If we have a CLICKING grab op we still need to send events to xwayland
so that we get them back for gtk+ to process thus we can't steer
wayland input focus away from it.
https://bugzilla.gnome.org/show_bug.cgi?id=726123
meta_wayland_seat_repick() can be called in various cases while mutter
has a GRAB_OP ongoing which means we could be sending wrong pointer
enter/leave events.
https://bugzilla.gnome.org/show_bug.cgi?id=726123
This ensures that we send the proper leave and enter events to wayland
clients.
Particularly, this solves a bug in SSD xwayland windows where clicking
and dragging on the title bar to move the window only works on the odd
turn (unless the pointer moves away from the title bar between
tries). This happens because xwayland gets a button press but doesn't
see the release so when it gets the next button press it discards it
because its pointer button tracking logic says that the button is
already pressed. Sending the proper wayland pointer leave event fixes
it since wayland clients must forget about button state at that point.
https://bugzilla.gnome.org/show_bug.cgi?id=726123
At one point, it was supported to run mutter without a compositor,
but we don't allow that any longer. A lot of code already assumes
display->compositor exists and doesn't check for a NULL pointer,
so just kill the rest of the checks.
This is specifically about managing X11 windows, not necessarily
running as an X11 compositor. By that I mean that this code is
still used for XWayland windows, and event handling is still and
modesetting / monitor management is still in core/.
This is also a fairly conservative move. We don't move anything
like screen.c or bell.c in here, even though those are really
only for X11 clients.
This is fairly simple and basic for now, with just skip_taskbar /
skip_pager, but eventually a lot of "WM policy" like this, including
move-resize, will be in subclasses for each individual surface.
In particular we need to know about all key events to keep the xkb
state reliable even if the event is then consumed by a global shortcut
or grab and never reaches any wayland client.
We also need to keep track of all pressed keys at all times so that we
can send an updated set or pressed keys to the focused client when a
grab ends.
https://bugzilla.gnome.org/show_bug.cgi?id=722847
Any given clutter event carries the modifier state as it was before it
occured but, for the wayland modifiers event, we want the state
including the current event.
To fix this, we'll keep our xkb_state instance around instead of the
serialized mods.
https://bugzilla.gnome.org/show_bug.cgi?id=722847
Currently touch events are ignored in the core event handler,
and hence dealt with within GDK. If those touch events were
emulating pointer events, GDK would attempt to convert back
those events to pointer events as the frame GdkWindow doesn't
have the GDK_TOUCH_MASK set.
This results in XI_TouchBegin events being initially processed
by GDK, converted to button events, and triggering a grab op
that subverts touch events into pointer events, so the touch
is never ever seen again by GDK. This leaves GDK in an
inconsistent internal state wrt pointer grabs, so future
pointer-emulating touches will refer to the same window forever.
Fix this by handling touch events minimally, just enough to
convert XI_TouchBegin to GDK_BUTTON_PRESS within mutter, so GDK
is bypassed for every touch event just like it is for pointer
events. This, and the XIGrabDevice() that keeps coercing pointer
events when the grab operation starts, are enough to fix window
drag and drop on touch devices.
https://bugzilla.gnome.org/show_bug.cgi?id=723552
We try to exempt CSD windows from being forced fullscreen if they are
undecorated and the size of the screen; however, we also catch almost
all windows that *do* need to be forced fullscreen in this check, since
they also have decorations turned off.
Identify actual CSD windows by checking whether _GTK_FRAME_EXTENTS is set -
GTK+ will always set this on CSD windows even if they have no invisible
borders or shadows at the current time.
We explicitly turn off the legacy-fullscreen check for native wayland windows
so we don't start legacy-fullscreening them if the new
meta_window_is_client_decorated() is later made more accurate.
https://bugzilla.gnome.org/show_bug.cgi?id=723029
We try to exempt CSD windows from being forced fullscreen if they are
undecorated and the size of the screen; however, we also catch almost
all windows that *do* need to be forced fullscreen in this check, since
they also have decorations turned off.
