* Added display parameter to many internal functions so video modes can be set on displays that aren't the public current one.
* The fullscreen mode is associated with fullscreen windows - not displays, so different windows more naturally have a mode associated with them based on their width and height. It's no longer necessary to specify a fullscreen mode, a default one will be picked automatically for fullscreen windows.
D3D renderer shall try mapping YV12 and I420 (IYUV) to D3D texture formats via FOURCC. This will enable HW acceleration for those formats when driver is capable (most of them are). Note that SDL's IYUV maps I420 FOURCC on Woe.
Currently SDL uses GL_RGB for internalFormat when GL_YCBCR_MESA is passed as format for glTextImage2D when using Linux Mesa's OpenGL. However this is wrong and makes glTextImage2D fail with invalid argument error. GL_YCBCR_MESA should be also internalFormat (not GL_RGB) there and this is what can be found googling various source codes using GL_YCBCR_MESA.
This one is quite puzzling. I found a partial workaround, but I don't fully understand the reasons yet.
First, the console is complaining about not finding a nib for MainWindow.
I tried removing the entry for this in the info.plist, and the message went away, but it didn't really change anything.
Second, I stepped through this with the debugger and broke up some lines. It seems that the basic act of calling
view = [SDL_uikitopenglview alloc];
or even
view = [SDL_uikitview alloc]
will crash the program. The debugger messages plus the stack trace make me think it's not finding the SDL_uikitview classes for some reason. But I don't understand why this would be.
view = [UIView alloc] will not crash the program.
For kicks, I added a new definition of a class called SDL_object which subclasses NSObject in the same files as SDL_uikitopenglview and then call
view = [SDL_object alloc];
This does not crash the program.
So, then I modified SDL_object to subclass UIView. No crash.
Next, I made SDL_object subclass UIView<UITextFieldDelegate> . This crashes.
So it is the act of conforming to the UITextFieldDelegate protocol that is crashing things.
I don't understand why it would crash on alloc though. I'm guessing either a delegate needs to be set somewhere or one of the required methods needs to be implemented. But in the former case, I would not expect a crash, but a silent message to nil and something else doesn't work. And in the latter case, I would expect a compiler warning and an exception thrown instead of a crash.
Anyway, my temporary workaround is to change the interface declaration for SDL_uikitview to look like:
#if SDL_IPHONE_KEYBOARD
@interface SDL_uikitview : UIView<UITextFieldDelegate> {
#else
@interface SDL_uikitview : UIView {
#endif
And then disable the keyboard support in the SDL_config_iphoneos.h file.
/* enable iPhone keyboard support */
#define SDL_IPHONE_KEYBOARD 0
-Eric
On Nov 23, 2009, at 1:43 AM, Sam Lantinga wrote:
> I ran into a blocking startup crash with the Happy demo on iPhone OS 3.1.2 on my new iPhone:
>
> #0 0x323fea14 in _class_isInitialized
> #1 0x323fea68 in _class_initialize
> #2 0x32403e92 in prepareForMethodLookup
> #3 0x32401244 in lookUpMethod
> #4 0x323fea10 in _class_lookupMethodAndLoadCache
> #5 0x323fe746 in objc_msgSend_uncached
> #6 0x323feb26 in _class_initialize
> #7 0x323fea58 in _class_initialize
> #8 0x32403e92 in prepareForMethodLookup
> #9 0x32401244 in lookUpMethod
> #10 0x323fea10 in _class_lookupMethodAndLoadCache
> #11 0x323fe746 in objc_msgSend_uncached
> #12 0x000554dc in UIKit_GL_CreateContext at SDL_uikitopengles.m:103
> #13 0x0004f89e in SDL_GL_CreateContext at SDL_video.c:3155
> #14 0x000579e8 in GLES_CreateRenderer at SDL_renderer_gles.c:282
> #15 0x0004d7b8 in SDL_CreateRenderer at SDL_video.c:1509
> #16 0x00002bc2 in SDL_main at happy.c:156
> #17 0x000571b2 in -[SDLUIKitDelegate postFinishLaunch] at
> SDL_uikitappdelegate.m:77
> #18 0x313f9ef2 in __NSFireDelayedPerform
> #19 0x32567bb2 in CFRunLoopRunSpecific
> #20 0x3256735c in CFRunLoopRunInMode
> #21 0x32912cbe in GSEventRunModal
> #22 0x32912d6a in GSEventRun
> #23 0x32b6276e in -[UIApplication _run]
> #24 0x32b61472 in UIApplicationMain
> #25 0x00057088 in main at SDL_uikitappdelegate.m:50
>
> Any ideas?
