We want TabBox to operate on key press, not on key release. So far
it only operated on release as the press was filtered out by the
global shortcuts filter. To prevent that the tab box filter is moved
before the global shortcuts filter.
Note: first usage of TabBox has current window selected instead of
next. This problem is also visible on X11.
The TabBox implements methods for those events and performs same
logic as on X11. Click outside of TabBox closes. If the event is on
the TabBox we don't filter the event out and let the internal filter
forward the event.
Drag'n'Drop on Wayland allows us to improve the drag'n'drop experience.
When entering a window during the drag'n'drop operation, KWin raises it.
BUG: 36065
FIXED-IN: 5.6.0 (Wayland only)
When starting effect mouse interception the current focused window
and or decoration should get a leave event. Similar when the effect mouse
interception ends the current pointer position needs to be evaluated and
a pointer enter be sent if needed.
So far updating the cursor image was not really defined. It was possible
to use the cursor image from the wayland seat or have a custom set cursor
image. But there are no rules in place to decide which one to use when.
With this change a dedicated CursorImage class is introduced which tracks
the cursor image changes on the seat, on the decoration, in the effects
and so on. In addition it tracks which is the current source for the
image, that is whether e.g. the cursor from the seat or from effects
override should be used. Whenever the cursor image changes a signal is
emitted, which is connected to the signal in AbstractBackend.
Based on that the backends can directly show the image. The existing
code in the backends to install a cursor shape or to install the cursor
from the server is completely dropped. For the backend it's irrelevant
from where the image comes from.
A new feature added is that the cursor image is marked as rendered. This
is then passed on to the frame rendered in the Surface and thus animated
cursors are finally working. Unfortunately animated cursors are broken in
Qt (see https://bugreports.qt.io/browse/QTBUG-48181 ).
Small regression: the command didn't get updated at all, so it was always
MouseNothing.
To prevent such a regression to sneak in again the change comes with
autotest for the action on inactive and active window.
As a Wayland server KWin does not have to emit additional key repeat
events (unlike X11). The clients are responsible for handling this based
on the provided key repeat information.
Internally KWin needs key repeat, though. E.g. the effects need key
repeat (filtering in Present Windows), window moving by keyboard needs
repeat, etc. etc.
This change introduces the internal key repeat. For each key press a
QTimer is started which gets canceled again on the key release. If the
timer fires it invoked processKey with a new KeyboardKeyAutoRepeat state.
This is handled just like a KeyPress, but states are not updated and
the QKeyEvent has autorepeat set to true.
The event filters check for the autorepeat state and filter the event
out if they are not interested in it. E.g. the filters passing the event
to the Wayland client need to filter it out.
Currently auto-repeat is bound to using libinput. This needs to be
modified. The only backend sending repeated events is X11, thus for
other backends it should be enabled.
Whether creating a timer on each key event is a good idea is something to
evaluate in future.
Reviewed-By: Bhushan Shah
Experimental testing in real world showed it's just a signing issue
in this specific case. The events passed to wayland clients scroll
in correct direction.
With that all the actions are implemented just like on X11.
There are two not yet implemented differences:
* hide splash window when clicking it
* replay event on special window
Implemented in the ForwardEventFilter: before forwarding the event
to the window we check whether a modifier is pressed and perform the
wheel command.
Possible improvements: each axis event triggers the same change, there
is no adjusted scaling.
This change implements the mouse command for modifier (alt/meta) plus
click in InputRedirection so that it also works on Wayland.
Modifier plus mouse wheel is not implemented yet.
For easier code in Options a new method is added which provides the
configured modifier as a Qt::KeyboardModifier instead of a Qt::Key code.
Test case is added which simulates all variants of modifiers plus
supported mouse buttons to trigger move.
So far the key handler in the InternalWindowEventFilter used the
PointerInputRedirection's internal window. This had the result that
key events were only delivered to an internal window if the window
was under the cursor.
