Previously, different rects were used to restore tiled and maximized
windows, creating edge cases for inconsistent restoration. All states
now restore m_floating_rect, which saves the last valid size and
location of a window while free-floating.
The point of a reference type is to behave just like the referred-to
type. So, a Foo& should behave just like a Foo.
In these cases, we had a const Vector. If it was a const Vector of Foo,
iterating over the Vector would only permit taking const references to
the individual Foos.
However, we had a const Vector of Foo&. The behavior should not
change. We should still only be permitted to take const references to
the individual Foos. Otherwise, we would be allowed to mutate the
individual Foos, which would mutate the elements of the const Vector.
This wouldn't modify the stored pointers, but it would modify the
objects that the references refer to. Since references should be
transparent, this should not be legal.
So it should be impossible to get mutable references into a const
Vector. Since we need mutable references in these cases to call the
mutating member functions, we need to mark the Vector as mutable as
well.
This change unfortunately cannot be atomically made without a single
commit changing everything.
Most of the important changes are in LibIPC/Connection.cpp,
LibIPC/ServerConnection.cpp and LibCore/LocalServer.cpp.
The notable changes are:
- IPCCompiler now generates the decode and decode_message functions such
that they take a Core::Stream::LocalSocket instead of the socket fd.
- IPC::Decoder now uses the receive_fd method of LocalSocket instead of
doing system calls directly on the fd.
- IPC::ConnectionBase and related classes now use the Stream API
functions.
- IPC::ServerConnection no longer constructs the socket itself; instead,
a convenience macro, IPC_CLIENT_CONNECTION, is used in place of
C_OBJECT and will generate a static try_create factory function for
the ServerConnection subclass. The subclass is now responsible for
passing the socket constructed in this function to its
ServerConnection base; the socket is passed as the first argument to
the constructor (as a NonnullOwnPtr<Core::Stream::LocalServer>) before
any other arguments.
- The functionality regarding taking over sockets from SystemServer has
been moved to LibIPC/SystemServerTakeover.cpp. The Core::LocalSocket
implementation of this functionality hasn't been deleted due to my
intention of removing this class in the near future and to reduce
noise on this (already quite noisy) PR.
Briefly flash the menubar menu containing the keyboard shortcut action
to give the user immediate visual feedback on their interaction with the
system.
I encountered a WindowServer crash due to null-pointer dereference in
this function, so let's protect against it by simply skipping over
nulled-out WeakPtrs.
I added a FIXME about how we ideally wouldn't be in this situation in
the first place, but that will require some more investigation.
This encapsulates what our multi-client IPC servers typically do on
startup:
1. Create a Core::LocalServer
2. Take over a listening socket file descriptor from SystemServer
3. Set up an accept handler for incoming connections
IPC::MultiServer does all this for you! All you have to do is provide
the relevant client connection type as a template argument.
These ones all manage their storage internally, whereas the WebContent
and ImageDecoder ones require the caller to manage their lifetime. This
distinction is not obvious to the user without looking through the code,
so an API that makes this clearer would be nice.
Everyone used this hook in the same way: immediately accept() on the
socket and then do something with the newly accepted fd.
This patch simplifies the hook by having LocalServer do the accepting
automatically.
235f39e449 secretly added an if check that ignores all the files that
couldn't be loaded into bitmaps. Unfortunately, we send an empty path
as a way to unset the wallpaper, which meant that we couldn't go back
to the default background color anymore.
With this change, System::foo() becomes Core::System::foo().
Since LibCore builds on other systems than SerenityOS, we now have to
make sure that wrappers work with just a standard C library underneath.
Render the window switcher with the same background and shadow as other
WindowServer overlays.
Note that we don't actually render it as a WindowServer::Overlay, as the
window switcher uses mouse and keyboard events, and there's currently
no way for an overlay to receive events.
This fixes an issue with menus not being immediately "usable" with the
the right mouse button after opening a context menu.
The bug was that we were clearing a local pointer to the active input
tracking window instead of the pointer in WindowStack.
We need to make sure the menu was pushed to the open menu stack before
calling set_visible, as this may trigger cursor re-evaluation, which
in turn expects the menu to be considered open.
Fixes#10836
This reverts commit 239520ae54.
The call to set_visible() is not redundant. Removing the call leads
to the "start" button in the taskbar not being painted as "pressed" even
when it is.
