GitLab

Populate only the fields we care about with designated initializers
to avoid breakage when the gdi_dc_funcs struct changes.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Mesa Wayland EGL, and Wayland in general, don't support front/single
buffer rendering. Emulate GL front buffer rendering by reading from the
front buffer on glFlush and applying the read pixels at window surface
flush time.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Some (older) games require display mode depths of 8bpp or 16bpp. Provide
such depths in the list of returned display modes in order to make such
games happy. Note that we don't really support depths other than 32bpp,
so modes with smaller depths are never applied, even if a program
requests it. This is not typically an issue, since Wine (e.g., wined3d)
takes care of any depth transformations.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Free the surface memory on destruction only at the very end, after we
don't need to access any of its members.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

The HeapFree() Win32 function, and thus the heap_free() Wine helper, can
gracefully handle NULL values.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Some surface may be both fullscreen and maximized. In such a case the
stricter requirements of the maximized state apply, so a fullscreen fit
is not appropriate, since it can lead to sizes smaller than the required
maximized size.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

The VK_ERROR_OUT_OF_DATE and VK_ERROR_SURFACE_LOST errors imply that
that the images have been released. Returning these errors without
actually invoking the host vkQueuePresent will keep the images acquired,
and thus may cause issues with applications that depend on the images
being released. So, always call vkQueuePresent even if we are going
to return an error value.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

When submitting a dummy buffer to ensure a surface is mapped, use
dimensions in wine coordinates.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Don't override handling of the SC_MAXIMIZE and SC_RESTORE SysCommands.
Let Wine core handle these and involve the driver to update the state
(e.g., through SetWindowPos handling) as needed.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

When the surface is in a transitional state, e.g., a new window state
has been requested but cannot be applied yet due to incompatibility with
the current Wayland surface state, block updating the contents of the
Wayland surface to avoid transient visual artifacts.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Win32 API like SetWindowPos may send messages to other threads, which
could access internal driver callbacks that require the same driver
lock, leading to a deadlock.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

In some cases, notably when an application calls UpdateLayeredWindow,
position and size changes may be emitted from a thread other than the
window thread. Since in the current implementation updating the wayland
state needs to happen in the context of the window thread to avoid racy
interactions, post a message to update the state in the right thread.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

When handling position and size changes (e.g., SetWindowPos())
update the wayland state from a single point at the end of
the driver callback sequence.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

This function is going to be needed in an upcoming commit to set
the format for wayland state updates.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Ensure that while handling the wayland surface for a window it
won't disappear from beneath our feet, e.g., some other thread
destroying it.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Set the xdg_toplevel title at creation time and update it when the
title of the win32 window changes.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Don't create xdg_toplevels for hidden windows, since some compositors
may list them in their window lists. Instead, create surfaces without a
role and recreate them with the proper role when they become visible
(and similarly recreate them without a role if they are hidden again).
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Request the minimization state from the compositor when an app is
minimized. Use the xdg_toplevel activation state to infer when the
xdg_toplevel, and thus the corresponding window, has been unminimized.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Checking only for a valid serial is not enough to guarantee that there
is a pending WM_WAYLAND_CONFIGURE message: we may have processed the
message but not acked the configure request due to surface size
incompatibilities (see window.c:update_wayland_state()). To solve this,
use a separate 'processed' flag to track WM_WAYLAND_CONFIGURE message
processing.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Some posted driver window messages assume the existence of a backing
wayland xdg_toplevel, which exists at the time of posting but may not
exist when the message is dispatched.

This can happen when between post and dispatch time the window changes
from a toplevel to become a child (through SetParent) or changes
ownership and is recreated as a wl_subsurface. This commit ensures we
handle this case gracefully, effectively ignoring the invalid messages.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Previously we were using GetParent() which returned the owner only if
the window was a WS_POPUP toplevel. Since we also want to find the
proper effective parent for non WS_POPUP toplevels, use GWLP_HWNDPARENT
which provides the owner for all kinds of toplevel windows, and still
returns the parent for true child windows.

GetParent: WS_CHILD => parent, WS_POPUP => owner, else => NULL
GWLP_HWNDPARENT: parent != desktop => parent, else => owner
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Remove a fix for fullscreen issues introduced in an earlier commit, as
it turned out that the behavior this fix was for is a problem with the
tested compositor (weston) rather than some valid Wayland protocol
expectation.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Some Wayland surface states (e.g., fullscreen, maximized) require strict
adherence to the compositor size constraints. Ensure that we only apply
the window configuration to a Wayland surface that can accept that
configuration.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Always use the output that Wine thinks our window is on as the main
output for the window. Do this even if we don't yet know if this output
contains the backing surface, since some compositors often send
wl_surface.enter events late (e.g., after configure events), leaving us
without a main output and thus unable to calculate correct sizes and
scaling.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Introduce a separate function to apply the new surface state after a
reconfiguration, to ensure all state is applied and also in the order
that causes the least issues for compositors.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

During thread wayland initialization we will get our initial output
information and init the display devices. There is no need to send out
WM_DISPLAYCHANGE in this case, since this is the initial display state.

Additionally, thread initialization may occur in a context that has
acquired the internal user32 lock, and sending messages would lead to an
internal user32 lock error.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

We delay clipboard window creation since the thread init function may be
invoked from within the context of a user32 function which holds the
internal Wine user32 lock. In such a case creating the clipboard window
would cause an internal user32 lock error.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

We currently notify each thread about wine mode changes by sending an
internal message to one of the thread's windows.

Going forward we won't be able to guarantee that all threads that need
to be notified about wine mode changes will have any windows. So,
instead of targeting a specific window, use the recently exposed
internal function to send the internal message to the thread directly.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Expose a function to drivers so that they can send internal messages to
specific threads, without the need to target a specific HWND.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

The GL/VK subsurface position is double-buffered state of the parent
surface. When positioning a GL/VK subsurface commit the parent surface
for the position to take effect.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

The subsurface position is double-buffered state of the parent
surface. When positioning a subsurface commit the parent surface for
the position to take effect.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Set the main output and scale of wayland subsurfaces at creation time
and ensure it remains up to date.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

We use heuristics to decide which toplevels are to be considered
transient (and thus be backed by wayland subsurfaces): the toplevel
needs to be undecorated and have a reasonably small size relative
to its main monitor.

In some cases transient windows using global coordinates are created
once and then moved around and shown as needed. This means that we
cannot decide the effective parent only at creation time, but rather
keep track of the current effective parent to be able to place such
transient windows at the correct surface-local positions.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

When notifying about a Wine mode change we need to send an internal
event to all threads. Use the per-thread clipboard window as the
recipient of the message since we know it always exists for all
Wayland/GUI threads.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Set the initial position and size of GL/VK subsurfaces to match the
client rectangle.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

We may need to create/destroy GL/VK wayland subsurfaces, as a result of
creating/destroying GL/VK surface types, from arbitrary threads. Handle
such operations in a more thread safe manner.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Implement support for the ChangeDisplaySettings(NULL, NULL, ...) call,
which instructs us to restore all monitors to registry settings.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>

Add a driver registry option "HiDPIScaling" to select between
"Application" scaling (the default) and "Compositor" scaling. Enabling
compositor scaling lets the Wayland compositor transparently scale
surfaces according to the output scale factor.
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>