There are use cases for the headers to be used, e.g. when implementing
wayland-specific workflows from an Effect.
In order to be able to use these, we also need to expose libkwin to be
imported as it carries the interfaces' symbols.
Signed-off-by: Victoria Fischer <victoria.fischer@mbition.io>
Static plugins being linked with shared libraries doesn't work smoothly.
POSITION_INDEPENDENT_CODE target property needs to be set for
every such plugin. However, there can be targets that we have no control
over, which need POSITION_INDEPENDENT_CODE to be set too. Asking such
projects to add this property and them refusing to do so would be
reasonable because kwin's entire static plugin pipeline is weird.
The test framework was made a shared library because kwin build
directory used to get really big (in 10s of gigabytes).
Due to the -fPIC issues, this change makes the test framework a static
lib again. Obviously, this brings back the big build directory problem.
But, it's not as terrible as it used to be. With this change, kwin build
directory is a couple of gigabytes in debug build, which is a lot but
not as bad as it used to be before.
The InputConfig was introduced to handle tests that don't have valid
kwinApp object. Those tests have been either reworked or dropped, so
let's drop the InputConfig to tidy the config stuff.
Interactions with KScreen cause settings to be reset to the default when the lid
gets closed, so handling lid closing in KWin needs to wait until KScreen no longer
writes configs
BUG: 471282
This change ports the drm backend to the GraphicsBuffer and
GraphicsBufferAllocator.
The main motivation is to unify graphics buffer abstractions across
various backends and to prepare it for output layers, which could be
nicer if we could have direct control over the buffers.
This change makes the EffectsHandler emit the windowAdded signal as soon
as the Workspace::windowAdded signal is received and minimize the
difference between the states observed in kwin core and effects.
The test is flaky. We work around it by passing --repeat until-pass, but
lately even it doesn't help us with this issue.
So skip the test when running it inside KDE CI for a few months until
it gets a newer kernel.
It makes decorations lighter as they don't need to bundle its
configuration logic.
Deprecates the kcmodule property in favour of kcmoduleName which instead
of assuming that the kcm is local to the plugin, it provides the plugin
name to find and load.
The Effects API has one signal screenEdgeApproaching with effects
monitoring the last signal to stay in sync.
If an Edge is destroyed whilst visible, effects currently do not get
notified and it can stay there forever.
This patch emits screenEdgeApproaching if appropriate with a factor of 0
when an edge is destroyed.
BUG: 403354
StrutsTest::testLeftScreenSmallerBottomAligned() used to pass because
one of the previous tests changes the maximize policy to maximize.
A window will be maximized only if it's larger than the maximize area
but smaller than the screen area. That's not the case and the test makes
an incorrect assumption about how X11Windows are placed.
Make the total padding between windows equal to the padding against the
screen borders by applying just half of it on the common borders.
Adjust tests accordingly.
BUG: 469720
FIXED-IN: 5.27.6
Use CursorSource::image() instead.
Cursor caching in the ScreenCastStream has been changed so
QImage::cacheKey() is not being used. This is rather a preparation for
making kwin grab the contents of the cursor scene.
This change introduces InputDevice::pointerFrame(). The main motivation
behind it is to allow batching multiple pointer events within a single
event frame.
BUG: 454428
Qt requires xcb-icccm 0.3.9. On the other hand, 0.3.9 contains all the
types and functions used by kwin, so remove the corresponding
XCB_ICCM_FOUND checks to simplify the code.
The quick tile test waits 1s to ensure that the quick tile combine timer
is not active. On the other hand, if the active window changes, it makes
sense to reset quick tile combine status. That also lets us get rid of
the QTest::qWait() in QuickTilingTest::testShortcut().
The test used to verify that kwin doesn't crash when ShellClient sets
new geometry. ShellClient used to access decoration borders without
checking whether decoration() is null.
On the other hand, we've added a bunch of new test cases in
testXdgShellWindow, so let's remove this one. If the bug is back, that
test will fail.
