QWindow test to hide
windows and check the resultant focus, but that test case does not
currently exist in the QWindow autotest. Since it seems controlled by
the window manager, I don't think it should be in the QQuickWindow
autotest either. https://codereview.qt-project.org/109631 should fix
the problem.

--
Alan Alpert