Removed internal variables _projectionMatrixArray and _viewMatrixArray from WebGLRenderer and render plugins.

These are not needed anymore as matrix elements are directly usable for uniforms.

Unfortunately we can't keep these settings always on: without "--jscomp_off=checkTypes" the amount of spam is just too crazy. It complains about tons of normal things, I had to manually dig out few useful warnings among 900 irrelevant ones.

Allow non-indexed BufferGeometries to render the entire buffer in a single draw call rather than a draw call per 2^16-1 vetexes

See #2638

Changed WebGLRenderer's sortFacesByMaterial to only split geometry if MeshFaceMaterial is used at the first render.

Fixes #2638

Before attributes were enabled just once, when creating a new program and then they were never disabled (with exception of UVs).

Now before rendering of the object existing enabled attributes are disabled and attributes required by program are enabled (these GL state changes are cached so they happen only when active geometry or material changes inside a frame).

There is still some gray zone for more exotic rendering methods (immediate render objects and render plugins do not disable outstanding enabled attributes, they just enable attributes they need) but we'll address these if they start to make problems.

Performance seems to be ok, at least for current examples, though there may be some use cases which could get dramatically slower with these changes (before optimizations naive attribute disabling / enabling every time halved the framerate in performance stress test).

Fixes #2633 (hope so)

Also, removed GeometryLoader. I'll try doing the JSONLoader to GeometryLoader change again in the future.

Patch to #2623

See discussion at https://github.com/alteredq/three.js/commit/5cb715520806a5afa1057a31a6777c2e454eeff0

Because that's what it is.

Now it should look the same no matter where the object is located.

Not yet 100% sure about all the details but this is something that should be useful for making helpers prettier.

It's now up to the application layer to deal with system specific shader limits.

Fixes #2477

Fixes #2581

Extra check doesn't seem to have much effect on performance. This may become a bigger concern if other bottlenecks would go away.

Fixes #2568

Not sure if helpers should be fogged or not. It does look pretty weird when they are not fogged. Also having helpers fogged helps to get a sense of the fog effect over larger space.

Maybe there could be a checkbox for enabling this, to be able to see things if they need to be modified while in the fog.

Also found and fixed a bug deep in WebGLRenderer, where shaders were not recompiled when changing fog types because fog type wasn't taken into account when creating shader code hash. Phew, that was a tough one to figure out. This Editor thingie really forces us to dust all nooks and cranies of the codebase ;)

See #1985

Not tested yet, but it should work in the same way like vertex colors in Ribbons: add one Vector3 with normal per vertex to "geometry.normals" array.

It's then up to the application layer to fill in proper values. It's use case specific how normals for triangle strips would get computed - for simple ribbon meshes it could be very similar like smooth normal computation for regular Mesh objects.

If degenerate triangles are used for pseudo-connected single ribbon mesh planes, application would need to take into account vertex adjacency.

Experimenting with "geometry.buffersNeedUpdate".

Removing unused face lookups in chunk buffering loops.

See discussion #2412

For the moment tried to keep it as dumb as possible: one string for define label, another string for define value.

I'm not yet fully sure of all the details but that's something I wanted to do for a long time, to be able to have user-configurable way for baked-in parameters (for use cases like the convolution filter).

See #2412

Fortunately here we don't need to flip the images.

Fixing reference to geometries counter.

Use two component normalScale and set eventual flipping per-model. It's quite ugly but it could do the job if we don't find a better solution.

Do not merge yet.

See #2358

Thanks to @toji for blog post and DDS utilities.

Fixes #2348

TODO: example textures for alpha formats (DXT3, DXT5).

See #2349

Particles are rendered using drawArrays, so there is no need for indices or offsets.

Particle sorting not implemented for BufferGeometry, will need to check if this is possible to do efficiently.

Normal maps store the perturbation of the surface normal (vector field), as opposed to bump maps, which are height fields (scalar field).

Work-in-progress, don't merge yet.

All examples should work now.

The only weird thing seems to be that for some reason bone texture doesn't like to be assigned to arbitrary texture unit (or more precisely to alternating texture units).

See #2340

As discussed in #2340

Do not merge yet, this is not finished.

Removed skinVerticesA/B arrays.

Specular map works like it used to for normal map shader. It is assumed to be a grayscale texture.

For Phong specular map value modulates specular term strength and environment map reflection strength.

For Lambert and Basic materials specular map affects just environment map reflection strength (as these have no specular term).

Still not very sure how environment mapping and specular term should interact with each other.