Change-Id: I4f83fdb59a60bb1f1720047f0b442d70192be5da
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: I75eda7e766182025b54f8df851ff1d5a72eefb17
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

The unconditional grab from 498946d6
did not have a corresponding ungrab. That only worked well as long as
there were only QWaylandSurfaceItems in the scene. Each of them just
stole the grab for itself at the first TouchBegin. Regular QML items
(some people have UI in the compositor process...) do not participate
in that mutual stealing so they'd sometimes not get any touch events
because the events still went to the last QWaylandSurfaceItem to
receive a TouchBegin.

Note: The grab would also go away if  touchEvent() didn't accept()
the event. Without the accept(),
QQuickWindowPrivate::translateTouchToMouse()
would be called at TouchEnd and do the ungrab.

Change-Id: Ia7d1b15fff366542d382eb97494b1e43dedb65a4
Reviewed-by: Andrew den Exter <andrew.den.exter@qinetic.com.au>
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

When calling ::exit() the socket notifier in the events thread may still
fire before the process actually exits, using objects that are being
destroyed and resulting in a segfault.
Stop the events thread before calling ::exit().

Change-Id: I187762da2a7efa83db1e62b0e28dfab89f478c7d
Reviewed-by: Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>

When hiding a window we attach a NULL buffer, which means that any
frame callback that is still waiting will not be sent. In the SHM
case we just destroy it, but we cannot do that in the EGL case, since
we don't have a hold of it. So destroy the EGLSurface, which will
in turn destroy the callback.

Task-number: QTBUG-46921
Change-Id: I20e0d1ca7b53fb88096101d46979996dd944d429
Reviewed-by: Gunnar Sletta <gunnar@sletta.org>

Instead throw a warning and keep using the old one.

Change-Id: I1533b3930d4d8c884db922b0d6dd567ef7e4f67c
Reviewed-by: Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>

If a transient parent of a surface is destroyed before the surface
itself, reset all references to that parent surface.

Change-Id: I5b748099d8a5502632d8846c5de5381b5f28b665
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

Change-Id: I51da06261acfb193a59db34bc2dd25e3452fce8d
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

LGPLv3 refers to it but does not include it in its body.

Change-Id: Ib056b47dde3341ef9a52ffff13eed18cf3504738
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>

It is a bit dangerous to call wl_display_dispatch() in the event thread,
since it may race with the dispatch called e.g. in QWaylandDisplay's
blockingReadEvents() and lead to a dead lock. Instead, use wl_display_prepare_read()
and wl_display_read_events() in the event thread, which doesn't block, and only
dispatch in QWaylandDisplay.
As a result we don't need the additional wayland queue anymore, so remove it.

Change-Id: I9fbbe5d2f38d06773beb7847df1a0212cca92c37
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>

QToolTip has the same problems as QMenu, that is you can create one
without a parent and with a global position. Use the same trick and
use the window that last received an input event as the parent.

Change-Id: I093c8da0d54110903f35670b01dea6fa96abecf4
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>

With a queue of just two buffers BufferAttacher can end up holding
references to both after the renderer is stopped starving the client
application and causing it to block in the glSwapBuffers until the
renderer restarts or the surface is destroyed.  Release the current
buffer to the client when the renderer is stopped so it can continue.

Change-Id: Ica0e13ef78f7e6058e273c26b517a88d07f958c7
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

Change-Id: I7000630632373cf7d81c6c0b307f1ab41c2e133f

Change-Id: Iaf9eaa171c141d79c52441de8a697c87e2e42407

Change-Id: Ia8a927c5bcfb45bc7a8c16d1c8ff75487fe71cdd
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

This occurs for example when a Qt Quick2 application calls
showFullScreen() while a rendering is already in progress.
The GUI thread was calling destroyEglSurfaces() from setGeometry() and
thus destroying the buffers, while the render thread was trying to use
the buffers from swapBuffers().

There was a division by zero in swapBuffers() with the modulo m_count
operation, because EGL surfaces were destroyed and m_count was 0.

This fix is quite fragile, proper locking on buffers would be better.

