When SIGCHLD is blocked, some executables (for example CMake) do not exit properly when executed using `SDL_CreateProcess` from any SDL thread (not main thread). `SDL_CreateProcessWithProperties` docs say that `SIGCHILD` should not be ignored or handled, therefore blocking it during thread creation is a likely reason for the bug. Should fix#15210.
Signed-off-by: Konstantin Tomashevich <konstantin.tomashevich@gmail.com>
If SDL_HINT_OPENGL_ES_DRIVER is enabled and a GLES context is requested,
don't try to load the WGL/GLX library. Doing so may fail, because
SDL_HINT_OPENGL_LIBRARY may have been set to a custom libGLESv2 (e.g.
ANGLE). This effectively restores the SDL2 behavior.
Other video drivers shouldn't be affected, because they always use EGL
for GLES.
Incidentally, this patch also fixes a missing GL_GetEGLSurface callback
in the X11 fallback path.
The first approach had the drawback that it accumulated handles while raw input was paused, eventually crashing the application.
Now we'll keep reading raw events from the queue, we just won't deliver them to the application.
When emulating display modes or using display scaling, custom cursors need to be scaled, or they can appear too large or small relative to the window size.
Added vids for a variety of xbox 360 devices, mostly for linux use.
Corrected wheel vid and pid being flagged as unknown. Internally, when XUSB is building vid and pids for wireless devices, it assigns 0x02a1 for devices with subtype 1 and 0x02a2 for devices with subtype 2, which are wheels.
Some of these vids and pids do rely on the following patch being merged into the linux kernel, as I have overhauled how VID and PIDs are handled there for wireless devices.
https://patchwork.kernel.org/project/linux-input/patch/20260314075034.1488655-2-sanjay.govind9@gmail.com/
Also added some lists for guitar and drum devices, so we can set controller types correctly for those controllers.
Previously we would spin up and shut down a thread every time we changed raw input mode, which can take several ms. Now we'll just put the raw input thread to sleep when disabling raw input and wake it up when enabling raw input.
The floating state is only set after the first configuration event arrives, so size changes before the window was initially shown could be dropped. Use the flags instead.
When using the display scaling mode, scaled outputs would overlap, as the origins are in scaled points, while the width and height are in pixels. Use a simple algorithm to push the overlapping rectangles apart, so they do not overlap. This may end up leaving gaps in complex layouts (which are already somewhat mangled from mixing coordinate systems), but it works well enough in most cases.
Technically we only want to do this for controllers that are opened, but we don't have a way to match up controllers using other APIs with individual GCControllers.
Some compositors will send a configure event immediately after a scale event, however, this event can contain the old logical size, and should be ignored, or the window will incorrectly change size when moved between displays with differing scale factors.
Store the last requested logical size as the last configure size when resizing a floating window due to a scale change, so a configure event immediately following with the old size will be ignored.
This PR makes the GameInput joystick backend ignore non-gamepad controllers if DirectInput or XInput backends are enabled. This is done to prevent possible bugs and/or regressions.
This PR includes small improvements to the GameInput joystick backend:
- Device subtypes
- Compatibility with DirectInput joystick mappings (if the controller is not a gamepad, i.e. `GAMEINPUT_InternalIsGamepad()` returns `false`)
- Fallback to DirectInput/XInput for currently unsupported devices (force feedback ones and the ones that are neither gamepads nor controllers, I'm not sure if that's possible, but maybe racing wheels and other device subtypes would count?)
We are locked to version 1 of the ABI, so don't load what might be a symlink to a newer and incompatible version.
Platforms like OpenBSD that only have .so versions of libraries can add a special case here like Android in the future.
Fixes https://github.com/libsdl-org/SDL/issues/15172
Read capabilities when using xinput controllers via the libusb backend
This gives us access to the subtype on linux and macOS, and gives us a lot of data we can use for handling more detailed device types when I look into a unified api for exposing instrument data later.
PS4/5 controllers put device specific data into a specific region in the report, so we have to extract it separately.
No known guitars use the right stick on the guitar, so to keep things working similarly to PS3, i have opted to map whammy and tilt the same way as the PS3 rb guitars.
Updated testgdk to demonstrate correct handling of suspend/resume and the new Render APIs, and updated the docs to explain the correct usage of these GDK functions.
Apparently, because PlaneSlice wasn't initialized, we couldn't use SDL_GPU_TEXTURETYPE_2D_ARRAY and SDL_GPU_TEXTUREUSAGE_COMPUTE_STORAGE_WRITE together, as the error d3d12:
"ID3D12Device::CreateUnorderedAccessView occurred: The PlaneSlice -858993460 is invalid when the resource format is R8G8B8A8_UNORM and the view format is R8G8B8A8_UNORM. Only Plane Slice 0 is valid when creating a view on a non-planar format. [ STATE_CREATION ERROR #344: CREATEUNORDEREDACCESSVIEW_INVALIDPLANESLICE]"
- Use exact numerator/denominator from spec for CMTime to avoid rounding errors.
- Wrap frame rate setting in @try-catch to prevent crashes on strict drivers.
Don't disable this functionality when defining SDL_LEAN_AND_MEAN, but if you need to, you can define SDL_DISABLE_STB if you want to compile this out.
Fixes https://github.com/libsdl-org/SDL/issues/15139
This Visual Studio 2019 compiler wrongly emits a warning.
test\testvulkan.c(271): error C2220: the following warning is treated as an error
test\testvulkan.c(271): warning C4090: 'function': different 'const' qualifiers
test\testvulkan.c(286): warning C4090: 'function': different 'const' qualifiers
Wayland is sometimes at-odds with clients that want to enforce an aspect ratio or min/max window size, as certain window states have dimensions that either must be obeyed (maximized), or will give terrible results if they aren't (tiled). Use a viewport and a masking subsurface to handle cases where surfaces are unable to match the exact window size.
The changes made to accommodate this also catches some additional windowing related edge-cases, simplifies synchronization, and prevents commits before a buffer has been attached to the surface.
We have an implementation that uses 128-bit math and one that handles large values without it (thanks @jessechounard and @rabbit-ecl!)
Fixes https://github.com/libsdl-org/SDL/issues/15042
V4L2 is able to advertise that a video device is able to display any frame
interval within a continuous range. SDL does not allow advertising this and
only exposes discrete frame intervals. To work around this, SDL attempted to
generate a subset of the range with a fixed interval. Unfortunately, the way
this was accomplish is inherently broken and led to attempting to allocate a
very large number of formats per resolution and colorspace. With the Magewell
Pro Capture HDMI, which can expose FRMSIZE_TYPE_CONTINUOUS as well, this can
expose a truly astronomical number of formats, exceeding 1 PB of RAM. This will
lead to an OOM kill for any process that tries to initialize the camera
subsystem.
This patch just tests to see if some common frame rates are within the
contiuous range and expose those. SDL still does not handle
FRMSIZE_TYPE_CONTINUOUS in a graceful way so it still uses over a gigabyte of
RAM for each possible combination of sizes, but with this patch it no longer
leads to an OOM kill. The API will need amending for proper support for both
continuous frame sizes and frame intervals.
This is only useful if the application knows to add the flag introduced by this
extension, and at that point the application can also include the extension
themselves.
Case in point: SDL_gpu_vulkan was already doing this!
DSV creation was missing a TEXTURE2DARRAY branch for array, cube, and
cube-array depth textures. It fell through to TEXTURE2D, so
FirstArraySlice was never set and all layers' DSVs targeted layer 0.
This caused incorrect rendering when using depth textures with multiple
layers, such as cubemap shadow maps.
Previously it would have reported success but not returned a valid string, and
didn't check for the (probably extremely unlikely) case of overflowing our
locale list array.
SDL_GetGrabbedWindow() can be called when the video system is uninitialized, and, since there is no window parameter, _this must be checked for validity to avoid a segfault.
It SDL_SYS_EnumerateDirectory was changing the path string and passing it to
the callback, causing chaos in the glob handler, which expected the original
string to pass through.
Fixes#15057.
This reverts commit 63c0650321.
This reverts a revert; leaving this in main, where we'll examine it in 3.6.0.
This remains reverted in release-3.4.x
Reference Issue #15058.
Avoids a null dereference of currentRegion->vulkanTexture->container in DefragmentMemory:
} else if (!currentRegion->isBuffer && !currentRegion->vulkanTexture->markedForDestroy) {
...
¤tRegion->vulkanTexture->container->header.info
... (among others)
by not "VULKAN_Submit"ting (and thus adding it to the defrag pool) before setting the container.
Although rare (defrag almost never ran), this crash happened in a real-world application.
Getting device names can hang for a long time on certain devices, so make sure this is done on a separate thread to avoid blocking initialization and the main loop.
Fixes https://github.com/libsdl-org/SDL/issues/12913
Rumble can often take a long time, and it is theoretically safe to simultaneously read and write hidapi devices on all platforms.
Fixes https://github.com/libsdl-org/SDL/issues/9441
FS.mkdir() will throw a javascript exception if the scratch directory already
exists, so catch/ignore that.
Wrap the rest of the scratch i/o in a try/catch block; the event will only
send if everything works out.
Wrap some calls from Javascript to the C runtime's free() in an
EMSCRIPTEN_KEEPALIVE function, so that the compiler doesn't optimize the
function out and crash at runtime.
Fixes#14999.
This controller has 3 interfaces, one for the Xbox gamepad protocol, and two HID interfaces. We should only handle the Xbox interface in the Xbox driver.
Removed obsolete documentation that shows this usage and macros that attempt it.
Also allow SDL_FUNCTION to be redefined by the application.
Fixes https://github.com/libsdl-org/SDL/issues/15004
currentX and currentY in L2263-L2264 are marked as conditionally uninitialized by MSVC and Clang
Clang: variable may be uninitialized when used here [-Wconditional-uninitialized]
In SDL_audio.c:OpenPhysicalAudioDevice, an attempt is made to SDL_CreateThread
if ProvidesOwnCallbackThread is false, but SDL_CreateThreadWithPropertiesRuntime
is not implemented for Emscripten, so this always fails. I'm not sure if/when
this ever worked, but it simply cannot work in its current state.
The patterns we were using don't match the official patterns, which fill up as
many lights as players instead of just using the nth player LED. Above 4,
there are some special patterns, as documented on Nintendo's Singaporean site:
https://www.nintendo.com/sg/support/qa/detail/33822
The default varies between OpenGL and OpenGL ES, so try to force it to what
the app actually requested with SDL_GL_FRAMEBUFFER_SRGB_CAPABLE.
Fixes#14898.
This hint is documented to not just turn off fullscreen windows going into a
new Fullscreen Space, but also to make the green button on a resizeable
window's title bar do a maximize/zoom instead of make the window fullscreen.
Previously, this only did the former and not the latter (or perhaps it worked
and the defaults changed in a newer macOS, we aren't sure).
Fixes#7470.
This also delays pen proximity out events to make sure that the pen is really gone before delivering them. On Android, you get a HOVER_EXIT event when the pen contacts the surface, which we don't want to treat as the pen leaving proximity.
- the connection type is already set correctly based on bluetooth connection state
- removed some verbose controller type logging
- device and ctx are guaranteed not to be NULL at the driver level
In this case, it's available via both libusb and hidraw, but the hidraw device isn't usable. If a device needs to be opened via libusb, don't expose it via hidraw enumeration.
On Windows there is a separate HIDAPI device for each slot. On Linux and macOS, there is a single HIDAPI device and the slot is included in the report.
When the OpenXR loader library fails to load, provide a detailed error
message explaining:
- On Windows: need openxr_loader.dll in app directory or PATH
- On Linux: install libopenxr-loader package or set LD_LIBRARY_PATH
- Can use SDL_HINT_OPENXR_LIBRARY hint to specify path
Also update D3D12 backend to include the error in its warning log,
and remove redundant SDL_SetError in Vulkan backend that was
overwriting the detailed message.
We still need the task to go through the IoRing, even though the flush
operation we use to get it there will always fail on a read-only file. So
check for this specific case and don't report failure.
Fixes#14878.
This reverts commit cd31381185.
Linking libbrcmEGL.so explicitly breaks (causes segmentation faults) in other code that links and runs the X11 EGL.
Reopens https://github.com/libsdl-org/SDL/issues/14749
At least in my Raspbian Stretch build environment, the vc_dispmanx symbols that were missing are found in libbcm_host.so, which is linked by default.
This avoids some misleading error messages when running unit tests with
the dummy driver: on a typical desktop Linux system, it's normal for
opening keyboards and mouse in `/dev/input/*` to fail with `EACCES`,
and in container technologies that share `/sys` but not `/dev` with the
container, we can find that we fail to open them with `ENOENT`.
Resolves: https://github.com/libsdl-org/SDL/issues/14872
Signed-off-by: Simon McVittie <smcv@collabora.com>
Make use of check_include_files to be able to also include
stdint.h when checking for the headers. Fixes detection of
usbhid.h on OpenBSD.
/usr/include/usbhid.h:40:2: error: unknown type name 'uint32_t'
40 | uint32_t _usage_page;
| ^
Otherwise, the quit event can be sent prematurely. The topmost status must be queried and cached before sending the close request event, as the window may be destroyed in an event handler.
As suggested in #14843, pass vertex tex_coord to vertex shader and to fragment shader even when no texture is bounded so they can be used by custom fragment shaders.
Stop the frame callback and flag the cursor for a refresh when the pointer re-enters the surface, but don't set a null cursor, as it may have already been set after entering a surface that is part of the window decorations, resulting in an unwanted invisible cursor.
Avoids UBSan warning (among other similar ones in SDL_thread.c):
src/thread/SDL_thread.c:109:13: runtime error: index 1 out of bounds for type 'struct (unnamed struct at src/thread/SDL_thread_c.h:70:5)[1]'
SUMMARY: UndefinedBehaviorSanitizer: undefined-behavior src/thread/SDL_thread.c:109:13
Requesting certain MIME types (e.g. EPS formats offered by KDE) can be *very* slow, on the order of multiple seconds, due to requiring significant processing. Only try to load image MIME types that SDL is known to support (BMP and PNG).
The default timeout value of 14ms is ideal when querying clipboard data while polling events, to prevent excessive lag if the source takes a long time to respond, however, when reading from SDL_GetClipboardData(), the timeout can be too short if a large amount of data must be processed or transferred. SDL_GetClipboardData() is not called while polling events, so using a longer read timeout to greatly increase the chance of success is acceptable.
