| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
| |
|
|
|
|
| |
AWTTree deadlock
|
|
|
|
|
|
|
|
| |
Throw ClassNotFoundException in Display/Screen/Window factory if neither custom
nor default class is available.
Suppress Warning of non existing custom class (in non DEBUG mode),
rely on later ClassNotFoundException (see above).
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1. Drawable, 2. GLContext
Remove deadlock situation where thread-1 (Animator Thread) holds the GLContext-Lock
and acquires the Surface-Lock, while thread-2 (UI/Main/EDT) holds the Surface-Lock
and attempts to create the GLContext and hence acquires the GLContext-Lock.
A GLContext-Lock and hence makeing the GLContext current requires to hold
the Surface-Lock. The prev. code acquired the locks in reverse order and
allowed the deadlock as described above.
This fix acquires the locks in the proper natural order
1 - Surface-Lock
2 - GLContext-Lock
This fix also renders the use of the non-synchronized behavior invalid,
since it is bogus not to wait for the GLContext lock where it waits for the
Surface lock. It also seems nonsense not to wait for any of both locks
and our code always waited for both (synchronized := true).
The GLContext [set/is]Synchronized(..) methods are removed
and waiting for the lock per default is the correct behavior.
|
|
|
|
| |
one holding the surfaceLock acquiring ctxLock where ctx acquires the surfaceLock
|
|
|
|
| |
cbc77718f01a8190e1a8aa0e9afdc2a3a3403358)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
ProjectFloat Matrix Multiplication of gluUnProject(..) impl.
ProjectFloat's previous gluMultMatricesf(..) used row-major order,
but the replacement multMatrixf(..) uses column-major order (like OpenGL, ..).
Note: The replaced 'gluMultMatrixVecf' by multMatrixVecf() already
used column-major order.
Fix: Reverse the arguments of matrix multiplication
m1 x m2 -> m2 x m1
Added proper API documentation in FloatUtil -> Column Major Order of Linear Matrix Layout
|
|
|
|
| |
commit a058e0a2f465a9bff3e32727edb55592f55c7b38)
|
| |
|
|
|
|
| |
ClassLoader to find JNI native libraries
|
| |
|
| |
|
|
|
|
| |
and classname
|
|
|
|
| |
window order on main-thread.
|
|
|
|
| |
recreation (detach window) on OSX (FF, Safari)
|
|
|
|
| |
instead of static final int values.
|
|
|
|
| |
ReparentOperation' in test.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Window.ReparentAction -> enum Window.ReparentOperation
Revert static/locked action instances due to possible deadlocks
- reverts commit: be59d561fd6ab8aa659e85cd962d38fffd1acb0a (partially)
- reverts commit: 5742b1faa210401470032ef129e56a83c47fd046
Even thought the idea of having no temp. objects is nice to have,
using a static instance requires locking which introduces a deadlock
in case the action is being issued from diff. threads.
class Window.ReparentAction -> enum Window.ReparentOperation (Minor API Change)
Clarifies reparent operations using enums
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- NEWT/WindowImpl:
- 'void windowDestroyNotify()' -> 'boolean windowDestroyNotify(boolean force)', allowing to signal a forced close,
as well as replying whether the window has been closed. (called by native code)
- destroy(): set states before releasing the window lock
- NEWT/X11: Pass windowDeleteAtom for reconfigure window, in case of reparenting child to top-level
- NEWT/OSX:
- Add 'BOOL windowShouldClose()' impl., ie. having a chance to reject the close attempt
- Common impl. for 'windowShouldClose' and 'windowWillClose' -> 'windowClosingImpl'
utilizing new 'windowDestroyNotify' code (see above).
Fixes bug 560.
- NEWT/JOGLNewtApplet1Run: Refine out-of browser window behavior for window-close button
- default: move NEWT window back to browser parent
- closeable: close NEWT window
- jogl-test-applets: Add NApplet-Closeable test (Applet out-of browser window is closable)
|
|
|
|
| |
determines whether it shall be executed.
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
disable support per default.
We have to disable support for ANGLE, the D3D ES2 emulation on Windows provided w/ Firefox and Chrome.
When run in the mentioned browsers, the eglInitialize(..) implementation crashes.
This behavior can be overridden by explicitly enabling ANGLE on Windows by setting the property
'jogl.enable.ANGLE'.
EGLDrawableFactory:
- destroy(): clear references and unregister factory, maybe triggered by GLProfile (ANGLE case)
- getAvailableCapabilitiesImpl(): return empty list in case EGL/ES is n/a (ANGLE case)
|
|
|
|
| |
allowing to remove EGL/ES based factory
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
window.
Since we cannot guarantee to have the pressed button information when receiving
the mouse move event, we synthesize the dragged event (move while mouse button pressed).
To simplify the situation and have a compromise,
we clear the mouse pressed states when mouse
enters or leaves the window to remove the dragged events at re-entering.
