WGI initialization fails on Win32 #5552
Assignees
Milestone
Comments
|
The WGI documentation says specifically to use |
Oh interesting, do you have a link to the page that said that? The comments for each class in Your info does seem to be accurate though because using
I can change |
I thought I read it elsewhere as well, but it may have been this:
That's a good point. Maybe we should add a hint for this, and default it to multi-threaded so SDL itself will work correctly? :) |
Upon further inspection, even SDL itself isn't fully MTA-safe. Our implementation of In addition, our IME code requires an STA, otherwise we fail with SDL/src/video/windows/SDL_windowskeyboard.c Line 1386 in 49a2e4b I'm going to poke around with WGI more tomorrow to see if maybe we're leaking a reference to something and that's why we crash when Windows.Gaming.Input.dll unloads. If not, we could be stuck having to do something annoying like running WGI on its own thread with an MTA. |
|
Good catches, thanks! |
|
OK, this rabbit hole goes deep, but I think I got somewhat to the bottom of what's going on in #5270. We only have a single apartment in our process, so the final call to It turns out I was wrong that using an MTA solves the problem. It appears that using an MTA in combination with the partial failure of the IME code (mentioned in my last comment) causes the MTA to stick around beyond our last Using time-travel debugging, I was able to figure out that the pointer WGI is trying to read used to belong to the vtable for Based on what the old vtable looked like prior to the unload, It appears that
WGI is directly loading I don't know enough about COM to know whether I'm not sure if there's a better workaround, but I did find that keeping the MTA around (even if nobody is using it) will prevent COM from uninitializing for the entire process and avoid unloading those DLLs. That can be done by calling |
|
That's awesome debugging, and we should implement that workaround, but it still doesn't solve the fact that SDL uses multiple COM threading models. Or does this mean we can actually initialize WGI with apartment threaded model? |
|
STA or MTA doesn't seem to make a difference as far as the crash is concerned. We still hit it either way, so I think it's safe to use either once we decide on a workaround for the bug. |
|
Let's go ahead and implement the |
|
@walbourn, @cgutman tracked down the CoUninitialize() crash in WGI, and it looks like it may be unrelated to whether we use RoInitialize(). The details are in #5552 (comment). |
cgutman
added a commit
to cgutman/SDL
that referenced
this issue
Apr 21, 2022
Yep, I updated PR #5554 with the workaround.
I made a small reproducer for the bug by just adding a couple WGI calls to the default Win32 application template that Visual Studio generates, so I can definitely confirm SDL isn't involved at all. If anyone at Microsoft wants to investigate, you can build and run the attached project. You must have a PS4 controller connected to the PC (possibly other non-Xbox gamepads may work) and you must leave the app running for at least around 30 seconds to hit the crash. |
slouken
pushed a commit
that referenced
this issue
Apr 21, 2022
icculus
pushed a commit
to icculus/SDL
that referenced
this issue
Apr 27, 2022
madebr
pushed a commit
to madebr/SDL
that referenced
this issue
Mar 19, 2023
* Make Win32 fullscreen and borderless windows minimizable * reduced a few ifdefs, fixed an unused warning if built w/o SDL_HAVE_YUV. * rawinput: Fix double detection of gamepads on some 3rd party X360 wireless receivers The name that the Raw Input joystick driver pulls from the HID stack comes from USB string descriptors contained on the device. For official wireless receivers, this always contains "Xbox 360 Wireless Receiver for Windows" which matches the friendly name that WGI provides. 3rd party Xbox 360 wireless receivers may have different strings in their USB string descriptors (one uses "XBOX 360 For Windows" instead). This fails to match WGI's name and causes Raw Input and WGI to both report the same gamepad. Since wireless Xbox 360 controllers seem to have a consistent VID/PID regardless of the adapter enumerating them, we can also match on that to catch these. The duplicate case reported to me was: Controller (XBOX 360 For Windows) - 030000005e040000a102000000007200 Xbox 360 Wireless Receiver for Windows - 030000005e0400000000000000007701 * Makefile.os2: fixed setting OS2DEBUG in CFLAGS. * minor os/2 clean-ups. * emscripten: Don't prevent default on filtered key events * IME Composition Truncation + SDL_IsTextInputShown + SDL_ClearComposition (libsdl-org#5398) * Fixes for IME Composition Truncation + Addition of SDL_ClearComposition, SDL_IsTextInputShown * Fixed: Documentation and code style issues raised