Use a 5 second timeout when reading from SDL_GetClipboardData() and GetPrimarySelectionText() to greatly increase the chances of a successful read, even if the requested format requires heavy processing.
Clients that defer repainting may hang in SDL_WaitEvent() while interactively resizing if they only redraw when an appropriate event is received, as resizing defers the new state until a frame callback is received, and if too much time elapsed since the last redraw, the last frame callback may have already occurred. Send an exposure event when deferring resizes so the client will make forward progress and trigger a frame callback to ack the pending configure state.
Prevent mouse and keyboard events from being processed twice by
skipping [super sendEvent:] for events SDL has already handled via
Cocoa_DispatchEvent. Other event types still go through AppKit's
normal handling.
The hybrid handling can still result in cases where a final event is dropped when the pointer leaves a surface. The spec says that all pointer events should be handled within a frame, so, do so.
When running SDL3 applications on tiling window managers like i3, moving a window to an invisible workspace does not trigger SDL_WINDOW_MINIMIZED or SDL_WINDOW_HIDDEN. Consequently, the application continues rendering at full speed (VSync dependent), consuming unnecessary GPU/CPU resources even when not visible.
When a workspace is hidden, i3(and possible other tiling WMs) unmaps the container and sets the client window state to WithdrawnState (via the WM_STATE atom). Previously, the SDL3 X11 backend ignored changes to WM_STATE during PropertyNotify events, failing to detect this transition.
Some devices with broken drivers hang when their name is queried, so added a workaround for applications that don't need input device details. The long term fix is to move the hotplug detection into a separate thread.
Older versions of XInput2 do not declare struct XIGesturePinchEvent
in XInput2.h, causing compilation failure in SDL_x11xinput2.c
Check for XIGesturePinchEvent in the test for enabling
SDL_VIDEO_DRIVER_X11_XINPUT2_SUPPORTS_GESTURE
hid.dll simply cannot send 7 bytes reports unlike other platforms
It enforces full length repots of 17 from the device's descriptor,
which does not work on the device.
This breaks ffb and led control, so we disable this by default on
windows.
Wireplumber now exposes a list of session services, so check the "session.services" property for an "audio" entry to determine whether Pipewire is configured for audio playback/capture.
CGDisplayPixelsHigh(kCGDirectMainDisplay) involves an IPC call to the
Window Server on each invocation. Cache the main display height in
SDL_CocoaVideoData and update it only when display configuration changes,
reducing overhead during high-frequency mouse event processing.
`target_get_dynamic_library` should not be called if `SDL_HIDAPI_LIBUSB_SHARED` is set to OFF, it otherwise causes a warning at best, or a build failure if libusb is provided by a parent project and not installed on the system
It turns out the reason this function was having so many overread issues was because our row copies were wrong - for compressed images we also need to reduce the row count based on the block size, similar to what we already do for pitch calculation - these copies are byte copies, not pixel copies!
The event core will do so automatically, and this may end up dropping events in rare cases when exiting fullscreen if an event with the final bordered window size is sent before the event notifying that the borders have come back on.
Don't decide if the button is pressed by the `buttons` bitmask, but rather by
event type. On macOS, the trackpad might produce a mousedown event with taps
instead of full clicks--if tapping is enabled in System Preferences--and in
this case might not set the flag in the bitmask.
Fixes#14640.
Events won't be delivered to the regular event watcher list callbacks if a client event filter discards events. Use the special window event watcher list to watch for resizes in the GPU renderers, as events are delivered to this list before a client can potentially discard them.
Anecdotally, some force-feedback wheels have been reported to experience a
reduced "definition", "texture", "precision", or "je ne sais quoi", which
appears to be caused by sending more update flags than necessary to
DirectInput.
This may be related to the fact that there are two USB PID packets that are
sent when updating a device: One contains the "general" force data, and the
other contains the "type-specific" data. My speculation is that many wheels
expect to only receive the latter, and misbehave when receiving both.
This has been tested and validated anecdotally by others who have received
a hacked-together version of PCSX2 that corrects the flags sent to DirectInput,
who noted a significant improvement in the "feeling" of the FFB effects.
The only way to validate this at a technical level is to grab a wheel that uses
the "generic" DirectInput FFB drivers (which map nearly 1:1 with the USB PID
specification), and inspect the USB packets (e.g. with USBPcap) to check whether
redundant data is being sent.
This upsets things, because you end up calling a javascript function with
more arguments than it expects, and if asserts are enabled, Emscripten notices
this and aborts the program when you hit this code.
Reference Issue #14670.
Wayland icons must be square, so scale non-square images to fit a square region instead of failing. This matches X11 behavior.
A warning that the image will be scaled will be logged.
As in the previous commit, loading GTK while setuid or setgid would
result in the process exiting. This is equally true if it's loaded
indirectly, for a libdecor plugin.
libdecor doesn't currently have any API by which it can be asked to
avoid specific plugins, but its GTK plugin declines to initialize if it
detects a non-main thread (because GTK documents that it must only be
used from the main thread), resulting in libdecor falling back to the
lower-priority Cairo plugin. We can make use of this by intentionally
initializing libdecor on another thread if we have been asked to avoid
GTK. This is a bit of a hack, but at worst it should be harmless.
Resolves: https://github.com/libsdl-org/sdl2-compat/issues/564
Signed-off-by: Simon McVittie <smcv@debian.org>
GTK explicitly doesn't support being used setuid or setgid, and if SDL
loads and initializes GTK, GTK will exit the process if it
detects a setgid executable. This is incompatible with the historical
practice of making game executables setgid in order to write out a
shared high-score table on multi-user systems (which is security theatre
at best, because typical game runtime libraries are not hardened against
an untrusted caller, but making it regress would be a user-observable
regression in sdl2-compat).
Helps: https://github.com/libsdl-org/sdl2-compat/issues/564
Signed-off-by: Simon McVittie <smcv@debian.org>
Currently, no compositor that supports the warp protocol sends a motion event after doing so, so SDL must synthesize one when using the warp protocol, as it did with the locking protocol.
This can be avoided once
https://gitlab.freedesktop.org/wayland/wayland/-/merge_requests/340 is completed.
If GL/Vulkan was loaded manually via a SDL_X_LoadLibrary() call instead of via window flags or through the GPU or renderer system, and there is no matching SDL_X_UnloadLibrary() call, the library instance won't be automatically unloaded while shutting down the video system, as the refcount won't go to zero, which can cause problems with some backends, and leaves a leaked DLL/SO handle when the global video object is destroyed.
Ensure that the libraries are unloaded after destroying the windows, but before shutting down the video backend, to prevent a leak and possible driver errors when shutting down the video backend.
An event flush while a window is being hidden may try to toggle the window borders. Don't poll the mapping state when toggling borders if the window is in the process of being hidden, or the window may already be unmapped, and the wait loop will hang forever.
This matches other platforms.
Also make sure Android explicitly sets vsync on restore, regardless of the
specific setting, to make sure it's consistent.
Fixes#14642.
Calling SteamAPI_InitEx() will set environment variables that SDL uses to properly support the Steam virtual gamepad. Make sure that we fall back to the real environment for the variables that Steam sets.
We might be setting an environment variable along with other application code setting environment variables (like code in SteamAPI_InitEx()) before initializing SDL.
Core X11 events are used for absolute motion while the mouse is grabbed, as XInput doesn't receive motion events from the master pointer, and the position from the slave devices lags one report behind. However, pointer events emulated from multitouch events can't be filtered out when using core events, as the XInput flags are not passed through.
Use the absolute position from slave devices in XInput motion events while the mouse is grabbed and touch events are active to allow filtering out pointer events emulated from touch events. Pointer events in this case may be a report behind the master device, but receiving input from both the pointer and touches simultaneously is not likely to be a common occurrence.
The script assumed Git's traditional loose refs or packed-refs format.
When cloning with extensions.refstorage=reftable, neither .git/refs/*
nor .git/packed-refs exist, causing cmake configuration to fail.
Fall back to git rev-parse when file-based ref lookup fails.
It turns out that some distros set SDL_IM_MODULE globally, which leads to broken text input with this check, and there really is no reason not to use the text input protocol when available.
Reverts 2b375d9704
Wayland cursors are surfaces sized in scaled points, so this hint wouldn't change the size, just prevent larger backbuffers from being used to draw sharper scaled cursors.
This gives us more robust handling of Bluetooth Xbox controllers which may vary the report format between firmware versions.
Firmware versions tested:
Xbox One S: 3.1.1, 4.8.1923, 5.13.3143
Xbox One S/X: 5.11.3118, 5.23.6
Xbox Elite Series 2: 5.22.16, 5.23.6
Fixes https://github.com/libsdl-org/SDL/issues/14597
We only want to change cursor display scale if the application or the user opts in by setting a hint. Otherwise cursors could change size unexpectedly when upgrading to SDL 3.4.0.
Windows may be destroyed in event handlers, so check the window validity after pushing window events onto the queue to ensure that the window and its properties are still valid.
This reverts commit 3ee29f2e7a, among other
changes. This means getting a scary warning on the javascript console in
Emscripten by default, but this is easily fixed by setting a GL swap
interval at startup or creating a 2D renderer with the appropriate property
(or call to SDL_SetRenderVSync()).
Fixes#14625.
The was duplicated in the XInput2 code, but taking an SDL_VideoData parameter instead of SDL_VideoDevice. Remove the duplicated XInput2 function, and use an SDL_VideoData parameter for X11_FindWindow to avoid an unnecessary dereference.
A common use case is asking someone to clone SDL and run a test. This simplifies the instructions so we don't also have to tell them how to enable the tests. Anyone savvy can just add -DSDL_TESTS=OFF if they don't want to have them.
The Emscripten support for setting window icons introduced in #14490 does not work when using pthreads. Using `SetWindowIcon` will cause the program to crash. The reason for that is that the `Blob` constructor does not accept shared memory (`SharedArrayBuffer`).
We need to create a local copy of the icon data if necessary.
* Implemented EnumerateStorageDirectory()
* Implemented RemoveStoragePath()
* Fixed init failure if the application previously crashed with a write batch in progress
* Allow creation of multiple Steam storage objects, the write batch is complete when the last storage is closed
* Check to see if a file exists in GetStoragePathInfo()
If you want access to the SDL_REVISION define,
you must explicitly include this header.
Language like this was present in the SDL2 version,
but disappeared in SDL3, even though it's still true.
This reverts commit a3992f504c.
The reason SDL_revision.h was not included by SDL.h or SDL_version.h is because SDL_revision.h changes with every git commit, causing a complete SDL rebuild for every trivial git commit.
* GPU: D3D12: Missing WinPixEventRuntime.dll now logs warning at device creation time.
* GPU: D3D12: Debug label functions are now a no-op when WinPixEventRuntime.dll is missing.
* Docs: GPU: Debug function documentation now notes WinPixEventRuntime.dll is required under D3D12.
In the unlikely case that the overflow check should fail,
the else clause would switch to the truncated stack message
without updating the len variable. This would contain the
return value from vsnprintf(), meaning it could point beyond
the buffer.
The subsequent code which trims NL and CR from the buffer,
would then read -- and possibly write -- out-of-bounds.
To fix this, we split the two joint conditions into separate
if-clauses, and adjust the len variable in the case where we
know the message buffer was truncated.
We already set the main thread when creating a thread or calling SDL_Init(), and applications can call SDL_SetMainReady() themselves for special cases, so there isn't a lot of value of forcing everyone through the SDL_RunApp() flow.
Fixes https://github.com/libsdl-org/SDL/issues/14580
This prevents validation errors on D3D12 caused by the previous solution
that utilized internal D3D12 functions. This is still the fallback
solution if WinPixEventRuntime.dll is not present in PATH, but a warning
is printed to let you know how to fix these errors.
SDL_SetWindowFullscreen now reports failure while a fullscreen transition is
already in progress on Emscripten, upsetting the test when it tries to come
back to windowed mode, but testautomation will never give correct results here
even when we fix the backend to queue pending reversal requests, because
you'll have to return to the Emscripten mainloop before these state changes
will occur.
It's just sort of luck this specific test passed before on this target.
At best, this is a no-op.
At worst, it might:
- Clobber a signal handler someone registered after us
- Overwrite the signal mask or flags
- Cause unregistration to fail (sigaction() isn't guaranteed to return the exact pointer passed to signal())
If video is initialized, the main thread is the one that initialized video, otherwise if events are initialized, the main thread is the thread that initialized events, otherwise the main thread is the one that called the main function.
Fixes https://github.com/libsdl-org/SDL/issues/14511
This lets applications choose which scene their windows are being created in. For example one window might be video output in a scene with a session role of UIWindowSceneSessionRoleExternalDisplayNonInteractive, and video controls in a scene with a session role of UIWindowSceneSessionRoleApplication.
Now other projects can use it with their headers, if they provide an options
file with specifics and are willing to follow some SDLish header conventions.
This shouldn't affect SDL satellite projects, etc, since they literally use
the SDL symbols, like `SDLCALL` and such, which are the defaults in the script
if a .wikiheaders-options file doesn't override them.
Grabbing the mouse interrupts touch events in progress, so if someone
enables relative mode while a button is pressed, wait for the button to
be released before grabbing the mouse.
It'll usually replace it with `{ "SDL_app", NULL }`, but things like Win32
can query the OS for the original command line arguments.
This allows apps/scripting languages that provide their own entry points to
use SDL_RunApp and not have to worry about how to compose an argv array on
things like Windows, when SDL was going to do it for them anyhow.
Most things won't experience any change with this commit, including apps that
that want extra control but originate in a standard main()-style entry point
and can just pass the existing argc/argv through to SDL_RunApp.
Windows isn't addressed here, since a previous commit already updated it.
GDK has a different fix here, but we'll unify that in a later commit.
Closes#12676.
This, strictly speaking, doesn't _need_ to be public domain any more, since
SDL3 no longer inserts this code into apps via a static SDL_main library; it's
just part of the normal zlib-licensed SDL library and called into through a
well-defined entry point at startup, now.
But removing a public domain notice isn't the done thing, so we'll leave it in.
(To a related point: even if others have rewritten good portions of this code,
they're doing it under the assumption that this specific file has a public
domain license.)
