If a after-paint was scheduled but nothing was painted, for example when
the it was scheduled by a subsurface wanting to update its position,
we'd still try to read back from the backfill cairo surface and update
the committed cairo surface reference even though no buffer was
attached.
Fix this by adding a new state, 'pending_buffer_attached', which is only
true if a buffer was attached during frame. Only when this is true will
the backfill be read back and the committed cairo surface reference be
updated.
https://bugzilla.gnome.org/show_bug.cgi?id=762120
Right now we use one buffer for both staged changes (freshly painted
changes waiting for the frame clock to send to the compositor) and
committed changes (changes actively being read by the compositor
process). This creates a problem in the event we need to stage updates
at the same time the compositor is processing committed updates: we
can't change what the compositor is actively processing.
The current solution for handling this contention is to allocate a
temporary buffer on the spot at the time the updates are staged, and to
copy that buffer back to the shared buffer later. The problem, though,
is that the copy to the shared buffer currently happens as soon as
the updates are finished being staged, not when the shared buffer is
done being processed by the compositor.
In order to address that problem, this commit changes the code to always
stage changes to a dedicated staging buffer. The staging buffer is
used exclusively by the client until the client is done with it, and then
once that staging buffer is committed, the client never writes to that
buffer again. If the client needs to stage new updates, it allocates a
brand new staging buffer, draws to it, and back fills the undrawn parts
of the buffer from a copy of the contents of the committed buffer.
As an optimization, the compositor has the option of releasing the
committed buffer back to the client. If it does so before the client
needs to stage new updates, then the client will reuse the buffer
for staging future updates. This optimization prevents having to allocate
a new staging buffer and the associated cost of back filling
that new buffer with a readback of the committed buffer.
https://bugzilla.gnome.org/show_bug.cgi?id=761312
Right now we handle buffer releases coming from the
compositor in a central place. We add a listener when
first creating the shared buffers.
This is problematic because a buffer can only have
one listener on it at once so users of the buffer
can't get notified when it's released.
This commit moves the buffer listener code from the
centrally managed display code to the cursor and window
code.
https://bugzilla.gnome.org/show_bug.cgi?id=761312
The client and compositor share access to the window
pixel buffers. After the client hands off (commits)
the buffer to the compositor it's not supposed to write
to it again until it's released by the compositor.
The code tries to deal with this contention by allocating
a temporary buffer and using that in the mean time. This
temporary buffer is allocated by a higher layer of the code
when begin_paint returns TRUE. Unfortunately, that layer of
the code has no idea when the buffer is released, so it ends
up blitting the temporary buffer back to the shared buffer
prematurely.
This commit changes begin_paint to always return FALSE.
A future commit will address the contention problem in
a different way.
https://bugzilla.gnome.org/show_bug.cgi?id=761312
There are a few places where we destroy a cairo surface and
then nullify it. This commit changes those to use
g_clear_pointer instead.
It also drops a cairo_surface_finish call that is unnecessary
https://bugzilla.gnome.org/show_bug.cgi?id=761312
The name surface is really overloaded when dealing
with wayland windows.
To alleviate ambiguity, this commit changes the name
of the "surface" and "subsurface" members to have
a wl_ prefix.
https://bugzilla.gnome.org/show_bug.cgi?id=761312
Temp windows without parent are used frequently in our testsuite;
using g_warning on them causes the tests to fail, which is not
useful. Reduce the warning to a g_message.
X11 has the notions of "transient for group", and while it's an ICCCM
violation, it's commonly used and documented that a window manager
would treat a window with transient_for set to None to transient for all
windows of its group.
gtk uses this when an application sets a dialog type window but does not
specify an explicit transient.
While this works on X11, there is no such thing as groups in Wayland and
the closest equivalent which is set_parent() in xdg-shell takes only one
parent. This is what is used for modal dialogs.
To get something similar in behavior to what is available on X11, a
solution is to update the parent() of the dialogs without transient when
the active surface changes.