We've already returned early if the menu is open, so there's no need to
verify that it isn't present in the stack of open menus before pushing
it onto said stack.
WindowServer returns {} on non-existing screen index,
however shot program hangs instead of retriving an empty
ShareableBitmap. With this change, the function returns an empty
ShareableBitmap and shot exits gracefully.
This makes the cursor update properly if it was above the window
switcher while it was visible, and something underneath it wants to use
something other than the default arrow cursor.
...to reevaluate_hover_state_for_window(). This name is not super great
either, but at least it doesn't sound like the window is necessarily
currently being hovered.
In 2e6bb987a3 the "did_construct" API in
Core::Object was removed, since it had only one user. For a replacement,
the Window would manually call the frame's "frame_was_constructed"
method. However, WindowServer::Window has two constructors, and only one
of them called this method. This caused windows to spawn without
buttons and various other breakage that spawned from this.
Derivatives of Core::Object should be constructed through
ClassName::construct(), to avoid handling ref-counted objects with
refcount zero. Fixing the visibility means that misuses like this are
more difficult.
This commit is separate from the other Servives changes because it
required additional adaption of the code. Note that the old code did
precisely what these changes try to prevent: Create and handle a
ref-counted object with a refcount of zero.
Derivatives of Core::Object should be constructed through
ClassName::construct(), to avoid handling ref-counted objects with
refcount zero. Fixing the visibility means that misuses like this are
more difficult.
There is also make_ref_counted(), which does not call did_construct(),
so the method was not guaranteed to be run. Since there is only a single
user, and `WindowServer::Window` is a final class anyway (so there is no
need to separate the constructor and post-constructor phases), let's get
rid of this concept.
(The following commits reduce the opportunities to call
make_ref_counted, but still.)
We create a base class called GenericFramebufferDevice, which defines
all the virtual functions that must be implemented by a
FramebufferDevice. Then, we make the VirtIO FramebufferDevice and other
FramebufferDevice implementations inherit from it.
The most important consequence of rearranging the classes is that we now
have one IOCTL method, so all drivers should be committed to not
override the IOCTL method or make their own IOCTLs of FramebufferDevice.
All graphical IOCTLs are known to all FramebufferDevices, and it's up to
the specific implementation whether to support them or discard them (so
we require extensive usage of KResult and KResultOr, together with
virtual characteristic functions).
As a result, the interface is much cleaner and understandable to read.
The Screen constructor already calls open_device(), so there's no need
to call it again right after creating a new Screen in apply_layout().
This makes the first screen compose happen ~50ms earlier on my machine.
This function only did one thing: call Screen::set_resolution().
We always call that function when opening the underlying device anyway,
so this was completely redundant.
This makes the first screen compose happen ~60ms earlier on my machine.
Add option to reverse primary and secondary buttons in Mouse Settings.
- WindowServer.ini: add default entry
- switch-mouse-buttons.png: new icon for settings entry
- Mouse.gml/MouseWidget.*: new settings dialog
- ClientConnection/WindowManager/Server: window message for settings
- EventLoop.cpp: swap buttons 1 and 2 if settings are on
If a mouse button was clicked, `EventLoop::drain_mouse()` would always
send the last MousePacket state to the screen input - even if that
state is equivalent to the last state sent as part of the button logic.
By remembering if the state was already sent, we prevent sending that
state a second time saving some resources in the process.
Currently, if there are not titlebar buttons, we fail to paint the title
because we treat the leftmost titlebar button as the empty rect. We will
now use the rightmost edge of the titlebar when there are no buttons.
This effectively makes us send a "mouse move" event to windows when they
become active, even if the mouse didn't actually move. By doing this, we
trigger hover/tooltip/etc logic immediately, instead of doing it on the
next 1px mouse movement.
It's a small detail but my goodness does it feel better this way. :^)
This fixes an issue for the magnifier that when the screen scaling is
increased to 2 the magnifier doesn't center around the cursor.
Since booting Serenity with multiple displays doesn't work at the moment
the rescaling is only added for the one display case.
This removes the awkward String::replace API which was the only String
API which mutated the String and replaces it with a new immutable
version that returns a new String with the replacements applied. This
also fixes a couple of UAFs that were caused by the use of this API.
As an optimization an equivalent StringView::replace API was also added
to remove an unnecessary String allocations in the format of:
`String { view }.replace(...);`
This allows any client to ask the WindowServer to give it the color
of the screen bitmap under the cursor.