>
> See ya!
> --
> -Sam Lantinga, Founder and President, Galaxy Gameworks LLC
Mason Wheeler 2009-11-23 06:59:48 PST
There's code in SDL_RecreateWindow specifically to handle SDL_WINDOW_FOREIGN,
but it appears to have been overlooked in the allowed_flags constant. This
causes the line
window->flags = (flags & allowed_flags);
to strip SDL_WINDOW_FOREIGN from the window's flags, which breaks some code in
WIN_WindowProc in SDL_win32Events.c that treats foreign windows differently.
This can be trivially fixed by defining allowed_flags as
const Uint32 allowed_flags = (SDL_WINDOW_FULLSCREEN |
SDL_WINDOW_OPENGL |
SDL_WINDOW_BORDERLESS |
SDL_WINDOW_RESIZABLE |
SDL_WINDOW_INPUT_GRABBED |
SDL_WINDOW_FOREIGN);
I updated SDL, and suddenly my SDL frames stopped working. They'd "initialize" full of gibberish, and I couldn't render anything to them. After a bit of digging, I found a problem: the renderer initialization routine in my SDL frame code wasn't getting called anymore.
procedure TSdlFrame.Paint;
begin
if SDL_SelectRenderer(FWindowID) = -1 then
CreateRenderer;
SDL_RenderPresent;
end;
function TSdlFrame.CreateRenderer: boolean;
const
pf: tagPIXELFORMATDESCRIPTOR = (nSize: sizeof(pf); nVersion: 1;
dwFlags: PFD_DRAW_TO_WINDOW or PFD_SUPPORT_OPENGL or PFD_DOUBLEBUFFER;
iPixelType: PFD_TYPE_RGBA; cColorBits: 24; cAlphaBits: 8;
iLayerType: PFD_MAIN_PLANE);
RENDERERS: array[TRendererType] of AnsiString = ('software', 'gdi', 'opengl', 'd3d');
var
pFormat: integer;
begin
if (SDL_SelectRenderer(FWindowID) = 0) then
begin
result := true;
Exit;
end;
if FRendererType = rtOpenGL then
begin
pFormat := ChoosePixelFormat(canvas.Handle, @pf);
if not SetPixelFormat(canvas.Handle, pFormat, @pf) then
outputDebugString(PChar(SysErrorMessage(GetLastError)));
if wglCreateContext(canvas.Handle) = 0 then
outputDebugString(PChar(SysErrorMessage(GetLastError)));
end;
if (SDL_CreateRenderer(FWindowID, SDL_RendererIndex(RENDERERS[FRendererType]), [sdlrPresentFlip3, sdlrAccelerated]) = 0) then
begin
SDL_ShowWindow(FWindowID);
assert(SDL_SetRenderDrawColor(0, 0, 0, 255) = 0);
FFlags := SDL_GetWindowFlags(FWindowID);
if assigned(FOnAvailable) then
FOnAvailable(self);
end
else outputDebugString(pChar(format('SDL_CreateRenderer failed: %s', [sdl_GetError])));
result := SDL_SelectRenderer(FWindowID) = 0;
end;
This is a critical issue. The Paint method gets called when the control receives a WM_PAINT message from Windows. I can't create the renderer before then, or it will fail and cause trouble. And when I do create it, it needs to be created with certain parameters. So imagine my surprise when I started debugging into the DLL and found that SDL_SelectRenderer was trying to be "helpful" by creating the renderer for me if it didn't already exist! Now not only does my initialization code not get called, I end up with the wrong renderer and so things don't render as expected when I try to use the window.