Bugzilla: https://bugzilla.gnome.org/show_bug.cgi?id=759161
Quite a few applications use GTK_WINDOW_POPUP to create various
temporary windows and place then on screen. That works fine on X11 but
on Wayland there is no global coordinate system for regular surfaces.
If the application is using a gdk temp window and set a parent with
gtk_window_transient_for(), the gdk wayland backend has all it needs to
create a subsurface that can be placed at will by the application.
Bugzilla: https://bugzilla.gnome.org/show_bug.cgi?id=759738
In most places, we can do with the pointer/keyboard of the default seat
instead of the client pointer. We can also remove some code from
gdk_input_init() because we know for sure there's no floating devices to
care about here.
Gdk Wayland backend walks up the transient windows tree, but does not
check for cycles when doing so.
As a result, if two or more windows are transient to each other, the
Wayland gdk backend will enter an infinite loop.
While this is clearly a bug in the application, gtk+/gdk should be more
robust and handle such errors more gracefully.
To avoid looping infinitely at various point in the code, check for a
possible loop when setting the transient relationship and deny the
request to set a window transient for another if that would create a
loop.
Bugzilla: https://bugzilla.gnome.org/show_bug.cgi?id=759299
We no longer need a grabbed seat, instead we'll just use the default
seat if this happens, not without first warning and recommending
gdk_seat_grab() for the operation.
https://bugzilla.gnome.org/show_bug.cgi?id=759309
In Wayland, the hotspot of a DND icon is set using the buffer offset in
wl_buffer.attach. To implement this, add a private API to cause the
next wl_surface.attach to offset the new buffer with a given offset.
Setting a DND icon hotspot sets this offset while also queuing a redraw
of the window to trigger the wl_surface.attach.
https://bugzilla.gnome.org/show_bug.cgi?id=759168
In Wayland, the hotspot of a DND icon is set using the buffer offset in
wl_buffer.attach. To implement this, add a private API to cause the
next wl_surface.attach to offset the new buffer with a given offset.
Setting a DND icon hotspot sets this offset while also queuing a redraw
of the window to trigger the wl_surface.attach.
https://bugzilla.gnome.org/show_bug.cgi?id=759168
gdk-wayland backend would not re-configure a surface when its size and
scale match the known size and scale.
But there might be a pending xdg_surface_configure() that would revert
this change so we should re-configure even if the currently known
size/scale match, otherwise we may end up with a wrong size after the
xdg_surface_configure() is received.
Bugzilla: https://bugzilla.gnome.org/show_bug.cgi?id=758901
Other backends take care of the cairo surface destruction in
GdkWindow::destroy. We must do the same here, or the cairo_surface
and its corresponding wl_buffer are left dangling.
https://bugzilla.gnome.org/show_bug.cgi?id=747295
Make sure the wayland backend sets a new geometry when the client
resizes itself, otherwise the compositor won't be notified and may
revert to the old size on state changes.
Thanks to Jasper St. Pierre <jstpierre@mecheye.net> who pointed out the
problem in gtk+.
bugzilla: https://bugzilla.gnome.org/show_bug.cgi?id=755051
If a GtkMenu (or something else that is mapped as a xdg_popup) tries to
use a subsurface window as a parent, it will be terminated by the
compositor due to protocol violation. So to avoid this, if a parent
window is not a xdg_popup or xdg_surface, i.e. a wl_subsurface, then
traverse up the transient parents until we find the right popup parent.
https://bugzilla.gnome.org/show_bug.cgi?id=756780
Tooltips tend to be placed on top of a parent surface with a given
relative coordinate, and without any input focus. So lets map them as
subsurfaces.
https://bugzilla.gnome.org/show_bug.cgi?id=756496
Restructure the mapping procedure so that its known up front what the
expected way mapping is to be done (subsurface, popup or stand alone),
and warn if it fails to actually map in such a way (for example a popup
without a parent or device grab, a tooltip without a parent).
https://bugzilla.gnome.org/show_bug.cgi?id=756496
Initially the subsurface will be in synchronized mode and we will leave
it like this until the first time the parent surface has been committed.