There's currently no way to get the screen bitmap *without* the
cursor already drawn on it, so for now we just take a pixel
beside the actual cursor position to avoid just getting the cursors
color.
This feature was problematic for several reasons:
- Tracking *all* the user activity seems like a privacy nightmare.
- LibGUI actually only supports one globally tracking widget per window,
even if no window is necessary, or if multiple callbacks are desired.
- Widgets can easily get confused whether an event is actually directed
at it, or is actually just the result of global tracking.
The third item caused an issue where right-clicking CatDog opened two
context menus instead of one.
There are a few places in the system where this could be useful,
such as PixelPaint and the MandelBrot demo. It seems general enough
that it is probably useful to have it as a system-wide cursor rather
than loading it manually each time.
This can be used immediately in PixelPaint (separate commit), but
I am adding this as a system-wide cursor since it may also be useful
for other applications that want to use it.
Only one place used this argument and it was to hold on to a strong ref
for the object. Since we already do that now, there's no need to keep
this argument around since this can be easily captured.
This commit contains no changes.
Applets and windows would like to be able to know when the applet
area has been resized. For example, this happens asynchronously after
an applet has been resized, so we cannot then rely on the applet area
position synchronously after resizing. This adds a new message
applet_area_rect_changed and associated Event AppletAreaRectChange,
and the appropriate virtual functions.
Previously, when `screen_index` was not provided when calling
`ClientConnection::get_screen_bitmap`, the bitmap that was created
was always the size of the bounding rect of the screen. The actual
screen bitmap was being cropped, but the bitmap being returned was
of the original size with just black pixels everywhere else.
Now you can specify a CursorTheme key in /etc/WindowServer.ini. The
cursors are loaded from /res/cursor-themes/<name> directory. This
directory contains a Config.ini file with format similar to previous
Cursor section, except it uses relative paths.
This commit adds also Default theme, which uses cursors being
previously in /res/cursors.
The WidgetGallery is updated to match the new cursor path format.
Otherwise, we emit a menu_item_left to the WindowServer client even
though the mouse never left the menu item (as is the case when a
disabled menu item is clicked).
Due to a bug in Clang 12, the compilation would fail with an 'unexpected
end-of-file' error when it encounters some of the nested generic lambdas
in `Compositor.cpp`.
Co-authored-by: Peter Bindels <dascandy@gmail.com>
Since C99 and C++20 have a standardized syntax for designated
initializer, we should use that instead of this GCC-specific extension.
While this currently works both in Clang and GCC, the former emits a
warning for it, while the latter has an [issue] open that plans to
deprecate it.
[issue]: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88144
Currently, any number of menubars can be plugged in and out of a window.
This is unnecessary complexity, since we only need one menubar on a
window. This commit removes most of the logic for dynamically attaching
and detaching menubars and makes one menubar always available. The
menubar is only considered existent if it has at least a single menu in
it (in other words, an empty menubar will not be shown).
This commit additionally fixes a bug wherein menus added after a menubar
has been attached would not have their rects properly setup, and would
therefore appear glitched out on the top left corner of the menubar.
We were missing to account for areas that are not covered by any
window. If any of these areas are covered by an overlay we need to
render the wallpaper into transparency and also render the overlay
over them.
This fixes not rendering overlays properly when e.g. the FileManager
(desktop) crashed as there is no longer any window underneath.
When changing the theme, there were two Core::ConfigFile instances
(one class scoped -- m_config and one function scoped -- wm_config)
fighting over the file, resulting in not saving the new theme name
to the config. :^(
This makes WindowServer remember selected theme from the menu
after reboot!
If a screen layout cannot be applied, instead of failing to start
WindowServer try to fall back to an auto-generated screen layout with
the devices that are detected.
Also, be a bit smarter about changing the current screen layout.
Instead of closing all framebuffers and bringing them back up, keep
what we can and only change resolution on those that we need to change
them on. To make this work we also need to move away from using an
array of structures to hold compositor related per-screen data to
attaching it to the Screen itself, which makes re-using a screen much
simpler.
We only need to re-draw the item being selected and the item being
deselected. We also don't care anymore if applets were added or
removed as we no longer have a global menu bar.
We were re-rendering areas that were considered transparency areas even
though they weren't transparency areas or were occluded by opaque
areas.
