makeKeyAndOrderFront: was sending three KVO transitions for isVisible,
for false -> true, true -> false, and then false -> true. This was
causing an infinite recursion.
We now suspend monitoring of the KVO before makeKeyAndOrderFront, then
resume afterwards and send any changes in isVisible's state.
This fixes an issue that would arise when you minimize / order out an
OpenGL window on Mac, where the window would lose it's window device.
Without a window device, you cannot render to the window.
It does so by making two changes:
- Windows are no longer "oneShot" (which caused their window device to
get destroyed when they're minified or ordered out)
- Windows are no longer "deferred" (which caused the OS to defer
window device creation until the window is shown, which meant that
we couldn't properly makeCurrent to it)
Thanks to http://www.mikeash.com/pyblog/nsopenglcontext-and-one-shot.html
This changes the old behavior of having values in the -32768 - 32767
range, like regular joystick axis. Now "button as axis" triggers (like
on Logitech controllers) and regular axis triggers (like on Xbox
controllers) have the same resting value, 0.
This has the benefit of ending the otherwise-bogus complaints that
SDL_GetError() reports "Passed a NULL mutex" if you call it instead of
checking if SDL_CreateWindow() actually succeeded. :)
Nitz
In Function,
MonitorInfo *
decode_edid (const uchar *edid)
In this function "info" is going out of scope and leaks the storage it points to, if the first if condition get true:
if (!decode_header (edid))
return NULL;
So while returning from this if statement there should be free for "info" pointer.
Nitz
In SDL_x11mouse.c file there is function named
static Cursor
X11_CreatePixmapCursor(SDL_Surface * surface, int hot_x, int hot_y)
{
// Some code
data_bits = SDL_calloc(1, surface->h * width_bytes);
mask_bits = SDL_calloc(1, surface->h * width_bytes);
if (!data_bits || !mask_bits) {
SDL_OutOfMemory();
return None;
}
// Some code
}
Here is the problem in if statement,
suppose if !data_bits is false and !mask_bits is true then,
data_bits will go out of scope and leaks the memory it points to.
Solution is that data_bits and mask_bits should be checked separately, not by using OR operator.
Philipp Wiesemann
There is a NULL pointer dereference in SDL_AllocRW() if the system is out of memory. The "type" field is always written. This may be fixed with an early return.
Or an else{} or not writing the field and using slower SDL_calloc().
This fault was recently introduced (http://hg.libsdl.org/SDL/rev/681820ca0e78).
Philipp Wiesemann
SDL_RWFromFile() sets an error to be queried with SDL_GetError() on Android although a valid SDL_RWops pointer is returned.
This happens if the fallback implemented in SDL_android.cpp is used to load compressed assets (see README.android in section "Loading assets") and results in a message like "java.io.FileNotFoundException: This file can not be opened as a file descriptor; it is probably compressed". I think this is confusing and not needed because the loading works as expected.
I attached a patch which changes SDL_android.cpp to not set an error if compressed assets are loaded. In this case also no Exception is queried and no additional string are created.
Philipp Wiesemann
SDL's Android project template has old default icons from Android while iOS project template has custom icons with SDL's logo.
There is a Wizard in the Android Developer Tools to create "Android Icon Sets". As an example I created icons from the iOS loading screen and attached them in a ZIP archive. They are named "ic_launcher.png" instead of "icon.png" because that is the new name used in Android projects. To use them the AndroidManifest.xml needs to be changed to have "@drawable/ic_launcher" instead of "@drawable/icon". I do not know why there was no icon created for ldpi. Maybe it is deprecated.
Philipp Wiesemann
SDL sets an error to be queried with SDL_GetError() for the initial touch on Android.
Android_OnTouch() in SDL_androidtouch.c uses SDL_GetTouch() to check if a touch device was already added. SDL_GetTouch() sets the error "Unknown touch device" for an out of range access because touch devices are added after initial touch. I think this error is confusing because it always happens by design.
I attached a patch which removes the call to SDL_GetTouch() and only uses SDL_AddTouch() which does the check (if already added) again and does not set an error (if not added yet).
q66
The SDL_opengl.h header contains this:
#ifdef __FreeBSD__ /* !!! FIXME: temp compiler warning fix... */
#define NO_SDL_GLEXT 1
#endif
However, I can't seem to find what kind of compiler warning it was and it makes it unusable to use on FreeBSD. If I comment out these lines on my machine, everything works fine - I use FreeBSD 9-STABLE (x86_64, gcc and clang both, the same in a x86 chroot). All I could find is that this was causing an error on FreeBSD 8, but I can't test that on my machine (maybe if I set up some FreeBSD 8 chroot).
I set up a 8.2 chroot and investigated the problem. Apparently this issue was fixed in Mesa 7.6 (and in Git, June 4 2009, but it didn't get into 7.5). By the time those lines were added, FreeBSD contained the libGL port version 7.4.4, which suffered from the issue, but on April 2012 the version was updated to 7.6, which is available for FreeBSD 8 and FreeBSD 9 alike, which means those three lines should be safe to remove (it'll work fine for everyone with sufficiently up to date ports).