The reason for this is because the subsurface position will be applied
as part of the parent surface state, and we need to synchronize the
initial position with the initial frame, so that we don't accidentally
draw the subsurface at the default position (0, 0) which would happen in
desynchronized mode if the subsurface content is committed before the
next parent surface commit.
https://bugzilla.gnome.org/show_bug.cgi?id=754839
If we are using gl for drawing, we don't have a shm surface,
so don't assert that we do. Instead, only call shm-specific
apis when they make sense.
This fixes a crash when showing popovers over a GtkGLArea,
as seen in gdkgears.
https://bugzilla.gnome.org/show_bug.cgi?id=754143
We call gdk_wayland_window_hide_surface when the window gets
destroyed, and in this case, the frame clock might not exist
anymore.
This was showing up in the displayclose testcase.
While we do not have subwindows in Wayland, we do create an
artificial root window. When the display is closed, the root
window gets destroyed, causing recursing to be true for the
toplevel windows.
Prior to this patch, the ID of the GtkApplication was always used for
clients which were GtkApplications. This would only be guaranteed to be
correct for D-Bus activatable programs. As a result, all
non-D-Bus-activatable applications would set the wrong ID making the
shell unable to find the corresponding .desktop file.
This change makes it so that the GDK backend always uses the name
passed to g_set_prgname, or the default value if not explicitly set, as
this more often corresponds to the .desktop file.
This means that in order to make D-Bus activatable applications set the
correct application ID, they must, for now, manually call
g_set_prgname() with their application ID (basename of the .desktop
file).
If g_get_prgname() returns NULL, fallback to gdk_get_program_class()
even though it will most likely never be correct according to the
xdg_surface.set_app_id specification.
https://bugzilla.gnome.org/show_bug.cgi?id=746435
We need to be mapped to have a gtk_surface and thus be able to do
requests on it so we need to save the modal hint and apply it when we
get mapped so that code that sets the hint before showing a window
doesn't get ignored.
https://bugzilla.gnome.org/show_bug.cgi?id=753138
-1 means that we have no specific preference for an initial
fullscreen monitor, and -1 is less than the number of monitors,
so we would end up accessing invalid memory. Prevent that.
https://bugzilla.gnome.org/show_bug.cgi?id=752875
Under Wayland, fullscreen/maximized windows may not cover the entire
area when a size increment is specified.
Ignore size increments for fullscreen/maximized windows just like most
window managers do under X11 so that windows with size increments can
still be fullscreen or fully maximized under Wayland as well.
https://bugzilla.gnome.org/show_bug.cgi?id=751368
A subsurface positioning operation only takes effect when the parent
surfaces state is applied. If a subsurface is mapped and positioned, but
the parent surface state is not immediately committed, the relative
position of the subsurface is undefined and may be placed incorrectly.
To avoid this undefined state, always request that the parent surface
should be committed after mapping a subsurface so that the position
operation will take effect.
https://bugzilla.gnome.org/show_bug.cgi?id=751098
If a menu was not attached to any widget, we try to calculate its
position given where the grabbed pointer is and what window has its
focus. Previously we failed to do so if a "transfer window" was used
for the grab, and this patch adds a code path that, if the menu window
itself didn't have the grab, look for the transfer window and get the
grab device from there.
https://bugzilla.gnome.org/show_bug.cgi?id=748951
If a position was already explicitly set, don't try to guess the
position of popup menus by looking at the pointer position, just use
the set coordinates.
https://bugzilla.gnome.org/show_bug.cgi?id=748951
According to the xdg-shell protocol specification the (x, y) coordinates
passed when creating a popup surface is relative to top left corner of
the parent surface, but prior to this patch, if the parent surface
was an xdg_surface, we'd position it relative to top left corner of the
window geometry of that xdg_surface.
https://bugzilla.gnome.org/show_bug.cgi?id=749717
The ordering of globals in connection setup under weston
is different from mutter, and we end up creating a the
dnd window before any outputs are present. Don't cause
a critical warning in that case.