This makes it possible to ensure type safety for enums, as each drm property
object can have its own type now, and it reduces the amount of typing needed
to access properties
The test needs OpenGL in order to ensure that the window view effect is
loaded and it registers a screen edge. On the other hand, we could
register a screen edge ourselves and thus allow running the test on
freebsd.
testXdgShellWindow already tests intricate subsurface size changes. The
surface pixmaps are handled differently now too, so the test is not
useful as it used to be 4 or 5 years ago.
Before Deleted merge, it used to be equivalent to waiting until the
window is closed.
This fixes tests waiting until the window closing animation completes
and the Window object is destroyed.
testPointerInput requires OpenGL compositing because it wants to test
cursor push back in the window view effect and the window view effect is
available only if OpenGL is supported.
On the other hand, ScreenEdgesTest::testPushBack() already tests similar
scenario, so let's drop relevant test in testPointerInput to allow
running it when using software rendering.
The virtual backend uses the surfaceless platform. On the other hand, we
move in a direction where the graphics buffer type is explicit, which
creates issues for the virtual backend.
This change ports the virtual backend to gbm so we could manually
allocate dmabuf buffers in order to unify buffer handling in kwin.
Its main drawback is that you won't be able to use the virtual backend
on setups without render nodes. On the other hand, given that the
compositor is meaningless without clients being able to share buffers
with it, it's reasonable to require some way to create and export prime
buffers.
WAYLAND_ONLY will create two tests: one with Xwayland, the other without
Xwayland. This is somewhat wasteful and it results in higher CI times.
On the other hand, Xwayland is started on demand. If a test doesn't need
Xwayland, it won't start.
So let's remove WAYLAND_ONLY in order to lighten kwin on CI resources.
If wayland only tests are needed, we can consider passing the operation
mode to the WAYLANDTEST_MAIN helper, but there aren't such tests afaik.
Window::desktop() is obsolete. On the other hand, X11 doesn't support
having a window on several virtual desktops, so we still need it. As a
compromise, this change moves it to X11Window instead.
Currently, managed and override-redirect windows are split in two types:
X11Window and Unmanaged. While looking at it strictly from type
perspective, this is great. But it creates other problems, e.g. we need
to put shared X11-specific code in the base Window class or mess with
"base" classes.
As an alternative solution, this change merges the Unmanaged class into
the X11Window class and disables some functionality based on the value
of isUnmanaged().
X11Window::manage() is used to create a managed Window. X11Window::track()
is used to create an unmanaged Window.
Currently, the normal window lifecycle looks as follows: create Window,
wait until it's shown, add it to Workspace, wait until it's closed,
create a Deleted, copy properties from the original window to the
deleted one, destroy the original window, wait until the last deleted
window reference is dropped.
There are a couple of issues with this design: we can't nicely
encapsulate X11 or Wayland specific implementation details if they need
to be accessed for closed windows; manual copying of properties is
cumbersome and error prone and we've had a dozen of cases where effects
worked incorrectly because some properties had not been copied.
The goal of this patch is to drop Deleted and extend the lifetime of the
original window, but with a special state set: Window::isDeleted().
The main danger is that somebody can try to do something with deleted
windows that they should not do, but on the other hand, such code needs
to be guarded with relevant checks too.
Workspace::outputAt() casts vectors to four rectangle corners and uses
the shortest one to decide which output is the closest to the given
point.
This works poorly on dual monitor setups where on the left side you have
a monitor with landscape orientation and one with portrait orientation
on the right hand side. In that case, outputAt() will prefer the left
monitor even though the right monitor is the closest one if you cast a
perpendicular from the given point to the right monitor.
In order to improve the handling of that case, this change makes
Workspace::outputAt() compute the closest point to the output geometry
rectangle and use the squared distance as the score.
The indirection contributes unnecessary complexity. The usage of
std::weak_ptr and std::shared_ptr complicates the things further, e.g.
---
Ideally, same should be done with TabBox and TabBoxHandler, but that can be done in another MR.
Tabbox supports two operation modes: switching between windows and
desktops. Switching between windows is more commonly used. Desktop
switching is not exposed in user settings and it requires some advanced
knowledge of kwin's internals to enable it.