Identify actual CSD windows by checking whether _GTK_FRAME_EXTENTS is set -
GTK+ will always set this on CSD windows even if they have no invisible
borders or shadows at the current time.
https://bugzilla.gnome.org/show_bug.cgi?id=723029
The user_rect represents the unconstrainted window size, and lots
of code in mutter assumes it can resize to the user_rect at any
time. If we wait for an attach to ACK and save the user rect, we'll
see lots of flickering as code is resizing to the old user_rect
at any time.
Make it a compile-time flag rather than a run-time flag, because
practically any time you're going to be debugging event spewing,
you're going to have to recompile anyway. Remove the WITH_VERBOSE_MODE
checks, too.
Which is used for Wayland popup grabs.
The issue here is that we don't want the code that raises or focuses
windows based on mouse ops to run while a client has a grab.
We still keep the "old" grab infrastructure in place for now, but
ideally we'd replace it eventually with a better grab-op infrastructure.
Clutter's input device initial position defaults to (-1, -1) on most
backends but for the evdev backend we changed it to be inside the
stage to prevent the pointer from wandering outside the stage until it
first enters, after which our constraining callback won't let it go
out.
This makes us be in sync with the real position from the start.
_SVID_SOURCE has been deprecated in newer versions of glibc breaking
-WError; the recommended replacement of _DEFAULT_SOURCE is fairly
new, so switch to _XOPEN_SOURCE instead.
_SVID_SOURCE has been deprecated in newer versions of glibc breaking
-WError; the recommended replacement of _DEFAULT_SOURCE is fairly
new, so switch to _XOPEN_SOURCE instead.
The "original coordinates" passed into meta_window_place() were the
coordinates of the client rectangle not the frame rectangle. When
meta_window_place() didn't place because the window was manually
positioned (e.g., 'xterm -geometry +x+y') that resulted in a window
being offset by the frame dimensions.
https://bugzilla.gnome.org/show_bug.cgi?id=724049
Since the introduction of frame sync in GTK+, updates to titlebar font and
colors haven't been working because GTK+ counts on the frame clock to
do style updates, and the frame clock doesn't run for an unmapped
GdkWindow. (It's possible that GtkStyleContext changes subsequent to
the introduction of the frame clock were also needed to fully break
things.)
We actually need to map the MetaFrames GdkWindow and let the
compositor code send out the frame sync messages in order to pick up
style changes.
Hopefully no bad side effects will occur from this - we make the window
override-redirect, 1x1, and outside the bounds of the screen.
https://bugzilla.gnome.org/show_bug.cgi?id=725751
The "original coordinates" passed into meta_window_place() were the
coordinates of the client rectangle not the frame rectangle. When
meta_window_place() didn't place because the window was manually
positioned (e.g., 'xterm -geometry +x+y') that resulted in a window
being offset by the frame dimensions.
https://bugzilla.gnome.org/show_bug.cgi?id=724049
Since the introduction of frame sync in GTK+, updates to titlebar font and
colors haven't been working because GTK+ counts on the frame clock to
do style updates, and the frame clock doesn't run for an unmapped
GdkWindow. (It's possible that GtkStyleContext changes subsequent to
the introduction of the frame clock were also needed to fully break
things.)
We actually need to map the MetaFrames GdkWindow and let the
compositor code send out the frame sync messages in order to pick up
style changes.
Hopefully no bad side effects will occur from this - we make the window
override-redirect, 1x1, and outside the bounds of the screen.
https://bugzilla.gnome.org/show_bug.cgi?id=725751
We need to resolve the keycode from the keysym again since the keycode
might have changed if there was a keymap switch between the grab and
the ungrab.
Before starting to use display_get_keybinding() we could compare
MetaKeyBinding.modifiers with MetaKeyCombo.modifiers directly. Now, we
need to resolve the virtual modifiers to match with the mask.
We need to resolve the keycode from the keysym again since the keycode
might have changed if there was a keymap switch between the grab and
the ungrab.
Before starting to use display_get_keybinding() we could compare
MetaKeyBinding.modifiers with MetaKeyCombo.modifiers directly. Now, we
need to resolve the virtual modifiers to match with the mask.
This allows us to look for a match with an O(1) search instead of O(n)
which is nice, particularly when running as a wayland compositor in
which case we have to do this search for every key press event (as
opposed to only when our passive grab triggers in the X compositor
case).