In order to fix this, we need to be a bit smarter about what is above
and below any given window. Even though a window may have transparent
areas, if those are occluded by opaque window areas on top they are
not actually any areas that should be rendered at all. And the opposite
also applies, opaque window areas for windows below that are occluded
by transparent areas, do need to be rendered as transparency. This
solves the problem of unnecessary transparency areas.
The other problem is that we need to know what areas of a window's
dirty rectangles affect other windows, and where. Basically any
opaque area that is somehow below a transparent area that isn't
otherwise occluded, and any transparent area above any other window
area (transparent or opaque) needs to be marked dirty prior to
composing. This makes sure that all affected windows render these
areas in the correct order. To track these, we now have a map of
affected windows and the rectangles that are affected (because not all
of that window's transparency areas may be affected).
This implements window stealing in WindowServer, which allows clients
to mark a window they own as 'stealable' by another client. Indicating
that the other client may use it for any purpose.
This also updates set_window_parent_from_id so that the client must
first mark its window as stealable before allowing other clients to
use it as a parent.
If the device requires a flush and we modify the front buffer, we need
to flush those changes to the front buffer. This makes the flashing
work using the VirtIOGPU.
Also fix a minor bug where we flushed the front buffer instead of
the back buffer after flipping, which caused the VirtIOGPU to not work
as expected when using the SDL backend and disabling buffer flipping.
The windows in the background are ignored when the window is fullscreen.
However, we still would like to see the background if that window is
transparent.
* LibGUI: Verify m_window_id is not-zero in set_maximized
Window::set_maximized requires non-zero window id to be a valid call,
i.e. calling Window::show beforehand. A verify statement before the
server call can help developers by hinting correct usage.
* LibGUI: Paint background when the fullscreen window is transparent
The windows in the background are ignored when the window is fullscreen.
However, we still would like to see the background if that window is
transparent.
* Userland: Add ability to capture rectangular region in shot
A click and drag selectable, transparent, fullscreen window is
displayed with the command line argument -r for screenshots.
This patch adds a missing minimize check for highligted windows in
WindowStack::for_each_visible_window_of_type_from_front_to_back().
Minimized windows should not be treated as visible in this context.
Previously, iterating through each visible Window when recomputing
occlusions in the Compositor would cause a crash if a highlighted
Window is also minimized at the same time. As the WindowSwitcher
currently highligts Windows even when they are minimized, opening
it while any Window is minimized would cause WindowServer to crash.
This patch removes the background behind window icons
in the WindowSwitcher which looked like it was being
rendered incorrectly (without alpha) previously.
While structs being forward declared as classes is not strictly an
issue, Clang complains as this is not portable code, since some ABIs
treat classes declared as `class` and `struct` differently.
It's easier to fix these than to reason about explicitly disabling
another warning.
It might be the case that we are passing non-movable/non-copyable things
through IPC. In this case, Clang will emit a warning as it can't
generate the requested default move/copy ctor for the IPC message.
To fix this, we use a `#pragma` to make the compiler silently ignore our
request.
The same was the case with the three-way comparison in `Screen`. Since
we don't use the three-way comparison operator anywhere else in our
codebase, we simply use the `==` operator instead.
This fixes a bug with menu keyboard navigation. If you pressed the right
arrow to enter a submenu, then the left arrow to exit the submenu, then
right and left again it would leave no menu item selected.
Because descending into the submenu wasn't making it the current menu,
when you press the left arrow it couldn't find the "current menu" in the
stack, so didn't know what menu to pop back to.
It was an accident that it worked the first time you navigated into the
menu. Selecting the parent item also opened the submenu, and opening
an already open menu sets it as the current menu. After closing the
submenu with the left arrow, it is no longer already open, so it wasn't
getting set as the current menu.
Before this change, invalidating any rect in a WindowFrame would cause
the entire window (including frame & drop shadow) to get invalidated,
leading to copious amounts of overdraw when mousing over menubars,
titlebars, and window buttons.
We now simply allow the partial frame invalidations through to the
window's dirty rects collection and the compositor takes care of it.
ec6debb changed item_index_at to return -1 when hovering over a
separator. The intent was to not send the separator to clients for
MenuItemEntered.
However, this had the unintented consequence of not closing the submenu
when you hover over a separator. Submenus ignore when the item index is
-1 in order to leave the menu open when you move the mouse outside. This
ends up leaving the submenu open without the highlight to show what menu
item the submenu belongs to.
