The _xgetbv intrinsic was being used in ARM builds of SDL/WinRT, which was
leading to linker errors. This commit limits _xgetbv use to the platforms on
which it is available, x86 and x64.
This will (eventually) make SDL_GetQueuedAudioSize() more accurate, and thus
reduce latency. Right now this isn't implemented anywhere, so we assume data
fed to the audio callback is consumed by the hardware and immediately played
to completion.
For versions of XAudio2 with an IXAudio2SourceVoice::GetState() that offers a
flags parameter, we can use XAUDIO2_VOICE_NOSAMPLESPLAYED, since we don't
need this information in our current calls. According to MSDN, this makes the
the call about 3x faster.
This requires at least Xcode 4.5 and the iOS 6 SDK to build, but it doesn't change the minimum supported runtime version (iOS 5.1). Less than 2% of iOS users are running iOS 5, so I hope developers aren't trying to build SDL using an SDK which doesn't support iOS 6/7...
Nitz
In GL_CreateTexture function:
if (GL_CheckError("glGenTexures()", renderer) < 0) {
SDL_free(data);
return -1;
}
Here only data is getting free but data->pixels getting leak.
So have to free data->pixels before free data.
Previously, SDL would always expose display modes and window dimensions in terms of pixels, and would add an extra 'fake' display mode on retina screens which would contain the non-retina resolution. Calling SDL_CreateWindow with the dimensions of that fake display mode would not work.
Now, SDL only exposes display modes and window dimensions in terms of points rather than pixels. If the SDL_WINDOW_ALLOW_HIGHDPI flag is passed into SDL_CreateWindow, then any OpenGL contexts created from that window will be sized in pixels rather than points (retrievable with SDL_GL_GetDrawableSize.) Window dimensions and mouse coordinates are still in terms of points rather than pixels even with that flag.
This matches the behavior of SDL in OS X more closely, and lets users choose whether to make use of retina displays and lets them handle it properly.
Damian Kaczmarek
Basically this bug is probably not a common use case. My goal is to allow rendering totally without a window, for example to a screenshot and I need to rely on SDL_SetRenderTarget to properly work for a purely software renderer created by SDL_CreateSoftwareRenderer.
Pablo Mayobre
When generating a signed app with SDL 2.0.3 an issue comes up, watching at the Error Log points out that the issue lies in the src/main/android/SDL_android_main.c where the process name is defined as "SDL_app", this name turns into an erroneous name so it should be changed to "app_process"
chasesan
When using SDL_RenderCopyEx, I get a problem on some platforms where the output is offset by +/-1 on other platforms and not on others. I tried it with a center of both 0,0 (and offsetting by width/height) and NULL (for centered).
The rotation involved is 90, and/or -90 rotation. The rotation was a constant, no arithmetic was involved when inputting it into SDL_RenderCopyEx.
This occurred with 32x32, 24x24, and 16x16 texture sizes. I apologize that I don't have more precise information, as I received the information as a bug report myself. But I have tracked the problem down to here.
My program requires pixel perfect alignment on several different platforms, so this is something of a showstopper for me.
--
Sylvain
It appears the RenderCopyEx is done as expected,
this is the red rectangle which is not correctly positionned !
So, here's patch with a 0.5 float increment, like for opengles2, for DrawLines, and also Draw Points.
binarycrusader
Since changeset 358696c354a8, SDL 2.0 has been broken on Solaris when compiling with the Solaris Studio compiler (which uses the pthread implementation of SDL_AtomicLock).
