The Xrandr dot clock value is declared as an unsigned long and the result when multiplying by 100 can overflow on a 32-bit system. Explicitly cast it to Sint64 to ensure that no overflow will occur.
Attempt to retrieve the display for fullscreen windows using the window position so that the correct display ID is returned if an exclusive fullscreen window is moved to another display.
* Setting the same mouse cursor twice is a no-op
* Cocoa: Call [NSCursor set] to change mouse cursor
The previous way, changing the mouse cursor was handled by invalidating
the mouse cursor rectangles and then recreating them (with the new
cursor) the next event loop. This is extremely slow; sometimes it can
take over a millisecond! With [NSCursor set] it happens instantly and
very quick performance-wise.
The downside is that it sets the cursor for the whole screen, so we
have some guards in place to change it to the system cursor if
the mouse moves outside the window or the window loses focus.
* Cocoa: Remove unneeded resetCursorRects: function
Without these mappings, this controller "kinda" works out of the box:
- `SDL_GameControllerMapping()` works because it will notice "Xbox" in
the name and use the default XInput mappings
- `SDL_GameControllerMappingForGUID()` will not find any mapping
lsusb:
```
ID 2dc8:2000 8BitDo 8BitDo Pro 2 Wired Controller for Xbox
```
In Linux this controller is supported by two drivers:
- `xpad` (built-in to the kernel), exposes the controller name from the
USB descriptor and the GUID starts with 03 (0x03 = BUS_USB)
- `xone` (https://github.com/medusalix/xone), exposes a virtual
controller which is always named "Microsoft X-Box One pad" and the
GUID starts with 06 (0x06 = BUS_VIRTUAL)
This commit adds the 2 GUIDs from both drivers so mappings will always
be found and the real controller name will always be reported.
(cherry picked from commit 4266cf8504)
Portrait displays may have native, physical resolutions that are taller than wide. Reverse the mode dimensions when dealing with these displays as well as those rotated via software means.
This makes it clear what the new versions are, and in the case of SDL_RenderDrawPoint() and SDL_RenderDrawLine(), the coccinelle script actually does the (float) casts for you.
This fixes rounding errors with coordinate scaling and gives more flexibility in the presentation, as well as making it easy to maintain device independent resolution as windows move between different pixel density displays.
By default when a renderer is created, it will match the window size so window coordinates and render coordinates are 1-1.
Mouse and touch events are no longer filtered to change their coordinates, instead you can call SDL_ConvertEventToRenderCoordinates() to explicitly map event coordinates into the rendering viewport.
SDL_RenderWindowToLogical() and SDL_RenderLogicalToWindow() have been renamed SDL_RenderCoordinatesFromWindow() and SDL_RenderCoordinatesToWindow() and take floating point coordinates in both directions.
The viewport, clipping state, and scale for render targets are now persistent and will remain set whenever they are active.
SDL uses window minimization to determine fullscreen window visibility and hide windows before changing the video mode back to the desktop. Wayland, however, does not have the concept of a minimized window and doesn't set the minimized flag (minimization can be requested, but what actually happens to the window is implementation dependent, and if a window is minimized via a desktop shortcut or decoration control, the application is not notified of any state changes). Make the video core mode setting a no-op so that the Wayland backend can handle reporting the display dimensions using its own internal logic.
Accommodate the new video core changes.
The new video core changes allow for some window geometry calculation refactoring that simplify the system:
- Removal of helper functions
- Eliminate some discrepancies between the libdecor and xdg-toplevel paths
- No need to short-circuit the video core window size event deduplication check
- Exclusive fullscreen windows will always end up on the correct output, even when fullscreen is initiated from the compositor
- Better handling of cases where the desktop is scaled, but does not expose the viewport protocol
- Return the display bounds for the emulated mode if an exclusive fullscreen window has focus
- Fixed cases where changing display properties during runtime wouldn't update the display mode lists
- General cleanup
The C specification states that passing a size of 0 to functions like memcpy is valid, but even if the size is 0 and the function is essentially a no-op, the result when passing any invalid pointers is considered undefined behavior. Don't rely on undefined behavior when copying the display or mode lists.
The display list can contain self-referential pointers if the current mode pointer points to the desktop mode or a fullscreen mode array element, and reallocating the display or fullscreen mode lists without updating the current mode pointer in these cases can leave them pointing to freed memory or garbage data. Manually copy the list items and update the self-referential pointers if necessary.
When adding a display, traverse the list of added fullscreen modes and ensure the display ID is set to its final, valid value, or the modes added before calling SDL_AddVideoDisplay() will have an invalid display ID.