during code review. * Sync wiki -> header * SDL_windowskeyboard.c (IME_IsTextInputShown): remove unused local vars. * cmake: Make test code use proper C main functions. Fixes libsdl-org#5021. * SDL_blit_N.c: removed duplicated const (fixes bug libsdl-org#5401) * SDL_endian.h: use endian predefs from newer gcc and clang versions. Closes: libsdl-org#5403 * Android: add comment to set render target to NULL when going to background (bug libsdl-org#4041) * Make SDL_VIDEO_OPENGL_EGL optional on Android * Fixed: Incorrect assumption that mouse button is released when window is allocated * SDL_audiocvt.c: Don't byteswap 8-bit streams Otherwise, this results an assert on big endian machines when attenpting to use SDL_LoadWAV_RW function to load 8-bit WAV files. * METAL: clip rect w/h must be <= render pass * Fix name of macOS platform and link to README file * Fixed bug libsdl-org#2199: make SDL_blit_slow handles SDL_PIXELFORMAT_ARGB2101010, storing as RGBA * Fixed bug libsdl-org#2691 - re-enable surface_testCompleteSurfaceConversion for ARGB2101010 * INSTALL.txt: Say "macOS" instead of "Mac OS X" in modern times. Reference Issue libsdl-org#5407. * SDL_windowskeyboard.c: fix build with SDL_DISABLE_WINDOWS_IME defined. Fixes libsdl-org#5408 * minor whitespace cleanup. * SDL_audiocvt.c: minor cleanup. * Adjust gradle dependencies to enable the build to position assets for the APK * emscripten: obey enable-misc/SDL_MISC settings * minor optimization (SDL_audiocvt.c) * imported two libtool mainstream commits 28fbcb6a and b55b1cc8 * Remove 'reserved identifier' warning * SDL_blit_slow: remove one nested 'if()' because of ARGB2101010 handling * SDL_triangle_blit_slow: sync code with SDL_blit_slow to handle ARGB2101010 * SDL_GetBasePath() fixes for OS/2 * use _Static_assert for SDL_COMPILE_TIME_ASSERT(), when available * testplatform.c: move static asserts out of TestTypes(). * simplify SetDSerror - no need to keep the error in a static variable - always print the error code - reduce the required stack-size - reduce the number of snprintf calls (and code size) * Fixed build when events are disabled Fixes libsdl-org#5413 * Removed problematic call to ISensor_SetEventSink() Fixes libsdl-org#5288 * Fixed warnings when building with cygwin Fixes libsdl-org#5025 * SDL_cocoawindow.m: update fullscreen toggle when SDL_SetWindowResizable called * Ignore unknown WM_SIZE types. According to MSDN, we can also get SIZE_MAXHIDE and SIZE_MAXSHOW, based on state changes to other windows. It's not clear under what circumstances this will happen (I saw some docs indicating it may require multiple application windows), but it doesn't seem right to treat them as RESTORED. * Fix relative mouse input for Unvanquished (unvanquished.net) Here's an IRC dump that hopefully explains the issue this fixes: > I'm debugging something odd where, for a libre game, unvanquished.net (a FPS), relative mouse input in fullscreen is buggy > it's like, working mostly ok, but it has a weird performance/cleanup bug > after some time in relative mouse input mode, some time as low as 15s, usually more, the SDL sends A LOT of relative mouse input per frame > almost all of which have xrel==0 && yrel==0 > by A LOT, I mean that after ~1min, it's usually in the thousands per frame > each frame, a while ( SDL_PollEvent( &e)) loop reads the inputs, but it seems SDL is not clearing the list. > one way to clear the list is to open the in-game console or menu, which switches the input mode to absolute, then close it which gets a working relative input mode (for some time at least) > I've shown the issue to be present with SDL2.0.20 but not with 2.0.14 on my system > some other players on Arch Linux (SDL2.0.20) report a possibly related issue, where some keys seem to be pressed at random > I've did some bisection on SDL master, and I've found that there are actually two commits involved, one breaking it totally (no input at all), and one fixing it partially (with the problem described above) First related commit that breaks it totally: commit 82793ac Author: Sam Lantinga <slouken@libsdl.org> Date: Thu Oct 14 14:26:21 2021 -0700 Fixed mouse warping while in relative mode We should get a mouse event with an absolute position and no relative motion and shouldn't change the OS cursor position at all Second related commit, that halfway fixes it: commit 31f8c3e Author: Sam Lantinga <slouken@libsdl.org> Date: Thu Jan 6 11:27:44 2022 -0800 Fixed event pump starvation if the application frequently pushes its own events Reverting the first commit did fix the issue for me, but would probably reintroduce the bug it was fixing(?). This patch should fix it for everyone hopefully. DaemonEngine/Daemon#600 is the upstream bug, and contains some early investigation. * Don't warn if anyone peeps for