A slightly less severe consequence is that pressing the up or down arrow
key in such a situation would now go the top or bottom of the menu
rather than the item above or below the separator.
We now push the special casing of separators into set_hovered_index so
that the rest of the code behaves as it did before ec6debb.
Previously, this mode would flash flush/repaint rects in yellow for
however it long it took for the compositor to replace the yellow with
the final image instead.
Now we usleep() for 10 ms when flashing, so you get a chance to see
the yellow. This immediately makes "flash flush" mode super useful. :^)
This patch adds the concept of a window being "Pinnable" (always drawn
on top of other windows). This can be toggled through a new checkable
action in the top left corner's window menu.
Because window states and various flags can affect the windows'
rendered areas it's safer to use the last computed occlusion rectangles
to invalidate areas on the screen that may have to be re-rendered due
to e.g. a window size change.
Fixes#6723
Depending on the driver, the second buffer may not be located right
after the first, e.g. it may be page aligned. This removes this
assumption and queries the driver for the appropriate offset.
Some devices may require DMA transfers to flush the updated buffer
areas prior to flipping. For those devices we track the areas that
require flushing prior to the next flip. For devices that do not
support flipping, but require flushing, we'll simply flush after
updating the front buffer.
This also adds a small optimization that skips these steps entirely for
a screen that doesn't have any updates that need to be rendered.
The only remaining sync call from client to server is now the call
that switches a window's backing store. That one actually relies on
the synchronization to hand over ownership of the backing stores,
so it has to stay synchronous for now.
Differentiates between normal minimization and hidden windows. A window
which is hidden is still minimized, but can be seen as another stage
of being minimized.
Also, make it return a reference as aside from only three special
situations (creating, destroying, and moving a window between stacks)
a window should always be on a window stack. Any access during those
brief situations would be a bug, so we should VERIFY this.
This solves two problems:
* A window was sometimes deemed occluded when the window rect was
entirely covered by other rectangles, transparent or opaque. This
caused a window to stop rendering even if a small portion was still
visible, e.g. when it was merely covered by a window shadow.
* The window switcher is interested in window updates even when a
window is entirely covered by another one, or when it is on another
desktop. This forces windows to be not occluded in those cases.
When using the Super+Tab hotkey then all windows will be displayed,
and we will switch to another virtual desktop if needed.
When using the Alt+Tab hotkey then only the windows on the current
desktop will be displayed.
This also adds the ability to query how many virtual desktops are
set up, and for the Taskbar to be notified when the active virtual
desktop has changed.
This creates a 2-dimensional array of WindowStack instances, one for
each virtual desktop. The main desktop 0,0 is the main desktop, which
is the desktop used for all stationary windows (e.g. taskbar, desktop).
When adding windows to a desktop, stationary windows are always added
to the main desktop.
When composing the desktop, there are usually two WindowStacks
involved. For stationary windows, the main desktop will be traversed,
and for everything else the current virtual desktop will be iterated.
Iteration is interweaved to preserve the correct order. During the
transition animation, two WindowStacks will be iterated at the same
time.
This removes StringView::find_first_of(char) and find_last_of(char) and
replaces all its usages with find and find_last respectively. This is
because those two methods are functionally equivalent.
find_{first,last}_of should only be used if searching for multiple
different characters, which is never the case with the char argument.
This also adds the [[nodiscard]] to the remaining find_{first,last}_of
methods.
Since this is always set to true on the non-default constructor and
subsequently never modified, it is somewhat pointless. Furthermore,
there are arguably no invalid relative paths.
The time interval for animations is most often described as `duration`
in animation contexts and the `WindowServer::Animation` class
should reflect that.
The menus always thought they were being outside of the main screen,
which caused them to be left and/or top aligned. This also fixes the
calculation of the available space by using the screen rectangle where
it will be displayed.
To make Assistant useful we need a way to quickly trigger it. I've
added a new specialized event coming from the window server for when a
user is holding down 'Super' and hits 'Space'.
The Taskbar will be able to listen for this event and spawn a new
instance of the Assistant if it's not already running.
Since being tiled means we restrict rendering a window to the screen it
is on (so that we don't "bleed" into an adjacent screen), we need to
untile it if the window either can't fit into the screen, or it is
detached from the screen edges.
The launch_origin_rect parameter to create_window() specifies where on
screen the window was launched from. It's optional, but if you provide
it, the new window will have a short wireframe animation from the origin
to the initial window frame rect.