On the other hand, over the past years, we've double downed on effects
like desktop grid and overview to provide graphical means to switch
between virtual desktops.
This change drops desktop switching because it's effectively unused to
simplify the tabbox code, which can be very handy for the future
refactorings of window switching.
Currently windows are scattered in a few separate lists. If you need to
go through the windows, you have to do it piece by piece. On the other
hand, with the overhaul of window types, we've started converging
towards one universal type: Window. Keeping windows in the separate
buckets goes against this design.
Workspace::stackingOrder() already contains all windows. This change
repurposes Workspace::allClientList() from a list of "normal" windows to
all windows, i.e. Workspace::windows(), to be consistent.
There's one API change though. Scripting API will expose other window
types too. This is an intentional change so scripted effects could
operate with all windows. It also matches the current behavior observed
in libkwineffects, which exposes all windows as well.
Instead of having every DrmProperty store pending values, store the data
for the next commit in a separate and temporary type. This simplifies the
code and makes it possible to do commits in a separate thread
Currently Deleted are destroyed with a delay to avoid dangling pointers
within the middle of painting.
On the other hand, it's reasonable to require not to delete windows when
kwin starts painting the screen.
Over the years, we refactored how deleted windows. They are always
unreferenced after finishing the current frame. So it should be fine to
destroy Deleted immediately now.
Qt::AA_UseHighDpiPixmaps has no any effect now.
We used to rely on the fact that Qt::AA_UseHighDpiPixmaps is disabled by
default in Qt 5 in kwin_x11. It's not clear what to do about it now.
It doesn't test anything useful. It's hard to make it test useful things
too due to needing to change the system time. Linking with it also
breaks the encapsulation and it won't work when using MODULE library.
Commit 88cf8355 changed the behaviour of Mali (Lima) / PinePhone devices by disabling GLSL
88cf8355 got backported in 5.27.1 and broke PinePhone devices (White rectangle on topright quarter of a black screen)
This patch restores the behaviour of 5.27.0
VirtualDesktopManager::currentChanged() can be emitted with
previousDesktop being null/0. It can happen only on startup. After that,
it's always valid. In order to make the code that uses the
currentChanged signal less trickier, ensure that the current virtual
desktop is initialized when the virtual desktop config is loaded.
This merges Window::clientMinimized() and Window::clientUnminimized()
signals with the Window::minimizedChanged() signal to simplify some
code.
The avoid animation flag has been dropped because its main usage is to
avoid playing animation during window initialization. But it's not possible
to trigger minimize animation at that moment. API-wise it's better to avoid
having such flags too and rely on surface role to decide whether to play a
given animation.
This signal exists as a convenience helper, but it's not always emitted
as it's advertised to work. Instead of fixing it, let's drop the signal
to simplify virtual desktop code. Its effects can be accomplished by
monitoring Window::desktopChanged() and VirtualDesktopManager::currentChanged()
signals in effects and scripts where needed.
Checking if the pointer needs to be confined before doing `std::clamp` on
the position is useless and causes problems. With this change, the pointer
will always be confined to exactly the screen, without any fractional offsets.
BUG: 461911
Use input device specific apis to change the position of the cursor. The
main reason to do so is to break the assumption that Cursor position is
the same as pointer position, which I would like to rely on later to
merge tablet and pointer cursors.
Even though the names seem to match, QtWayland maps button values
to enum values in ascendung order (as it does on X11). The wrong
mapping is usually not a problem because we send the native button
events to clients. However when the Qt names or values are used
for communication between KWin and a client this leads to
misunderstandings.
BUG:465463
FIXED-IN:5.27.1
This installs a socket notifier onto our xwayland socket, when a user
connects we launch xwayland. The client then connections once kwin has
established itself as the compositor.
For a full desktop plasma session this patch effectively does nothing
too useful as we still start kcminit and make xrdb calls on startup
which in turn will launch X, but for the same reason this patch should
be harmless now as we're still processing the xrdb calls before any
clients will connect.