Notably, it gets stuck at the MemoryBarrierRelease in SDL_GetErrBuf:
6585 # 218
6586 if (!tls_errbuf && !tls_being_created) {
6587 SDL_AtomicLock_REAL ( & tls_lock );
6588 if (!tls_errbuf) {
6589 SDL_TLSID slot;
6590 tls_being_created = SDL_TRUE;
6591 slot = SDL_TLSCreate_REAL ( );
6592 tls_being_created = SDL_FALSE;
6593 { SDL_SpinLock _tmp = 0 ; SDL_AtomicLock_REAL ( & _tmp ) ; SDL_AtomicUnlock_REAL ( & _tmp ) ; };
^^^ loops forever above
6594 tls_errbuf = slot;
6595 }
6596 SDL_AtomicUnlock_REAL ( & tls_lock );
6597 }
Running: testthread
(process id 28926)
^Cdbx: warning: Interrupt ignored but forwarded to child.
signal INT (Interrupt) in __nanosleep at 0xfe52a875
0xfe52a875: __nanosleep+0x0015: jae __nanosleep+0x23 [ 0xfe52a883, .+0xe ]
Current function is SDL_Delay_REAL
204 was_error = nanosleep(&tv, &elapsed);
(dbx) where
[1] __nanosleep(0xfeffe848, 0xfeffe850, 0xfe75a5ac, 0xfe5169d8), at 0xfe52a875
[2] nanosleep(0xfeffe848, 0xfeffe850), at 0xfe516a3b
=>[3] SDL_Delay_REAL(ms = 0), line 204 in "SDL_systimer.c"
[4] SDL_AtomicLock_REAL(lock = 0xfeffe88c), line 104 in "SDL_spinlock.c"
[5] SDL_GetErrBuf(), line 225 in "SDL_thread.c"
[6] SDL_ClearError_REAL(), line 216 in "SDL_error.c"
[7] SDL_InitSubSystem_REAL(flags = 0), line 116 in "SDL.c"
[8] SDL_Init_REAL(flags = 0), line 244 in "SDL.c"
[9] SDL_Init(a = 0), line 89 in "SDL_dynapi_procs.h"
[10] main(argc = 1, argv = 0xfeffe948), line 65 in "testthread.c"
As far as I can tell, this is because pthread_spin_trylock() always returns EBUSY for this particular lock; since it works in other places, I'm suspicious.
Different Solaris Studio compiler versions seem to make no difference.
I've verified this is broken on Linux as well if SDL_spinlock.c is modified to use the pthread implementation.
This appears to be because pthread_spin_init() and pthread_spin_destroy() are not used with the locks as required.
Wei Mingzhi
surface->map should be invalidated in SDL_SetSurfacePalette(), otherwise the palette would not be effective when blitting to another non-8bit surface which we previously blitted to.
Alex Szpakowski
Some minor changes to the Mac-specific backend code:
- Fixed up some code style issues (mostly brace style inconsistencies).
- Fixed a compiler warning in SDL_cocoaevents.m.
- Removed some useless code now that the 10.7 SDK is required to build SDL.
- Removed Gestalt(gestaltSystemVersion, ...) call and switched to NSAppKitVersionNumber for version checking code. Using Gestalt with gestaltSystemVersion will give 0x1090 in Mac OS 10.10+, and the whole Gestalt function was deprecated in Mac OS 10.8.
Alex Szpakowski
Since this commit https://hg.libsdl.org/SDL/rev/59b543340d63 , calling SDL_DestroyWindow will crash the program if the window has an active OpenGL context.
This is because the Cocoa_DestroyWindow code sets the window's driverdata to NULL and then calls [context setWindow:NULL], which tries to access the window's driverdata, resulting in a null pointer dereference.
I have attached a patch which fixes the issue by moving the line which sets the driverdata to NULL to after the lines which call functions that use the driverdata pointer.
Notes:
- Support for the 'Guide' button does not seem to be possible, as
XInputGetStateEx is not available on WinRT.
- Haptic support appears to be working on WinRT now!
- SDL/WinRT does not allow calls to LoadLibrary or LoadLibraryEx. The calls
to those were removed by this change, but only when compiling for WinRT.
Non-WinRT Windows will continue to detect and load XInput via LoadLibrary and
GetProcAddress calls.
If the EGL extension EGL_KHR_create_context is available, we can use it to
set the core/compatability profile and the minimum OpenGL version.
Use this if it is available to get the context requested by the GL attributes.