Change-Id: I41fcb2fcebec84c0dfc82ce56c6e323d3722b30a
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: I0c6b59eb9d59d6db4a71811562cdf740bfb0a06e
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

They were not destroyed until the client quit, so this could be
significant after many resize operations.

This was also leaked server-side as a result.

Change-Id: If9b84a06e3f9c79af92a949bc67e81ef9e00a230
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: I8944c176f12932eb51e867bc39820b2553a5d3ba
Reviewed-by: Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: I28200698706168308e450b2cd3cfa99df517f9b7
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: Ie02e6667e5867b729cf5a519705ded4f79b9f4b1
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: I7ad508cb07a73d6699e9d2742ea029409f498a87
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

Make it a scoped pointer.

Change-Id: Iaeef87cad2022496ebe3a7c746b02584fac9a485
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Also fix a comment alongside.

Change-Id: I0091a89ff8e7ed46e769ea47a0225bc87db2c1f5
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

Change-Id: I5d6eedeb3dfb80ca9df7ca27e29ad6604dbfe683
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

We must offset the wheel events' position, as we do with the other
types of mouse events.

Change-Id: If85e93ffe95304c7dee4c2a3ff195a73243a8182
Reviewed-by: Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: I5a161119b20301a405bdde9bbea6bfbdcc4b9fa0
Reviewed-by: Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

QWaylandDisplay::flushRequests() does not only flush them, but also
dispatches the pending events. If we have two wl_data_device.selection
events in the queue we must not dispatch the second one while we are
dealing with the first one because that will replace the data offer
and delete us while we're in QWaylandMimeData::retrieveData_sys().

Change-Id: Ib58ca571867faa2633daa9ec94fe7094df02e9fd
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

By calling destroy() on the active wl_data_source before replacing
it with a new one and calling set_selection() we trigger a spurious
selection(null) event before the one with the offer for the source
we are going to set.

Change-Id: I6c2f2fd029fa523312b9892c6a5050805dfa83b0
Reviewed-by: Pier Luigi Fiorini <pierluigi.fiorini@gmail.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Change-Id: If5ae8749ccbddfc878ae4e0adaa40f072527d425
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>

Task-number: QTBUG-44503
Change-Id: I6932df57bb4560b6dceb72a5cb7c536cd090e92a
Reviewed-by: Martin Gräßlin <mgraesslin@kde.org>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

Conflicts:
	.qmake.conf

Change-Id: Iafa184859c6610a39f4251e6ca59d0863eed3ae0

One should always use Q_SLOTS/Q_SIGNALS in then public headers

Change-Id: Ibbba0d3ff402bd320049bcd58ced07ad1342ef3b
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

::modifiers() might be invoked before being initialised. E.g. when
a pointer button event is emitted before the keymap is sent.

Change-Id: Ib95f9866d808b2a32ddfaa0862cdb3226fab1938
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

eglBindAPI's docs says "defines the current rendering API for EGL in the
thread it is called from". We were instead just calling it in the thread
the context was created in, not in the thread used for rendering.

Change-Id: Iba8ffe75a6f4f8b9d1bba59c0e7cce34499e9c48
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

This allows calling init() on the wayland object again after destroy()

Change-Id: I7856cc8ec676eee7575e15a5d1f81cf1b89537c8
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

Change-Id: I61d7aa4b6a37922eb3beefd983154ee267cae0ef
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

Change-Id: I0a2b56043235666bc57b46ae4ac78ca5e24e736d
Reviewed-by: Giulio Camuffo <giulio.camuffo@jollamobile.com>

If the compositor sent a wl_keyboard.leave and a new wl_seat.capabilities
without the keyboard bit we would end up deleting the Keyboard object
with a roundtrip in flight, ending up using freed memory later.
Destroy the callback when destroying the keyboard, and notify the
window system the active window is active no more.

Change-Id: Ie415ca62b46e0b8ef0fe4d09c8e7a8130928df90
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>

By passing "" instead of hardcoded defaults  we let xkbcommon decide what
the default parameters for the keymap should be. The user can change that
with environment variables, as described here:
http://xkbcommon.org/doc/current/structxkb__rule__names.html



Change-Id: I8fb037201aeababaa0aa4755cae339ae3d9800b9
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>