This new implementation only parses the command line into an argv when
the provided argv is NULL. This lets programs that don't want to/can't
include `SDL_main.h` to do their own custom argument processing before
explicitly calling `SDL_RunApp()` without having SDL clobber the argv.
If the user includes `SDL_main.h` as normal, the behavior remains the
same as before (because `SDL_main_impl.h` passes a NULL argv).
In addition, this new implementation performs fewer allocations and no
longer leaks on failure.
Instead of only using it for the side effect of making MsgWaitForMultipleObjects block on the next call.
This has the added benefit of avoiding an extra MsgWaitForMultipleObjects call if there was actually new raw input in the queue already.
This still had a bunch of text that wasn't updated for SDL3's ability to deal
with pre-user-interaction pages. Same idea, but changed text that suggested
you have to wait to open the audio device.
Same deal as the current Windows code: this "fixes" proximity but limits you
to a single pen device. But this is probably okay for most reasonable use
cases.
Now everything will attempt to track pens through proximity changes (instead
of removing the pen entirely). testpen.c has been updated to reflect this.
Some platforms and devices are better at this than others, but this seems like
a significant usability improvement across the board.
Fixes#12992.
This reverts commit 964bedfdd9.
The vendorName is more expressive and lets the user distinguish between different controllers of the same type, e.g. "PlayStation VR2 Sense Controller (R)" and "PlayStation VR2 Sense Controller (L)" vs "Spatial Controller"
If the product category is necessary, we can add a property to store that as additional information on the controller.
Normally the gamepad is used for navigation on visionOS, but when the controller subsystem is enabled we want to receive gamepad input as gamepad events instead.
An active cursor being destroyed may still have a frame callback referencing the thread queue, so clean it up before destroying the thread queue to ensure that no references to the queue remain when it is destroyed.
This PR adds support to the cmake build scripts so to allow building SDL with the Tiny C Compiler (tcc).
TinyCC supports the subset of C99 used by SDL and will complete the build once the --version-script linker flag is removed. The changes have been tested with various build configurations, including X11 and Wayland, and using tcc version 0.9.28rc 2025-10-27 mob@f4e01bfc on x86_64 Linux.
Some caveats:
- the fake pen will leave proximity only when relative mode is disabled
- unsure if detecting proximity is even possible from raw mouse input
Fixes#12324.
This changes the default for SDL_GL_FRAMEBUFFER_SRGB_CAPABLE to -1 (for
"don't care"). Minor ABI break, but technically this never worked correctly.
Fixes#13055.
The OpenGL headers are not always implicitly available, so this improves the check by calling `find_package` and using the `OPENGL_INCLUDE_DIRS` or `OPENGL_INCLUDE_DIR` var for the `check_c_source_compiles` test.
The minimum CMake version currently set is 3.16, `OPENGL_INCLUDE_DIRS` was only added in 3.29, so the code is set to choose `OPENGL_INCLUDE_DIRS` if it exists. If the minimum CMake version is ever set to >= 3.29 this check can be removed and just the `OPENGL_INCLUDE_DIRS` variable can be chosen.
If this application is being run under Wine but Steam doesn't know that, Steam won't set STEAM_COMPAT_PROTON. So we'll use wine_get_version() to detect that we're running under Wine instead.
This block was a sloppy way of trying to avoid prioritizing lavapipe over another device, but truthfully the only way to guarantee avoiding CPU drivers is to add a property to allow apps to require hardware acceleration.
If a dispatch event when showing a window returns a failure code, handle the display disconnected condition and break out of the loop, otherwise, it will hang forever.
Compositors that send multiple resize events per frame could result in the most recent configuration requests being left un-acked when the resize ends. Ack the most recent resized state on frame callbacks, so as to always ack the most recent configuration within a reasonable timeframe.
When running on 32-bit Windows, DefWindowProc resolves to a function in ntdll.dll, but after the window is created the actual window proc is a function in user32.dll. We solve this by using our own custom default window proc and looking for that instead.
Fixes https://github.com/libsdl-org/SDL/issues/1442
This fixes a crash if a surface is RLE encoded, then locked and unlocked.
We also mark the surface as no longer needing to be locked after undoing RLE encoding
We don't need to aggressively disable RLE acceleration anymore because the original pixels remain available for operations that aren't supported by the RLE blitter.
When the upload needs realignment, a new buffer is created to do the
upload, and the source data is copied to the new buffer. This commit
fixes the issue where the memcopy can read off the end of the source
buffer since it is reading based on destination pitch instead of source
pitch.
The HID descriptor maps the C buttons as a right thumbstick, but this mapping matches the HIDAPI Nintendo 64 mapping.
The start button resets the controller in both USB and Bluetooth mode on macOS, but I've kept the mapping assuming this is a macOS bug that will be fixed.
It happens fairly often that people enable X11 support but don't install the extension libraries that SDL depends on. This makes those required if not explicitly disabled, which should help prevent accidentally broken builds.
Recalculate the backing viewport dimensions in the resize handler, otherwise, this data can be out-of-sync with the logical window size if queried during transition animations.
This makes sure that anything that hooks window creation to set up window proc hooks will call DefWindowProc instead of infinitely recursing when we set up our window proc afterwards.
Fixes https://github.com/libsdl-org/SDL/issues/10529
The controller can be in XInput mode or enhanced mode, so it will always show up as an XInput controller, and the enhanced mode controller will come and go as enhanced mode is enabled or disabled in the FlyDigi Space Station app.
Store the images in order from smallest to largest to simplify lookup and be able to early-out of the search loop even if there is no exact match for the scale.
Some compositors don't implicitly use the pointer cursor when the tablet cursor is not set, and the presence of a tablet doesn't necessarily guarantee pointer capability. Set the cursor for tablet tools independently of pointer cursors.
This required refactoring of cursor state handling, as well as some tablet related structures.
Cache the cursor image data at creation time, and use a viewport to render scaled custom cursors, instead of generating new cursor images for every scale.
The OS uses the presence of the UILaunchScreen key to determine whether to display the application in compatibility mode or display using the whole screen.
Allowing GCC/Clang to auto-vectorize with AltiVec support breaks support
for non-AltiVec compatible processors. Adding this option allows the
existing AltiVec specific code to function as it always did, but
prevents the compiler from inserting AltiVec instructions into arbitrary
codepaths that SDL can't gate off via the existing CPUInfo code.
pa_channel_map_init_auto() with PA_CHANNEL_MAP_WAVEEX does the wrong
thing as it just takes the lest significant bits of
WAVEFORMATEXTENSIBLE's dwChannelMask in order. This doesn't match SDL's
chosen channel ordering.
The implementation here mirrors what we do for PipeWire.
Setting AudioCategory_GameChat breaks audio on several devices, including Behringer U-PHORIA UM2 and RODE NT-USB Mini. We'll disable this for now until we understand more about what's happening.
This means we exercise the non-dlopen()-based code paths for these,
as used in Debian and the Steam Runtime.
Signed-off-by: Simon McVittie <smcv@collabora.com>
libdecor 0.3, which changes the signature of libdecor_new and
libdecor_decorate to use a const pointer to the interface struct, has
not yet been released. In the latest release, libdecor 0.2.4, the
interface struct is a mutable pointer.
This doesn't affect typical upstream builds with
SDL_WAYLAND_LIBDECOR_SHARED=ON, in which case we're casting a pointer
returned by dlsym(); but Linux distributions that want tighter control
over dependencies often prefer to link them in the normal way, in which
case the build will fail if the signature doesn't match.
Fixes: 33834360 "wayland: Fix libdecor incompatible pointer types"
Signed-off-by: Simon McVittie <smcv@collabora.com>
Stopping the thread also destroys the queue used by pointer surfaces, and if any seats still have a surface associated with the queue at that time, it will cause a libwayland warning. Destroying the seats first ensures that all surfaces associated with the thread queue are destroyed before the queue is destroyed.
DBus does not require a video backend, and DBus messages still need to be processed even when a video backend is not initialized. Move the DBus pump call to the main event message pump function so that DBus events are still dispatched, even when a video backend has not been initialized.
This allows changing out `__FILE__` in general and the file path reported in
SDL_assert specifically to not leak build machine path info, or just simply
not include the absolute path if you swap it out with `FILE_NAME` on Clang/GCC.
Fixes#14290.
In this case, it means a newly-created window on a specific display won't
get mouse/keyboard input (which, presumably, will continue to go to a window
on a different physical display instead).
This also makes SDL_SetWindowFocusable() functional on the kmsdrm backend, to
change this flag later, but to be clear, there are no window focus events to
jump between displays in this backend, so this is only useful to manually
tweak things later.
Fixes#14289.
There is no guarantee that libdecor will always dispatch the main queue (a future plugin may run on its own queue), so ensure that the main queue is still dispatched when showing a libdecor window, as is already the case when pumping events.
The queue running on the cursor thread might flush color events before their queue has been set. Use proxy wrappers for their parent objects to assign the queue atomically at creation time.
If the main event handling thread runs slowly, so will cursor animations. Use a dedicated thread for cursor surface events, so that animations will always run at a consistent rate.
Clean up and implement some best practices for event polling and handling:
- Ensure that events are still pumped, even if existing events are in the queue, to ensure that the most recent data is always processed.
- Handle wl_display_flush returning EAGAIN by trying to poll the socket until it can be written to.
- Ensure that events are always pumped on interrupts and broken pipe errors to handle quit events.
- Minimize double-pumping of events, since returning a success code from a wait will implicitly pump events again.
- Wake to poll when using a DBus IME, but the text input protocol is not enabled, to ensure that text events are delivered.
The higher-level functions only convert the base image, so alternate scaled images for icons and cursors need to be converted when being copied to the buffers.
If the system doesn't have Zenity installed, message boxes can fall back to the X11 toolkit while the Wayland driver is in use, in which case, a Wayland window may be set as a parent to an X11 dialog. Ensure that the parent window for an X11 dialog is actually an X11 window.
Allows for finer grained timeout values, and fixes a FIXME.
This also drops the legacy select() fallback path in favor of presuming that poll() is always available. poll() is part of the POSIX.1-2001 standard, has been available in Unix since some time in the 1980s, the BSDs since at least the early 90s, and Linux since kernel 2.1, which predates kernel support for Pthreads. glibc also has its own emulation using select(), if necessary.
The syntax errors in the enum lists tell us that no one ever actually
compiled this `WINVER < _WIN32_WINNT_VISTA` fallback path. 2f5bc17 made
it increasingly unlikely for retro Windows builds to hit this path by
adding a new set of `WINVER` override rules to `SDL_windows.h` that
check for the presence of certain headers. `HAVE_MMDEVICEAPI_H` in
particular will be `#define`d for any CMake build with a ≥Vista-era
SDK, such as Visual Studio 2017's `v141_xp` toolset that 2f5bc17
targeted, and it's also unconditionally defined in the default
`SDL_build_config_windows.h` configuration used by the ready-made
Visual Studio project.
These fixes work for a XP-era configuration that undefines all of the
`HAVE_*_H` macros and `SDL_VIDEO_RENDER_D3D12` to make `SDL_windows.h`
fall through to the `#define _WIN32_WINNT _WIN32_WINNT_WINXP` case, on
both Visual Studio 2022 and Visual Studio 2017's `v141_xp` toolset.
Adds support for animated cursors on Cocoa, Wayland, Win32, and X11.
testcursor can take a semicolon separated list of filenames and load an animated cursor from them.
015cc6a broke texture format selection for this kind of surface if the
renderer's first texture format doesn't happen to have an alpha
channel. This happens for on-screen software renderers on Windows,
whose preferred format is the window's own `SDL_PIXELFORMAT_XRGB8888`.
This fixed check also cover the intent behind 015cc6a, assuming that
indexed and alpha formats are mutually exclusive.
Convert SDL_d3dmath.c functions to static inline in SDL_d3dmath.h to make
it less likely to conflict when static linking SDL. raylib's SDL backend
does not work with a "normal" upstream binary static link. It has these
errors:
/usr/bin/ld: /usr/local/lib64/libSDL3.a(SDL_d3dmath.c.o): in function `MatrixIdentity':
SDL/src/render/SDL_d3dmath.c:35: multiple definition of `MatrixIdentity'; CMakeFiles/raylib.dir/rcore.c.o:rcore.c:(.text+0x18470): first defined here
/usr/bin/ld: /usr/local/lib64/libSDL3.a(SDL_d3dmath.c.o): in function `MatrixMultiply':
SDL/src/render/SDL_d3dmath.c:44: multiple definition of `MatrixMultiply'; CMakeFiles/raylib.dir/rcore.c.o:rcore.c:(.text+0x18540): first defined here
collect2: error: ld returned 1 exit status
It works if these functions aren't emitted, (i.e.
!SDL_VIDEO_RENDER_D3D(11|12|GPU|VULKAN)
Or, if SDL is not linked static. Which I know is preferred anyway.
In raylib's case, it doesn't use this code anyway so it's not a problem
to build it yourself with them disabled. But it's a minor hassle to be
incompatible with static linking libraries from upstream.
I can't see any good reason for these functions to not be static inline
already, so I just put them in the header and deleted the .c.
Raylib's conflicting case is a library with a public function exposed to
users. But SDL's case is not.
- Moved all matrix functions from SDL_d3dmath.c to SDL_d3dmath.h as static inline
- Removed SDL_d3dmath.c from all project files (Visual Studio and Xcode)
- Functions affected: MatrixIdentity, MatrixMultiply, MatrixScaling,
MatrixTranslation, MatrixRotationX, MatrixRotationY, MatrixRotationZ
Changed SDL_TRY_ACQUIRE and SDL_TRY_ACQUIRE_SHARED success value
from 0 to true for functions that now return bool instead of int.
This fixes false positives/negatives in Clang's thread safety analysis.
Detect times that cross forward/backward across the zero timepoint, and handle them appropriately, so that events with an old timestamp don't mangle the time base offset, or have their timestamps mangled.
For SInput, the paddle order is left/right, which conflicted with the paddle order 1/Right, 2/Left.
The order has been fixed to properly apply the mapping strings generated to be 2/Left, 1/Right, 4/Left, 3/Right.
gcc < 3.1 doesn't support 'used' attribute at the least.
Also change the SDL_ELF_NOTE_DLOPEN variadic macro for disabled case to
be compatible with old gcc versions: tested with gcc-2.95.
Old MSVC cases, e.g. VS 2003 and older, are still not resolved.