This change tries sending the event to the latest created and visible
window. Thus e.g. with nested context menus it goes to the current
sub menu as expected. The return value of sendEvent is used to filter
out the event.
So far the DrmOutput connected to all input events when going into
power saving. As we now have the input filters it's better to just
install a filter when an output goes into powersave and remove the
input filter again when all outputs are enabled again.
To make this work InputRedirection gains a new method to add a new
filter as the first filter. This is a potentially dangerous method
as it allows to have a filter before LockScreenFilter gets the
events. But in case of DPMS it's something we actually want.
A nice new feature possible with the input filter is that we can
filter out the event which re-enables the outputs. Thus when getting
on a system with output off and screen locked, the first key hit
doesn't go to the lock screen.
Reviewed-By: Bhushan Shah
BUG: 341201
Fixed-in: 5.6.0 (Wayland-only)
Similar to the change regarding pointer and touch a
KeyboardInputRedirection is created. The Xkb class is also moved to
the new files keyboard_input.h and keyboard_input.cpp.
Just like in the case of PointerInputRedirection no signals are added,
but the existing signals in InputRedirection are directly invoked.
All pointer related code is moved into a new class called
PointerInputRedirection.
The main idea is to simplify the code and make it easier to maintain.
Therefore also a few changes in the setup were performed:
* before init() is called, no processing is performed
* init() is only called on Wayland and after Workspace is created
* init property is set to false once Workspace or WaylandServer is
destroyed
Thus code can operate on the following assumptions:
* Workspace is valid
* WaylandServer is valid
* ScreenLocker integration is used
The various checks whether there is a waylandServer() and whether
there is a seat are no longer needed.
Some of the checks have been reordered to be faster in the most common
use case of using libinput. E.g. whether warping is supported is first
evaluated by the variable bound to whether we have libinput and only if
that is false the backend is checked.
The new class doesn't have signals but invokes the signals provided
by InputRedirection. I didn't want to add new signals as I consider
them as not needed. The areas in KWin needing those signals should
be ported to InputEventFilters.
So far the area based edges connected directly to global pointer pos
changed in InputRedirection. This didn't allow proper checking whether
the edge was triggered (e.g. missing timestamp).
This change merges the functionality into the new input filter mechanism.
There is now a dedicated input filter for screen edges, installed after
lock screen and before effects. It always passes events on, but also passes
all events through ScreenEdges to handle the activation. As it's installed
after the lock screen filter we don't need to check for screen locked any
more.
The code is now similar strucutured to the existing X11 based variants
and maybe will allow to also merge the X11 variant with the new one.
The logic should not be tied to whether libinput is used. It's relevant
for all Wayland backends whether they use libinput or not.
In addition this should generate a pointer motion event, so that proper
processing can take place and we get proper pointer enter events.
If the pointer is warped the position change should be treated like
a change coming from the input device. Our normal processing should
take place.
A problem in this case is the timestamp to pass to the wayland server.
Normally our timestamps come from the backend/libinput and we don't
know the next one. As an intermediate solution we just use the last
timestamp on the seat. In future a solution could be to not use the
backend's timestamp at all, but have our own timestamp handling.
When the screen gets locked any existing sequence gets cancelled
and the focused touch surface gets reset. While screen is locked
touch events are filtered to only go to lock screen or input methods.
Test case is added for touch event during lock screen.
Reviewed-By: Bhushan Shah
Instead of only making the active client the focused keyboard surface,
the method now also performs the lock screen security restriction.
Also just like udatePointerWindow the method becomes public, so that
it can be used from the LockScreenEventFilter and is connected for
lock state changes. This means as soon as the screen locks the current
focused keyboard surface will get a leave event and get an enter event
once the screen unlocks.
The auto test is adjusted to verify these new conditions.