Before
Executed in 10.79 secs fish external
usr time 4.85 secs 918.00 micros 4.85 secs
sys time 0.62 secs 119.00 micros 0.62 secs
After
Executed in 1.45 secs fish external
usr time 941.63 millis 486.00 micros 941.14 millis
sys time 578.99 millis 62.00 micros 578.93 millis
Before
Executed in 17.39 secs fish external
usr time 422.91 millis 1.92 millis 420.99 millis
sys time 586.41 millis 0.22 millis 586.19 millis
After
Executed in 1.23 secs fish external
usr time 395.63 millis 0.00 millis 395.63 millis
sys time 228.55 millis 2.13 millis 226.42 millis
The test waits longer than needed to determine whether effects stopped
playing animations
Before
Executed in 22.41 secs fish external
usr time 997.42 millis 722.00 micros 996.70 millis
sys time 744.81 millis 79.00 micros 744.73 millis
After
Executed in 9.71 secs fish external
usr time 894.42 millis 1.03 millis 893.38 millis
sys time 689.80 millis 0.11 millis 689.69 millis
testLockScreen uses !eventSpy.wait() expression to check whether the
given signal is emitted. QSignalSpy::wait()'s default timeout is 5s,
which is too much. We don't need to wait that much, in fact, even
wl_display_sync() would suffice. In either case, let's pass explicit
timeout of 10ms which should be good enough.
Before
Executed in 60.49 secs fish external
usr time 3.92 secs 1.03 millis 3.92 secs
sys time 2.29 secs 0.16 millis 2.29 secs
After
Executed in 19.38 secs fish external
usr time 1.64 secs 347.00 micros 1.64 secs
sys time 1.15 secs 54.00 micros 1.15 secs
* Allow to do quick tiling to custom tile geometries, windows will be snapped to tiles when dragged with the shift modifier pressed.
* Tile geometries are screen specific.
* The global shortcut Meta+T will trigger a fullscreen configuration ui as a QML effect for the tiles which allows to add, remove and resize tiles
* UI and behavior is a bit similar to the Windows Fancy Zones addon: https://docs.microsoft.com/en-us/windows/powertoys/fancyzones
* Its main scope is to help the workflow with very big monitors, especially ultra wide ones, where most application don't make sense maximized to the full screen (eventually also support games to be full screened to a given tile instead of the whole screen)
* it should get also some bindings for scripting, as its ain goal is not to replicate other popular tiling window managers, but should give the popular kwin tiling scripts to have a more robust infrastructure
* it will eventually get support for a set of predefined layouts, but this is for a second phase
BUG: 438788
We use KWIN_NAME, KWIN_INTERNAL_NAME_X11 and KWIN_INTERNAL_WAYLAND
properly only in a few places. In other, we use hardcoded names.
Let's not bother and hardcode kwin names everywhere rather than have one
foot in and one foot out, it's simpler.
Due to being a compositor, kwin has to conform to some certain
interfaces. It means a lot of virtual functions and function tables to
integrate with C APIs. Naturally, we not always want to use every
argument in such functions.
Since we get -Wunused-parameter from -Wall, we have to plumb those
unused arguments in order to suppress compiler warnings at the moment.
However, I don't think that extra work is worth it. We cannot change or
alter prototypes in any way to fix the warning the desired way. Q_UNUSED
and similar macros are not good indicators of whether an argument is
used too, we tend to overlook putting or removing those macros. I've
also noticed that Q_UNUSED are not used to guide us with the removal no
longer needed parameters.
Therefore, I think it's worth adding -Wno-unused-parameter compiler
option to stop the compiler producing warnings about unused parameters.
It changes nothing except that we don't need to put Q_UNUSED anymore,
which can be really cumbersome sometimes. Note that it doesn't affect
unused variables, you'll still get a -Wunused-variable compiler warning
if a variable is unused.
Makes it possible to follow the dbus service for locale configuration,
making it possible to have third parties integrate with Plasma.
This is done opt-in for now, it can be adopted generally in the future,
if necessary.
