Move GLTFLoader, OrbitControls, and BufferGeometryUtils to core

Is your feature request related to a problem? Please describe.

The GLTFLoader, OrbitControls, and BufferGeometryUtils are the three most imported things from the examples/jsm (and the postprocessing-related things are the fourth, I think). I also think that they are stable enough to be moved to the core.

Describe the solution you’d like

I suggest they would be moved to the core.

Describe alternatives you’ve considered

They can remain in the examples and then users will still import them from examples/jsm.

[donmccurdy]

… But most users are using the three.js’ built-in one - and for them it would be just more understandable to import it from the core and not the examples.

I’m not sure I agree with this – isn’t it also confusing if some official loaders and controls come from one place, others from another? Why should switching from OrbitControls to MapControls change how you import the module?

It is a real source of confusion that we provide both examples/js and examples/jsm, and CDNs add another layer of complexity. I think Spotify’s concept of a golden path (“the opinionated and supported path to build something”) is a good thing, and we don’t really have that for our examples today. To me that is the underlying problem that leads people to want things moved into core, but I’m not sure it’s the right solution.

Keeping the core build small also makes it easier to avoid gradually slipping into complicated interdependencies. For example, GLTFLoader’s spec/gloss materials are not (and shouldn’t be) core three.js materials.

[donmccurdy]

The threshold for “three” vs. “three/examples/jsm/…” should probably not be stability or popularity, but:

1. Is it a building block for the three.js library, or key libraries in the ecosystem?
2. Is it valuable for three.js to offer exactly one obvious way to solve the problem?

Personally I don’t think these modules satisfy those two goals. They’re popular, but not foundational to other things. There are many ways to implement or customize controls, of which OrbitControls is the most popular but not the “best” in any objective way. There are at least two other glTF loaders for three.js, with their own pros/cons.

I could see components related to post-processing or node materials needing to be pulled into the core build at some point, for these reasons. Stability is necessary but not sufficient, and popularity is orthogonal.