Surprisingly the effect on performance is minimal, despite doing quite much less operations. Yet again raw math performance doesn't seem to matter.

Deferred shading: replaced special case handling of emitters with emissive color in G-buffer and full-screen emissive pass. Added handling of directional lights.

Refactored EffectComposer, added setSize method. Changed examples with postprocessing to use setSize.

Also now sharing numericalSort between particles and morphs.

Fixes #2710

Still need to deal with internal emitter / light proxy scenes. These should be per-scene not common.

Also while doing this I managed to catch another bug (in OrthographicCamera), short variable names are indeed dangerous.

See #2700

See #2700

The last loaded geometry mapping is used by mistake
causing undefined behaviour when using loaderParams.

This got broken after GeometryLoader got removed, using instead JSONLoader.

Another step towards WebGLDeferredRenderer. It's still quite a mess but at least now code lives outside of the examples (and is reusable, to a degree).

These are first steps toward WebGLDeferredRenderer. This will take a while.

Not needed now that these are in G-Buffers.

This allows to have different per-object Phong materials in deferred shading.

To use FBX converter you must now install FBX SDK Python bindings yourself from here:

http://usa.autodesk.com/adsk/servlet/pc/item?siteID=123112&id=10775847

See #2686

Using vec3 <=> float conversions from @benaadams

For the moment just testing how it works (had to fix blending in WebGLRenderer for this). Seems with two RGBA float passes we could eventually get quite decent material range.

See discussion in https://github.com/mrdoob/three.js/commit/d951fcbcfa8efa8b7cecc10075e5e78be4027bac

This happens when you use MeshFaceMaterial for a single material model (that's what was going on in Lucy example).

Alternatively, we could set face.materialIndex by default to 0 instead of "undefined". This kinda makes sense, just would need to be checked if it wouldn't break things.

This was causing a major headscratching in deferred rendering where I just couldn't get rid of data alpha seeping into color channels.

Watch out for unexpected changes. I believe NoBlending is correct for opaque pass, but there could be some demos that depend on incorrect behavior with NormalBlending (getting transparency without setting "material.transparent" flag).

Most likely troubles could come from using transparent PNGs without setting "material.transparent".