When Xwayland confines, the surface dimensions will include the server
side window manager decorations. We don't want the decorations to be
included in the constraint region so intersect the calculated input
region with the parts of the buffer rect that is not part of the window
frame.

https://bugzilla.gnome.org/show_bug.cgi?id=771859

Put the conditions for enabling a pointer constraint in a helper
function, and use that in both maybe_enable() and maybe_remove(). The
constraint region checking is still only done in maybe_enable()
however.

This changes the conditions for maybe disabling the constraint on focus
change and other trigger points, namely it makes constraints by Xwayland
not disable when they shouldn't due to the constraining window being an
override-redirect window.

https://bugzilla.gnome.org/show_bug.cgi?id=771859

When the grab is cancelled, for example because of an Alt-tab, VT
switch etc, disable or remove (depending on the constraint type) the
constraint. This avoids a re-entry issue when the focus is returned and
the focus listener tries to re-enable a disabled constraint.

https://bugzilla.gnome.org/show_bug.cgi?id=771859

Dismiss the popup when the grab is cancelled, so that if the grab is
ended for whatever reason (such as VT switching or the last pointer
being disconnected), it doesn't try to end the grab when it isn't
active.

This fixes a crash when VT switching back and forth while a popup grab
is active.

https://bugzilla.gnome.org/show_bug.cgi?id=771858

Previously a grab could suddenly end without the grabber knowing
anything about it. Some grabs assume they won't suddenly end without
notice, and can use then new 'cancel' vfunc to be notified.

Currently a grab is cancelled when a new one is started (i.e. in
meta_wayland_pointer_grab_start()), when a non-popup compositor wide
event route is initiated, and when the seat looses the pointer
capability.

https://bugzilla.gnome.org/show_bug.cgi?id=771858

Instead of using meta_wayland_pointer_end_grab() which focuses the new
grab, add a new helper mean to be used to reset the grab state without
changing the pointer focus. When using this function, the call site is
supposed to explicitly manage focus.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Some very long lines that stood out were shortened, and an old naming
convention from weston was removed.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Previously the focus was reset implicitly by a memset() on the whole
MetaWaylandPointer struct. When MetaWaylandPointer was turned into a
GObject, this was not possible any more, and the focus was not updated
properly.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Make the caller of focus setting check whether there is a pointer to
update the focus state of. It makes it more obvious what to expect, as
the call would be a no-op in when no pointer is present.

Grabbing is still allowed without the presence of a pointer because it
is used by popups even on touch-only systems.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Unify next event serial retrieval using a helper function.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Make the caller of focus setting and grab starting check whether there
is a keyboard to update the focus state or start grabbing. It makes it
more obvious what to expect, as the call would be a no-op in when no
keyboard is present.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Initialize on init(), unlink and reinitialize the list headers on
disable() so that any delayed resource destruction doesn't affect future
state.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Initialize on init() and just end grab on disable().

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Initialize and cleanup properly in a _init()/_destroy() function pair.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Some very long lines that stood out were shortened, and an old naming
convention from weston was removed.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

https://bugzilla.gnome.org/show_bug.cgi?id=771646

The variable name 'l' usually refers to a GList iterator, but here it's
just a short hand for a specific list. Stop using this shorthand, since
it just makes it harder to read what list is used.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

Use the same method everywhere for getting the next event serial number.

https://bugzilla.gnome.org/show_bug.cgi?id=771646

A xdg_popup, when active, always has a parent surface. However, a popup
created may immediately become invalid, for example when it is not
granted a grab, in which case it won't be assigned a parent since it
will never be mapped.

This case needs to be handled elsewhere, as one cannot assume a
MetaWaylandXdgPoup that is processed (via wl_surface commit handling
etc) will have a parent_surface.

https://bugzilla.gnome.org/show_bug.cgi?id=771495

If a X11 client would initiate a Xdnd session after it had lost pointer
focus (for example when the Xdnd event starting the drag happens after
the implicit pointer grab is already broken due to the button being
released), just end the drag operation instead of dereferencing the
non-existing focus surface.

Also avoid using a native Wayland surface as a drag origin, as that can
never happen, but allow any arbitrary Xwayland client, since there is
no way to find out the actual drag origin.

https://bugzilla.gnome.org/show_bug.cgi?id=770940

We cannot check 'has-target' or 'has-seat' when verifying a
wl_data_offer.finish request is valid or not, since the source may have
effected 'has-target' or whether theh source has a seat or not when the
finish request was already on the wire.

Instead of checking against the source state, keep track whether the
required operations has been done on the offer in question (i.e.
whether an action has been sent, or a mime type been accepted).

This fixes incorrectly raised error when dragging from gtk+'s testdnd
via Xwayland onto gtk+'s testdnd using Wayland.

https://bugzilla.gnome.org/show_bug.cgi?id=770937

Now with the existance of offscreen view framebuffers the buffer age
damage regions are only valid if the view in question doesn't doesn't
have an intermediate offscreen. So, for views that doesn't have buffer
age, return the dirty pixel (0,0).

https://bugzilla.gnome.org/show_bug.cgi?id=770672

Otherwise if buffer age is not supported the dirty pixel would be
shifted potentially in unpaintable position.

https://bugzilla.gnome.org/show_bug.cgi?id=770672

This wont fix anything, it just one small step away from using
deprecated cogl functions.

https://bugzilla.gnome.org/show_bug.cgi?id=770672

Use G_DECLARE_FINAL_TYPE instead of the set of macros.

https://bugzilla.gnome.org/show_bug.cgi?id=770672

Since wl_surface.set_buffer_transform() is not supported, until it is
added, pretend outputs are never transformed, so that clients are less
likely to attach pre-transformed buffers.

https://bugzilla.gnome.org/show_bug.cgi?id=770672

When blitting an offscreen onto an onscreen, the whole offscreen should
always be drawn on the whole onscreen. Thus, don't try to convert
between coordinate spaces, just draw the whole offscreen on the whole
onscreen.

https://bugzilla.gnome.org/show_bug.cgi?id=770672

Also maybe remove a constraint when the pointer focus changes. This is
needed because when Xwayland has a constraint focus may change, the
constraint object will not receive a 'appears-focused' event on its
window since it never changed.

This happens for example when an override-redirect window (which never
appears focused) holds the constraint, and alt-tab happens. In this case
focus changes, but from the constraint's point of view, none of the
windows it knows about changed its focus appearance.

https://bugzilla.gnome.org/show_bug.cgi?id=771345

We should never enable a pointer constraint for an unmanaging window.

https://bugzilla.gnome.org/show_bug.cgi?id=771345

Instead of initializing the default grab when the device class is
enabled, initialize it on object initialization. This way other device
classes can still grab the pointer, as if there was one. This may be
useful for example if a touch grab is active and a mouse is connected.

This also makes it possible for popup grabs, which currently use a
pointer grab for controlling, to be triggered by touch devices, while
still holding an active pointer grab.

https://bugzilla.gnome.org/show_bug.cgi?id=771305

Make the device <-> seat association permanent, and move it into
MetaWaylandInputDevice. A device will never be disassociated with a
seat, so there is no point in unsetting it.

https://bugzilla.gnome.org/show_bug.cgi?id=771305