If we have to, we can use vertical line and horizontal line textures for vertical and horizontal lines, and then create custom textures for diagonal lines and software render those. It's terrible, but at least it would be pixel perfect.
This shader implements the software renderer mask semantics where the source pixel is multiplied by the color and alpha modulation values and then any pixel with non-zero alpha is fully opaque.
The OpenGL renderer on Mac OS X now passes all the automated render tests! :)
Hello Sam!
You have reverted back my patches for OpenGL renderer :)
To reproduce an issue, compare graphics output while running these tests:
testdraw2 --renderer opengl --blend mask --cyclealpha
and
testdraw2 --renderer software --blend mask --cyclealpha
You will see, that software renderer output is different from opengl renderer output.
Thanks!
Roger Willcocks 2009-10-25 08:45:37 PDT
appDidFinishLaunching is triggered before all the setup's complete. The easiest
fix is to arrange for another event to be sent when it's really ready, and run
SDL_main from there. Ref.
http://blog.rightsprite.com/2008/11/iphone-applicationdidfinishlaunching.html
SDL-1.3.0-4563/src/video/uikit/SDL_uikitappdelegate.m:
- (void)postFinishLaunch {
/* run the user's application, passing argc and argv */
int exit_status = SDL_main(forward_argc, forward_argv);
/* free the memory we used to hold copies of argc and argv */
int i;
for (i=0; i<forward_argc; i++) {
free(forward_argv[i]);
}
free(forward_argv);
/* exit, passing the return status from the user's application */
exit(exit_status);
}
- (void)applicationDidFinishLaunching:(UIApplication *)application {
/* Set working directory to resource path */
[[NSFileManager defaultManager] changeCurrentDirectoryPath: [[NSBundle
mainBundle] resourcePath]];
[self performSelector:@selector(postFinishLaunch) withObject:nil
afterDelay:0.0];
I think something is wrong (look at the fix patch below):
(1) NSRect rect shouldn't be initialized with contentRectForFrameRect: at the top of the function, contentRectForFrameRect is called in non-fullscreen case anyway (1st @@)
(2) I think you've left two lines that should be removed completely (2nd @@)
When using deployment in Xcode we use 10.4 SDK for PPC & i386 and 10.6 SDK for x86_64 unfortunately MAC_OS_X_VERSION_10_6 is defined only in 10.6 SDK, in older SDKs it is undefined which makes >= comparison return always TRUTH, so the GCC complains about undefined <NSWindowDelegate>, even if the original intention was to omit <NSWindowDelegate> on older SDKs.
Solution, don't relay on MAC_OS_X_VERSION_10_6 but use OSX revision number 1060 directly as SDL does in many other places.
Sending a patch for fullscreen Mac OS X SDL 1.3 (SVN) Cocoa mouse position handling. In fullscreen mouse coordinates should be relative to SCREEN not to the window, which doesn't really occupy fullscreen.
Without this patch mouse position (especially Y) was totally incorrect (shifted) in fullscreen.
Ludwig Nussel 2009-10-18 05:34:18 PDT
src/joystick/linux/SDL_sysjoystick.c has some problems:
- test_bit() might break with strict aliasing
- test_bit() assumes array is Uint32 but its actually "unsigned long"
on 64bit systems sizeof(long) != sizeof(Uint32).
- the keybit array is too small
- the arrays are unitialized so the number of
detected buttons is quite random
Eric Wing to Sam
I've been sitting on this too long. I need to push.
It's untested because of the unrelated crashing bug I've been experiencing.
Also have a fix for SIZEOF_VOIDP in the config for both iPhone and Mac.
Daniele Forghieri 2009-09-30 15:40:53 PDT
To compile the source in libm the variable huge must be renamed, I choose
huge_val
The patch attached change it so it compiles
Since current DirectFB implementation is incomplete for YUV surfaces (actually causes segmentation faults when trying Lock and use YUV planar textures) I decided to fix it a bit.
Here's a patch that should make DirectFB properly support YUV both packed and planar (3 planes).