GUI::Window looks for the "__libgui_launch_origin_rect" environment
variable. Put your launch origin rect in there with the format
"<x>,<y>,<width>,<height>" and the first GUI::Window shown by the app
will use that as the launch origin rect.
Also it looks pretty neat, although I'm sure we can improve it. :^)
This patch adds the WindowServer::Animation class, which represents
a simple animation driven by the compositor.
An animation has a length (in milliseconds) and two hooks:
- on_update: called whenever the animation should render something.
- on_stop: called when the animation is finished and/or stopped.
This patch also ports the window minimization animation to this new
mechanism. :^)
We regularily need to flush many rectangles, so instead of making many
expensive ioctl() calls to the framebuffer driver, collect the
rectangles and only make one call. And if we have too many rectangles
then it may be cheaper to just update the entire region, in which case
we simply convert them all into a union and just flush that one
rectangle instead.
This fixes a regression where the geometry label isn't updating even
though the window geometry had changed because the geometry label's
location isn't changing.
An Overlay is similar to a transparent window, but has less overhead
and does not get rendered within the window stack. Basically, the area
that an Overlay occupies forces transparency rendering for any window
underneath, which allows us to render them flicker-free.
This also adds a new API that allows displaying the screen numbers,
e.g. while the user configures the screen layout in DisplaySettings
Because other things like drag&drop or the window-size label are not
yet converted to use this new mechanism, they will be drawn over the
screen-number currently.
If a window which has an active modal window is focused, the modal
window starts blinking. In this case, the window (and modal) should
still be focused. For this, the order of the checks in
process_mouse_event_for_window has to be changed.
This fixes#8183.
Since MultiScaleBitmaps only loads icons for the scales in use, we need
to unconditionally reload them so that we pick up the correct bitmaps
for a scale that hasn't been previously used.
This enables the shot utility to capture all screens or just one, and
enables the Magnifier application to track the mouse cursor across
multiple screens.
This enables rendering of mixed-scale screen layouts with e.g. high
resolution cursors and window button icons on high-dpi screens while
using lower resolution bitmaps on regular screens.
This sets the stage so that DisplaySettings can configure the screen
layout and set various screen resolutions in one go. It also allows
for an easy "atomic" revert of the previous settings.
If there are any screens that are detached from other screens it would
not be possible to get to them using the mouse pointer. Also make sure
that none of the screens are overlapping.
We were calculating the old window rectangle after changing window
states that may affect these calculations, which sometimes resulted
in artifacts left on the screen, particularily when tiling a window
as this now also constrains rendering to one screen.
Instead, just calculate the new rectangle and use the window's
occlusion information to figure out what areas need to be invalidated.
When a window is maximized or tiled then we want to constrain rendering
that window to the screen it's on. This prevents "bleeding" of the
window frame and shadow onto the adjacent screen(s).
This allows WindowServer to use multiple framebuffer devices and
compose the desktop with any arbitrary layout. Currently, it is assumed
that it is configured contiguous and non-overlapping, but this should
eventually be enforced.
To make rendering efficient, each window now also tracks on which
screens it needs to be rendered. This way we don't have to iterate all
the windows for each screen but instead use the same rendering loop and
then only render to the screen (or screens) that the window actually
uses.
Some paths of the mouse event processing code will upgrade the event
from a regular MouseDown to a MouseDoubleClick. That's why we were
passing `MouseEvent&` everywhere.
For the paths that don't need to do this, passing `MouseEvent const&`
reduces the cognitive burden a bit, so let's do that.
Instead of plumbing a Window* through the entire mouse event processing
logic, just do a hit test and say that the window under the cursor is
the hovered window.
It's funny how much easier this is now that we have a way to hit test
the entire window stack with one call.
Move the logic for processing a mouse event that hits a specific window
into its own function.
If the window is blocked by a modal child, we now get that out of the
way first, so we don't have to think about it later.
Even if a window is in fullscreen mode, we still want hit testing to
walk the window stack. Otherwise child windows of the fullscreen
window will not receive mouse events.
The button widgets internally rendered by WindowServer are only used
in titlebars, and require a bit of mouse event handling. Instead of
mixing it with the window-oriented mouse event handling, get the
button event stuff out of the way first.
We were forgetting to preserve the m_drag and m_mime_data members of
WindowServer::MouseEvent when making a translated copy.
This didn't affect any reachable code paths before this change.