We actually need two hash tables. On one we keep all the keybindings
themselves which allows us to add external grabs without constantly
re-allocating the array we were using previously.
The other hash table is an index of the keybindings in the first table
by their keycodes and mask which is how we actually match the key
press events. This second table thus needs to be rebuilt when the
keymap changes since keycodes have to be resolved then but since we're
only keeping pointers to the first table it's a fast operation.
https://bugzilla.gnome.org/show_bug.cgi?id=725588
Instead of looping over an array of keybindings to find the correct
binding, just use display_get_keybinding().
In the next commit, we'll change the array to be a hash map, so this
helps the patch be cleaner.
https://bugzilla.gnome.org/show_bug.cgi?id=725588
This allows us to look for a match with an O(1) search instead of O(n)
which is nice, particularly when running as a wayland compositor in
which case we have to do this search for every key press event (as
opposed to only when our passive grab triggers in the X compositor
case).
We actually need two hash tables. On one we keep all the keybindings
themselves which allows us to add external grabs without constantly
re-allocating the array we were using previously.
The other hash table is an index of the keybindings in the first table
by their keycodes and mask which is how we actually match the key
press events. This second table thus needs to be rebuilt when the
keymap changes since keycodes have to be resolved then but since we're
only keeping pointers to the first table it's a fast operation.
https://bugzilla.gnome.org/show_bug.cgi?id=725588
Instead of looping over an array of keybindings to find the correct
binding, just use display_get_keybinding().
In the next commit, we'll change the array to be a hash map, so this
helps the patch be cleaner.
https://bugzilla.gnome.org/show_bug.cgi?id=725588
Creating a new cogl texture may fail, in which case the intent to
free it will crash. While something is clearly wrong (insanely
large window, oom, ...), crashing the WM is harsh and we should
try to avoid it if at all possible, so carry on.
https://bugzilla.gnome.org/show_bug.cgi?id=722266
Do to a bad mixup, the surface listener was never actually fired.
This was accidentally fixed as part of a refactoring in a27fb19,
but the surface listener was broken, and we started crashing. To
fix, just remove the surface listener, as we've mostly been testing
without it.
This is not needed since the instance is being destroyed and in fact
actively harmful when code called from other handlers disconnects us
for other reasons. In that case we might crash because the
disconnection doesn't prevent other handlers from running in the
current signal emission and thus we try to remove ourselves from an
empty list.
Creating a new cogl texture may fail, in which case the intent to
free it will crash. While something is clearly wrong (insanely
large window, oom, ...), crashing the WM is harsh and we should
try to avoid it if at all possible, so carry on.
https://bugzilla.gnome.org/show_bug.cgi?id=722266
This changes the user data of all surface extensions resources to be
the MetaWaylandSurface instead of the MetaWaylandSurfaceExtension,
which means that we no longer need all these pesky wl_container_ofs
in implementations.
Don't set the surface actor to a new buffer if it's becoming unmapped.
This is also technically wrong since we'll send out the release event,
but oh well.
We should probably decouple MetaWaylandBuffer from the CoglTexture
at some point, so we can send out releases on-demand.
We don't want to match the keysym so that e.g. an accelerator
specified as "<Super>a" works if the current keymap has a keysym other
than 'a' for that keycode which means that the accelerator would
become inaccessible in a non-latin keymap.
This is inconvenient for users that often switch keyboard layouts, or
even have different layouts in different windows, since they expect
system-level keybindings to not be affected by the current layout.
https://bugzilla.gnome.org/show_bug.cgi?id=678001
We don't want to match the keysym so that e.g. an accelerator
specified as "<Super>a" works if the current keymap has a keysym other
than 'a' for that keycode which means that the accelerator would
become inaccessible in a non-latin keymap.
This is inconvenient for users that often switch keyboard layouts, or
even have different layouts in different windows, since they expect
system-level keybindings to not be affected by the current layout.
https://bugzilla.gnome.org/show_bug.cgi?id=678001
The only events we handle as XIEvents are FocusIn/Out, Enter and
Leave. Motion, ButtonPress/Release, KeyPress/Release are handled
through clutter instead.