events after quitting the event subsystem Fixes libsdl-org#5013 * video: Add a hint to allow Vulkan surfaces on foreign windows * Fixed compile warning and comment typo * Added a hint to mark a foreign window as usable with OpenGL Fixes libsdl-org#2942 * When updating grab state, only activate windows that are grabbed, fullscreen, and shown. Fixes libsdl-org#5371 * Try not forcing activation when grabbing the mouse in fullscreen windows * Workaround for bug in Microsoft WGI support Fixes libsdl-org#5270 * Use SDL_Log instead of printf * Added a hint to capture the mouse when mouse buttons are pressed, defaulting on Fixes libsdl-org#5301 * Fixed memory leak in WIN_CreateBlankCursor() * Fixed freeing the Windows blank cursor * Start rumbling once a raw input controller has been correlated Fixes libsdl-org#5351 * Modern CMake doesn't need "LANGUAGE C" for Objective-C CMake 3.19 fails to compile Objective-C with that property set Fixes libsdl-org#5418 * Added the hint SDL_HINT_MOUSE_RELATIVE_MODE_CENTER, controlling whether the mouse should be constrained to the center of the window or the whole window in relative mode. For further info about the pros and cons, check out the discussion in libsdl-org#5271 * Added a note that SDL_RenderReadPixels() should be called before SDL_RenderPresent() * Sync wiki -> header * Fix "SDL_TRUE is not defined" runtime error for emscripten. * SDL_UnionRect: If both rects are empty, zero out the result struct. * SDL_Rect: minor code cleanups. * SDL_Rect: Added floating point versions of all the rectangle APIs. Fixes libsdl-org#5110. * Fix "SDL_FALSE is not defined" runtime error for emscripten. * Fixed Android crash when SDL_HIDAPI_DISABLED set to 1 The Java code needs the native functions to be implemented, even if they're not surfaced via the C API. Fixes libsdl-org#5326 * Relative mouse mode is tied to the window with keyboard focus This isn't obvious, but makes sense when thinking about how games actually use it. This is also in line with how Windows mouse relative mode is implemented. Fixes libsdl-org#5340 * Fixed parameter operation ordering for ease of reading * audio: Set error message on dsp init failure. if SDL_EnumUnixAudioDevices() fails to find any devices, set an error message on the exit path. Without this, SDL_Init() could fail without any message available in SDL_GetError(). * Sync wiki -> headers. * Fixed bug libsdl-org#2426 - SDL_RenderReadPixels result is unspecified and fails testautomation Call SDL_RenderPresent after calling SDL_RenderReadPixels. From "include/SDL_render.h": "If you're using this on the main rendering target, it should be called after rendering and before SDL_RenderPresent()." * Fixed bug libsdl-org#2962 - when SDL_RenderReadPixels format = 0, used format of the target texture include/SDL_render.h, format: "0 to use the format of the rendering target " * blit-auto optimizations * regenerated SDL_blit_auto.c. * Don't try to hide foreign windows when destroying the SDL window representation Fixes libsdl-org#5432 * hints: Added SDL_HINT_QUIT_ON_LAST_WINDOW_CLOSE. Fixes libsdl-org#2349. * SDL_render.c: remove unsused case, since renderer target has been forced to NULL previously (see bug libsdl-org#4213) * Add mapping for Logitech Precision Gamepad * Also fix Android crash when SDL_HIDAPI_DISABLED set to 1 The Java code needs the native functions to be implemented, even if they're not surfaced via the C API, therefore, a stub version of functions were made only to the purpose of "fill the gaps" when SDL_HIDAPI_DISABLED set to 1. * use != '\0' comparison instead of SDL_strlen in the functions of SDL_getenv * make SDL_strcasecmp standard compliant * optimize the SDL_str(case)cmp functions * Avoid trying to use texture framebuffers on emscripten * stdlib: just cast iconv()'s 2nd arg to void *. This makes the compiler happy (enough) regardless of whether the C runtime headers think this argument should be const or not. Fixes libsdl-org#4966. * Fixed freeing a constant string (thanks @gnrlwart) * Android: add SetWindowResizable() implementation - which also enable/disable the orientation lock status. This is only provided when the window is not SDL_WINDOW_FULLSCREEN (see SDL_video.c). Final orientation also depends on SDL_HINT_ORIENTATIONS. * Video: do not use hardcoded VIVANTE and VDK libraries * Add SDL_RenderGetWindow() API to get the window associated with a renderer (libsdl-org#5440) Add SDL_RenderGetWindow() API to get the window associated with a renderer * Sync wiki -> header * wayland: Bump minimum requirement to wayland-client 1.18 or newer * wayland: The rest of the wayland-client 1.18 requirement... Git, please * wayland: Basic support for zwp_tablet_*v2 protocol * video: wayland: Use xdg-output for retrieving the desktop dimensions Using