SDL/WinRT currently uses a separate XInput backend from SDL/Win32, as WinRT
has no support for DirectInput. This change makes SDL/WinRT's XInput
code snag some recently-changed bits from the Win32-specific,
DirectInput + XInput backend, in order to get the SDL_GameController API
working again on WinRT, insofar that axes map to the correct parts.
TODO:
- test all buttons, making sure WinRT maps buttons the same way that Win32 does
- consider making the Win32 and WinRT codebases share more stuff, minus
the sort of duplication happening via this change. Maybe simulate, or
stub-out, DirectInput calls when on WinRT?
Todd Seiler
Call Stack:
#0 0x0000000101c29291 in Cocoa_StartTextInput at /Users/Todd/Desktop/codes/sources/SDL/src/video/cocoa/SDL_cocoakeyboard.m:512
#1 0x0000000101c110c5 in SDL_SetKeyboardFocus at /Users/Todd/Desktop/codes/sources/SDL/src/events/SDL_keyboard.c:643
#2 0x0000000101c32be4 in SetupWindowData at /Users/Todd/Desktop/codes/sources/SDL/src/video/cocoa/SDL_cocoawindow.m:981
#3 0x0000000101c32d2a in Cocoa_CreateWindowFrom at /Users/Todd/Desktop/codes/sources/SDL/src/video/cocoa/SDL_cocoawindow.m:1092
#4 0x0000000101c99999 in SDL_CreateWindowFrom_REAL at /Users/Todd/Desktop/codes/sources/SDL/src/video/SDL_video.c:1338
#5 0x0000000101ce1484 in SDL_CreateWindowFrom at /Users/Todd/Desktop/codes/sources/SDL/src/dynapi/SDL_dynapi_procs.h:547
#6 0x0000000100018a5e in SceneRenderer at /Users/Todd/Desktop/codes/sources/tseiler_Todds-MacBook-Pro_3405/AppName/src/SceneRenderer.cpp:138
#7 0x0000000100017ca5 in SceneRenderer at /Users/Todd/Desktop/codes/sources/tseiler_Todds-MacBook-Pro_3405/AppName/src/SceneRenderer.cpp:145
#8 0x000000010000cd96 in App::execute(int, char**) at /Users/Todd/Desktop/codes/sources/tseiler_Todds-MacBook-Pro_3405/AppName/src/App.cpp:28
#9 0x0000000100004402 in main at /Users/Todd/Desktop/codes/sources/tseiler_Todds-MacBook-Pro_3405/AppName/src/main.cpp:8
This issue occurred when using Ogre3D Graphics engine on Mac (cocoa) to create the window. Then handing the window handle off to SDL_CreateWindowFrom().
In Ogre3D application you do the following:
window_ = root_->initialise(true, "Ogre Window 2");
loadOgreResources();
Ogre::WindowEventUtilities::addWindowEventListener(window_, this);
#if OGRE_PLATFORM == OGRE_PLATFORM_APPLE
NSWindow* Data = 0;
window_->getCustomAttribute("WINDOW", &Data);
sdl_window_ = SDL_CreateWindowFrom((void*)Data);
#endif
It results in a crash in this function:
SDL_cocoakeyboard.m
void
Cocoa_StartTextInput(_THIS)
{
SDL_VideoData *data = (SDL_VideoData *) _this->driverdata;
NSAutoreleasePool *pool = [[NSAutoreleasePool alloc] init];
SDL_Window *window = SDL_GetKeyboardFocus();
NSWindow *nswindow = nil;
if (window)
nswindow = ((SDL_WindowData*)window->driverdata)->nswindow;
// ...
}
The crash occurred because "driverdata" was nil. Before this function call, a call to SetupWindowData is called:
SDL_cocoawindow.m
static int
SetupWindowData(_THIS, SDL_Window * window, NSWindow *nswindow, SDL_bool created)
{
// ...
if ([nswindow isKeyWindow]) {
window->flags |= SDL_WINDOW_INPUT_FOCUS;
SDL_SetKeyboardFocus(data->window);
}
/* Prevents the window's "window device" from being destroyed when it is
* hidden. See http://www.mikeash.com/pyblog/nsopenglcontext-and-one-shot.html
*/
[nswindow setOneShot:NO];
/* All done! */
[pool release];
window->driverdata = data;
return 0;
}
As you can see: "window->driverdata = data" is performed after the "SDL_SetKeyboardFocus()" call, which eventually leads to "Cocoa_StartTextInput()" where the crash occurs.