The entire approach behind that test is wrong. Even slight changes in
one pixel can make the test fail. There's no human-friendly way to
determine what the difference between reference and actual frames are.
The reference frame is generated in code rather than from an image on
the disk, which makes the tests even more error-prone.
The test is flaky too and constantly blocks merging patch submissions.
We need a different way to test rendering results, not what
testSceneQPainter currently does. Normally, I would say that we need
to add such an infrastructure before dropping the test, but it makes the
merging process so inconvenient and the test doesn't really bring any
value, it tests really basic cases which are noticeable right away after
launching kwin.
With the addition of KGlobalAccel::globalShortcutActiveChanged(),
there's a way to monitor all triggered actions without adding any
boilerplate code.
This change removes the KGlobalAccel bits from the Platform class. This
reduces the amount of boilerplate code; currently we need to be careful
about ensuring that Platform::setupActionForKGlobalAccel() is called.
Another reason behind this change is to simplify the api of Platform and
remove things that don't have anything to do with outputs.
Use isNull on QSizeF to check for a zero delta instead of comparing it
with a default-constructed QSizeF, which in practice initializes to
(-1.0,-1.0). This caused relative motion events to be omitted if the
delta happened to be equal to (-1.0,-1.0), causing mouse jumping in some
applications.
BUG: 444510
Signed-off-by: John Brooks <john@fastquake.com>
The signal will be emitted if text-input-v3 applies a new enable state
if it's already enabled.
In Gtk's implementation, sending text_input_v3.enable for the second
time has the similar semantics of text-input-v2.showInputPanel. Try to
connect them together and add corresponding test.
See also: a95cfb1c45/gtk/gtkimcontextwayland.c (L555)
Writing is what our guard is supposed to prevent, let's try to mmap it
for writing. It seems FreeBSD still allows PROT_READ for sealed-for-writing
mmaps.
When text input gets enabled, the sequence should be
1. enable
2. request to show the panel
3. panel gets added and shown
And showing the panel again after hiding is an important case to cover as
well.
QXkbCommon fails to map `Ё` to `~` when pressing `Shift+grave` and
sequenceEdit seems to have focus now.
On the other hand, does it make sense to bind global shortcuts with key
sequences such as "Shift+`"? The system settings doesn't even allow typing
in such key sequences.
testDbusInterface fails with Qt 6 build because QCOMPARE detects the
type difference between QString and QByteArray.
Since these properties being byte arrays has caused reasonable amount of
discomfort, let's make them QStrings. It will fix the test and make some
scripted effect code more cleaner.
PointerInputTest::testWarpingBetweenWindows() compares
QVariant(QPointF) with QPoint. While this works in Qt 5, Qt 6 seems to
take the type into account when comparing the two.
TestXdgShellWindow::testPointerInputTransform() compares
QVariant(QPointF) with QPoint. While this works in Qt 5, Qt 6 seems to
take the type into account when comparing the two.
In some cases, stateCommitted may fire after a new surface being focused
and in correctly set input method to be inactive. This can be reproduced
by switch client between an active text input v3 client and text input
v2 client.
The Workspace requires at least one output and no output has 0x0 size.
This fixes a crash in confineToBoundingBox() where qBound() asserts that
the min value is less than the max value, in our case the x coordinate
of the left edge of the workspace is less than the x coordinate of the
right edge of the workspace.
Client-side wrappers for input-method-unstable-v1 fail to build because
wl_keyboard_interface is referenced in the header file generated by
wayland-scanner.
Unfortunately, qt6_generate_wayland_protocol_client_sources() forces
--include-core-only argument, this is addressed in Qt 6.4.1, but in
meanwhile let's ship a copy of Qt6WaylandClientMacros.cmake file until
the required Qt version is out.
In libinput 1.19, three new pointer axis events were added in order to
provide support for high-resolution scrolling.
LIBINPUT_EVENT_POINTER_AXIS is de-facto deprecated and new users of
libinput should use instead SCROLL_WHEEL, SCROLL_FINGER, and
SCROLL_CONTINUOUS.