wl-output to get the desktop display dimensions and dividing by the integer scale factor will not return the correct result when using a desktop with fractional scaling (e.g. a 3840x2160 display at 150% will incorrectly report the scaled desktop area as 1920x1080 instead of 2560x1440). Use the xdg-output protocol, if available, to retrieve the correct desktop dimensions and offset. Versions 1 through 3 of the protocol are supported. * Vita: support audio format fallback * Vita: add native YUV textures support. * Fail if texture init fails. * Refactor and cleanup. * static analysis: Fixed several complaints from codechecker. There are still some pending Objective-C specific issues. Reference issue libsdl-org#4600. * add some Thrustmaster wheels * Fixed bug libsdl-org#1650: X11 doesn't set KMOD_NUM and KMOD_CAPS to system state * x11: Minor cleanups and corrections in X11_ShowMessageBox. Reference issue libsdl-org#3254. * Avoid depending on libwayland 1.20 unnecessarily When using shared linking (linking in the normal way with -lwayland-client) rather than loading Wayland libraries dynamically at runtime, listing symbols that don't exist in the current version results in a build failure. We don't actually call wl_proxy_marshal_flags() or wl_proxy_marshal_array_flags() directly; the reason we need them is that they're called by the code generated by wayland-scanner >= 1.20. If we're building against an older Wayland library, then we'll have its corresponding version of wayland-scanner (mismatched versions are not supported), so we won't need those two symbols, and can avoid generating a dependency on them. Conversely, if we're building against a newer Wayland library, the generated code will call them unconditionally, so we cannot treat them as optional and gracefully fall back: that would result in a crash. Instead, treat them as a mandatory part of the Wayland library, so that if they are not found at runtime, we can fall back to X11 without crashing. libwayland 1.18 is in several LTS distributions (Ubuntu 20.04, Debian 11, RHEL 8) so avoiding a hard dependency on 1.20 is quite useful. Signed-off-by: Simon McVittie <smcv@collabora.com> Resolves: libsdl-org#5376 * wayland: Don't overwrite error message from SDL_EGL_CreateSurface SDL_EGL_CreateSurface sets a more specific error message, so overwriting it would lose information. Signed-off-by: Simon McVittie <smcv@collabora.com> * wayland: Minor fixes for old compilers * wayland: Remove some now-redundant casts * iOS >= 10.0 silence GLes deprecation warnings * wayland: Add an xkb_keysym_t->SDL_Keycode mapping for backspace * wayland: Enforce text capitalization manually, for remapped keymods * wayland: Use xkb_keymap_mod to set mod state * YUV: fix invalid read on last line when converting from SDL_PIXELFORMAT_YUY2 (see bug libsdl-org#4841) * Vita: Fix NULL-pointer dereference * wayland: Optimize keyboard_handle_modifiers. 1. Mod index values are (mostly) constant, so can be done with xkb_state_new 2. Mods can change without the group changing, avoid remap events if possible Lastly, as a bonus, I added braces to the locale check, because I was nearby. * Initialize compile status variable and check also program link status * testshader: use SDL_malloc instead of SDL_stack_alloc. * README-windows: Convert to actual Markdown text. * README-windows: Wordwrap text file. This makes for easier reading outside of a formal Markdown viewer. * README-windows: Notes on building with Visual Studio/LLVM. Fixes libsdl-org#5186. * wayland: Work around a GNOME xdg_output scaling issue * wayland: Avoid overwriting xdg_output position with wl_output position * wayland: Relax the check for mismatching output scales * configure: improvements to libdecor discovery : use PKG_CHECK_MODULES, and use DECOR_LIB with find_lib. Closes: libsdl-org#5460 Co-authored-by: Trigan2025 <trigan2025@hotmail.fr> * direct3d: Implement missing blend operations. This is only for Direct3D 9; Direct3D 11 already had this implemented. Fixes libsdl-org#5375. * Remove URLs from markdown headers in README-visualc.md * Clean up formatting in README-directfb.md * testgles2: Call correct function to get shader info log and add link status checking * Sync wiki -> header * Sync wiki -> header * Vita: Use preallocated memory pool for textures * Sync wiki -> header * Correct default structure packing on Windows ARM64 See issue libsdl-org#5454 for details * Fix potential memory leak in QueueCmdFillRects * Vita: add hint to select which touchpad generates mouse events * emscriptenaudio: proxy calls to main thread * video: wayland: Use wp-viewporter for fullscreen with non-native resolutions Wayland doesn't support mode switching, however Wayland compositors can support the wp_viewporter protocol, which allows for the mapping