SurfaceViewWidget is now always a child of a GtkGraphicsOffload and
handles the GdkTexture on its own. This way the extra GtkPicture is no
longer needed and also the frame_callback of the Wayland server can be
implemented.
This replaces our custom subsurface implementation and has the
advantages, that the subsurface gets repositioned automatically and it
also supports underlay surfaces, so that the application can put control
widgets on top of the SurfaceView.
Our custom subsurface implementation will still be used for GL rendering
Now that we have a reasonably well working SurfaceView implementation,
it is actually cleaner to just implement GLSurfaceView the way AOSP
does. In fact, their code doesn't have any weird dependencies, and
can mostly be used as-is.
The AOSP code is pure Java, which means we had to implement some
EGL wrappers.
This change fixes issues with Wayland (it only ever worked because
the pbuffers were allocated using an XWayland EGLDisplay), and
with resizing (which we simply didn't support), all while getting
rid of quite some (arguably not very readable) LoC.
GtkWidgets extend GInitiallyUnowned and are automatically freed when
removing from parent widget. We need to add an extra reference, to make
sure the object keeps alive as long as the java widget has a reference
to it
Makes it easier to overwrite behavior in subclasses. Have a fallback
implementation for ViewGroup.
Save some _GET_LONG_FIELD / _SET_LONG_FIELD calls by directly passing
the native pointers to and from native methods.
NOTE: the main addition in this commit is WIP support for apps which
render on an EGL surface obtained using ANativeWindow_fromSurface
currently, this EGL surface is obtained by creating a 700x700 pixel
window with GLFW (the 700x700 size is hardcoded in several places)
and only Wayland is supported
ideally, we'd want to use a wayland subsurface to position the EGL
surface above the Surface widget it's associated with (and do
whatever for X11)