Add two new requests to the gtk_surface interface: set_modal and
unset_modal. The server will currently not do anything special with
input focus, and its up to the client to ignore events on the parent
surface.
This commit bumps the gtk_shell interface version to 2. By connecting to
a Wayland server with another gtk_shell interface version any features
depending on the gtk_shell protocol will not be available.
https://bugzilla.gnome.org/show_bug.cgi?id=745721
The "app_id" of a xdg_surface should be the ID that can potentially be
used to get the DBUS name or the .desktop file.
For GtkApplication programs this is often the ID passed when creating the
GtkApplication object, so when available lets use that.
As fallbacks, first try g_get_prgname as it often corresponds to the
basename part of the .dektop file for non-GtkApplication programs.
Otherwise use gdk_get_program_class, even though that string usually
doesn't conform to the expectations of xdg_surface.set_application_id.
https://bugzilla.gnome.org/show_bug.cgi?id=746435
When a window is hidden, its surface and all its roles are destroyed,
if this happens when we already issued a wl_surface_commit and are
awaiting for a frame callback, the clock will remain frozen for the
next time the window is shown.
To avoid this, keep track of the wl_surface_frame() calls issued,
and ensure the clock is thawed after hiding. If we happen to receive
the frame callback, it is just ignored.
https://bugzilla.gnome.org/show_bug.cgi?id=743427
We were just throwing the request away if the app asks to
fullscreen or maximize a window before it has been mapped.
This is something the GdkWindow API explicitly supports,
so make it work by saving the state until the surface exists.
This fixes things under weston. There are bugs in mutter
that keep this from working correctly with gnome-shell.
https://bugzilla.gnome.org/show_bug.cgi?id=745303
Before this patch, we'd always allocate a full size SHM buffer via
the wl_shm_pool, even though it would never be used. Instead allocate a
logical 1x1 cairo image surface.
https://bugzilla.gnome.org/show_bug.cgi?id=745076
In order to support window scales for EGL windows, resize the
wl_egl_window to the window dimension multiplied with the window scale,
just as with SHM window buffers.
https://bugzilla.gnome.org/show_bug.cgi?id=745076
When the preferred surface scale changes, for example when entering a
wl_output with a higher scale than any previous entered output, recreate
the shm surface and redraw the window content with the new window scale.
Before this patch, the internal scale would be changed, but the shm
surface would not be recreated given the new scale, i.e. we'd attach a
buffer for a different scale than wl_surface.set_scale specified.
https://bugzilla.gnome.org/show_bug.cgi?id=745076
If the compositor is too old for handling surface buffer scales, never
tyr to set change it. This will effectively always leave it to its
initial state, i.e. 1.
https://bugzilla.gnome.org/show_bug.cgi?id=745076
Use xdg_surface_set_window_geometry() to tell the compositor about the
shadow widths, this makes some gnome-shell/mutter features (edge resistance,
frames around windows in the overview, side maximization, ...) work alright
with GTK+.
In order to add this, some other places in gdkwindow-wayland had to gain
some knowledge about margins:
- xdg_surface_configure() now syncs the shadow after applying the state,
and gdk_wayland_window_set_shadow_width() possibly reconfigures the
window in order to preserve window geometry. This is necessary to keep
shadows in sync with state/geometry changes, as this does not happen
all at once.
- xdg_popups relative to an xdg_surface are shown relative to buffer
coordinates, so the left/top margins must be added there.
https://bugzilla.gnome.org/show_bug.cgi?id=736742
It's unused. At the same time, rename "begin_paint_region" to
"begin_paint". This will help us clean up how GDK painting works
in the future to allow more creative use of double-buffering.
Instead of possibly calling wl_surface_commit() out of
GdkFrameClock::after-paint, tick the transient parent clock so ::after-paint
can be eventually run.
This ensures that the subsurface coordinates (considered part of the state
of the parent) aren't committed untimely, and guaranteed to be orderly with
the wl_subsurface-relative state.
