Fixes Bugzilla #2798.
Fixes Bugzilla #2212.
Fixes Bugzilla #2826.
Fixes Bugzilla #2661.
Fixes Bugzilla #1885.
Fixes Bugzilla #1578.
Fixes Bugzilla #2751.
(whew!)
Notable changes, from Alex's notes:
- The SDL_WINDOW_ALLOW_HIGHDPI flag is now needed (along with SDL_GL_GetDrawableSize or SDL_GetRendererOutputSize) to use Retina / high DPI resolutions, bringing SDL?s Retina-related behavior on iOS in line with Mac OS X. Window dimensions and display modes are now in the ?points? (non-high DPI) coordinate system rather than pixels, whereas SDL_GL_GetDrawableSize is in pixels.
- Reworked the custom extended launch screen code:
- It now hides after the first SDL_PumpEvents call rather than SDL_CreateWindow, and it fades out in a similar manner to the system launch screen behavior.
- It now mirrors the system launch screen behavior when deciding which image to display: it falls back to using the Launch Images dictionary in Info.plist if the iOS 8+ launch screen nib isn?t available, and if the Launch Images dictionary doesn?t exist it uses the old standard launch image names.
- The extended launch screen can now be disabled via the SDL_IPHONE_LAUNCHSCREEN define in SDL_config_iphoneos.h.
- Added support for SDL_HINT_ACCELEROMETER_AS_JOYSTICK.
- Added access to a window view's renderbuffer and framebuffer to syswm.
- Added OpenGL ES debug labels for the Renderbuffer and Framebuffer Objects created with SDL_GL_CreateContext.
- Added support for sRGB OpenGL ES contexts on iOS 7+.
- Updated OpenGL ES contexts to support native-resolution rendering (when SDL_WINDOW_ALLOW_HIGHDPI is enabled) on the iPhone 6 Plus, i.e. 1080x1920 rather than 1242x2208.
- Updated SDL_GL_CreateContext, SDL_GL_SwapWindow, SDL_GL_MakeCurrent, and SDL_GL_DeleteContext to be more robust.
- Updated SDL windows to display a UIView at all times, even when an OpenGL context is not active. This allows rotation, touch events, and other windowing-related events to work properly without an active OpenGL context. It also makes it easier to use SDL_GetWindowWMInfo after creating a SDL window.
- Updated the iOS-specific Objective-C code to use cleaner and more modern language features and APIs, including ARC instead of manual reference counting.
- Updated SDL_HINT_ORIENTATIONS to allow disabling custom orientations if the hint is set with no valid orientation names.
- Fixed several rotation and orientation bugs with windows and display modes, especially in iOS 8+.
- Fixed SDL_SetWindowFullscreen failing to update the status bar visibility on iOS 7+.
- Fixed the orientation of the offset applied to the window?s view when the onscreen keyboard is shown in iOS 8+.
- Fixed SDL_IsScreenKeyboardShown (patch by Phil Hassey.)
- Fixed several major memory leaks caused by missing autorelease pool blocks in the iOS-specific Objective-C code.
- Removed several dead code paths.
- The iOS 7 SDK (Xcode 5) or newer is now required to build SDL for iOS.
This avoids a hint lookup for each mouse event we get by setting a static Java
variable from native code during our hint watcher callback.
Also attempts to do the right thing with mouse buttons if you happen to be
on an API14 (Ice Cream Sandwich, Android 4.0) or later device. We still
target API12 (Honeycomb MR1, Android 3.1) for SDL 2.0.4 though.
This isn't tested, so I'm pushing to see what the Android buildbot says. Stand
back, I'm a professional!
Clarify that grabbing the mouse only works with one window at a time; this was
always true at the system level, though SDL could previously get confused
by multiple simultaneous grabs, so now we explicitly break any existing
grab before starting a new one and document it as such.
Also track the window that is currently grabbed, and provide an API to query
for that window. This makes it easy to automate mouse ungrabbing at
breakpoints with gdb7's scripting, since the scripts can now know which window
to ungrab.
In 2.1, we should probably change this API to SDL_GrabInput(win) and
SDL_UngrabInput(void), or something.
We continued looping while maxlen > 0, but maxlen was unsigned, so an overflow
would make it a large number instead of negative. Fixed.
Some snprintf() implementations might return a negative value if there isn't
enough space, and we now check for that.
Don't overrun the SDL error message buffer, if snprintf() returned the number
of chars it wanted to write instead of the number it did.
snprintf is a portability mess, we should just never use the C runtime for it.
Fixes Bugzilla #2049.
Ryan C. Gordon
Philipp Wiesemann
Alex Szpakowski
Sam Lantinga
Edward Rudd
Emmanuel Gil Peyrot
Joseba Garc?a Etxebarria