Reference issue: https://github.com/libsdl-org/SDL/issues/14214.
SDL_PumpEvents() was freeing the temporary memory in the clipboard event, so if the application was iterating over the mime types in the event and retrieving the clipboard data, it would crash after the first entry.
When descriptor leaks overflow, the D3D12 SDL GPU renderer fetches more from the pool, but never returns them, which eventually causes an "out of memory" crash.
On Android, if you create a window with SDL_WINDOW_OPENGL, you can't create a Vulkan surface. The error message has been improved to reflect this, and the error is propagated back up to the application.
Also added warn level logging if the renderer couldn't be created.
This was added in 2016, presumably to help address the move/resize issues on Windows, which have been since been addressed by the live-resize functionality.
Fixes https://github.com/libsdl-org/SDL/issues/14079
Test code:
---
int main( int argc, char *argv[] )
{
SDL_Surface *orig = SDL_LoadPNG("testyuv.png");
SDL_Surface *surf16 = SDL_ConvertSurface(orig, SDL_PIXELFORMAT_RGB565);
SDL_Surface *surf32 = SDL_ConvertSurface(surf16, SDL_PIXELFORMAT_ARGB8888);
Uint64 then = SDL_GetTicks();
for (int i = 0; i < 100000; ++i) {
SDL_BlitSurface(surf16, NULL, surf32, NULL);
}
Uint64 now = SDL_GetTicks();
SDL_Log("Blit took %d ms\n", (int)(now - then));
return 0;
}
---
Results on my system:
BlitNtoN: Blit took 34522 ms
Blit_RGB565_32 (3 LUT): Blit took 9316 ms
Blit_RGB565_32 (1 LUT): Blit took 5268 ms
Blit_RGB565_32_SSE41: Blit took 1619 ms
This beats the previous 3-LUT version and even beats SSE on my system.
Test code:
---
int main( int argc, char *argv[] )
{
SDL_Surface *orig = SDL_LoadPNG("testyuv.png");
SDL_Surface *surf16 = SDL_ConvertSurface(orig, SDL_PIXELFORMAT_RGB565);
SDL_Surface *surf32 = SDL_ConvertSurface(surf16, SDL_PIXELFORMAT_ARGB8888);
Uint64 then = SDL_GetTicks();
for (int i = 0; i < 100000; ++i) {
SDL_BlitSurface(surf16, NULL, surf32, NULL);
}
Uint64 now = SDL_GetTicks();
SDL_Log("Blit took %d ms\n", (int)(now - then));
return 0;
}
---
Results on my system:
BlitNtoN: Blit took 34522 ms
Blit_RGB565_32 (3 LUT): Blit took 9316 ms
Blit_RGB565_32 (1 LUT): Blit took 5268 ms
Blit_RGB565_32_SSE41: Blit took 6399 ms
Tray events on *nix platforms usually run over DBus, and the events subsequently aren't delivered via the window event queue. As a result, SDL_WaitEvent() won't unblock when tray events arrive, particularly if there is no currently active window.
Wake up periodically to poll when tray items are active to avoid blocking the delivery and processing of tray events.
While in a modal loop, the size in WM_WINDOWPOSCHANGING/WM_WINDOWPOSCHANGED may only be updated if the window is being resized interactively. Set the SWP_NOSIZE flag if the size hasn't changed from the last move/size event, or a size set programmatically may end up being overwritten by old size data.
When iterating over the keymap entries, a valid xkb state object has already been allocated, so use that instead of allocating/destroying a new state object for every lookup, which avoids a calloc/free operation inside libxkbcommon. Any state set by the level lookup function will be overwritten with valid state after keymap iteration has completed.
The spec doesn't guarantee that a modifier event won't arrive before a keymap event, or that it will always be sent after a keymap change if the modifiers and layout index haven't changed, so restore any valid state after allocation when building a new keymap.
The uclibc version of SDL_snprintf() is currently limited to values
<ULLONG_MAX due to the `unsigned long long` cast in SDL_PrintFloat(),
but it makes sense to at least ensure that it does support the full
unsigned 64-bit range. This also covers the one defined case where a
32-bit MSVC build can't assume that _ftoul2_legacy() == _ftol2().
cl.exe versions ≥v19.41 call this builtin for double → uint64_t
conversions on x86. SDL currently needs such conversions in:
* MainCallbackRateHintChanged()
* SDL_PrintFloat()
* WIN_ApplyWindowProgress()
This seems enough to justify implementing this function rather than
trying to work around it, as it was done in sdl12-compat:
https://github.com/libsdl-org/sdl12-compat/issues/352
This implementation was taken from ReactOS:
https://git.reactos.org/?p=reactos.git;a=commitdiff;h=f637e6b809adb5e0ae420ef4f80c73b19172a2e7
Passes the stdlib testautomation, and also matches the behavior of
Microsoft's 64-bit libc for the currently implementation-defined case
of calling SDL_PrintFloat() with values >SDL_MAX_UINT64.
This is probably something we already cleaned up that has something running
in an unexpected order now that we've moved disconnect work to the main thread.
We can tell it's SDL 3 from the version number and git revision,
so there's no need to duplicate that in the prefix.
Signed-off-by: Simon McVittie <smcv@collabora.com>
SDL_CreateGPURenderer() now allows passing in an existing GPU device and passing in a NULL window to create an offscreen renderer.
Also renamed SDL_SetRenderGPUState() to SDL_SetGPURenderState().
This function is only available since version 1.0.0, but the SDL minimum is 0.5.0, and Steam Runtime 2 'soldier' uses 0.8.2, so add a fallback function with similar functionality for older versions.
xkb_keymap_key_get_mods_for_level() is more efficient, so it is still favored when available.
Full object validation was always done before this hint and there are a number of shipping products that relied on this, so don't change it by default.
The "as-installed" tests have metadata files in the .desktop-like format
used by gnome-desktop-testing, which uses a subset of shell syntax for
the Exec field. The list of arguments is represented as a
semicolon-separated list by CMake, but we need to convert that into a
space-separated list for the Exec field.
Strictly speaking we should be quoting the arguments with the equivalent
of Python's shlex.quote, but I couldn't find a way to do that in CMake,
and currently none of the tests have arguments that need quoting.
Signed-off-by: Simon McVittie <smcv@collabora.com>
You must set SDL_PROP_GPU_DEVICE_CREATE_FEATURE_DEPTH_CLAMPING_BOOLEAN to false on this platform, since setDepthClipMode is not supported on Apple headsets (as of visionOS 2.3) and clipping is the default.
VUID-VkPipelineRasterizationStateCreateInfo-depthClampEnable-00782(ERROR / SPEC): msgNum: -316906200 - Validation Error: [ VUID-VkPipelineRasterizationStateCreateInfo-depthClampEnable-00782 ] | MessageID = 0xed1c6528 | vkCreateGraphicsPipelines(): pCreateInfos[0].pRasterizationState->depthClampEnable is VK_TRUE, but the depthClamp feature was not enabled. The Vulkan spec states: If the depthClamp feature is not enabled, depthClampEnable must be VK_FALSE (https://vulkan.lunarg.com/doc/view/1.3.268.0/windows/1.3-extensions/vkspec.html#VUID-VkPipelineRasterizationStateCreateInfo-depthClampEnable-00782)
Each texture gets its own palette, the palette is updated when it is rendered if necessary, rendering is flushed if a texture render is already queued up.
Trying to share palettes and handle the different cases of queued rendering and changing palettes in the middle of the frame ends up being very complicated, so we'll keep it simple for now.
The old names made it difficult for people searching or otherwise trying to understand the codebase, as they were somewhat mangled from their expected names (e.g. configure_toplevel_xdg vs xdg_toplevel_configure, the latter of which would be expected by someone reading the spec).
This avoids situations like:
- PulseAudio holds its own lock in the hotplug thread.
- The hotplug thread notices a device went away.
- The hotplug thread calls SDL_AudioDeviceDisconnected().
- SDL_AudioDeviceDisconnected() tries to grab the device lock.
- The device thread is holding the device lock...
- ...but is currently waiting on PulseAudio's lock to release.
In short: deadlock. It's better to queue this work to run on the main thread,
where we can guarantee a start with _none_ of the audio subsystem locks held.
If multiple threads are calling SDL_UpdateJoysticks(), then an overlapped read can be initiated on one thread (read_pending set to true) and GetOverlappedResult() called on another thread. This results in ERROR_OPERATION_ABORTED. This is harmless so we shouldn't return an error in this case, we'll just retry the read on the next call.
Fixes https://github.com/libsdl-org/SDL/issues/14033
If a client takes a long time to present the first frame after creating the window, a configure event to set the suspended state may arrive with libdecor increasing the content size by the decoration dimensions, which should be ignored.
As documented, the contained_signature is to be passed in as a
nul-terminated C string.
For basic types that are represented by a single character, on
little-endian platforms, putting the type in the least significant
byte of an int and casting its address to `char *` happens to result in
a valid string, because the int's in-memory representation looks like
`(char []){ 'b', 0, 0, 0 }`. However, on big-endian platforms, the int's
in-memory representation is `(char []){ 0, 0, 0, 'b' }` which is not
a valid type for a D-Bus variant to hold (it is interpreted as an empty
string, and variants are not allowed to be empty).
Instead, do this the straightforward way, with a mnemonic string and
no casts (in the same style used in `SDL_portaldialog`).
Fixes: 3f2226a9 "Add progress bar support for Linux"
Resolves: https://github.com/libsdl-org/SDL/issues/13953
Bug-Debian: https://bugs.debian.org/1115705
Signed-off-by: Simon McVittie <smcv@collabora.com>
Slave pointer devices report raw button values, while the master pointer device reports button values with remapping applied. Manually apply the remapping table to slave device buttons to eliminate multiple button events from one press, and allow button remapping to function when relative mode is active.
Fixes api availability compiler warnings when building the joystick subsystem for macOS versions older than 11.0. The compiler expects the version number passed into @available and API_AVAILABLE macros to be greater than or equal to 11.0, which 10.16 is not. (Even though 10.16 and 11.0 are semantically equivalent, the compiler doesn't know this.)
fix 'JVM object referenced by 'context' is of type 'Context' and it does not have access to method 'registerReceiver(BroadcastReceiver, IntentFilter)' declared in class 'ContextWrapper'.'
Full paths are used as-is, relative paths are prepended with SDL_GetAndroidInternalStoragePath(), and operations fall back to the asset system as appropriate.
This is consistent with the behavior of opening files on Android.
If the application is waiting in SDL_WaitEvent(), we still need to run event loop maintenance in between calls to Android_PumpEvents().
Fixes the testautomation events_mainThreadCallbacks() test on Android.
SDL_WaitEvent() will pump OS events if no poll sentinel is pending, so flush events before inserting a user event to ensure that the user event will be next in the queue.
SDL supports the following use cases:
* Normal operation with fast parameter checks (default):
SDL_SetHint(SDL_HINT_INVALID_PARAM_CHECKS, "1");
* Object parameters are checked for use-after-free issues:
SDL_SetHint(SDL_HINT_INVALID_PARAM_CHECKS, "2");
* Enable full validation, plus assert on invalid parameters:
#define SDL_ASSERT_INVALID_PARAMS
* Disable all parameter validation:
#define SDL_DISABLE_INVALID_PARAMS
VMware relies on the cursor being set to NULL to optimize relative mouse motion for games.
We should also revisit whether current RDP works better with a NULL cursor or a transparent cursor.
Fixes https://github.com/libsdl-org/SDL/issues/13700
This architecture is deprecated and is now failing with:
Windows SDK 10.0.22621.0 : 'C:\Program Files (x86)\Windows Kits\10\include\10.0.22621.0\um' not found or was incomplete
Fixes https://github.com/libsdl-org/SDL/issues/13949
Previously the test would kill the child process while it was in the process of initializing (loading DLLs, etc) and this would cause the test to fail.
This made sense when switching between a window and an exclusive fullscreen video mode of the same dimensions, but it causes user confusion when the window changes size, and doesn't work well with Wayland. I've gotten this feedback from several sources, so we'll just go ahead and remove this behavior for now.
Fixes https://github.com/libsdl-org/SDL/issues/11577
This happens to work here because we don't have anything too complex, afaik.
Rejecting strings with underscores was crucial for other things, like video
with all its X11 variables, so this can't be applied to the other subsystems.
(Also, renaming the `SDL_VIDEO_RENDER_OGL_ES2` variable might be nice, too,
but let's keep this simple for now.)
Fixes#13938.
* SInput: version capabilities compression
This commit includes additions relating to SInput generic device reporting capabilities in a bit more detail, to automatically choose the best input map possible for the given device.
Thanks to Antheas Kapenekakis (git@antheas.dev) for contributing the neat compression algorithm, this is pulled from the PR Draft here: https://github.com/libsdl-org/SDL/pull/13565
Co-authored-by: Antheas Kapenekakis <git@antheas.dev>
This works without it, but generates an extra redirect on the web server to
the same a URL with the '/' appended, so we might as well supply it ourselves
and remove the extra roundtrip to the server.
It always fails, because we don't set the needed link library, i.e. shcore.
Besides, SDL3.dll doesn't link to shcore.dll, it links to it at runtime, so
it is unnecessary to check for that symbol here.
Reference issue: https://github.com/libsdl-org/SDL/pull/13904
The software renderer now matches the 3D renderer behavior by rendering a reversed rectangle (extending width or height in the opposite direction) when width or height is negative. It also now renders no rectangle if the width and height are 0.
Fixes https://github.com/libsdl-org/SDL/issues/13400
(buttons above 12 all shifted down by four, since the dpad buttons now look
like a hat...the only one above the dpad buttons on a standard layout is the
guide button, index 16. This change moves it to 12.)
snd_pcm_recover() puts the device back in SND_PCM_STATE_PREPARED state; you
have to explicitly restart the device afterwards with snd_pcm_start().
Fixes#13761.
This now relies on the implementation to set these flags on short reads/writes
instead of the higher level checking if SDL_SetError() was called.
Additionally (and crucially), this now sets ERROR or EOF on all short reads,
across all backends, not just when we get a zero-byte return value.
Fixes#13720.
Otherwise, groff will interpret it as a macro, causing the man page
for SDL_GetPrefPath() to be mis-rendered. Escape unescaped backslashes
as `\(rs` ("reverse solidus") before escaping other characters with
macros that, themselves, contain backslashes.