Among other things, this means we don't need to fake motion compression
by peeking over gdk event queue...
All WM events (passive button grabs and passive keyboard grabs)
are handled through clutter now, so we must make sure we spoof
them even if they happen on frames (because that's where we
grab on)
Weirdly, clutter stopped segfaulting when we call clutter_x11 methods
and the backend is not right, but this is correct anyway, and
probably fixes some BadDrawable errors in mutter-wayland on x11,
caused by mixing windows of the outer X and windows of Xwayland.
Mouse event handling was duplicated, resulting in weird interactions
if clutter was allowed to see certain events (for example under
wayland, where it gets all events). Because now clutter sees all
X events, even when running as an x11 compositor, we can handle
everything using the clutter variants.
At the same time, rewrite a little the passive button grab code,
to make it clear what is being matched on what and why.
meta_ui_window_is_widget() returns FALSE for frame windows, so we
must filter those explicitly (by letting the event go to gtk
and from there to MetaFrames). Also, for proper gtk widgets
(window menus) we want to let gtk see all events, including
keyboard, otherwise we break keynav in the window menu.
This means that having a window menu open disables keybindings
(because the event doesn't run through clutter)
We must spoof events to clutter even if they are associated
with a MetaWindow, because keyboard events are always associated
with one (the focus window), and we must process keybindings
for window togheter with the global ones if they include Super,
because we're not going to see them again.
... and individually. It turns out that updating the opaque region
was causing the shape region to be updated, which was causing a new
shape mask to be generated and uploaded to the GPU. Considering
GTK+ regenerates the opaque region on pretty much any focus change,
this is not good.
For decorated windows, we don't want to apply any input
shape, because the frame is always rectangular and eats
all the input.
The real check is in meta-window-actor, where we consider
if we need to apply the bounding shape and the input shape
(or the intersection of the two) to the surface-actor,
but as an optimization we avoid querying the server in
meta-window.
Additionally, for undecorated windows, the "has input shape"
check is wrong if the window has a bounding shape but not an
input shape.
... and individually. It turns out that updating the opaque region
was causing the shape region to be updated, which was causing a new
shape mask to be generated and uploaded to the GPU. Considering
GTK+ regenerates the opaque region on pretty much any focus change,
this is not good.
The handler pointer is dangling in MetaKeyBinding until
rebuild_key_binding_table() is run, so we can't dereference it.
Because we only need the flags at ungrab time, store a copy
in the MetaKeyBinding structure.
https://bugzilla.gnome.org/show_bug.cgi?id=724402
The handler pointer is dangling in MetaKeyBinding until
rebuild_key_binding_table() is run, so we can't dereference it.
Because we only need the flags at ungrab time, store a copy
in the MetaKeyBinding structure.
https://bugzilla.gnome.org/show_bug.cgi?id=724402
At some point meta_window_actor_cull_out stopped calling
meta_cullable_cull_out_children which caused the unobscured region
to never be set for the stex.
https://bugzilla.gnome.org/show_bug.cgi?id=725216
For decorated windows, we don't want to apply any input
shape, because the frame is always rectangular and eats
all the input.
The real check is in meta-window-actor, where we consider
if we need to apply the bounding shape and the input shape
(or the intersection of the two) to the surface-actor,
but as an optimization we avoid querying the server in
meta-window.
Additionally, for undecorated windows, the "has input shape"
check is wrong if the window has a bounding shape but not an
input shape.
We need a MetaWaylandSurface to build a MetaSurfaceActor, but
we don't have one until we get the set_window_xid() call from
XWayland. On the other hand, plugins expect to see the window
actor right from when the window is created, so we need this
empty state.
Based on a patch by Jasper St. Pierre.
Turns out we only ever need to freeze/thaw whole windows, not
surfaces or subsurfaces.
This will allow removing the surface actor without losing
the count.
This time, to make way for MetaSurfaceActorEmpty. This also fixes
destroy effects as a side effect. It still has issues if we try
to re-assign an actor that's already toplevel (e.g. somebody
re-popping up a menu that's already being destroyed), but this
will be fixed soon.
