Thank for all the nice feedback, guys! I'm extremely happy that you like what you see! 🙂
How well will mobile devices handle this?
That depends on the GPU: all the shading is done in the fragment shader, which is thus a little more complex than the standard texture rendering. I just tried it out on the iPhone 4S, and I still get about 800 objects @30 fps. On more modern hardware, such shaders shouldn't pose any problem. On the CPU side, it's not significantly more expensive than standard textured objects.
If the context is lost, will the normal maps need manual restoration?
The standard Starling logic for context losses applies here. In other words, if you loaded the texture with the AssetManager, you don't have to do anything special. It will just work. (Note that the extension does not contain any context-loss logic at all, that's all handled automatically by the base classes.)
Is there any difference in performance v2.0 vs v1.7 ?
There's no easy answer for that, since the architecture has changed quite a bit. In a pure brute-force benchmark, Starling 2.0 will not be faster than the old version, maybe even a little slower, to be perfectly honest. But in a typical game, it will be faster, because static parts of the display tree are rendered much faster.
Remember the old "flatten" method? That's gone in Starling 2.0, because that stuff is done automatically, for the complete display tree. 🙂
We want dynamic shadows!!!
Haha, I knew that something like that would come! 😉
The nice thing about Starling 2.0: people can create such extensions now — and I know a few developers who are already waiting for the new architecture to do just that. 🙂