This is a gtk-side fix for https://bugzilla.gnome.org/show_bug.cgi?id=738887
cairo_region_copy(NULL) will effectively return an empty region, as this
function is always meant to return valid memory. This however inverts the
meaning of the NULL region and results in entirely non-clickable windows.
We make user facing gl contexts not attached to a surface if possible,
or attached to dummy surfaces. This means nothing can accidentally
read/write to the toplevel back buffer.
A surface may be hidden when a frame is already scheduled, which may cause
crashes on on_frame_clock_after_paint() when calling commit() on a NULL
surface. To fix this, ensure commit_pending is also set to FALSE when the
surface is gone.
https://bugzilla.gnome.org/show_bug.cgi?id=735226
This has been made to work similarly to X11, requests for the data device
contents are notified through GDK_SELECTION_REQUEST events, the data stored
in the GDK_SELECTION property as a reaction to that event is then stored
into the wayland selection implementation, and written to the fd when
requested/available.
https://bugzilla.gnome.org/show_bug.cgi?id=697855
Subsurface position is deemed part of the state of the parent surface, so
ensure wl_surface_commit() happens on the parent surface if none is
scheduled, so the repositioning takes place.
The latest implicit grab serial is used in order to start the compositor
grab, If it belongs to a touch event, remove that touch sequence, as the
rest of the sequence will be gone for good.
This avoids stale sequences (and implicit grab info) after a window is
moved/resized.
https://bugzilla.gnome.org/show_bug.cgi?id=731380
_gdk_wayland_device_get_button_press_serial() has been replaced by
_gdk_wayland_device_get_implicit_grab_serial(), which takes a touch/pointer
event and figures out the relevant serial, and
_gdk_wayland_device_get_last_implicit_grab_serial() which returns
the most recent serial.
The button press serial was currently used when operating popping up
xdg_shell/surface popups and window menus, so this is now touch aware, of
some sort.
https://bugzilla.gnome.org/show_bug.cgi?id=734374
To all effects each window has its own "root" coordinates system, so set
toplevels at 0,0 in that coordinate system, so root coordinate calculations
are locally right.
https://bugzilla.gnome.org/show_bug.cgi?id=729215
xdg-shell has moved on and replaced set_margin with set_window_geometry.
To properly support set_window_geometry requires a full rewrite of how
we've been dealing with toplevel windows for now, so just don't set any
margin until we can have a proper toplevel window abstraction in GTK+.
Traditionally, the way painting was done in GTK+ was with the
"expose-event" handler, where you'd use GDK methods to do drawing on
your surface. In GTK+ 2.24, we added cairo support with gdk_cairo_create,
so you could paint your graphics with cairo.
Since then, we've added client-side windows, double buffering, the paint
clock, and various other enhancements, and the modern way to do drawing
is to connect to the "draw" signal on GtkWidget, which hands you a
cairo_t. To do double-buffering, the cairo_t we hand you is actually on
a secret surface, not the actual backing store of the window, and when
the draw handler completes we blit it into the main backing store
atomically.
The code to do this is with the APIs gdk_window_begin_paint_region,
which creates the temporary surface, and gdk_window_end_paint which
blits it back into the backing store. GTK+'s implementation of the
"draw" signal uses these APIs.
We've always sort-of supported people calling gdk_cairo_create
"outside" of a begin_paint / end_paint like old times, but then you're
not getting the benefit of double-buffering, and it's harder for GDK to
optimize.
Additionally, newer backends like Mir and Wayland can't actually support
this model, since they're based on double-buffering and swapping buffers
at various points in time. If we hand you a random cairo_t, we have no
idea when is a good time to swap.
Remove support for this.
This is technically a GDK API break: a warning is added in cases where
gdk_cairo_create is called outside of a paint cycle, and the returned
surface is a dummy that won't ever be composited back onto the main
surface. Testing with complex applications like Ardour didn't produce
any warnings.
Weston releases buffers almost immediately after they're done, which
means that GTK+ doesn't use a temporary surface and instead paints
directly onto the SHM backing store that Weston will use.