The idea here is that MetaWindowActor will do the unparenting of
the surface actor when it itself is destroyed. To prevent bad issues
with picking, we only make the surface actor reactive when it's
toplevel.
It triggers too often, making G_DEBUG=fatal-warnings quite useless.
Owen is going to rewrite this code sometime in the near future, so
I'm just gonna kill this warning for now.
If the last reference of a MetaIdleMonitor is held by the caller, it may
happen that the last reference is lost when calling the GDestroyNotify,
if this happens when the watched DBus name vanishes, the object (and the
watches hashtable) are destroyed while manipulating the watches hashtable,
so bad things may happen then.
Fix this by wrapping the operation by a ref/unref pair, so the object would
be destroyed after operating on the hashtable.
https://bugzilla.gnome.org/show_bug.cgi?id=724969
gnome-shell has some complex tracking to set the X input focus
correctly, assuming various things about how the stage is set up in X11.
For instance, it assumes that all actors that get key focus are
gnome-shell Chrome actors that will get events through the stage, so
when one of them is focused, it will try to set the focus back to the
stage.
In Wayland, windows are considered chrome actors that will get key
events through the stage, so this only has the result of unfocusing any
windows that have just received key focus.
We should probably move this input focus moving to mutter instead of
gnome-shell so we can better use mutter's internal state and heuristics.
If the last reference of a MetaIdleMonitor is held by the caller, it may
happen that the last reference is lost when calling the GDestroyNotify,
if this happens when the watched DBus name vanishes, the object (and the
watches hashtable) are destroyed while manipulating the watches hashtable,
so bad things may happen then.
Fix this by wrapping the operation by a ref/unref pair, so the object would
be destroyed after operating on the hashtable.
https://bugzilla.gnome.org/show_bug.cgi?id=724969
We cannot intersect the the complete volume with the unobscured bounds
because it does not include the shadows. So just intersect it with the
windows's shape bounds and union it with the shadow bounds.
This also matches what the comment in the code says:
"We could compute an full clip region as we do for the window texture,
but the shadow is relatively cheap to draw, and a little more complex to clip,
so we just catch the case where the shadow is completely obscured
and doesn't need to be drawn at all."
We cannot intersect the the complete volume with the unobscured bounds
because it does not include the shadows. So just intersect it with the
windows's shape bounds and union it with the shadow bounds.
This also matches what the comment in the code says:
"We could compute an full clip region as we do for the window texture,
but the shadow is relatively cheap to draw, and a little more complex to clip,
so we just catch the case where the shadow is completely obscured
and doesn't need to be drawn at all."
It turns out XKB has keysyms for these for no real reason, and I
was hitting those instead of the <Primary><Alt>F1 path. This is
ridiculous, but key, so is the entirety of XKB.
This took an embarassingly long time to figure out and debug.
The rendering logic before was somewhat complex. We had three independent
cases to take into account when doing rendering:
* X11 compositor. In this case, we're a traditional X11 compositor,
not a Wayland compositor. We use XCompositeNameWindowPixmap to get
the backing pixmap for the window, and deal with the COMPOSITE
extension messiness.
In this case, meta_is_wayland_compositor() is FALSE.
* Wayland clients. In this case, we're a Wayland compositor managing
Wayland surfaces. The rendering for this is fairly straightforward,
as Cogl handles most of the complexity with EGL and SHM buffers...
Wayland clients give us the input and opaque regions through
wl_surface.
In this case, meta_is_wayland_compositor() is TRUE and
priv->window->client_type == META_WINDOW_CLIENT_TYPE_WAYLAND.
* XWayland clients. In this case, we're a Wayland compositor, like
above, and XWayland hands us Wayland surfaces. XWayland handles
the COMPOSITE extension messiness for us, and hands us a buffer
like any other Wayland client. We have to fetch the input and
opaque regions from the X11 window ourselves.
In this case, meta_is_wayland_compositor() is TRUE and
priv->window->client_type == META_WINDOW_CLIENT_TYPE_X11.
We now split the rendering logic into two subclasses, which are:
* MetaSurfaceActorX11, which handles the X11 compositor case, in that
it uses XCompositeNameWindowPixmap to get the backing pixmap, and
deal with all the COMPOSITE extension messiness.