If a window is currently actively tracking input events (because
sent it a MouseDown and haven't sent it a MouseUp yet), we now simply
send mouse events to that window right away before doing any other
event processing.
This makes it much easier to reason about mouse events.
Instead of just answering hit/no-hit when hit testing windows, we now
return a HitTestResult object which tells you which window was hit,
where it was hit, and whether you hit the frame or the content.
This feature had been there since early on and was not actually useful
for anything. I just added it because it was fun. In retrospect, it's
not a very good feature and I only ever activated it by accident.
This patch moves the window stack out of WindowManager and into its own
WindowStack class.
A WindowStack is an ordered list of windows with an optional highlight
window. The highlight window mechanism is used during Super+Tab window
switching to temporarily bring a window to the front.
This is mostly mechanical, just moving the code to its own class.
Remove the confusingly-named inflate_for_shadow() function and inline
its logic into render_to_cache(). And remove the m_shadow_offset
member variable since it was only needed locally in one place.
Also improve some variable names to make it more understandable what
is going on.
Let clients manage their own window ID's. If you try to create a new
window with an existing ID, WindowServer will simply disconnect you
for misbehaving.
This removes the need for window creation to be synchronous, which
means that most GUI applications can now batch their entire GUI
initialization sequence without having to block waiting for responses.
This patch moves the magnifier rect computation over to the server side
to ensure that the mouse cursor position and the screen image never get
out of sync.
By moving the logic to determine what window areas (shadow, frame,
content) into WindowFrame::opaque/transparent_render_rects we can
simplify the occlusion calculation and properly handle more
arbitrary opaque/transparent areas.
This also solves the problem where we would render the entire
window frame as transparency only because the frame had a window
shadow.
This replaces ctype.h with CharacterType.h everywhere I could find
issues with narrowing conversions. While using it will probably make
sense almost everywhere in the future, the most critical places should
have been addressed.
Previous to this commit, if a `Window` wanted to set its width or height
greater than `INT16_MAX` (32768), both the application owning the Window
and the WindowServer would crash.
The root of this issue is that `size_would_overflow` check in `Bitmap`
has checks for `INT16_MAX`, and `Window.cpp:786` that is called by
`Gfx::Bitmap::create_with_anonymous_buffer` would get null back, then
causing a chain of events resulting in crashes.
Crashes can still occur but with `VERIFY` and `did_misbehave` the
causes of the crash can be more readily identified.
Use the configured desktop background color, if defined, otherwise
default to the current theme's background color. If a user chooses
a background color via "desktop settings", then this new color
will always be used.
Switching themes will delete the user-defined background color, so
the background color resets to the theme's defined color.
Instead of using a low-level, proprietary API inside LibGfx, let's use
Core::AnonymousBuffer which already abstracts anon_fd and offers a
higher-level API too.
This functionality, while neat, isn't really something you need enabled
all the time. Let's make it opt-in instead. Pass MakeInspectable::Yes
to the Core::EventLoop constructor if you want your program to become
inspectable.
Changes to the system font settings are now persisted in /etc.
Note that you still need to restart the system for changes to fully
apply in all programs.
This patch adds a set_system_fonts() IPC API that takes the two main
font queries as parameters. We'll probably expand this with additional
queries when we figure out what they should be.
Note that changing the system fonts on a live system mostly takes
effect in newly launched programs. This is because GUI::Widget will
currently cache a pointer to the Gfx::FontDatabase::default_font()
when first constructed. This is something we'll have to fix somehow.
Also note that the settings are not yet persisted.
Instead of everybody getting their system fonts from Gfx::FontDatabase
(where it's all hardcoded), they now get it from WindowServer.
These are then plumbed into the usual Gfx::FontDatabase places so that
the old default_font() and default_fixed_width_font() APIs keep working.
Problem:
- `static` variables consume memory and sometimes are less
optimizable.
- `static const` variables can be `constexpr`, usually.
- `static` function-local variables require an initialization check
every time the function is run.
Solution:
- If a global `static` variable is only used in a single function then
move it into the function and make it non-`static` and `constexpr`.
- Make all global `static` variables `constexpr` instead of `const`.
- Change function-local `static const[expr]` variables to be just
`constexpr`.
Instead of doing a full IPC round-trip for the client and server to
greet each other upon connecting, the server now automatically sends
a "fast_greet" message when a client connects.
