When hiding a window, libdecor can report bogus content region sizes that are +/- the height of the title bar. Ignore any size values from libdecor when hiding a window, or the size may be incorrect when restored.
The compositor can arbitrarily move windows between displays, including fullscreen windows. Update the internal state when a fullscreen window is moved so the internal SDL state accurately reflects the window location, and resize the window to fit the new display.
This also fixes an edge case where the compositor can make a window fullscreen on a different display than SDL thinks it will be on (usually when a window is made fullscreen by the compositor while straddling multiple displays), which can result in the window being incorrectly sized.
If additional fullscreen requests are received when the window is already fullscreen, it is typically due to the fullscreen flags or emulated video mode being changed. A commit must be explicitly triggered or the requested changes won't take effect until some other event, such as a resize or focus change, causes the changes to be committed.
The compositor can toggle the fullscreen state (via a hotkey or otherwise), so the internal SDL state must be updated accordingly when it does.
When toggling fullscreen via the compositor, SDL will attempt to use the last fullscreen flag explicitly set. If no flag was previously set, SDL_WINDOW_FULLSCREEN will be used if a window video mode was set, otherwise SDL_WINDOW_FULLSCREEN_DESKTOP will be used. If the previous flag was SDL_WINDOW_FULLSCREEN and the window video mode was cleared, it will revert to SDL_WINDOW_FULLSCREEN_DESKTOP.
If libdecor performs a commit with the frame title being undefined, a crash can occur within the library or its plugins. Always ensure that the title is set to a valid string to avoid this.
Eliminate excessive calls to SetFullscreen by removing the calls in the libdecor and xdg-toplevel config callbacks.
These calls were being made there in case something explicitly called the window minimization function from within SDL, which unsets fullscreen, and as minimizing a window in Wayland is just a suggestion to the compositor and doesn't actually change the window state or communicate anything back to the application, it was necessary to call SetFullscreen in every call to the config functions just in case something minimized a window via SDL_MinimizeWindow() and later needed to restore it. GNOME in particular had issues when fullscreen set/unset operations were being hammered, leading to overlapping acks and commits when switching to fullscreen.
With the new video system flag to disable unsetting fullscreen when minimizing a window, these calls in the configuration functions are no longer needed and can be removed. This significantly reduces calls to the SetFullscreen() function, reverts #6044 while fixing the issue, and fixes a similar problem when hiding and showing a window initially created with fullscreen flags.
Don't call the roundtrip in ShowWindow unless restoring a previously hidden window. This fixes a regression in GNOME when creating a window with the fullscreen flag set, as the fullscreen window will be positioned down the screen by the height of the top bar if the window is made fullscreen on the primary display and the roundtrip is called when initially displaying the window.
The video core assumes that window->w/h will be updated before returning
from SetWindowFullscreen(). This is needed to generate a resize event
with the correct window size when exiting fullscreen.
The roundtrip allows us to receive the configure callback that informs
us of the new window size before returning.
Fixes#6043
The current method of toggling the libdecor window visibility by destroying and recreating the frame results in a race where a use-after-free bug can manifest itself within libdecor when window visibility is toggled quickly. Instead, use the libdecor function for toggling visibility instead of destroying and recreating the frame every time.
We actually request CSD mode with xdg-decoration for borderless
windows, so we get what we wanted there and there's no point in going
into fallback paths.
Currently, the SDL_WINDOWEVENT_RESIZED event is sent before the actual
window is resized (and various internal state, such as the desired
GL/Vulkan backbuffer size, are updated). This makes sense, as SDL will
discard a no-op resize, which would be the case if we had resized before
sending the event (indeed, there are existing hacks to prevent this).
However, this means that SDL_{GL,Vulkan}_GetDrawableSize() will still
use the old size in the SDL_WINDOWEVENT_RESIZED handler. In the case of
SDL_Renderer, this means the drawable size it uses will be wrong, and
the viewport will get "updated" to the old value.
This then results in bug #5899.
Use SDL_lroundf() to round fractional backbuffer sizes halfway away from zero, as this is the rounding method recommended by the forthcoming Wayland fractional scaling protocol.
XDG-toplevel min/max size values are double-buffered data and must be committed before entering the fullscreen state, or a max window size value smaller than the display dimensions may cause the compositor to incorrectly configure the fullscreen window size. This fixes windowed->fullscreen transitions on GNOME, where, previously, certain combinations of window flags and min/max size values could cause entering fullscreen mode to fail with odd window sizes and/or offsets due to the new max size values not being committed before entering fullscreen, causing the compositor to clamp to the old values.
In the case of libdecor, it has its own layer of buffering on top of the xdg-toplevel surface for the min/max window dimensions, so both a frame commit and surface commit are required to set the state properly.