* MetaSurfaceActorWayland, which handles the Wayland compositor case
for both native Wayland clients and XWayland clients. XWayland handles
COMPOSITE for us, and handles pushing a surface over through the
xf86-video-wayland DDX.
Frame sync is still in MetaWindowActor, as it needs to work for both the
X11 compositor and XWayland client cases. When Wayland's video display
protocol lands, this will need to be significantly overhauled, as it would
have to work for any wl_surface, including subsurfaces, so we would need
surface-level discretion.
https://bugzilla.gnome.org/show_bug.cgi?id=720631
It's mostly equivalent to the case where we've already detached
the pixmap, *except* for the x11_size_changed case. We can simply
detach the pixmap at the time the window changes size, though.
https://bugzilla.gnome.org/show_bug.cgi?id=720631
We guarantee ourselves that a valid pixmap will appear any time
that the window is painted. The window actor will be scheduled
for a repaint if it's added / removed from the scene graph, like
during construction, if the size changes, or if we receive damage,
which are the existing use cases where this function is called.
So, I can't see any reason that we queue a redraw in here.
With the split into surface actors, we don't have an easy place
we can use to queue a redraw, and since it's unnecessary, we can
just drop it on the floor.
https://bugzilla.gnome.org/show_bug.cgi?id=720631
If the client destroys the pointer resource, we shouldn't unfocus the
surface, and we should regrab it when the client gets the pointer
resource again.
This also fixes a crash at surface destruction because of the unchecked
wl_link_remove that will happen on both pointer and surface destroy.
We can never have a window actor that represents either the X root
window or the stage window, so it doesn't make sense to bail out
early in case we do.
I'd imagine that this came from a much earlier version of the code
where the compositor was much separate and had its own MapNotify
handling.
This reverts commit 283a81eac0.
This can't be done yet, as it will crash when we try to do a destroy
effect from a plugin. The surface actor needs to outlive the surface
in this case.
Though, the unparenting happening is wrong anyway for a destroy effect.
We need to figure out a sane way of doing this unparenting only after
all effects have finished.
As resource destruction can happen in any order at shutdown, we
need to be flexible here. A client disconnecting without cleaning
up all its resources should not assert fail.
Since the unredirected window MetaWindowActor is stacked on top, it
will naturally get culled out of the process, so we can remove the
special casing here. Unfortunately, with the way that the code is
currently structured, it's too difficult to actually prevent setting
the clip / visible regions if the window is redirected, so just let
those be set for unredirected windows for now.
The input region was set on the shaped texture, but the shaped texture
was never picked properly, as it was never set to be reactive. Move the
pick implementation and reactivity to the MetaSurfaceActor, and update
the code everywhere else to expect a MetaSurfaceActor.
I implemented pinging, but never actually enabled the feature
properly on Wayland surfaces by setting the net_wm_ping hint to
TRUE, causing the fallback path to always be hit.
Rename net_wm_ping to can_ping so it doesn't take on an
implementation-specific meaning, and set it for all Wayland windows.
I tested this with meta_window_get_input_rect and decided to change
it at the last minute. Turns out meta_window_get_rect is unlike all
the others. Just access window->rect directly instead.
Now that we call set_custom_frame_extents, the frame rect corresponds
to the "visible window geometry" used for constrainment, while the
x/y fields in get_xdg_popup instead are relative to the surface.
Something noticed on code inspection. If we have a popup grab,
it will always return FALSE. The code here clearly meant to continue
if we had an existing popup grab from an existing client.
Both the pointer/keyboard resource and surface resource can be destroyed
at any point in the destruction process, so we need to have destroy
listeners on both. To make the code easier to follow, rename ->focus
to ->focus_surface at the same time, and rearrange the code so that
the two of them are always grouped together.
To prevent the MetaSurfaceActor from being destroyed, we normally
unparent it before we unmanage the window. However, this doesn't
work for XWayland windows, which we unmanage when we get UnmapNotify
or DestroyNotify, not when we get the wl_surface_destroy.
To solve this, add an early hook in meta_window_unmanage that
unparents the surface actor if we have one. At the same time, clean
up the destruction code to remove old comments and assumptions about
how wl_shell behaves.