Resolves: https://github.com/libsdl-org/SDL/issues/13039
Signed-off-by: Simon McVittie <smcv@collabora.com>
There is an escaped pipe character, because this ends up in a Markdown table
on the wiki, and the wiki software (incorrectly, I think) sees the pipe inside
a code-block and thinks it's the end of the table cell if not escaped.
But this escape char looks wrong everywhere else.
Keeping the parameter table terse is always a good idea, so I moved the
detail out to the Remarks section, which doesn't have a problem with this
character appearing in the text.
When linking directly to libX11 as a hard dependency, we assign the
addresses of functions like XDestroyImage to function pointers like
X11_XDestroyImage. However, by default Xutils.h doesn't declare
XDestroyImage as a function: it only provides a macro which looks into
the XImage struct and calls a function pointer directly, similar to
a virtual method.
SDL_x11framebuffer.c was already relying on being able to call the macro
without explicitly linking to libX11, so do the same here.
Fixes: d14cbd7b "Introduce X11 toolkit and make message dialogs use it"
Signed-off-by: Simon McVittie <smcv@collabora.com>
Needed to generate debug symbols for both release and debug builds and explicitly set the format so all arches actually generate the dSYMs.
This allows distributing an SDL3 app to the AppStore without the need to generate the dSYMs via script when using either the xcFramework or embedding the SDL3 Xcode project.
Fixes#13573
clang doesn't support -mcrc32 until version 14.0, therefore drop -mcrc32
use in sse 4.2 test, and change it to check _mm_cmpgt_epi64() instead of
_mm_crc32_u32(). without this, sse4.2 check was failing with clang <= 13
Fixes https://github.com/libsdl-org/SDL/pull/12223
Found when running in valgrind looking at another issue.
- RenderPass' depth_stencil_target
Tripped in SDL_BindGPUFragmentSamplers when not binding a DS target
- VulkanCommandBuffer's swapchainRequested
Tripped in VULKAN_Submit for the end transition barrier when
creating an image. The field is only reset when reused, not on first
use
When attaching a renderer (GL based specifically) to a window that was not created with the appropriate flags, the window would be destroyed and recreated to configure it for the desired rendering backend. While most of the issues with this have been mitigated over time, there can still be some undesirable side effects from doing so on certain platforms.
If the window was just created and was never configured for any graphics context, it is possible that the reconfiguration can be done without destroying the window first.
The Wayland implementation fixes an issue when creating a window with the fullscreen flag on wlroots based Wayland compositors, and can likely be extended to other platforms to avoid unnecessarily destroying/recreating a window in the very common case where a window is created, followed immediately by attaching a renderer.
This is a dirty hack, but it currently works for our purposes. We just refuse
to link an eligible string (like, say...`SDL_shadercross`), if there's a '['
character right before it.
This WILL NOT WORK if the string is in the middle of a link! For example, if
you had `[please refer to SDL_shadercross](https://example.com/)`, this will
still fail in the usual ways, but we're not building out a robust parser here
at the moment, and this catches the most common problem.
Fixes#13160.
Fixes https://github.com/libsdl-org/sdlwiki/issues/752
Closes https://github.com/libsdl-org/sdlwiki/pull/725
If a bound keymap is being destroyed, unbind it to ensure that the keyboard state won't be left pointing to an invalid keymap. Backends now no longer need to manually do this when destroying keymaps during keyboard removal and shutdown.
The dongle appears to stop sending input reports with some combination of rumble patterns, easily reproduced with Forza Horizon 5. We'll disable rumble temporarily until @8BitDo can investigate.
This correctly sets the xdg portal fields for targeting a specific
new filename or existing file.
"current_name" sets the dialogs placeholder name.
"current_file" targets an existing file.
"current_folder" for when the target is a folder.
We previously thought this wasn't possible because constant buffer offsets and
partial updates were unavailable, but we were reading the wrong table - this is
only the case for D3D11...
https://learn.microsoft.com/en-us/windows/win32/direct3d11/overviews-direct3d-11-devices-downlevel-intro
... while 12 doesn't list this feature at all:
https://learn.microsoft.com/en-us/windows/win32/direct3d12/hardware-feature-levels
We double checked and Jesse Natalie confirmed that this feature is required for
D3D12 even for 11_0 drivers. (Thanks Jesse!)
Additionally, D3D12 requires that UAVs are accessible from all shader stages,
meaning Tier 2 is enough to support the number of UAVs we need. Tier 1 could be
a property to lower the requirements, but that can be done later.
The X server maintains the desktop, but XRandR sends disconnect notifications for all displays. In this case fall back to the generic X11 desktop as a display.
- In previous firmwares for my gamepads, the buttons were sent in a static order for ABXY.
- To better be 'transparent' to the driver and save myself from future fragmentation, I am updating the mappings to be consistent, where the HID report should always expect to receive the button inputs based on cardinal button directions rather than button labels.
- This better aligns with the existing 'fallback' behavior of a generic device.
- Coincides with firmware library update: https://github.com/HandHeldLegend/HOJA-LIB-RP2040/pull/30
* Update NDK version to 28 and add 16kb page size linker flags to x86_64
* Remove Android Linker Options
16kb page size is now the default since NDK r28c
* Update Android CI to use NDK 28
SDL has been building on GNU/Hurd for a long time, using either drivers
based on external libraries (e.g. X11, pulseaudio, sndio, etc) or dummy
drivers. This commit introduces it explicitly as platform, so it can be
recognized, and tweaked as needed. In particular:
- introduce the SDL_PLATFORM_HURD define
- tighten/improve the platform detection in cmake, and use "Hurd" as
identifier
- return the platform name in SDL_GetPlatform()
- tweak the CFLAGS/LDFLAGS so pthreads can be used properly
- implement SDL_GetExeName(), using /proc/self/exe as provided by the
basic Linux-like procfs
- enable GLES 2 in tests (mostly for consistency with Linux)
In certain cases when moving fullscreen windows in scaled desktop configurations, the window origin might overlap two displays at once. Check if the window is at the origin of a specific display before falling back to the generic window rectangle check.
Fixes rare fullscreen window misplacement when moving fullscreen windows via a desktop shortcut while using the Wayland scale-to-display mode.
Use the existing SDL_GetExeName(), available for all the UNIX
platforms, in the internal GetAppName(); this has few advantanges:
- SDL_GetExeName() (and SDL_GetAppID() that builds on top of it) are
used in various places already; since it caches the executable name,
this may remove one extra read of the application name
- SDL_GetExeName() has a non-dummy implementation in more OSes than
GetAppName(), thus providing a small improvement for this IME
As drive-by change: since SDL_GetExeName() provides a constant string,
there is no more need to allocate a new string in GetAppName(), which
is used as constant string anyway. Hence, return a constant string in
GetAppName() too.
I assume the demoninator is a typo, rather than an indication that
someone has been playing too much Doom :-)
Signed-off-by: Simon McVittie <smcv@collabora.com>
"This thing allows to do something" is not really grammatically correct.
The closest rephrasing would be "allows one to do something" or "allows
the user to do something", but I think the passive voice reads more
naturally here.
Detected by Debian's lintian QA tool.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This reverts commit dc2c83c383
If you need to support the Google Play 16 kiB page size requirement, the recommendation is to use NDK r28c or newer, which automatically aligns binaries correctly.
Trying to determine where a window should be made fullscreen from the size and position can be unreliable if the window exceeds the display bounds. Add additional checks with the following priority:
- If the window was positioned with a macro that explicitly passes a display ID, store and use the requested display as the explicit fullscreen target.
- Check if the window position is an exact match for any display origins, and use that display if found, as positioning a fullscreen window by moving it to the origin of the destination display is common behavior.
- Fall back to the existing center point check if the previous checks were not successful, as it is known behavior, and won't risk breaking existing clients that rely on it.
The currently used way to determine the endianness (i.e. include
<endian.h> and use the __BYTE_ORDER macro) is provided in general by
GNU libc. Thus, extend that to any platform/OS based on GNU libc.
- Use modern Xkb functions where appropriate and cleanly separate the modern and legacy paths.
- Remove the deprecated XKeycodeToKeysym function in favor of directly querying the keymap on the legacy path.
- Look up virtual modifiers by name on the Xkb path to better handle remapping (equivalent to the modifier handling under Wayland).
- Optimize keymap creation on the Xkb path to cut keymap build times and enable fast group switching (equivalent to keymap handling on Wayland).
- Enable and handle Xkb events to handle changes to the group, mapping, and modifier states. This is more reliable than using the legacy events (group changes may not arrive if the window lacks pointer focus), and better handles cases where modifiers are latched, locked, or activated externally rather than physically pressed.
Prefer axes with the 'Rel X'/'Rel Y' labels, followed by 'Abs X'/'Abs Y', and only fall back to the old behavior of using the first two enumerated axes if no others are found.
Fixes a FIXME when determining which axes to use for relative motion.
- Removed GTK signal handler in x11settings. XSettings events are now properly dispatched to X11_XsettingsNotify.
Previously events were not being passed to xsettings-client as no SDL xsettings_window was created. Now all events
are filtered through xsettings_client_process_event allowing it to process the external window events that are selected.
Global content scale is updated for changes to any recognized dpi settings.
- X11_GetGlobalContent now reads the current RESOURCE_MANAGER prop off of the root window to ensure it sees the
current value. XResourceManagerString is now only used if getting the current prop fails as it caches the current resource
manager value per-display connection.
- Clean up some warnings in SDL_gtk.
In addition to hiding the border on bordered windows that will immediately become fullscreen, The combination of flags used in STYLE_BORDERLESS_WINDOWED will still show the borders on borderless windows if the initial window size exactly matches the desktop, so STYLE_BORDERLESS must be used instead.
If attempting to switch to an exclusive mode while a fullscreen spaces transition is active, wait until the transition is complete before trying to apply the changes, or the window can wind up in a weird, broken state if a mode switch occurs while in a fullscreen space.
This reverts commit ee8f2861e7.
It turns out that the problem is elsewhere, related to needing to block mode changes until spaces transitions are complete.
- Removed gtk-xft-dpi read from GetGlobalContentScale. Xrm either returns either the same value as gtk-xft-dpi or the integer
scale value in cases where gtk-xft-dpi is 1.
- Refactor SDL_x11settings handlers to defer to GetGlobalContentScale
- GetGlobalContentScale is now exported and the XSettings and Gtk signal handlers now use it for consistency. This involves
a bit of extra work reading from Xrm rather than the setting notification but ensures consistent handling based on signal
origin and hints enabled.
- Hook both gtk-xft-dpi in SDL_x11settings. This should generally result in only one being called based on which is updated.
Since both signal handlers defer to X11_GetGlobalContentScale this will cause the same content scale to be applied multiple
times. The gtk-xft-dpi signal is now only used to trigger content scale updates when the XSettings notification does not occur.
This reverts commit 47d8bdd1c3.
There are runtime reasons why creating a tray can fail, so the correct approach is not to assume that just because a platform supports a tray that trays are available. Instead, you should create a tray at application startup, for the lifetime of the application, and handle failures at that point.
Closes https://github.com/libsdl-org/SDL/pull/13632
ABXY Face Style Application
- The abxy face style extraction now applies to any SInput device.
- The use-case here is that gamepads that allow for changing styles can dynamically adjust the button labels firmware-side to have the physical location better represent the actual button being pressed.
SInput Sub-Type Clarification
- The code has been updated to better reflect the intention behind the sub-type field of the GUID byte 15.
- Any SInput device may utilize the Sub-Type field to indicate variants of the same device (Same device that has rear paddles or additional physical features but is otherwise identical, etc.).
"Firebird" Company Name Clarification
- "Bonjiri" was a wrong translation. Got clarification from the company and official PID registration for the official English spelling (See 0c5234cd92).
- Fixed this spelling for all references (official mapping strings still pending and this will return NULL until this is provided).
Code Notes
- Various notes have been updated to have more clear indications.
- I have moved the face-type and sub-type debug messages to be paired with the extraction for easier location.
- I've removed an incorrect note left-over from a previous commit regarding command response information.
When skipping presentation due to the window being hidden, presentDataCount is not incremented on the command buffer, and subsequently the submitted command buffers will not be cleaned up as long as the window is hidden. This results in a lag spike when showing the window due to all previously submitted command buffers suddenly being cleaned up at once, and lag at shutdown due to an equivalent number of fences needing to be destroyed.
Instead of relying on presentDataCount to determine whether a command buffer should be cleaned up, use a flag, which is set under the appropriate circumstances.
- gtk_init_check claims it will not terminate the program on failure, however this is does
not apply to the internal check it performs to see whether another GTK library is already
loaded, the detection of which triggers a g_error which glib treats as fatal.
- gtk_progress_get_type and gtk_misc_get_type are checked to match the checks done
by _gtk_module_has_mixed_deps.
- Clean up gtk/gdk loading. There are no library fallbacks so the one option available can
be loaded rather than iterating through a list of one item.
There is a bug where SInput checks for player led capability before
setting RGB. This means that if a controller does not have a player led,
RGB commands are not sent.
There may have been a reason for not attaching a null buffer when destroying a popup at some point in the past, but that is unnecessary now, as is the comment about ShowWindow crashing, as ShowWindow assures that a null buffer is attached before (re)creating the window.
The previous calculation could result in a window whose original position was positioned exactly corner-to-corner with the parent not being adjusted to be adjoining, and thus subject to spurious closure.
- Android defines SDL_PLATFORM_UNIX but does not compile core/unix sources
so add checks for SDL_PLATFORM_ANDROID where SDL_gtk is used.
- Fix missing void parameter list in SDL_Gtk_EnterContext.
- Remove unneeded SDL_gtk.h include in SDL_events.c
- This is to support dynamic updates of content scale when running in XWayland. The GTK signal is
preferred over the XSettings watcher and Xrm database if supported as it will trigger and update
for both native X11 and XWayland on changes to Xft.dpi.
- SDL creates and requires usage of a specific glib context. This context is set as the global
glib context with SDL_Gtk_EnterContext and the previous context is restored with
SDL_Gtk_ExitContext.
- To avoid changing the behavior of SDL_tray, which is the only consumer of SDL_gtk currently,
the SDL_UpdateTray function now unconditionally runs a single glib frame iteration and is
responsible for dispatching glib events for all consumers in SDL_PumpEvents.
- Cleaned up some error handling in SDL_tray.
