We had a few methods that needlessly traversed whole TCastleTransform hierarchy. Reworking them was an optimization and also an API simplification (IOW, a “no-brainer” 🙂 ). Changes:

1. TCastleTransform.Press, TCastleTransform.Release are now only called when TCastleTransform.ListenPressRelease is true.

   The need to process keys in TCastleTransform is very seldom (we only had 2 use-cases for it in the engine, one is in soon-to-be-deprecated CastlePlayer, the other was for X3D key sensors which are not very useful for normal engine usage).

2. TCastleTransform.CameraChanged, TCastleTransform.UpdateGeneratedTextures, TCastleTransform.VisibleChangeHere virtual methods removed (their use-case is now implemented as an internal mechanism, tailored to factual needs, not needing to be recursive).

3. TCastleTransform.Dragging removed (not needed, a simple Boolean internal field that tracks “whether something is between PointingDevicePress and PointingDeviceRelease” is enough).

CastleWindow GTK backend (default on Linux) can now initialize OpenGLES context using EGL. (Note that it was already possible with Xlib backend previously.)

This allows to test OpenGLES rendering on desktops, just define symbol OpenGLES at compilation (e.g. by adding it to CastleEngineManifest.xml, or just tweak one line of src/base/castleconf.inc). And make sure you have OpenGLES installed (libgles2 package on Debian and similar).

Updated documentation about CastleWindow backends.

Optimized bounding box calculation for meshes (boxes are now a little less optimal, but much faster calculated; undefine FAST_MESH_BOUNDING_BOXES if you want to try again the old method).

Optimized GeneratedCubeMap with update=NONE (should take zero time, was accidentally consuming some).

Fixed LOD node behavior under transformation animation (testcase: navigation/lod_test.x3dv in demo-models).