Alex Szpakowski
SDL's code for exposing the accelerometer as a joystick on iOS currently uses UIAccelerometer, which was superseded by the CoreMotion framework and deprecated since iOS 5.
The UIAccelerometer code still works (for now), but it also throws deprecation warnings whenever SDL is built for iOS, since SDL's deployment target is no longer below iOS 5.
I've created a patch which replaces the old UIAccelerometer code with a replacement based on the CoreMotion framework. It has identical functionality (to SDL users), however iOS apps are now required to link to the CoreMotion framework when using SDL.
This adds support for all XInput devices, exposed through the SDL joystick API.
The button and axis reporting for XInput devices has been changed to match DirectInput and other platforms.
The game controller xinput mapping has been updated so this change is seamless.
There is a new hint, SDL_HINT_XINPUT_USE_OLD_JOYSTICK_MAPPING, for any applications that have hardcoded the old xinput button and axis set. This hint will be removed in SDL 2.1.
Rainer Deyke
If 'SDL_OpenAudio' is called with 'obtained == NULL', 'prepare_audiospec' performs a bad 'memcpy' with the destination and source pointing to the same block of memory. The problem appears to be on in 'SDL_OpenAudio', which calls open_audio_device with 'obtained = desired' when 'obtained == NULL'. 'open_audio_device' cannot deal with 'desired' and 'obtained' pointing to the same block of memory but can deal with 'obtained == NULL'
Testing:
* For each theme in Windows 7, Windows 7 Basic, and Windows 7 Classic:
- Ran testsprite2
- Pressed Ctrl-G to grab the mouse
- Alt-tabbed away, verified mouse is no longer grabbed
- Alt-tabbed back, verified that mouse was grabbed
- Alt-tabbed away
- Clicked in the window, verified mouse was grabbed
- Alt-tabbed away
- Grabbed the title bar and dragged the window around successfully, verified that mouse was grabbed when move modal loop completed
- Alt-tabbed away
- Clicked the minimize button on the title bar, the window was successfully minimized
- Clicked on the icon in the task bar, the window was restored and the mouse grabbed again
- Alt-tabbed away
- Clicked the close button on the title bar, the window was successfully closed
I added -Wshadow and then turned it off again because of massive variable shadowing in the blit macros.
Feel free to go through that code and fix these if you want. Just uncomment CheckWarnShadow in configure.in if you want to try this.
bill
In SDL_wave.c, BEXT wave files with "bext" instead of "fmt " are choked on
if (chunk.magic != FMT) {
SDL_SetError("Complex WAVE files not supported");
was_error = 1;
goto done;
}
BEXT files http://en.wikipedia.org/wiki/Broadcast_Wave_Format actually playback the same as regular waves. All they have is (A LOT OF) extra header info.
To open them, just SKIP the "bext" chunk, and the "fmt " chunk will be a couple of hundred bytes later.
The "fmt " chunk is also bloated, but if you skip past the extra information to the "data" chunk, there is nothing different about a BEXT wave file than a "normal" one.
You can then load the data and proceed as normal.
callow.mark
Compiling with SDL_VIDEO_RENDER_OGL=0, SDL_VIDEO_OPENGL=0, SDL_VIDEO_OPENGL_WGL=0, SDL_VIDEO_RENDER_OGL_ES2=1, SDL_VIDEO_OPENGL_ES2=1 and SDL_VIDEO_OPENGL_EGL=1 set in SDL_config_windows.h fails.
A patch is attached. See bug #2570 for reasons you might want to compile this way.