Discrete deltas were replaced with v120 delta values. 120 corresponds to
a single discrete delta. Smaller values correspond to "partial" wheel
ticks.
https://gitlab.freedesktop.org/wayland/wayland/-/merge_requests/72
This change extends the OutputChangesTest so it also covers the cases
where a maximized and a fullscreen window is moved back to its original
output when it's hotplugged.
We use the PMF syntax so the isValid() check is unnecessary as the
compiler will notify about wrong signal at compile time. It makes
writing autotests feel less boilerplaty.
It adds more test cases in OutputChangesTest, particularly swapping
outputs.
Swapping outputs is an interesting case because outputs can temporarily
overlap so workspace()->outputAt() can return wrong output and the
window is going to stick to wrong output.
Currently the Workspace processes output updates as they occur, e.g.
when the drm backend scans connectors, the Workspace will handle
hotplugged outputs one by one or if an output configuration changes the
mode of several outputs, the workspace will process output layout
updates one by one instead of handling it in one pass. The main reason
for the current behavior is simplicity.
However, that can create issues because it's possible that the output
layout will be temporarily in degenerate state and features such as
sticking windows to their outputs will be broken.
In order to fix that, this change makes the Workspace process batched
output updates. There are several challenges - disconnected outputs have
to be alive when the outputsQueried signal is emitted, the workspace
needs to determine what outputs have been added or removed on its own.
Separate trigger progress and semantic progress in gesture.
Move effect activation and desktop switching over to semantic progress.
Allow semantic progress to exceed 1 for overshoot in animations.
I've added VerticalAxis, HorizontalAxis, DirectionlessSwipe and BiDirectionalPinch gestures directions.
These are all combinations of other gesture directions that semantically work well together.
I've implemented these gestures as well as changed some labels and improved documentation,
Also,
Add vector signal to SwipeGesture
- Now only 1 GestureDirection enum
- Now only 1 registerGesture() call
- The 4 kinds of gesture (Pinch/Swipe) and (Touchpad/Touchscreen) in globalshortcuts.h/cpp are merged into 1 GestureShortcut
- Change from range to set of finger counts in gestures
No behavior should change, just a refactor.
This might be the root cause of random ASAN errors in testQuickTiling.
From commit 617291c6974d232ee99c4c49e891ce16863e3d6e:
The internal EventQueue is a child of the registry object. This means
that after the registry is destroyed, all proxy objects in that event
queue are going to have invalid reference to it, which is not a problem
as long as the wl_display_dispatch() function is not called.
The wl_display_dispatch() function uses wl_proxy's queue reference to
enqueue incoming events to that queue.
Unfortunately, during teardown, the internal ConnectionThread may
dispatch events right after the registry object has been destroyed,
which can lead to a crash.
In order to fix the crash, we need to destroy all proxy objects and only
after that we can destroy the event queue. It's okay if wayland events
are dispatched in between.
i.e. the EventQueue object must be destroyed last to ensure avoid hitting
dangling pointers.
This enables again the crossfade between the old window picture and the new one in the maximize and morphingpopup effects.
It does that with the OffScreenEffect redirect() feature.
BUG:439689
BUG:435423
Things such as Output, InputDevice and so on are made to be
multi-purpose. In order to make this separation more clear, this change
moves that code in the core directory. Some things still link to the
abstraction level above (kwin), they can be tackled in future refactors.
Ideally code in core/ should depend either on other code in core/ or
system libs.
Other policy enums are declared in options.h so let's do the same for
placement policy. Besides consistency, another advantage of moving the
enum in kwin namespace is that the enum could be forward declared.
An application that does not support text-input has no way of
communicating with the input method, so even if you show the input
method the application receives nothing. As a fallback, instead send
fake key events so the application still gets something at least.
The key events are synthesised based on the text string that the
input method sends, which may result in things that do not actually
correspond to real keys. Unfortunately I do not see a way around that.
CCBUG: 439911
Currently, the main user of these two functions is the X11 standalone
platform.