The current implementation uses the returned address of the `dlsym` function
directly to load the `environ` symbol. But this function doesn't return the
address to the symbol itself, instead it returns the address to the location
where the actual address is stored, i.e. it's an additional indirection.
Consequently, the implementation fails to load and process the environment
variables successfully.
One example where this error shows up is in the `Dialog API`: in an `X11`
environment, the `zenity` driver requires access to the user's `DISPLAY` and
`XAUTHORITY` environment variables. Because these variables aren't transfered
to the `zenity` process, no dialogs are shown. This can be exercised in the
`test/testdialog.c` testprogram.
The fix changes the indirection level of the `dlsym` call from `char **` to
`char ***`, does a `NULL`-check in case the call failed, and returns the
dereferenced actual adress to the `environ` symbol.
xpad has a series of questionable design choices when it comes to button
mapping. Notably, BTN_TRIGGER_HAPPY1-4 are used for the D-pad on dance mats
instead of the typical BTN_DPAD_*, and maps the paddles to
BTN_TRIGGER_HAPPY5-8. This commit plumbs through driver detection for a udev
device and adds special exceptions for xpad's quirks.
Newer kernels (6.17+) add the BTN_GRIP* mappings for paddles on controllers. We
should prefer that if available, as its meaning and mapping is unambiguous.
However, since it's only in new kernels, we need to maintain the older mappings
where we know they exist.
It also exposes KEY_RECORD as the share button regardless of vendor, which had
previously been gated behind being a Microsoft controller. All Xbox Series
controllers can include this button, and many third party ones do.
- Implements 'handheld' flag to indicate whether the SInput device is a gamepad or handheld (ROG Ally etc)
- Implements generic mapping fallback along with face style setting ability
Building SDL for armeabi-v7a gives this warning:
```
SDL/src/audio/SDL_audiotypecvt.c:541:14: warning: '#pragma FENV_ACCESS' is not supported on this target - ignored [-Wignored-pragmas]
541 | #pragma STDC FENV_ACCESS ON
```
This reverts commit 608f706a95.
Didn't end up using this in SDL3_mixer, and it's a super-awkward API if we
don't need it.
I _might_ bite the bullet and let people lock a physical audio device, though,
as I could see that being useful but less awkward for the same reasons I
originally wanted it.
Keymap construction is an expensive process, so keymaps are cached to facilitate fast switching, as they are static after initial construction, and do not need to be rebuilt every time.
This still logs backend choices, but we might use it for other things. For
example, what Android device is being used, or all the devices we enumerated,
etc.
Ideally this eventually logs all the stuff we often have to ask followup
questions about.
...so the backends don't have to do it.
Also added a stern warning about `org` being omitted, but leaving it as
allowed so as not to break existing apps (more than they are already broken,
at least).
Fixes#13322.
Only use the specified output if an exclusive mode is being used, or a position was explicitly requested before entering fullscreen desktop. Otherwise, let the compositor handle placement, as it has more information about where the window is and where it should go, especially if fullscreen was requested before the window was fully mapped, or the window spans multiple outputs.
When changing the text input mode on a window, only update the seats that currently hold keyboard focus on that window, otherwise, text input might be inadvertently enabled or disabled on a seat focused on another window.
Build all the available keyboard layouts at once; this adds a negligible bit of overhead when initially handling the keymap (which was already significantly lowered by previous commits), but reduces the later cost of changing layouts to just swapping a pointer.
Additionally, handling of unknown keysyms, particularly when dealing with virtual keyboards, is improved, as keys generating valid Unicode values with no corresponding scancode will be dynamically added to the keymap with reserved scancodes, allowing for proper round-trip lookup behavior.
Initializing the compose table is a very expensive operation, and only necessary if the locale envvar changed, which is often not the case when just changing the keymap, so don't destroy and recreate it whenever the keymap changes. The state only needs to be reset in this case.
Older Android phones have a kernel bug where time is not properly calculated
when calling `clock_gettime(CLOCK_MONOTONIC_RAW, ...)`. The returned time
either has nanoseconds out of range (outside of [0, 999999999]) or the returned
time is not monotonic in regards to previous call or both.
The issue is reproducible in Android Studio on arm64 emulators from at least
Android 7.0 (didn't try older versions) up to including Android 8.1 (kernel
3.18.94). The issue is in the kernel, these are just the versions of Android
emulator with buggy kernels.
The kernel commit that fixed this is [1]. Because this fix was backported to
various LTS releases of kernels it's hard to find out exactly which kernels are
buggy and which not. Therefore always use `CLOCK_MONOTONIC` on Android.
The `CLOCK_MONOTONIC` is slowly adjusted in case of NTP changes but not for
more than 0.5ms per second [2]. So it should be good enough for measuring
elapsed time.
---
An option would be to dynamically select
`CLOCK_MONOTONIC_RAW`/`CLOCK_MONOTONIC` at runtime by checking at init time
that `clock_gettime(CLOCK_MONOTONIC_RAW, ...)` returns nanoseconds in range.
Testing showed that nanosecons are out of range for 999/1000 cases. I guess
this could be good enough for support on older phones.
But because this would introduce a small perf hit by always reading a global
variable for first argument to `clock_gettime` and because `CLOCK_MONOTONIC`
does not have that high time deviation, this commit uses `CLOCK_MONOTONIC` on
Android always. It is also less burden to maintain.
---
[1] dbb236c1ce
[2] https://github.com/torvalds/linux/blob/master/include/linux/timex.h#L136
It _must_ have the format set for the opposite side from the device (so
playback needs the src format set, and recording needs the dst format set),
since the stream gets mangled by the device thread if not. So if it has never
been set (stream created with NULL audiospec), just set it to match the device.
If the stream is just meant to buffer and not convert, this is desired
behavior, even if it didn't also fix a bug.
Binding the audio stream will always set the device side's format, as usual;
this does not need to be set by the caller at all.
Fixes#13363.
This allows us to avoid browser mouse emulation for touches, since we provide
our own anyhow.
The other option is to "prevent default" in the legacy touch event handlers we
historically used, to tell the browser not to supply emulation, but we can't
currently tell Emscripten to mark those handlers as not "passive," so as it
stands they are unable to prevent default. Using Pointer Events bypasses this
problem entirely.
Fixes#13161.
macOS sends a focus loss event when the dialog is created, which causes SDL
to try to minimize the window, which confuses the entire system. So in this
special case, don't do the minimization.
Fixes#13168.
This specifically affects SDL_EnumerateDirectory and SDL_GetPathInfo. Android
assets are read-only, so no need to do this for things like
SDL_CreateDirectory, etc, and the POSIX SDL_CopyFile() uses SDL_IOStream
behind the scenes, which already supports Android assets.
Fixes#13050.
It apparently works better (or can work better?) on multimonitor setups
than SwapBuffers.
This should be available back to Windows 95, but just in case, it falls
back to standard SwapBuffers if not available.
Fixes#13269.
Wayland previously didn't specify that the seat name preceded the capabilities, but it is now specified that the name event must always come first. Remove the 'SDL_Set<device>Name()' functions that were only added to accommodate the case of compositors sending the name after the seat capabilities, as this clarification means that they are no longer needed.
The internal key repeat mechanism already disables itself if the key repeat interval is 0, and SDL tracks and handles the flagging of repeated keys itself, so just map the 'repeated' event to 'pressed'.
The spec states that xdg_surface must have seen an initial configure event before attaching a buffer, however, this was only being done when initially showing the window, and not after show->hide->show cycle.
Always wait for the initial configure event when (re)creating an xdg_surface as part of the show window sequence.
- Re-enable FPS counter, which is hidden by default anyway - but essential while working on an actual project.
- Fix link in the platform's README.
- Re-order list of supported platforms (alphabetical order to maintain consistency).
If the pointer should be confined, keep trying until a confine/lock signal is received. This considerably improves locking/confinement reliability on compositors where confining can be a racy operation, or may not take effect until the pointer is actually in the confinement region.
A pointer lock is used to special-case 1x1 confinement regions, as otherwise, the pointer can still exhibit jitter at the subpixel level, particularly on scaled desktops.
Pointer events in seat version 5 and higher should be grouped and sent together when a pointer frame event occurs. Store pending pointer motion events and dispatch them together when a frame event is received.
This helps avoid spurious pointer motion events that some compositors generate with no associated frame (e.g. from some tablet events).
This adds several minor changes to the gyro instruments.
* The HID Sensor Time display is now throttled to 10hz.
* Calibration for the gyro is now time based, not sample count based. Different polling rates will have drift calibrated over the same space of time.
* Pitch/Yaw/Roll readout: Yaw is prioritized, and then pitch, and then roll. This gives a more human-readable pitch/yaw/roll display, closely matching game engines.
* Pitch/Yaw/Roll text is colorized to match the axes in the 3D gizmo.
* Added set of axes to the 3D gizmo to show the "Left Hand Space" positive axis directions.
On newer systems, the trick isn't necessary, and if you do it, if the user is
moving the mouse when launching the app, it'll show a hidden Dock.
Fixes#10340.
Always update the border sizes on frame extent events, or they can incorrectly still be zero if followed by a PropertyNotify event when leaving fullscreen.
Fixes sending the correct restored window size when leaving fullscreen in fvwm.
The Xbox Adaptive Controller reports the pressed guide button state as 2
instead of 1. This is contrary to the spec, but it's needed for the guide
button to work.
This also adds a bit more documentation on observed-but-not-known GUIDs.
This is a special-case piece of functionality, generally these are expected
to go away during shutdown, but maybe someone is switching between audio
subsystems or something...
SDL2 would set a high bit in the mouse button flags to indicate when raw input had been read from, without this, if you hold down a mouse button and left raw input mode (leaving relative mode) the button would remain partially stuck, and would require two clicks to start producing mouse down events again.
SDL3's raw input code was refactored to not use the mouse button flags, but forgot to invalidate the flags, causing this bug to manifest.
- Check for specific availability instead of waiting for "device ready."
- Don't use snd_pcm_wait, just use a simple SDL_Delay (nothing to recover).
- Fixed SDL_max call that should have been SDL_min (now using SDL_clamp).
- ALSA_RecordDevice() now returns 0 if no availability, which kicks us back
to WaitDevice to try again, as an extra safety check.
Also made accelerometer threshold for drift calibration more lenient for very noisy accelerometers.
The testcontroller tool could eventually be used to come up with a better way to profile an IMU's "stationary" noise so that this threshold can be as tight as necessary for the sake of automatic drift calibration.
(thanks @HilariousCow!)
Flydigi IMU rate now matches observed rate of packets in both dongle and wired connection.
Flydigi Vader 4 Pro IMU rate correction was set at a fixed 125hz. In actuality rate is 1000hz over dongle and 500hz when wired.
Gathered correct IMU polling rate in wired mode for good gyro synchronization.
Wireless: different models had different amounts of Bluetooth packet loss.
USB_PRODUCT_8BITDO_ULTIMATE2_WIRELESS: Solid 120hz over bluetooth (note: only appears via).
USB_PRODUCT_8BITDO_PRO_2_BT: Lossy - 80-90hz registered.
SB_PRODUCT_8BITDO_SN30_PRO_BT & USB_PRODUCT_8BITDO_SN30_PRO_BT: Very Lossy - 60-90hz registered
* Added tools to Test Controller for evaluating gyroscope accuracy and IMU polling rates.
This adds a visual suite to the testcontroller tool to help validate IMU data from new gamepad drivers and HID implementations.
The 3D gizmo renders accumulated rotation using quaternion integration of gyroscope packets. If a controller is rotated 90° in real space, the gizmo should reflect a 90° change, allowing quick detection of incorrect sensitivity or misaligned axes.
Also includes:
- Euler angle readout (pitch, yaw, roll)
- Real-time drift calibration display with noise gating and progress
- Accelerometer vector overlay
- Live polling rate estimation to verify update frequency
Intended for developers working on controller firmware or SDL backend support to confirm correctness of IMU data processing.
The pointer warp protocol allows us to warp the pointer to a different position on
the surface, without any hacks like locking and unlocking the pointer.
XMonad ignores size hints and shrinks the client area to overlay borders on fixed-size windows, even if no borders were requested, resulting in the window client area being smaller than requested. Calling XResizeWindow after mapping seems to fix it, even though resizing fixed-size windows in this manner doesn't work on any other window manager.
When querying the keycode produced by a scancode with a certain set of modifiers, it would fall back to defaults if a key hash value with the exact set of modifiers wasn't found, which resulted in certain modifier combination returning incorrect keycodes on non-ANSI keyboard layouts. For example, querying SDL_SCANCODE_Y with the alt modifier on a QWERTZ layout returns SDLK_Y instead of SDLK_Z on most platforms, as the backends don't generate a specific entry for this key + modifier combo, so the lookup would fall back to the default ANSI layout.
Adding additional key+modifier combinations when building the keymap is one solution, but it makes an already expensive operation even more so, pushing the time needed to build the keymap into double-digit milliseconds in some cases due to the large amount of key combos that need to be queried, most of which are redundant.
Instead, falling back to searching through the shift levels for the given modifier state when querying the keymap will ensure that the most appropriate keycode is returned. This does add some overhead to lookups if the key doesn't have an entry with the exact set of modifiers, but it is minimal as hash table lookups are an inexpensive operation, and unnecessary lookups are avoided. In my own testing of an optimized build, the difference between best-case and worst-case performance (the latter of which is highly unlikely in real-world usage) is only a few hundred nanoseconds. Additionally, the unmodified keys are queried when pumping events, so there is no additional overhead in that case.
Windows seems to implicitly enable IME text input on windows created while an IME is active, which causes the IME suggestion window to pop up when keys are pressed, even if a client never explicitly enabled it. Ensure that IME support is initially disabled on new windows; SDL will enable it at a later time, if required.
Previously, all transparent pixels were rendered as opaque due to the limitations of the 4K color mode. Replaced Gc()->BitBlt() with Gc()->BitBltMasked() and updated the mask during copy operations to correctly respect the alpha channel of textures, while maintaining good performance.
The first public beta that had a corresponding arm64 version has now
been promoted to stable, and subsequent releases will be for both
x86_64 and arm64.
Signed-off-by: Simon McVittie <smcv@collabora.com>
This makes sure we get reliable mouse enter/exit events from the system on
older macOS releases.
Newer releases don't have this problem--my assumption is that Cocoa has a
more aggressive default tracking area installed for some newer UI feature.