(1) Removed SDL_BYTESPERPIXEL at all in favor of DFB_BYTES_PER_PIXEL(SDLToDFBPixelFormat(fmt)) which does return always proper BPP for YUVs too, coz SDL_BYTESPERPIXEL returns incorrect values for FOURCCs
(2) Fixed data->pixels allocation for planar YUVs in CreateTexture, it should allocate 150% more space
(3) Copy other planes for planar YUVs in UpdateTexture
(4) Moved checking if format is supported at all with DirectFB on CreateTexture at the beginning of the code
Waiting for comments,
--
Adam Strzelecki | nanoant.com
Mason Wheeler to sdl
When I tried to render an image using something other than NULL for srcrect, it got horribly distorted. I traced it down to the fact that the math in the rectangle adjustments performed just before the call to renderer->RenderCopy is written inside out. It should look like this:
if (dstrect->w != real_dstrect.w) {
int deltax = (dstrect->x - real_dstrect.x);
int deltaw = (dstrect->w - real_dstrect.w);
real_srcrect.x += (deltax * real_srcrect.w) / dstrect->w;
real_srcrect.w += (deltaw * real_srcrect.w) / dstrect->w;
}
if (dstrect->h != real_dstrect.h) {
int deltay = (dstrect->y - real_dstrect.y);
int deltah = (dstrect->h - real_dstrect.h);
real_srcrect.y += (deltay * real_srcrect.h) / dstrect->h;
real_srcrect.h += (deltah * real_srcrect.h) / dstrect->h;
It looks like newer kernels do the logical device mapping in the driver,
so this code crashes. I don't have one of these nor do I have remote
access to debug this, so I'm disabling the logical mapping for now.
Armin Burgmeier 2009-03-15 04:35:45 PDT
When I hold a mouse button down on another (non-SDL) window, then move the
mouse over an SDL window and releasing the mouse button there, then the
application does not receive an SDL_ACTIVEEVENT with state SDL_APPMOUSEFOCUS
and gain 1.
Furthermore, SDL_GetAppState() reports the application not having mouse focus
until moving the mouse out of the window and back in again.
Ryan C. Gordon 2009-09-19 08:25:21 PDT
This line in SDL_iconv_string (src/stdlib/SDL_iconv.c) ...
if (!fromcode || !*fromcode) {
tocode = "UTF-8";
}
Is probably supposed to assign to "fromcode" and not "tocode".
Mason Wheeler 2009-07-05 09:28:33 PDT
This patch fixes two issues with SDL_CreateTextureFromSurface.
1. If no renderer is available, the function will return 0 without calling
SDL_SetError. (It does this in other places as well, but it appears that in
these cases, SDL_SetError was already called by a previous function call.)
2. Removal of a dead code block that checks for an impossible return value.
Nicholas Phillips 2009-04-26 21:34:05 PDT
I am using x64 Linux (using Intel Core 2 DUO), and I have noticed that there is
an error in SDL_cpuinfo.c, function CPU_getCPUIDFeaturesExt for my platform.
SDL_SemWaitTimeout in src/thread/generic/SDL_syssem.c line 179 (SVN trunk):
--sem->count;
should be
if (retval == 0) {
--sem->count;
}
Without this, sem->count will underflow on timeout effectively breaking the
semaphore. It appears that the implementation has been wrong since the initial
revision.
Hi,
so here we go: GL_LINE_SMOOTH bad for business! ;)
I checked the code within SDL_renderer_gl.c / GL_CreateRenderer(..)
and have found the reason for blurred lines when using OpenGL.
I have attached 2 images here.
I don't know whether this has to be changed and I don't know if there
is an option to turn it off/on...just wanted to say that with GL_LINE_SMOOTH
enabled there is a difference when rendering with opengl compared to directx.
The linux code appears to be complete and *should* be the base of all Unix and GCC based versions. The macosx and win32 versions
are currently just copies of the dummy code. I will begin working on the windows version as soon as this check in is done. I
need someone to work on the Mac OS X version.
I'm afraid that this check in will break QNX (Sorry!)
I noticed in trunk/SDL/src/video/win32/SDL_win32events.c, in this code here...