This seems more sensible, since dragging after re-entering the mouse shall not be expected.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use window-static instances for runnable actions
NEWT General/X11: Decouple setFullscreen() and setAlwaysOnTop()
- X11 fullscreen/above: Don't assume 'always-on-top' if switch to fullscreen.
- WindowImpl: Remove relation between 'always-on-top' and fullscreen
when quering and switching.
Use window-static instances for runnable actions
- Removes temp objects for EDT runnables
- Uses synchronization on action instance to avoid concurrency
|
| |
|
|
|
|
| |
native mask.
|
|
|
|
| |
button-mask func.
|
|
|
|
| |
debug mode).
|
| |
|
|
|
|
| |
.. also reverse AWT rotation sign, since NEWT uses the reverse oriantation.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Based on Andres Colubri's initiative and commit 218d67fc0222d7709b21c45792d44501351939c4.
- Reading real screen refresh rate ('stolen' from NEWT)
- Properly handling swap-interval and vsync-to in native code
- Increasing accuracy vsync-to to microseconds
Tested manually w/ TestGearsES2AWT.
|
| |
|
| |
|
|
|
|
| |
and custom lib path; MovieCube: '-wait' to wait for debugger.
|
|
|
|
|
|
|
|
|
|
|
| |
upcoming version (master branch)
We need native structure access methods to deal with API changes
in the libav headers, which break binary compatibility!
Currently we are binary compatible w/ [0.6 ?, ] 0.7 and 0.8 but not w/ trunk.
Tested: Linux, Windows and OSX (self compiled binary)
|
|
|
|
|
|
|
|
| |
case it's already in destruction (destroyNotifySend via windowWillClose())
This fixes the double release crash of the NSWindow, at the end of an application.
Tested on OSX 10.6.8 and 10.7.3.
|
|
|
|
| |
finished, mimics proper Java behavior.
|
|
|
|
|
| |
Even thought we didn't set the thread to be a daemon,
if the [parent] thread instantiating the Animator the attribute would be passed along.
|
|
|
|
| |
one of a set must be available.
|
| |
|
|
|
|
| |
implementations.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The Java classes already slipped through in
commit 10935e1ec0d8ed677bc3fddfaa8cd73898a3bcbf - oops.
Since we cannot provide a Libav binary (even though Google does in Android and Chrome)
due to legal uncertainities .. we dynamically link to an existing Libav / FFmpeg library
in a 'relaxed' manner. Ie. we allow certain recent functions to be absent to be able
to run on a wider range of Libav versions.
Currently tested on Debian Linux and Windows7 64bit/32bit
Binaries for Win/OSX:
- Windows http://ffmpeg.zeranoe.com/builds/
- OSX http://www.ffmpegx.com/
Features:
- Dynamic relaxed linking to Libav (see above)
- YUV420P texture lookup function shader stub (conversion to RGB)
- 1-copy only (decoder buffer to texture)
- simple
- uses libavformat's network streaming
- fixes some odd PTS values
TODO:
- Audio output (Should use OpenAL, duh)
- Seek works poorly
- Offthread multi-texture fetching for smoother animation
- Maybe more pixelformat conversions
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
General enhancments.
For details about TextureSequence/GLMediaPlayer shader collaboration w/ your own shader source,
see TextureSequence and TexCubeES2 / MovieSimple demo.
TextureSequence allows implementations to provide their own texture lookup function
which may provide color space conversion (YUV) .. or other runtime hw-accel features.
Have a look at the next commit, which provides an Libav/FFMpeg implementation w/ YUV/RGB shader conversion.
MovieCube adds keyboard control (Android: firm touch on display to launch keyboard, don't break it though :)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
translation. Minor fixes for mouse as well.
- AndroidNewtEventFactory
- Android -> NEWT KeyCodes, leave out control keys like HOME, BACK, .. TBD ..
- We don't consume the key event, in case we cannot map it,
so the Android OS has a chance to handle the control keys.
- Key-Release generates a Key Typed event.
- Key/Mouse translate Android time-base (upTime) to Unix.
- AndroidWindow
- set our onKeyListener, using above tranlator
- setFocusable(true) and setFocusableInTouchMode(true) on our SurfaceView,
otherwise no key event will be delivered.
Note: 'requestFocus()' must also be called on the view for key input!
- Test: NEWTGearsES2Activity, which launches the soft key via:
InputMethodManager mgr = (InputMethodManager) win.getAndroidView().getContext().getSystemService(Context.INPUT_METHOD_SERVICE);
mgr.toggleSoftInput(InputMethodManager.SHOW_FORCED, 0); // shows keyboard ..
and requests the focus of the view for input.
|
|
|
|
|
|
|
| |
notify method being called.
This happens on android, where we just react on an already happened screen rotate.
W/o animation suspend GL rendering could be disturbed or the thread could starve.
|