This change ports that code to Workspace::geometry(), which is not great
but the X11 backend already depends on the Workspace indirectly via the
Screens. Not sure if it's worth making the standalone X11 backend track
the xinerama rect internally.
This class can be used to create an anonymous file, for instance
to pass data between compositor and clients, through means of a
file descriptor, as is done in various Wayland protocols, notably
the keymap exchange.
It also implements sealing the file, so that it can be shared
between multiple clients without them being able to modify it.
If supported, memfd_create is used, otherwise a `QTemporaryFile`
is used.
Signed-off-by: Victoria Fischer <victoria.fischer@mbition.io>
Since the screen number is well-known, we can look up the default
screen on demand. Note that xcb_get_setup() is pretty cheap as it
simply returns a const pointer to pre-allocated data.
At the moment, a platform should provide two output lists - one that
lists all available outputs, and the other one that contains only
enabled outputs. In general, this amounts to some boilerplate code and
forces backends to be implemented in some certain way, which sometimes
is inconvenient, e.g. if an output is disabled or enabled, it will be
simpler if we only change Output::isEnabled(), otherwise we need to
start accounting for corner cases such as the order in which
Output::isEnabled() and Platform::enabledOutputs() are changed, etc.
In X11 when a window is maximised if the client is unable to fufill the
space provided we centre align the window.
With the new floating point geometry behaviour of centreing changes.
Instead of a 1 pixel gap at the top, we get a 0.5 pixel gap either side.
When we get into the codepath to "fix" the window in `closeHeight` we
only move the top, giving us an invalid buffer size.
We don't really want to change the logic here; on xwayland with the
scaling opt-out path it's feasible for a floating sized logical size to
still be representable. This code rounds to the native unit after all
the logic has taken effect.
It's important for tablet devices to be able to specify to which section
of the display we'll be fitting the tablet. This setting allows to
specify this by providing some options that will do so relative to the
output size.
CCBUG: 433045
The Session can be useful not only to the platform backend but also
input backends and for things such as vt switching, etc. Therefore it's
better to have the Application own the Session.
Platform backends are provided as plugins. This is great for
extensibility, but the disadvantages of this design outweigh the
benefits.
The number of backends will be limited, it's safe to say that we will
have to maintain three backends for many years to come - kms/drm,
virtual, and wayland. The plugin system adds unnecessary complexity.
Startup logic is affected too. At the moment, platform backends provide
the session object, which is awkward as it starts adding dependencies
between backends. It will be nicer if the session is created depending
on the loaded session type.
In some cases, wayland code needs to talk to the backend directly, e.g.
for drm leasing, etc. With the plugin architecture it's hard to do that.
Not impossible though, we can approach it as in Qt 6, but it's still
harder than linking the code directly.
Of course, the main disadvantage of shipping backends in a lib is that
you will need to patch kwin if you need a custom platform, however such
cases will be rare.
Despite that disadvantage, I still think that it's a step in the right
direction where the goal is to have multi-purpose backends and other
reusable components of kwin.
The legacy X11 standalone platform is linked directly to kwin_x11
executable, while the remaining backends are linked to libkwin.
The original intention behind creating plugins before the workspace was
to handle the case where kwin_wayland may need to wait until outputs are
available. However, since things have changed a lot in that regard,
plugins can be loaded after the workspace now.
The main benefit behind this is that plugins can be simpler, they won't
need to track when the workspace is created.
On X11, plugins are already loaded after the workspace is instantiated.
This change adjusts the window management abstractions in kwin for the
drm backend providing more than just "desktop" outputs.
Besides that, it has other potential benefits - for example, the
Workspace could start managing allocation of the placeholder output by
itself, thus leading to some simplifications in the drm backend. Another
is that it lets us move wayland code from the drm backend.
We gain nothing with it. XCB setup logic in the Xwayland server has to
be moved to the workspace layer anyway. For example, this move of
responsibilities will be needed to support running more than just one
instance of Xwayland. Architecture-wise, it would be cleaner too.
Unfortunately, it breaks encapsulation of the Application, but this can
be taken care later.