of arbitrarily sized buffer regions to output surfaces. Use this functionality, when available, for fullscreen output when using non-native display modes and/or when dealing with scaled desktops, which can incur significant overdraw without this extension. This also allows for the exposure of arbitrarily sized, emulated display modes, which can be useful for legacy compatability. * wayland: Minor fix for old compilers * video: wayland: Expose more resolutions for mode emulation Expose as many emulated display modes as possible. They will currently display stretched to the display's native desktop aspect, but if an application requires a hardcoded resolution, it will work at minimum. Aside from the change in the emulated display mode list, the Wayland event handling code had to be updated to support separate scaling for the x and y axes, as square pixels are no longer guaranteed. * video: wayland: Use viewports for non-fullscreen windows with fractional scaling Use viewports for non-fullscreen windows when the desktop uses fractional scaling and the window is flagged as DPI-aware to provide a backbuffer mapped as close to 1:1 output as possible. In the cases of odd window sizes the backbuffer may be a pixel off of scaling perfectly into the window size due to its scaled size being rounded off, but a minute amount of scaling during output is likely preferable to the large amounts of overdraw needed with integer scaled buffers. * Vita: add audio capture support * Vita: add SDL_GetPreferredLocales support * stdinc: SDL_COMPILE_TIME_ASSERT defines shouldn't have a semicolon. * x11: Catch X11 errors in X11_SetWindowPosition and X11_SetWindowSize. The functions can go south if other operations are in progress, like X11_SetWindowBordered, which might be doing something traumatic behind the scenes of the window manager. We can't make these tasks totally synchronous, which would fix the problem, because not only can the window manager block however long it wants, it might also decide to deny our requests without any notification, so we'd be waiting forever for a window change that isn't coming. :( Fixes libsdl-org#5274. * joystick: Fix rumble issues on PS5 HIDAPI controllers We were returning the report size from HIDAPI_DriverPS5_RumbleJoystick() rather than 0 upon success, causing SDL_JoystickRumble() (and callers) to think that rumbling failed. This didn't cause major problems until 1868c5b, when it started preventing rumble state from being persisted in the joystick core, even though it was successfully sent to the hardware. This led to all sorts of strangeness, including broken rumble duration and attempts to stop rumble being discarded. * Desktop OpenGL 1.X/2.X PSVita Support * Update README-vita.md * Cleanup Spaces * Vita: PVROGL: fix indentation and ifdef guards * Vita: fix readme * x11: Ignore BadValue for extremely small XRRSetScreenSize resolutions. Reference Issue libsdl-org#4840. * SDL2 thread proxying fixes This PR uses new APIs added in [emscripten-core/emscripten#9336](emscripten-core/emscripten#9336) to improve compatibility with USE_PTHREADS=1. Original PR: emscripten-ports/SDL2#127 By: @jakogut Reviewed by: Daft-Freak * emscriptenmouse: remove old extra `_INT` * fix formatting and cast warnings Co-authored-by: Charlie Birks <charlie@daftgames.net> * emscriptenmouse: remove useless return statement * emscriptenframebuffer: fix formatting * emscripten: Proxy SDL_GetUsableDisplayBounds to the main thread. * Fixed loading 32-bit BMP files * Move SDL_List functions to SDL_list.c to avoid more merge with eventual PR * Add SDL_list.c/h * Add SDL_list to macosx xcode * Use RoInitialize/RoUninitialize for Windows.Gaming.Input Thanks @walbourn! Fixes libsdl-org#5270 * fix build against older SDKs after commit 8ebef12. * check for HAVE_ROAPI_H in cmake and autotools, and update SDL_config_windows.h and SDL_config_winrt.h * attempt to fix uwp build * render: Fix setting the scale mode for non-native textures * Visualise scroll wheel events in testmouse * Minor cleanup * x11: Don't unload libGL.so to prevent a crash in XCloseDisplay() libGL.so may register callbacks that can be invoked upon XCloseDisplay(). If XCloseDisplay() is called after libGL.so is unloaded, the callback pointer will point at freed memory and invoking it will crash. The texture framebuffer check optimized out in f37e4a9 was causing libGL.so to never be unloaded as a side-effect. Skipping it exposed this bug by allowing libGL.so to actually unload. * x11: Wait a bit to see if window pos changes when changing fullscreen. Helps prevent window from moving to 0,0 when leaving fullscreen. Fixes libsdl-org#4749. * testgles2: Fix buffer object sizes * Added a define VERBOSE_MOTION_EVENTS to show mouse and finger motion events * Handle interaction between