Zachary L
SDL_hapticlist and SDL_hapticlist_tail are not set correctly when quitting the subsystem. This matters because they are represented as global variables. In the case you quit and reinitialize the subsystems, problems with dangling pointers arise.
For instance, SDL_hapticlist_tail will not be null on second initialization and because of the check on line 298, it will fail to set SDL_hapticlist appropriately. This can cause a few things to go wrong, like feeding SDL_strcmp a null fname which can cause a segfault.
Alex Szpakowski
Now that SDL for iOS requires at least iOS 5.1 at runtime, there are several old codepaths in the UIKit backend which can be removed. I've attached a patch which does so.
Alvin
The new OpenGL ES 2.0 YUV Texture support does not correctly display padded, non-contiguous YUV data.
I am using SDL2 95bd3d33482e (as provided by 'hg id --id') from Mercurial.
The YUV data I am using is provided by the FFMPEG family of libraries. According to FFMPEG's documentation, "The linesize [pitch] may be larger than the size of usable data -- there may be extra padding present for performance reasons."
The dimensions of the video file that I am using are 480x360. What I get from FFMPEG is a Ypitch of 512, and Upitch and Vpitch are both 256.
When I pack new Y, U and V buffers with only the "usable" data (Ypitch is 480 and Upitch and Vpitch are both 240), and use those new buffers, the image is display correctly.
It appears that the Ypitch, Upitch and Vpitch parameters are not being used by SDL_UpdateYUVTexture().
I use SDL_PIXELFORMAT_YV12 for my YUV texture, however, the same results are seen when I use SDL_PIXELFORMAT_IYUV.
Not sure if this is related or not, but when I render the YUV texture (padded and unpadded) to a RGB24 texture, the resulting image is greyscale (or could by just the Y channel).
The URL field for this bug entry is set to my email (SDL mailing list archive) which includes an example image of what I see when rendering padded, non-contiguous YUV data.
sfalexrog
On systems with vsnprintf call SDL_SetError fails when passed a NULL as an argument. SDL's implementation checks for NULL (as seen in the commit: https://hg.libsdl.org/SDL/rev/835403a6aec8), but system implementation may crash.
Melker Narikka
Local files that are dropped onto a window under X11
are not going through a URI decoding step, resulting in the following
in my test application:
Dropped file /home/meklu/Pictures/Screenshot%20from%202013-10-30%2014:04:50.png
Couldn't load /home/meklu/Pictures/Screenshot%20from%202013-10-30%2014:04:50.png
Expected result:
Dropped file /home/meklu/Pictures/Screenshot from 2013-10-30 14:04:50.png
Loaded /home/meklu/Pictures/Screenshot from 2013-10-30 14:04:50.png successfully
I've attached a patch that fixes the issue by doing URI decoding in-place on
the file string buffer.
Ronie Salgado
The GL Renderer current context tracking fails when one window is used with an SDL renderer but another separate window is used with a user handled OpenGL context.
Attached is a small program that reproduces this bug, at least in some Linux machines where an OpenGL renderer is provided by default.
Expected Output:
-"First window" should be blue.
-"Second window" should be green.
Gotten Output:
- "First window" black.
- "Second window" blue.
What happened:
The renderer created for the "first window" ends rendering into the "second window" OpenGL context.
Bug location:
SDL_render_gl.c - line 286 on hg:
static SDL_GLContext SDL_CurrentContext = NULL;
When making SDL_GL_MakeCurrent from the user perspective, that variable or the GL renderer is not notified about the OpenGL context change.
Solution proposal:
- Move the current GL context cache into another place global.
Brad Smith
Attached is patch from the OpenBSD ports tree to add 24-bit support to the sndio backend and to make use of the sio_open() option SIO_DEVANY.
Diego
The Xcode Instruments Leak tool reports a leak from IODisplayCreateInfoDictionary in Cocoa_GetDisplayName.
This happened after upgrading to Xcode 5.
Frank Praznik
Add a gamepad mapping entry for Bluetooth DualShock 4 controllers on Linux.
The button mapping is the same as the USB controller, but the GUID is
different.