Normally, after painting to the temporary surface, GTK+ *replaces*
the existing backing surface with CAIRO_OPERATOR_SOURCE. However,
if we immediately paint to the backing surface, it might have junk
from the last paint in it. So clear out the backing surface whenever
somebody calls begin_paint_region().
Maybe we should just always use the temporary surface like the X11
codepath, since that prevents us from having to do weird things like
this, but oh well.
wl_surfaces can't switch roles, so destroying the xdg_surface but not
the wl_surface means that we could get an error when trying to re-map
the surface.
We could fix this by not destroying the xdg resource and only do it at
finalization time, but it's just as easy to just create a new wl_surface.
Since the xdg roles are a special case of the surface, some compositors
like Weston destroy them automatically when the wl_surface is destroyed.
Thus, we need to destroy these first.
The Wayland compositor is completely allowed to send us configure
events for the same size, and this validly happens if we're changing
states. Fizzle these out.
Since you can't take grabs on unmapped windows, GtkMenu takes a grab on
the menu in a convoluted way: it first grabs another window, shows the
menu window, and then transfers the grab over to the GtkMenu widget.
For normal menubars, this is perfectly fine, as the first window it grabs
is our toplevel, and that gets picked up in our transient path. For
GtkMenuButton or other spurious uses of gtk_menu_popup, it creates a new
temporary input-only window which it takes the grab on, known as the "grab
transfer window". Since this window isn't a transient-for of our new menu
widget window, the grab isn't noticed when we go to show it, and thus the
menu ends up as a new toplevel.
Add a special hack to GtkMenu and the Wayland backend which lets us notice
this "grab transfer window", and include it in our grab finding path.
It's sort of terrible to have to hack up the widgets instead of just the
backend, but the alternative would be an entirely new window type which is
managed correctly by GDK. I don't want to write that.
It seems that some backends implemented get_root_origin wrong
and returned the client window coordinates, not the frame window
coordinates. Since it's possible to implement generically for all
windows, let's do that instead of having a separate impl vfunc.
window->x / window->y are in "root window coordinates", e.g. relative
to the topmost toplevel. However, the coordinates in get_xdg_popup are
relative to the passed-in surface, so we need to do the reverse
translation here.
GtkWindow calls set_shadow_width then maps the window, meaning
that we never set the margin. Save it when we set and then set
it when we create the XDG surface.
Instead of destroying the surface in the backend if this is
unable to resize, let the core code do it, and do it properly.
Based on a patch by Benjamin Otte.
https://bugzilla.gnome.org/show_bug.cgi?id=725172
The code in GDK is incredibly broken and nobody is quite sure what's
right-side-up and what's upside down, but this breaks mutter-wayland
now, so let's remove it. It might leak, but we should probably do a
full restructuring of GDK drawing to fix it.
Like in other backends (except X) we can't resize cairo image surfaces
so let's sync the code here with what the other backends do.
This prevents the painting machinery above us to paint on the wrong
buffer.
https://bugzilla.gnome.org/show_bug.cgi?id=724968
We can't destroy buffers if they're in-use by the compositor. Well,
technically we can, but that is considered undefined by Wayland and
mutter won't cope with it very well -- it simply kills the client.
To solve this, we need to delay the destroy operation until the
compositor tells us that it's released the buffer. To do this, hold
an extra ref on the cairo surface as long as the surface is in-use
by the compositor.
This prevents warnings like
(gtk3-demo:14948): Gdk-CRITICAL **: _gdk_frame_clock_thaw: assertion 'GDK_IS_FRAME_CLOCK (clock)' failed
(gtk3-demo:14948): Gdk-CRITICAL **: gdk_frame_clock_get_timings: assertion 'GDK_IS_FRAME_CLOCK (frame_clock)' failed
We need to do this, as the compositor might have already sent us a frame
event, in-flight, at the same time we destroy our window. In this case, we'll
receive the then-in-flight "done" event, and then warn as we try to look
up the frame clock on a destroyed window.