Reviewed-By: Bhushan Shah
InputRedirection connects to lockStateChanged to udate the current
pointer window. This way we can ensure that the current pointer
surface gets reset as soon as the screen locks (c.f. the expect
fail in the autotest) and also that it restores to the surface under
the mouse once the screen is unlocked.
The relevant code was not yet lock screen aware and performed an
early exit. Part of the code was fine, e.g. findToplevel is lock
screen aware. So this change adjusts the methods for updating the
internal window and decoration to be lock screen aware, that is they
get reset. With that updatePointerWindow is also lock screen aware.
Thus the LockScreenFilter can also use updatePointerWindow just like
the normal handling and does not need to reimplement parts of it. As
it now relies on other code being correct it has an additional check
to verify that the current pointer surface is a surface which is allowed
to get events. If it isn't the events are not forwarded.
Reviewed-By: Bhushan Shah
The main motivation of this change is to remove the spaghetti code
in the input event handling. Each area of processing (e.g. lock screen)
is moved into a dedicated event filter. Processing the events now just
means calling a virtual method on each of the filters. As soon as the
method returns true, the processing is stopped.
This allows to have the security for the lock screen just in one place:
whenever the screen is locked the event filter can ensure that the events
are not further processed.
Currently all event filters are implemented directly in input.cpp and
are registered by InputRedirection itself. In future it would be better
to have those moved to the area they belong to and get registered from
there. E.g. the input filter for EffectsHandlerImpl should be created
by EffectsHandlerImpl. This requires an improved API to ensure that the
filters are installed in the correct sequence.
If the screen is locked and no lock screen is shown yet we unset
the focused keyboard surface on key event. Similar we restore when
screen is unlocked.
This should hopefully fix the broken lockscreen unit test which hits
the special condition as the greeter doesn't show up on build.kde.org.
InputRedirection uses the inputTransformation() to pass to SeatInterface
for focused pointer surface. This prepares for proper input
transformation including scaling and rotation.
The implementation of VirtualTerminal is too linux specific and doesn't
compile on e.g. freebsd. Currently the most usage is in combination with
libinput. Only usage is:
* libinput related functionality in InputRedirection
* backends without custom input handling
Thus binding the feature to whether libinput is available is currently
the least invasive approach to get it compile on non-Linux.
In the long run this needs a different solution. The functionality
provided by VirtualTerminal is required and without the backends don't
work. It's needed to get notified about VT switches, when KWin needs to
stop rendering. So a solution for non-Linux needs to be found if
non-Linux wants to provide Wayland in future.
REVIEW: 126182
We need to update the pointer position, also if the screen is locked.
Otherwise the pointer doesn't move on the locked screen with libinput.
In addition we need to use the m_globalPointer for finding the correct
lock screen window as updatePointerWindow also does sanity checking on
the coordinates.
Also we need to introduce security checks where we use the signal.
REVIEW: 126103
When screen is locked,
- No window other then screenlocker or inputmethods gets rendered
- Only screenlocker gets keyboard events
- Only screenlocker and inputmethods get mouse events
Things that are not secured/tested are :
- Touch events
- Global shortcuts for screenlocker
- Fallback/emergency screen not yet working
REVIEW: 126015
The way it was implemented it allowed an X11 unmanaged window to become
a key logger. Basically as soon as there was an unmanaged window it got
all key events. This problem was discovered through the xembed-sni-proxy
which broke key input to all Wayland windows in a Plasma/Wayland session.
With this change Unmanaged windows don't get any key events at all. This
might break some applications as e.g. context menus are using override
redirect windows. A test with Qt applications shows that the menus are
still functional and the events are delivered correctly internally.
If applications show problems with this change, we might need to weaken
the restriction.
The Connection thread fills the event queue, it gets read from the
main thread. In order to properly support the threaded approach the
setup is changed to delegate into the own thread.
Properly handle the mouse press/release events in InputRedirection
while we move windows. If it's the last mouse release event we end
the move resize of the window. For that we reuse the code written
in Client.
Martin Gräßlin
Bhushan Shah