... if the device handle isn't found in mice[], which it won't be if
the mouse was plugged in after SDL_Init, then you end up with an
uninitialized value in index, which is then passed to various
SDL_SendMouse* functions.
Unfortunately the audio and cdrom systems do not build at all, and if you
disable those, SDL still fails to link with these errors:
Undefined symbols:
"_OBJC_IVAR_$_NSScreen._frame", referenced from:
-[NSScreen(NSScreenAccess) setFrame:] in SDL_cocoamodes.o
"_KLGetKeyboardLayoutProperty", referenced from:
_UpdateKeymap in SDL_cocoakeyboard.o
_UpdateKeymap in SDL_cocoakeyboard.o
"_KLGetCurrentKeyboardLayout", referenced from:
_UpdateKeymap in SDL_cocoakeyboard.o
ld: symbol(s) not found
When i call the "SDL_Quit()" function, it call the func "SDL_QuitSubSystem(SDL_INIT_EVERYTHING)", which call the "SDL_VideoQuit()" function.
It's where i'm lost, this function "SDL_VideoQuit()" will recall the "SDL_Quit()" function (i don't find where it is called), which in turn recall the "SDL_QuitSubSystem(SDL_INIT_EVERYTHING)", which in turn recall "SDL_VideoQuit()", which cause a "double free" seg fault since it has been already called.
I have set the num_displays to 0 instead of 1, so it wont free it twice in the videoquit function loop.
Hello sam, me again.
I'v fixed another bug in the pandora code, it's on the no-X videodriver, i forgot to add some code to destroy the window so it was impossible to reload SDL inside the same application (eglTerminate).
You'll find the svn diff attached.
See you,
David.
I fixed a bug in the "SDL_DestroyRenderer()" function in the pandora x11 renderer (in fact the bug was in the "X11_GLES_MakeCurrent()" function) that was causing a crash when exiting SDL.
There was a problem while terminating the x11 egl window, that was preventing to quit/close SDL/SDL-window then reload SDL/SDL-window inside the same application.
/test/testatomic.c performs absolutely basic tests to show that the function work as expected. Need a second test to do more detailed tests.
/include/SDL_atomic.h provides declarations for all included functions.
/src/atomic/linux/SDL_atomic.c provided all the functions. On a generic built the 64 bit functions work, but they are emulated. On a build for -march=pentium and above the 64 bit functions use native instructions
/src/atomic/dummy/SDL_atomic.c emulates all the operations using SDL_mutex.h.
/src/atomic/win32/SDL_atomic.c is a copy of dummy
/src/atomic/macosx/SDL_atomic.s is a copy of dummy
These versions of SDL_atomic.c provide a frame work for building the library with a mixture of native and emulated functions. This allows the whole library to be provided on all platforms. (I hope.)
I hope this fits with the SDL philosophy of either providing a common subset or emulating when the platform is missing a feature.
I have not added dummy, macosx, or win32 to the build. They are there as place holders for future work.
I have modified congifure.in to compile sources in /src/atomic/linux. (The SDL configure.in file is an amazing piece of work and I hope I didn't mess it up. :-)
It also update configure.in to compile the linux version of the library. The three versions are all dummies
implementations that do nothing. They are being checked in as place holders. Mostly, I just wanted to get
place holders and the configure.in checked in.
Why not to use hardware vertex processing instead of software one if
it is available in D3D render driver? With hardware processing
testsprite2 runs three times faster on all videocards which I could
test.
Since many different event structures include windowID it should be placed near
the beginning of the structure (preferably right after type) so it's position
is the same between different events.
This is to avoid code like this:
if (event.type == SDL_WINDOWEVENT)
win = event.window.windowID;
else if ((SDL_EVENTMASK(event.type) & SDL_KEYEVENTMASK) != 0)
win = event.key.windowID;
else if (event.type == SDL_TEXTINPUT)
win = event.text.windowID;
else if (event.type == SDL_MOUSEMOTION)
win = event.motion.windowID;
else if ((SDL_EVENTMASK(event.type) & (SDL_MOUBUTTONDOWNMASK |
SDL_MOUBUTTONUPMASK)) != 0)
win = event.button.windowID;
else if (event.type == SDL_MOUSEWHEEL)
win = event.wheel.windowID;
...
in favor of:
win = event.window.windowID;
2. SDL_CreateTextureFromSurface() now has best texture format search for non-alpha pixel formats.
3. Added comparision for pixels packed order to the video mode sorting callback to avoid mixing 1555/565/4444 pixel formats.