For 3.2.16, we'll use the explicit tracking area on older macOSes only, but
I'll remove that check in revision control for newer OSes and see what
happens.
Fixes#12725.
Use the proper types for xkb mod masks and layout indices, rename the mod masks to reflect that they are bitmasks and not indices, and use the 'layout' nomenclature instead of the deprecated 'group' nomenclature.
xkbcommon 1.10.0 declared certain modifier names to be deprecated, and the current plan is to remove them in 1.12.0. Use the new recommended names and modifier mask retrieval function when building against version 1.10.0 and higher.
Windows can get in a state when it stops reporting raw input events for game controllers until reboot.
The downside of this change is that we lose support for trigger rumble and are limited to 4 controllers again, but if that's important for your application you can use SDL_SetHint(SDL_HINT_JOYSTICK_RAWINPUT, true) to enable this functionality.
Fixes https://github.com/libsdl-org/SDL/issues/13047
Using processed timestamps can result in anomalies that cause excessive repeat events, and hard caps can cause issues.
In the key event handler, use the raw elapsed time to calculate any remaining repeat events to avoid the artifacts that can result from using processed timestamps.
The Wayland key repeat rate ranges from 0 to 1000 per second, so millisecond resolution doesn't lose any precision.
More non-compositing window managers than just openbox seem to need this, so always force sending the window position and dimensions when entering/leaving fullscreen. If they are wrong, they will be immediately overwritten by the correct dimensions from a subsequent ConfigureNotify event.
This is disabled on XWayland, as it seems to cause hitching on some compositors.
This mostly fixes stylistic issues, but also fixes an issue with some
controllers where an erroneous check would clear the DLI flag, offsetting the
share button index so it wouldn't work.
Cap the elapsed time to something sane in case the compositor sends a bad timestamp, which can result it in it looking like the key has been pressed for a *very* long time, bringing everything to a halt while it tries to enqueue all the repeat events.
The function SDL_GetCurrentKeymap() would return null instead of the actual bound keymap for non-latin layouts if certain mapping options were set. Add a parameter to ignore the keymap options and always return the actual bound keymap, which is needed in the Wayland backend to avoid spamming keymap changed events on every keystroke with certain layouts.
The text input cursor should reflect the cursor position, not the entire text input rect. Set it correctly so that IME chooser dialogs appear in the correct location.
SDL_PrivateJoystickAdded was called before setting the InstanceId in the adapters ctx->joysticks array. This would eventually broadcast the SDL_EVENT_JOYSTICK_ADDED event with the new InstanceId, if your program listens for the added events and opens joysticks at that point it would always fail because there would be no matching InstanceId in the ctx->joysticks array.
This can be toggled per-window, so use the individual window flags instead of the global toggle to selectively enable it only for the relevant window in a multi-seat scenario, as is already done with keyboard and pointer grabs.
The main view always reflects the size of the output, so don't use the dimensions of the currently bound render target texture when updating it, or it will reflect an incorrect size when the render target texture is unbound.
This and its related commits introduced some bugs and quirks such as duplicated and delayed/missed key events that can't be easily worked around, so revert this for now.
This reverts commit 47162a4168.
The keys and values of device_hash are pairs
`(SDL_AudioDeviceID devid, void *dev)` where dev can be either a
`SDL_AudioDevice *` or a `SDL_LogicalAudioDevice *`, depending on
bit 1 of devid.
We can confirm that we have got this right by looking at the
instance_id member, because logical audio devices happen to start with
the devid, whereas physical devices start with a pointer which is
unlikely to match the devid by chance.
Signed-off-by: Simon McVittie <smcv@collabora.com>
Compositors typically send keyboard, pointer, touch, and tablet leave events when a window is destroyed, however, this is not guaranteed behavior, and at least one compositor in widespread use doesn't always send pointer leave events immediately upon destroying a window. Ensure that all references held by seats to a focused window are removed before the underlying window surface and structs are destroyed to prevent potential segfaults if the seats are immediately destroyed after the window.
This loop breakout was originally from an attempt to parse coalesced packets.
Breaking out early does more harm than good, and no devices coalesce packets,
so this is unnecessary.
By moving the message sending onto the main thread and adding a small timeout
we can now ensure the ACKs are sent fast enough to avoid the firmware bugs in
these controllers.
At the moment, only the ThrustMaster T.Flight Hotas One has full support. The
documentation says you can query the extra buttons via a specific command, but
the stick appears to reject the command. Further investigation is needed for
automatically querying this state.
This is needed for future work bringing up things like the chatpad.
This commit also fixes a few minor things, such as still sending motor packets
to devices that don't support it, enabling quirks that hide trigger rumble on
devices that are marked as not having it, and fixing #12942.
The surrounding code in all of these instances expects the Unicode
variants. Previously, this code mixed Unicode and ANSI/ASCII calls if
`UNICODE` was undefined, which caused type and logic errors. Explicitly
spelling out the W removes any reliance on that macro.
IME text events can result in sending duplicate key press events, which will result in undesired repeated key presses. Since the events are exact duplicates, compare the serials to filter out redundant key down events.
The Module interface is the one used by the outside world. This code
is inside the module itself so can use the internal name, avoiding the
need to export this function on the Module at all.
See https://github.com/emscripten-core/emscripten/pull/24269
You can't have two sections with the same name (so it can generate unique
page anchors), so fix one of these, and tweak another thing that _isn't_ a
header but wikiheader's quick-and-dirty parser thinks is one.
Using a 0 sequence number is always allowed and avoids having to synchronize sequence numbers with the controller (and potentially confusing other things talking to the controller)
Also fixes occasional long running rumble at controller connection.
The xserver will still send EnterNotify events while the pointer is captured, and the grab shouldn't be updated in these cases, as it will cause the capture to be lost.
Otherwise, the message box thread goes back to blocking, but the app's main
thread progresses to SDL_WaitThread and won't pump the event queue anymore.
On iOS, this means the messagebox won't pop back up (because it uses
SDL_RunOnMainThread, which needs the event loop to pump), but it's just bad
practice on any platform to stop pumping the event queue for indefinite
lengths of time.
If it is known that the window will immediately enter fullscreen upon being shown, set the borderless style when showing the window to hide the borders, or they may linger in the background if the client takes some time to draw the first frame.
Unnecessarily calling ShowWindow with SW_RESTORE when applying the window flags must be suppressed in this case, or the borders can reappear in a weird, partial state.
Otherwise the window won't have focus until you click on it again. Calling
makeKeyAndOrderFront isn't enough to fix it, either.
This trick comes from a similar problem we solve in our
applicationDidFinishLaunching implementation: activate (give app focus to) the
system Dock, as something that definitely exists that isn't us and is harmless
to activate, and then activate us right afterwards. This unconfuses whatever
is getting confused inside Cocoa.
Fixes#12684.
So something like "index.pb" will now accept any file with a ".pb" extension,
to make macOS happy. This seems like a reasonable tradeoff.
Other minor cleanups.
Fixes#12778.
sdl2-compat will call SDL_GetDisplayForWindow() when querying the display to use for fullscreen, which won't always be correct if the backend can't actually reposition the window. When calling this function, get the ideal fullscreen display and store it in a property for retrieval by sdl2-compat.
This new driver is based on official documentation released by Microsoft in
September, though it still lacks several important features, notably the
Security handshake for wireless dongles and audio support. It is, however, more
reliable and extensible than the old driver.
This is for internal use to signal that a mapping uses positional GameCube buttons. It's set so we know whether a mapping uses the older style labeled buttons or the newer style positional buttons. If a positional mapping is used with SDL2, then it will be ignored, since the hint is marked as defaulting true and the mapping conditional is that the hint is false.
Mouse button events that trigger a hit test are not passed to the client, but the client may still query the global mouse button state from within the hit test handler, so the reported buttons need to be accurate. Query the buttons directly from the seat instead of the higher global mouse state to match the behavior of other platforms.
The low 16-bits of the message extra info is an event sequence number when using the Wacom tablet with Windows Ink disabled. The high bits of normal mouse motion when using touch input match the touch signature, 0xFF515700. The high bits of raw input mouse motion when using touch input do not match that signature, so we have to check for the touch bit in that case.
Fixes https://github.com/libsdl-org/SDL/issues/12927
The problems are two-fold. When this happens a WM_POINTERDOWN event is sent with IS_POINTER_INCONTACT_WPARAM() evaluating as true. So when SDL_SendPenButton() is sent for the barrel button, there is no pen in contact yet, so the right mouse button is sent. Then SDL_SendPenTouch() is sent, which generates a left button press event.
Fixes https://github.com/libsdl-org/SDL/issues/12926
These events fire for other things, such as pressing a barrel button while
the pen is hovering.
The correct thing to do is check IS_POINTER_INCONTACT_WPARAM in the event.
If the pen is already touching or not, SDL_SendPenTouch() will do the right
thing, so it's safe to call it even if we're already in the right state.
By default, popups are automatically constrained to be completely within display bounds, so as not to cut off information and result in an unusable menu, or unreadable tooltip. In some cases, however, this is not wanted, so a property to toggle this behavior is added.
There are also cases where the client may not want a popup menu to implicitly grab the keyboard focus, as is the default behavior, so popup menus now respect the focusable flag/property, as well as being able to toggle focus grabbing via SDL_SetWindowFocusable().
The GameCube controller has a different face button layout than the Xbox or Nintendo Switch style controllers. It has the B button on the left and the X button on the right, so we should map those to SDL_GAMEPAD_BUTTON_WEST with SDL_GAMEPAD_BUTTON_LABEL_B and SDL_GAMEPAD_BUTTON_EAST with SDL_GAMEPAD_BUTTON_LABEL_X respectively.
Fixes https://github.com/libsdl-org/SDL/issues/12847
When showing or hiding a popup menu, manually check and set the focus if the new topmost window under the cursor is an SDL window. Otherwise, the focus won't be updated until the cursor is actually moved.
There is a quirk with XInput2 mouse capture that causes a leave event to be sent if the pointer moves out->in->out, which breaks mouse tracking outside the window. If the mouse leaves the window with buttons pressed, continue tracking it until the buttons are released.
- IsIconic/IsZoomed must be checked after sending SDL_EVENT_WINDOW_SHOWN as that may trigger window operations if any are pending from when
the window was hidden. e.g. the window may be shown, which triggers SDL_MaximizeWindow and a new WM_WINDOWPOSCHANGED where
SDL_EVENT_WINDOW_MAXIMIZED is sent, then control returns to the original WM_WINDOWPOSCHANGED which would not think the window is zoomed
and send SDL_EVENT_WINDOW_RESTORED.
This info shows up in `pw-top` ...Gnome's sound preferences (etc) would
already show something more app-specific, but this command line tool shows
specific stream names.
Previously, we would WaitDevice first, but that would feed a silent buffer
to AAudio upfront, introducing latency. When this change was made, the
semaphore count should have been adjusted, since we're waiting on one less
buffer.
Fixes#12882.
This variable is always defined when targeting Android:
- when using the toolchain file shipped along the ndk
(using -DCMAKE_TOOLCHAIN_FILE=$ENV{ANDROID_NDK_HOME}/build/cmake/android.toolchain.cmake)
- when using the toolchain file provided by CMake
(using -DCMAKE_SYSTEM_NAME=Android)
- Add checks that ObtainPhysicalAudioDevice() was previously doing
(is subsystem initialized, is device valid).
- Remove optimizations that copy string to stack to release device_hash_lock
before SDL_GetPersistentString is called. Probably not necessary, and made
the code more complex.
Previously, touchpad events were not reported to the application.
This patch exposes two touchpads and forwards x/y/pressure information
from the HIDAPI packet.
This fixes issue #12855
According to the documentation and testing the `emscripten_webgl_create_context` function returns `0` on error as oppposed to a negative number. This was causing the error message to be empty when the later `emscripten_webgl_make_context_current` call fails given the invalid context.
See https://emscripten.org/docs/api_reference/html5.h.html#c.emscripten_webgl_create_context
Xfce, unlike every other window manager in existence, sends ConfigureNotify events before PropertyNotify events when toggling the fullscreen and maximized window state. Check the window state when handling ConfigureNotify events, and defer emitting SDL size/position events until the corresponding PropertyNotify event arrives, since SDL and clients expect to get the window state before the new size and position.
This warning is actually wrong, as the variable is only used if the string pointer is valid, in which case it was initialized, but this is needed to quiet MSVC.
This was already present for regular rumble to ensure that controllers would
continue rumbling for extended periods, but was missing for trigger rumble. I
don't know if this affects any controllers at the moment, but it's helpful for
future-proofing.
For a depth buffer in D3D12 that is also going to be used in a texture sampler, the creation (on an Intel HD 5500 igpu) fails. e.g. SDL_GPUTextureCreateInfo type = TEXTURE_2D, format = D32_FLOAT, usage = DEPTH_STENCIL_TARGET | SAMPLER
The error messages are:
D32_FLOAT
D3D12 ERROR: ID3D12Device::CreateShaderResourceView: The Format (0x29, R32_FLOAT) is invalid when creating a View; the
Resource was already created with a fully qualified Format, which is not castable (0x28, D32_FLOAT).
D24_UNORM
D3D12 ERROR: ID3D12Device::CreateShaderResourceView: For the resource format D24_UNORM_S8_UINT, when making a D3D view, the format name for the view can't be R24_UNORM_X8_TYPELESS.
I found this is because the texture format needs to be created as _TYPELESS, then the views (depth stencil view, shader resource view) should then be created as their respective types - e.g. texture = R32_TYPELESS, dsv = D32_FLOAT, srv = R32_FLOAT
Tested and working on:
NVidia RTX 3050 (D3D12 feature set 12_2)
Intel HD 5500 (D3D12 feature set 11_1)
Wayland compositors may send recursive clipboard offers to the client, which need to be filtered out to avoid clearing local data. Previously this was worked around with a hack, but this caused the ownership flag to be set incorrectly, which broke some clients.
This introduces a metadata MIME type of application/x-sdl3-source-id to be sent with SDL3 selection offers, which contains a string that is a unique identifier for the instance, and can be used to detect if a received selection offer is originating from the same instance that generated it.
If DBus is available, the unique identifier string is the unique name of the connection, otherwise, the process ID is used.
If a window isn't resizable from specific directions, the compositor can inform clients of the current edge constraints, so they don't display resize cursors for non-resizable edges.