It was simpler to just have the polling (actually: hotplug detection)
functions return immediately if it's not an appropriate time to poll.
Note that previously, if any joystick/controller was opened, we would poll
every time anyhow, skipping this function.
This makes the X error handler used for GL context creation handle *all* errors
and provide the user with specific error messages when SDL_GL_CreateContext
fails.
CR: icculus@icculus.org
Alex Szpakowski
Patch to fix the y component of the position of fullscreen windows in OS X.
In Mac OS X with the latest Mercurial code, when a window is in exclusive-fullscreen the y component of its position is offset by the same amount that is normally taken up by the menubar, resulting in a black bar at the top of the screen.
The recent changes to the internal ConvertNSRect function make it treat the bottom of the menubar as 0 for the y component of window positions, even when the window is fullscreen and 'above' the menubar.
I have attached a patch which fixes the issue by only making the window position relative to the menubar in windowed modes.
Eric Wasylishen
Steps to reproduce:
- Run testwm2 app in the SDLTest Xcode project
- Press Control+R to enable relative mouse mode. The mouse cursor should disappear.
- Press Control+Enter to enter fullscreen.
- Expected: a black screen with no cursor visible. Observed: a black screen, but the mouse cursor is visible in the middle of the screen. It doesn't move when I move the mouse.
Reproduced with latest sdl2 hg (changeset f6010ead184f) on OS X 10.9.2. Can't reproduce the problem on OS X 10.6.8 or 10.7.5.
I'm speculating that this really an Apple bug.. but anyway, the attached workaround seems to fix it for me, and I think it's fairly safe.
A more obvious idea, sticking a call SDL_SetCursor(NULL) at the end of Cocoa_SetWindowFullscreen, didn't work.
Eric Wasylishen
The problem seems to be the spaces handling code in -setFullscreenSpace: (SDL_cocoawindow.m) is incorrectly reporting that the SDL_WINDOW_FULLSCREEN -> windowed transition has already happened.
i.e. I saw this case was getting hit when trying to leave SDL_WINDOW_FULLSCREEN:
"else if (state == isFullscreenSpace) {
return YES; /* already there. */
}"
With the attached patch, both Control+Enter (SDL_WINDOW_FULLSCREEN toggle) and Option+Enter (SDL_WINDOW_FULLSCREEN_DESKTOP toggle) work in an sdl test app (I tried testwm2). Tested on OS X 10.9.2.
snake5creator
When starting application with the usual "double click on file" method on Windows, only holding the last click, an unnecessary MOUSEBUTTONDOWN event is sent before the initial MOUSEMOTION event, and mouse button state is stuck in the sense that it takes a subsequent button release, followed by another press for the system to resume sending events (beginning with the next button release / MOUSEBUTTONUP event).
Input event log with held double-click startup: http://i.imgur.com/nypGKR2.png
Without: http://i.imgur.com/yaIqAvV.png
From Melesie
I noticed that when user switches from fullscreen mode to windowed mode and exits application while in windowed mode, Windows performs an additional change of display settings, even though desktop resolution is the same as current one. This causes short black screen to show up. The only way I know of avoiding this is to explicitly switch to default display settings found in registry. MSDN documentation for ChangeDisplaySettingsEx states:
Passing NULL for the lpDevMode parameter and 0 for the dwFlags parameter is the easiest way to return to the default mode after a dynamic mode change.
This would have been a one-line patch to the documentation (specifying that
captures only work as long as the left mouse button is pressed), but I didn't
like that, so I got a little crazy about this instead.
There were several good arguments for this: it's how Windows works with
WM_NCHITTEST, SDL doesn't need to manage a list of rects, it allows more
control over the regions (how do you use rects to cleanly surround a circular
button?), the callback can be more optimized than a iterating a list of
rects, and you don't have to send an updated list of rects whenever the
window resizes or layout changes.
This should be a "long" which on a 64-bit system is likely to be > 32-bits,
causing XGetICValues() to write past the end of the variable (and stack).
Fixes Bugzilla #2513.