4. Added sorting call for current video mode list after each new video mode addition, because SDL_GetClosestDisplayMode() requires video modes to be sorted, and this is one place only where we can detect video modes addition.
I have prepared a set of patches to readd WindowsCE support to SDL 1.3.
I've created a new GAPI/Rawframebuffer and a DirectDraw renderer.
Both renderers are work in progress and there are several unimplemented
cases. (Notably
RenderLine/RenderPoint/RenderFill/QueryTexturePixels/UpdateTexture and
texture blending )
Nevertheless I am successfully using these renderers together with the
SDL software renderer. (On most devices the SDL software renderer will
be much faster as there are only badly optimized vendor drivers available)
I send these patches now in this unpolished state because there seems to
be some interest in win ce and someone has to start supporting SDL 1.3
Now on to the patches:
wince_events_window_fixes.patch
fixes some wince incompatibilities and adds fullscreen support via
SHFullScreen. NOTE: This patch shouldn't have any side effects on
Windows, but I have NOT tested it on Windows, so please double-check.
This patch doesn't dependent on the following ones.
wince_renderers_system.patch
This patch does all necessary modifications to the SDL system.
- it adds the renderers to the configure system
- it adds the renderers to win32video
SDL_ceddrawrender.c
SDL_ceddrawrender.h
SDL_gapirender_c.h
SDL_gapirender.c
SDL_gapirender.h
these files add the new render drivers and should be placed in
src/video/win32
Some notes to people who want to test this:
- I have only compiled sdl with ming32ce, so the VisualC files are not
up to date
- As mingw32ce has no ddraw.h this file must be taken from the MS SDK
and modified to work with gcc
- I had to modify line 2611 in configure.in to
EXTRA_LDFLAGS="$EXTRA_LDFLAGS -lcoredll -lcommctrl -lmmtimer
-Wl,--image-base -Wl,0x10000"
otherwise GetCPinfo wouldn't link. If someone knows whats causing this
I'd be happy to hear about it.
It would be great if these patches could make their way into SVN as this
would make collaboration much much easier.
I'm out of office for the next week and therefore will be unavailable
via email.
Regards
Stefan
Betreff: [SDL] [PATCH] Make static variables const
Datum: Tue, 19 May 2009 19:45:37 +0200
Hi,
this is a set of simple changes which make some of SDL's internal static
arrays constant. The purpose is to shrink the number of write-able
static bytes and thus increase the number of memory pages shared between
SDL applications.
The patch set is against trunk@4513. Each of the attached patch files is
specific to a sub-system. The set is completed by a second mail, because
of the list's 40 KiB limit.
The files readelf-r4513.txt and readelf-const-patch.txt where made by
calling 'readelf -S libSDL.so'. They show the difference in ELF sections
without and with the patch. Some numbers measured on my x86-64:
Before
[13] .rodata PROGBITS 00000000000eaaa0 000eaaa0
0000000000008170 0000000000000000 A 0 0 32
[19] .data.rel.ro PROGBITS 00000000003045e0 001045e0
00000000000023d0 0000000000000000 WA 0 0 32
[23] .data PROGBITS 00000000003076e0 001076e0
0000000000004988 0000000000000000 WA 0 0 32
After
[13] .rodata PROGBITS 00000000000eaaa0 000eaaa0
0000000000009a50 0000000000000000 A 0 0 32
[19] .data.rel.ro PROGBITS 0000000000306040 00106040
0000000000002608 0000000000000000 WA 0 0 32
[23] .data PROGBITS 0000000000309360 00109360
0000000000002e88 0000000000000000 WA 0 0 32
The size of the write-able data section decreased considerably. Some
entries became const-after-relocation, while most of its content went
straight into the read-only data section.
Best regards, Thomas