auto capture and the SDL_CaptureMouse() API Fixes libsdl-org#5457 * Minor cleanup * IBus should use ev keycode instead of X keycode See: https://github.com/ibus/ibus/blob/5a455b1ead5d72483952356ddfe25b9e3b637e6f/client/gtk2/ibusimcontext.c#L468 * Ignore focus change messages that contradict GetForegroundWindow. On Wine, when a window is programmatically minimized in response to losing focus, we receive a WM_ACTIVATE for the deactivation, but GetForegroundWindow still indicates that our window is focused. This causes an incorrect SDL_WINDOWEVENT_FOCUS_GAINED. This is probably a Wine bug, but it may take a while to fix and then for the fix to make its way to users. * Don't resize fullscreen windows when hiding or minimizing them (thanks @madewokherd!) This has the benefit of window previews (mousing over the icon) having the correct size and contents. Fixes libsdl-org#5320 * x11: when waiting on fullscreen changes, not window position _and_ size. This makes sure the window doesn't have outdated values if you try to access them (or call something that does, like SDL_SetWindowMinimumSize). Fixes libsdl-org#5233. * Send key release event to input method. (libsdl-org#5281) Co-authored-by: Ethan Lee <flibitijibibo@gmail.com> * Cache the fact that a device didn't look like a joystick Fixes libsdl-org#5211 * x11: Try to keep SDL_WINDOW_FULLSCREEN* in sync with window manager. So if Gnome/KDE/etc have a keyboard shortcut or titlebar decoration to make any window go fullscreen (with the _NET_WM_FULLSCREEN flag on the _NET_WM_STATE property), we update the SDL window flag. Fixes libsdl-org#5390. * Sort controllers by the js* index on Linux Also fixed the initial scan to directly scan devices instead of using udev so they can be sorted, as intended. Fixes libsdl-org#4688 * wayland: Evaluate WINDOWPOS_CENTERED_DISPLAY for move events Partially fixes the mouse cursor in UE5 editor. Imperfect because UE5 uses window position and global mouse state to get position, but of course we don't have global mouse and this is just to get the right display index so this still fails overall. We really need to make global mouse support a feature query... * wayland: Try to detach at the beginning of ShowWindow, just in case. It's possible that an external component (probably a GL/VK context) committed, so we need to cover our bases and detach in both HideWindow and ShowWindow. Fixes a crash in UE5 editor's pop-ups. * wayland: Add a bug link for the detach FIXME * wayland: Pin the fake window position at (0, 0). I kind of thought it'd be nice to have it in the center, but this is an issue for applications that still assume global mouse and window positions are accessible. For example, this fixes cursor offset issues in UE5. * Makefile.in: added missing CXX variable. * Compile with recursive mutexes for emscripten Emscripten actually does support recursive mutexes, so no need to use SDL's fake recursive code. Background: libsdl-org#5428, libsdl-org#5479 * x11: Treat WM setting the window "fullscreen" like FULLSCREEN_DESKTOP. Fixes libsdl-org#5390. * Only update modifier state for keys that are pressed in another application Fixes libsdl-org#4432 * minor adjustment to os/2 watcom makefile * Make sure the UIKit message box is being handled on the main thread Potentially fixes libsdl-org#4865 * render: Update the size/scale/viewport on moves, in addition to resizes. For OpenGL this means resetting the viewport state shadowing flag too. Fixes libsdl-org#1504 * direct3d11: Set the swapchain target immediately after creating it. Fixes libsdl-org#4782 * video: wayland: Set the surface damage region when using fullscreen viewports When using emulated display modes, the output size is often larger than the drawable buffer. As the surface damage region is automatically calculated from the smaller drawable buffer size, the damage region needs to be manually set to cover the entire viewport region or visual repaint artifacts can result. * Fixed typo * Updated the patch notes with API changes for 2.0.22 * Update version to 2.0.22 for release * audio: SDL_ConvertStereoToMono_SSE3 missed an unaligned load. * UWP: Require Windows 10 16299 or newer. This is required to build with WGI support. Thanks for @FrozenChameleon for the fix! Fixes libsdl-org#5504 * Vita: fix VIDEO_VITA_PVR flag * fix os2 timer in fallback mode * testevdev: Recognise touchpads as such At the time I contributed this unit test, SDL didn't understand Linux touchpads, but now it does. Fixes: 373216a "Added support for touchpads in the Linux evdev code" Signed-off-by: Simon McVittie <smcv@collabora.com> * testevdev: Adapt to a broader definition of keyboards At the time I contributed this unit test, SDL had a relatively narrow definition of what is a keyboard, approximately matching udev