The client simply waits for that message to arrive before proceeding.
(Waiting is necessary since LibGUI relies on the palette information
included in the greeting.)
This was only synchronous since WindowServer managed the ID allocation.
Doing this on the client side instead allows us to make create_menu()
an asynchronous IPC call, removing a bunch of IPC stalls during
application startup.
Since applications using Core::EventLoop no longer need to create a
socket in /tmp/rpc/, and also don't need to listen for incoming
connections on this socket, we can remove a whole bunch of pledges!
This was already being used asynchronously by LibGUI, which meant that
WindowServer would generate a response, and the client would ignore it.
This patch simplifies the WindowServer side so it no longer generates
the unnecessary response.
We were not substituting the window modified marker ("[*]") in the
title strings we were sending to WM clients. This caused the Taskbar
to show pre-substitution window titles for the Text Editor application.
This patch moves the window title resolution to Window::compute_title()
which is then used throughout.
Also update the window switcher for good measure. The window switcher
doesn't visualize this information at the moment, but we generally do
this when any window state changes.
Instead of trying to update only the little bit that changes, let's
have a function that updates all the window menu items in one go.
It's just a couple of string and boolean assignment, and the real
cost is performing the subsequent menu redraw, which remains the same.
Without this change, window buttons would get stuck in the "pressed"
state as long as the left mouse button was pressed, even if you moved
the mouse cursor out of the button rect.
Make the taskbar 27 pixels tall instead of 28. This makes the button
icons and applets vertically centered.
On a related note, this required touching *way* too many places..
This enables us to use keys of type NonnullRefPtr in HashMaps and
HashTables.
This commit also includes fixes in various places that used
HashMap<T, NonnullRefPtr<U>>::get() and expected to get an
Optional<NonnullRefPtr<U>> and now get an Optional<U*>.
This ignores unhandled mouse clicks for the window buttons. Right now
right-clicking on the window buttons animates them as if some action
were to occur when the mouse button is released.
Most of the IPC that happens between clients and WindowServer when
creating and configuring windows can be asynchronous. This further
reduces the amount of ping-ponging played during application startup.
Creating a menu/menubar needs to be synchronous because we need the
ID from the response, but adding stuff *to* menus (and adding menus
to menubars, and menubars to windows) can all be asynchronous.
This dramatically reduces the amount of IPC ping-pong played by
each GUI application during startup.
I measured how long it takes TextEditor to enter the main event loop
and it's over 10% faster here. (Down from ~86ms to ~74ms)
This enables support for automatically generating client methods.
With this added the user gets code completion support for all
IPC methods which are available on a connection object.
This commit unifies methods and method/param names between the above
classes, as well as adds [[nodiscard]] and ALWAYS_INLINE where
appropriate. It also renamed the various move_by methods to
translate_by, as that more closely matches the transformation
terminology.
Windows that are marked as modified will now have another (themable)
close button. This gives an additional visual clue that some action
will be required by the user before the window gets closed.
The default window-close-modified icon is an "X" with "..." underneath,
building on the established use of "..." in menus to signify that
additional user input will be required before an action is completed.
It's possible that the backing store hasn't been updated yet, so
when performing an alpha hit-test make sure the bitmap actually
contains it.
Fixes#6731
When a new Window instance is added to the WindowManager, it does not
yet have an updated value for `m_frame->rect()` and we're not checking
if there is a new candidate for the hovered window, which we need to do
since the mouse cursor might hover above the newly opened window.
This fixes both issues: as soon as a Window frame's rect is changed,
ask the WindowManager to reevaluate its hovered window. This takes care
of newly opened windows _and_ windows that are programmatically changed
in size.
This works because when a Window becomes hovered, the WindowManager
sends out an enter event. This event in turn triggers the Window to
evaluate the cursor type under the mouse position and to update it when
necessary.
Fixes#4809.
This patch removes the IPC endpoint numbers that needed to be specified
in the IPC files. Since the string hash is a (hopefully) collision free
number that depends on the name of the endpoint, we now use that
instead. :^)
Additionally, endpoint magic is now treated as a u32, because endpoint
numbers were never negative anyway.
For cases where the endpoint number does have to be hardcoded (a current
case is LookupServer because the endpoint number must be known in LibC),
the syntax has been made more explicit to avoid confusing those
unfamiliar. To hardcode the endpoint magic, the following syntax is now
used:
endpoint EndpointName [magic=1234]