Wayland environments can expose more than one seat for multiple collections of input devices, which can include multiple, simultaneously active, desktop pointers and keyboards with independent layouts. The Wayland input backend previously presumed that only one seat could exist, which caused broken behavior if the compositor exposed more than one, which is possible on wlroots based compositors such as Sway. This introduces support for handling multiple seats, including proper handling of dynamically added and removed seats and capabilities at run time.
The SDL Wayland input system was accreted over time, and the assumption that only one seat will ever exist resulted in state and related objects not always being tied to their most appropriate owner in a multi-seat scenario, so refactoring was required to manage several bits of state per-seat, instead of per-window or globally.
As Wayland keyboards can have per-seat layouts, fast keymap switching is required when multiplexing input from multiple seats to the global SDL keyboard device. A parameter was added to the keymap creation function to specify if the keymap lifetime should be externally managed to facilitate keymap reuse, and some layout info was moved from the global keyboard state to the keymap state to avoid unnecessarily redetermining it whenever a reused keymap is bound. This reduces the overhead of switching keymaps to setting a single pointer.
Multiple seats also means that multiple windows can have keyboard and/or mouse focus at the same time on some compositors, but this is not currently a well-handled case in SDL, and will require more work to support, if necessary.
Fix IMU timing stamp - simulate the timing stamp from the controller hardware's perspective, as opposed to the receiving computer's perspective. Do this by advancing a time stamp by a fixed rate, based on observation of average rate over bluetooth.
This reverts commit 5a59b5f321.
The strong ABI guarantee only applies if bool is a byte (we repurposed a Uint8 field as a bool in SDL_gpu.h), and 32-bit PPC isn't a platform we support, so I'm going to back this patch out.
The window may be initially maximized or made fullscreen by the window manager for various reasons, such as automatically declaring a window that precisely fills the usable desktop space as maximized, or a "kiosk-mode" automatically making the window fullscreen.
Don't redundantly make restored or unset fullscreen calls when initially showing a window, or the expected state can be unset.
These callbacks run in a thread we don't control, but might call SDL_SetError,
which will claim a TLS slot that never gets otherwise deallocated. To prevent
the leak, finish each callback by explicitly freeing any thread-local data.
Fixes#12732.
It doesn't work well with Vulkan, and Vulkan windows may not have the Vulkan flag set in all circumstances, so only enable it if we explicitly know that OpenGL is being requested instead.
Moved the event handlers from SDL_CreateWindow to SDL_VideoInit
emscripten_set_mouseup_callback
emscripten_set_focus_callback
emscripten_set_blur_callback
emscripten_set_pointerlockchange_callback
emscripten_set_fullscreenchange_callback
emscripten_set_resize_callback
We may want to extend this with additional properties in the future.
Also removed SDL_PROP_GPU_DEVICE_DEBUG_VULKAN_CONFORMANCE_STRING. If we need feature level queries we can add them in the future.
On non-compositing window managers, XIconifyWindow can trigger map/unmap events, which would toggle the window hidden/shown state. This should not be done, as a hidden window in SDL is equivalent to a withdrawn window in X, and SDL will try to set/reset state when it is shown again.
Unless the window is explicitly being withdrawn via X11_HideWindow(), set unmapped windows to the minimized/occluded state, and send a restored event when mapped again.
This hint needs to persist outside of the normal application flow, so use the environment to set the initial value, and then save the value set via SDL_SetHint() after that.
Fixes https://github.com/libsdl-org/SDL/issues/12677
This fixes the feature detection in cases where linking the final executable would fail.
This is fine since SDL only opens these libraries using dlopen as well.
The taskbar list is invalid after OleUninitialize(), so make sure we clean it up first.
Also don't bother creating the taskbar list if we don't have any progress state
this was causing problems when i was specifying non-zero buffer slots in
`SDL_BindGPUVertexBuffers`, `SDL_GPUVertexAttribute` and
`SDL_GPUVertexBufferDescription`.
`firstSlot + 1` is simply copied over from metal and d3d12 backends.
We'll default this off for now, since we don't detect whether the controller is running firmware v1.03 and don't have a fallback for controllers running firmware v1.02 (the out-of-box firmware)
For some reason this locks up the Windows compositor when called by Steam. I'm disabling it for now until we understand why and whether this can cause issues for other applications as well.
The window failing to enter/exit fullscreen notifications don't necessarily imply anything about the actual state of the window. On failure, dump pending events, and don't presume anything about the current window state, as it will be handled by subsequent enter/leave notifications.
SDL still manages hotplug and default device management here, but if a user
wants to use an external tool, such as `pactl move-sink-input`, to put it
onto a different device, they probably know what they're doing and we
shouldn't stop them.
If this turns out to have some unexpected consequences, though, we can revisit
the change.
Fixes https://github.com/libsdl-org/sdl2-compat/issues/367
GNOME requires this to allow keyboard grabs on XWayland. Otherwise, XGrabKeyboard will still report success, but shortcuts won't be inhibited.
See 5f132f3975
Otherwise, in the time it takes the thread to start and other init tasks to
complete, we tend to get an underrun on some systems, which ALSA logs to
stderr.
So this is moved to an InitThread implementation, which runs from the device
thread, right before it begins its main loop.
Reference PR #12632.
The accelerated relative coordinates are always relative to the desktop resolution, and need to be scaled with emulated fullscreen resolutions for mouse movement to remain stable. Otherwise, pointer speeds will be too fast on emulated resolutions below the desktop resolution, and too slow on those above.
The GPU API doesn't currently support transparent windows (transparent swapchain effects doesn't seem possible on D3D12) so we should explicitly fail so users don't expect transparency and then not get it.
Fixes https://github.com/libsdl-org/SDL/issues/12410
If the window is flagged with SDL_WINDOW_TRANSPARENT, create the associated swapchain with a composite alpha value that supports blending, if available.
Fixes transparent windows on the Vulkan backend, and prevents possible validation errors by ensuring that the composite alpha value is always a valid bit in VkSurfaceCapabilitiesKHR::supportedCompositeAlpha.
* Allow custom message boxes with colors and multiple buttons to work if Asyncify is enabled
* Keep old functionality of using alert when Asyncify is not available
* Update testmessage to allow for setting random colors as the color scheme of the message box
We can't use keyboard input as a signal about whether a keyboard is attached. There might be keyboard input from any number of generated inputs or non-keyboard devices.
The tm_gmtoff field of the broken-down time 'tm' struct wasn't formally standardized until POSIX.1-2024, but practically it has been available on desktop *nix platforms such as Linux/glibc, FreeBSD, OpenBSD, NetBSD, OSX/macOS (NextStep before that), and others since the 1990s. The notable exception is SunOS/Solaris, where the timezone offset must still be retrieved in the strictly POSIX.1-2008 compliant way.
SDL_IsGamepad() calls SDL_GetJoystickTypeForID(), which will call SDL_IsGamepad() if it's not a known controller type. The new code which is breaking was added to prevent Logitech FFB wheels from showing up as gamepads, which we check for separately.
On Windows 11, apparently HidD_GetManufacturerString() and HidD_GetProductString() can return TRUE without actually filling in any string data.
Fixes https://github.com/libsdl-org/SDL/issues/12566
These changes enable the Logitech G29 wheel to run on hidapi with both SDL_Joystick and SDL_Haptic interfaces.
While it is already possible to use the wheel on Linux in WINE + SDL2 thanks to the in-tree evdev driver as well as new-lg4ff, these set of changes allow the G29 to be used with WINE under MacOS and FreeBSD
These wheels should also be supported, but I can only test them from G29's compat modes: G27, G25, DFGT, DFP, DFEX
Haptic and led support are ported from https://github.com/berarma/new-lg4ff
SDL 2.0.18 added preciseX/Y to mouse wheel events, which we cannot
emulate in sdl2-compat without a mechanism to control integer position
and scroll deltas separately.
The existing behavior helps clients that don't expect exclusive fullscreen windows to move by maintaining a consistent size and mode, however, some are aware that this can occur and want to handle mode selection themselves.
Add a hint to disable auto mode switching when an exclusive fullscreen window moves to accommodate this use case, and don't override fullscreen changes that may occur in an event watcher between the display changed event being posted and SDL running the display changed handler, as the mode switch may have already been handled there by the client.
Move hints used by Emscripten to window properties. This change will be necessary if multiple windows for Emscripten will be supported in the future.
- Added Window Create Property SDL_PROP_WINDOW_CREATE_EMSCRIPTEN_CANVAS_ID
- Added Window Create Property SDL_PROP_WINDOW_CREATE_EMSCRIPTEN_KEYBOARD_ELEMENT
- Use hint SDL_HINT_EMSCRIPTEN_CANVAS_SELECTOR as override to SDL_PROP_WINDOW_CREATE_EMSCRIPTEN_CANVAS_ID
- Use hint SDL_HINT_EMSCRIPTEN_KEYBOARD_ELEMENT as override to SDL_PROP_WINDOW_CREATE_EMSCRIPTEN_KEYBOARD_ELEMENT
- WIN_SetWindowAlwaysOnTop is always called as part of window creation (normally setting HWND_NOTOPMOST) which also set the z-order for the parent
window when child windows are created. Since HWND_NOTOPMOST is above all non-topmost windows this had the effect of reordering but not focusing
the parent window.
With ASAN enabled the existing code triggered null-pointer-dereference
in the strcmp since EMSCRIPTEN_EVENT_TARGET_WINDOW is a "fake" pointer
with the value ((const char*)2). In fixing this it was also noticed
that using the #window/#document/#screen targets as mentioned in the
docs fails since document.querySelector() returns null for them and we
should instead be mapping them to the special HTML targets provided.
External windows may call this to update the renderer size, but not pump SDL events, either by syncing or running an event loop, so the size event always needs to be synthesized for external windows. If it is wrong, the true size will be sent if/when events are processed.
The bug this appeared to fix turned out to be randomly occurring and has other causes, and the read pipe is already polled for readiness, so this extra overhead isn't necessary.
The platforms that needed this (Windows Phone and Windows RT) are no longer supported, and if this is needed in the future it should be done at the renderer level, not here.
Fixes https://github.com/libsdl-org/SDL/issues/12475
External windows often handle their own events, and SDL implicitly processing events during synchronization can be unexpected and cause issues. For requests that don't change the window state beyond size or position, use the legacy synchronization method that synthesizes events from the window attributes without processing events.
This reverts commit 0c8ddc1f0a.
The real bug here is that XWayland isn't sending us a display disconnected event, which will be handled in the next commit.
Openbox doesn't send size events when entering fullscreen, so they must be synthesized. This is not desirable on any other window manager, as it can break fullscreen positioning on multi-monitor configurations.
People are adding present calls while rendering to render targets, not understanding that this doesn't make sense, and wondering why they get flicker on some systems. If there are programs that relied on the previous behavior we can add a hint to control this.
Fixes https://github.com/libsdl-org/SDL/issues/12432
This is true if fullscreen is pending or currently active. This is a better check in SDL_SetDesktopDisplayMode() because a fullscreen mode may be pending and complete asynchronously and the window hasn't been set to fullscreen yet.
The Debian 12 'bookworm' stable release only has 0.3.65, which is earlier
than the introduction of that symbol. Similarly, Steam Runtime 3 'sniper'
and Steam Runtime 2 'soldier' have a backport of 0.3.65 from Debian 12,
so they have the same limitation.
steamrt/tasks#686
Signed-off-by: Simon McVittie <smcv@collabora.com>
This helps the Pipewire camera driver to access cameras
in a sandboxed environment without host Pipewire socket access.
Unlike other platforms, no event is sent when the user rejects
camera access. This is because there is no mechanism to query
cameras through the portal, and we only obtain access to the
Pipewire fd if the user accepts the request. The Pipewire driver
will attempt to open the host socket instead.
Changing the mode triggers a display reconfiguration event, which will overwrite the desktop mode with the set fullscreen mode, preventing proper restoration when leaving fullscreen. Don't overwrite the desktop mode if the reconfiguration is due to a fullscreen mode switch.
Fullscreen spaces and miniaturization will always give us some notification that they succeeded or failed, so the timeout isn't required. This prevents errant timeouts when live-resize is active.
"steamrt-sniper":JobSpec(name="Steam Linux Runtime (Sniper)",os=JobOs.UbuntuLatest,platform=SdlPlatform.Linux,artifact="SDL-slrsniper",container="registry.gitlab.steamos.cloud/steamrt/sniper/sdk:beta",),
"steamrt3":JobSpec(name="Steam Linux Runtime 3.0 (x86_64)",os=JobOs.UbuntuLatest,platform=SdlPlatform.Linux,artifact="SDL-steamrt3",container="registry.gitlab.steamos.cloud/steamrt/sniper/sdk:latest"),
"steamrt3-arm64":JobSpec(name="Steam Linux Runtime 3.0 (arm64)",os=JobOs.Ubuntu24_04_arm,platform=SdlPlatform.Linux,artifact="SDL-steamrt3-arm64",container="registry.gitlab.steamos.cloud/steamrt/sniper/sdk/arm64:latest"),
"steamrt4":JobSpec(name="Steam Linux Runtime 4.0 (x86_64)",os=JobOs.UbuntuLatest,platform=SdlPlatform.Linux,artifact="SDL-steamrt4",container="registry.gitlab.steamos.cloud/steamrt/steamrt4/sdk:latest",more_hard_deps=True,),
"steamrt4-arm64":JobSpec(name="Steam Linux Runtime 4.0 (arm64)",os=JobOs.Ubuntu24_04_arm,platform=SdlPlatform.Linux,artifact="SDL-steamrt4-arm64",container="registry.gitlab.steamos.cloud/steamrt/steamrt4/sdk/arm64:latest",more_hard_deps=True,),
"vita-pvr":JobSpec(name="Sony PlayStation Vita (GLES w/ PVR_PSP2)",os=JobOs.UbuntuLatest,platform=SdlPlatform.Vita,artifact="SDL-vita-pvr",container="vitasdk/vitasdk:latest",vita_gles=VitaGLES.Pvr,),
envvardesc = SDL3 can be controlled by the user, externally, with environment variables. They are all operate exactly like the [hints you can get and set programmatically](CategoryHints), but named without the `_HINT` part (so `"SDL_HINT_A"` would be environment variable `"SDL_A"`).\n\nThis list matches the latest in SDL3's revision control.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