ID_INPUT_KEYBOARD. Now it uses the equivalent of udev ID_INPUT_KEY, which matches anything with keyboard keys, and not just reasonably complete alphanumeric keyboards. Fixes: 040bd7a "Fix udev not detecting ID_INPUT_KEY devices when udev is not running" Signed-off-by: Simon McVittie <smcv@collabora.com> * test: Copy utf8.txt to build directory testiconv wants this. Signed-off-by: Simon McVittie <smcv@collabora.com> * Initialise scandir argument 'scandir' does not initialise 'entries' on error * Wayland: Add SDL_HINT_VIDEO_WAYLAND_PREFER_LIBDECOR This hint allows libdecor to be used even when xdg-decoration is available. It's mostly useful for debugging libdecor, but could in theory be used by applications which want to (for example) bundle their own libdecor plugins. * WhatsNew.txt: Added SDL_HINT_VIDEO_WAYLAND_PREFER_LIBDECOR * WhatsNew.txt: Document new dependency on libwayland-client 1.18.0 Signed-off-by: Simon McVittie <smcv@collabora.com> * CI: update os2.yml to use open-watcom/setup-watcom also remove os2-buildbot.sh -- not needed anymore. * avoid NullPointer in SDL_GL_MakeCurrent * avoid NullPointer in SDL_GetRenderTarget * video: Wayland: Clamp fullscreen window dimensions to desktop A scaled fullscreen window may exceed the bounds of the desktop. Clamp the window size to the desktop dimensions in fullscreen mode. * video: Wayland: Always round scaled pointer coordinates down Rounding up can cause the pointer coordinates to exceed the window boundaries at the right and bottom edges. * x11: Update the display when the WM changes a window's fullscreen state. Fixes libsdl-org#5390. * Added CMake option to disable the installer * cmake: Fixed indenting and some oldschool `endif(TEXT)` things. * hints: Make SDL_VIDEODRIVER and SDL_AUDIODRIVER formal hints. They were just environment variables before. Fixes libsdl-org#5528. * Revert "video: Prefer Wayland over X11" This reverts commit 8ceba27. SDL Wayland support is stable, but there are a number of issues with third-party software (NVIDIA drivers, libwayland event overflow, libdecor not handling plugin load failures, Steam overlay not working with Wayland, etc.) that make it better to default to X11 at this time. Games which would like to prefer wayland when available can use the following code before SDL_Init(): SDL_SetHint(SDL_HINT_VIDEODRIVER, "wayland,x11"); Fixes libsdl-org#5527 * Certain audio drivers, like the RME "Pro" Audio driver, have resampling quality issues when using WASAPI. We'll use SDL's resampling algorithm so we have consistent quality between platforms and drivers. Fixes libsdl-org#5538 * wayland: Add support for TOOLTIP/POPUP_MENU * video: Don't minimize fullscreen windows on focus loss by default when mode switching is disabled When mode switching is disabled in a video backend, fullscreen windows are basically just fullscreen desktop windows with different internal scaling. As no mode switching occurs, there's no need to minimize them on focus loss by default. This can still be overridden by explicitly setting the internal hint for minimizing on focus loss. This has the side effect of fixing a bug on GNOME, where, when a fullscreen Wayland window has it's focus lost and restored via alt+tab followed by switching back to windowed mode, the top portion of the window won't end up being obstructed by GNOME's top bar. * Added controller mapping for the Thrustmaster Dual Analog 3.2 on Linux * Enable evdev-based gamecontrollerdb on FreeBSD * simplify SDL_GetBasePath on windows - use GetModuleFileName directly (as recommended) * use explicit GetModuleFileNameW * Remove HWND_TOPMOST for fullscreen windows Fixes libsdl-org#5509 * SDL_render.c: internally change viewport/cliprect type from SDL_FRect to SDL_DRect (double precision). (see bug libsdl-org#5547) * Don't call scandir() inside of scandir() This works around a crash in address sanitizer * Fixed compile warnings * x11: Revert "Fix keymap updating for X11 backend" This reverts commit de6d290. This patch had multiple issues, discussed in libsdl-org#5520. * Fixed build * Fix recent changes in VitaSDK * SceKernelMemBlockType to unsigned int * Fix a -Wshadow warning * Remove an unused function prototype * WhatsNew.txt: Add SDL_FRectEqualsEpsilon() * Add SDL_FLT_EPSILON. * SDL_Rect: Use a default epsilon in SDL_FRectEquals() Add SDL_FRectEqualsEpsilon() for when more control over equality test is required. * test: Basic tests for SDL_FRectEquals Based on the integer version. These tests mostly check that input isn't mangled and that invalid input gives the expected negative result. * hidapi: Wrap CopyHIDDeviceInfo in define checks. The purpose of this is to silence a 'defined but not used' warning. * os2: SDL_DestroyMutex should ignore NULL mutexes. Every other backend does this, so this should