Previously, the surface damage region was being set in the same callback used for preventing render hangs in the GL backend when the surface was not visible. This was not ideal, as the callback was never fired in the case of using a different render backend or having a swap interval of 0. Use a separate frame callback for setting the surface damage region to ensure that it fires reliably, regardless of the backend being used or swap interval.
Previously, scale values used by the displays and surfaces were always integers, with fractional scale values only being calculated when the backbuffer and viewport sizes were being determined. Now, if xdg-output is available, the fractional scale of output displays is calculated when the displays are enumerated and the true scale values of the output devices are used whenever possible.
This unifies the integer and fractional scaling systems, allows for the use of more accurate scale values that minimize overdraw when windows straddle multiple outputs, and lays the groundwork for the pending Wayland scaling protocols that will report the preferred scale values per-surface instead of per-output.
Compartmentalize the fullscreen mode emulation code blocks, unify the windowed/fullscreen viewport logic, consolidate all window geometry code into a central function to eliminate blocks of duplicate code and rename related variables and functions to more explicitly reflect their purpose.
Don't be fooled by the diff size - this ended up being a big refactor of the
shell surface management, masked only by some helper macros I wrote for the
popup support.
This change makes it so when xdg_decoration is supported, but CSD is requested,
the system bails on xdg support entirely and resets all the windows to use
libdecor instead. This transition isn't pretty, but once it's done it will be
smooth if decorations are an OS toggle since libdecor will take things from
there.
In hindsight, we really should have designed libdecor to be passed a toplevel,
having it manage that for us keeps causing major refactors for _every_ change.
When using emulated display modes, the output size is often larger than the drawable buffer. As the surface damage region is automatically calculated from the smaller drawable buffer size, the damage region needs to be manually set to cover the entire viewport region or visual repaint artifacts can result.
I kind of thought it'd be nice to have it in the center, but this is an issue
for applications that still assume global mouse and window positions are
accessible. For example, this fixes cursor offset issues in UE5.
It's possible that an external component (probably a GL/VK context) committed, so we need to cover our bases and detach in both HideWindow and ShowWindow.
Fixes a crash in UE5 editor's pop-ups.
Partially fixes the mouse cursor in UE5 editor. Imperfect because UE5 uses window position and global mouse state to get position, but of course we don't have global mouse and this is just to get the right display index so this still fails overall. We really need to make global mouse support a feature query...
Use viewports for non-fullscreen windows when the desktop uses fractional scaling and the window is flagged as DPI-aware to provide a backbuffer mapped as close to 1:1 output as possible. In the cases of odd window sizes the backbuffer may be a pixel off of scaling perfectly into the window size due to its scaled size being rounded off, but a minute amount of scaling during output is likely preferable to the large amounts of overdraw needed with integer scaled buffers.
Expose as many emulated display modes as possible. They will currently display stretched to the display's native desktop aspect, but if an application requires a hardcoded resolution, it will work at minimum.
Aside from the change in the emulated display mode list, the Wayland event handling code had to be updated to support separate scaling for the x and y axes, as square pixels are no longer guaranteed.
Wayland doesn't support mode switching, however Wayland compositors can support the wp_viewporter protocol, which allows for the mapping of arbitrarily sized buffer regions to output surfaces. Use this functionality, when available, for fullscreen output when using non-native display modes and/or when dealing with scaled desktops, which can incur significant overdraw without this extension.
This also allows for the exposure of arbitrarily sized, emulated display modes, which can be useful for legacy compatability.
If we get a SDL_SetWindowSize() call right after SDL_SetWindowFullscreen() but
before we've gotten a new configure event from the compositor, the attempt to
set our window size will silently fail (when libdecor is enabled).
Fix this by remembering that we need to commit a new size, so we can do that
in decoration_frame_configure().
This caused some weird stuff to happen in the libdecor path, probably because
the window hasn't actually been mapped yet. It ends up calling stuff that
should not yet apply, and so fullscreen in particular would have a really
messed up titlebar.
The good news is, libdecor is good about tracking fullscreen state, so we can
let the callback do this for us. Keep this for xdg_shell because we actually
map the window ourselves, so we know this call is valid for that path.
The xdg_shell spec seems to state[1] that xdg_toplevel_configure events can
always provide a 0×0 width/height to signal that the compositor doesn't
care. SDL previously assumed the provided width/height was always valid
for fullscreen windows, and so applied it as-is.
This broke SDL applications on KDE/KWin 5.23, which now sends 0×0
configure events (and, in 5.23.3, 1×1 events for some reason), breaking
all SDL applications in fullscreen[2].
[1]: https://gitlab.freedesktop.org/wayland/wayland-protocols/-/issues/6
[2]: https://bugs.kde.org/show_bug.cgi?id=444962#c6