match, now. It's possible this was harmless, but we can avoid the system call and the (likely?) debug message when it fails, though! * windows: Fix calling convention for RoInitialize/RoUninitialize Fixes libsdl-org#5563 * windows: Fix RoInitialize() failure after a CoInitializeEx() call using apartment threading This mirrors the same codepath in WIN_CoInitialize() which handles STA and MTA. * WGI: Keep a reference to the MTA to avoid crashing on COM teardown Fixes libsdl-org#5552 Fixes libsdl-org#5270 * WGI: Only call RoUninitialize() if RoInitialize() succeeded * Handle potential out of memory condition when working with hints * SDL_Rect: Add \sa documentation block to SDL_FRect * Return a correlation error when trigger rumble is attempted without correlation * Fixed build * SDL: sometimes the PS5 controller doesn't report having to power even when connected over USB. Possibly related to being completely charged? Either way we already know that it's USB or BT so let's use the driver's knowledge instead. * Added support for the Backbone One controller on iOS * Revert "Fix relative mouse input for Unvanquished (unvanquished.net)" This reverts commit 3318590. Fixes libsdl-org#5569 * hidapi, libusb: import mainstream commit 536bad201e * Fixed logical size synchronization issue on macOS https://discourse.libsdl.org/t/sdl-2-0-22-prerelease/35306/6 * x11: revert checks for _NET_WM_STATE_FULLSCREEN changes. This reverts commit 8597735. This reverts commit 0249df9. Fixes libsdl-org#5572. Reopens libsdl-org#5390. Co-authored-by: Jo Bates <jo@valvesoftware.com> Co-authored-by: Ozkan Sezer <sezeroz@gmail.com> Co-authored-by: Cameron Gutman <aicommander@gmail.com> Co-authored-by: Charlie Birks <charlie@daftgames.net> Co-authored-by: Zach Reedy <86971250+zreedy@users.noreply.github.com> Co-authored-by: SDL Wiki Bot <icculus-sdlwikibot@icculus.org> Co-authored-by: Ryan C. Gordon <icculus@icculus.org> Co-authored-by: Sylvain <sylvain.becker@gmail.com> Co-authored-by: Guus Waals <_@guusw.nl> Co-authored-by: Zach Reedy <zreedy@opera.com> Co-authored-by: Wohlstand <admin@wohlnet.ru> Co-authored-by: Sam Lantinga <slouken@libsdl.org> Co-authored-by: staphen <staphen@gmail.com> Co-authored-by: pionere <pionere@freemail.hu> Co-authored-by: uyjulian <uyjulian@gmail.com> Co-authored-by: Esme Povirk <esme@codeweavers.com> Co-authored-by: Antoine Fontaine <antoine.fontaine@epfl.ch> Co-authored-by: Ethan Lee <flibitijibibo@gmail.com> Co-authored-by: Christian Kündig <kuendig@scandit.com> Co-authored-by: Eddy Jansson <eddy@klopper.net> Co-authored-by: Egor <egor@opensrc.club> Co-authored-by: Jonatha Gabriel <jonathagabrielns@gmail.com> Co-authored-by: Joao Paulo Magalhaes <dev@jpmag.me> Co-authored-by: Florian "sp1rit" <sp1ritCS@protonmail.com> Co-authored-by: Frank Praznik <frank.praznik@gmail.com> Co-authored-by: Ivan Epifanov <isage.dna@gmail.com> Co-authored-by: Kimplul <kimi.h.kuparinen@gmail.com> Co-authored-by: Simon McVittie <smcv@collabora.com> Co-authored-by: DominusExult <domiman+github@gmail.com> Co-authored-by: capehill <juha.niemimaki@gmail.com> Co-authored-by: Trigan2025 <trigan2025@hotmail.fr> Co-authored-by: Cameron Cawley <ccawley2011@gmail.com> Co-authored-by: Mathieu Eyraud <70028899+meyraud705@users.noreply.github.com> Co-authored-by: Connor Clark <cjamcl@gmail.com> Co-authored-by: Jaylon Gowie <jaylongowie46@gmail.com> Co-authored-by: Reinhold Gschweicher <pyro4hell@gmail.com> Co-authored-by: Christoph Reichenbach <creichen@gmail.com> Co-authored-by: Weng Xuetian <wengxt@gmail.com> Co-authored-by: David Gow <david@ingeniumdigital.com> Co-authored-by: Semphris <semphris@protonmail.com> Co-authored-by: Jan Beich <jbeich@FreeBSD.org>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
While retesting #4915, I discovered that WGI initialization now deterministically fails (since 8ebef12) unless the application calls
CoInitializeEx(NULL, COINIT_MULTITHREADED)prior to initializing the joystick stubsystem.Our
RoInitialize()call fails due to incompatible concurrency model (RPC_E_CHANGED_MODE) with a previousCoInitializeEx()call.The first call to
CoInitializeEx()comes from DirectInput:WIN_CoInitialize()tries to use an apartment threading model if COM is not already initialized:So we will end up with apartment threading unless the application has already initialized COM outside of SDL.
Next we will try to call
RoInitialize(RO_INIT_MULTITHREADED), which is a conflicting concurrency model with apartment threading that we used earlier.This fails with
RPC_E_CHANGED_MODEand causes us to fail to initialize the WGI joystick driver. It looks likeRO_INIT_SINGLETHREADEDis equivalent toCOINIT_APARTMENTTHREADED, so we should probably try that likeWIN_CoInitialize()does.The text was updated successfully, but these errors were encountered: