meshes vs instanceSkin

[RemiArnaud]

This is related to #100, I thought it would be better to create an new issue to discuss a specific design matter rather than adding to #100.

In node, we can have meshes and instanceSkin

Design wise, I think we can improve this bringing better uniformity. Basically, I think we either should have instanceMesh and instanceSkin, or we can have meshes and skins in node.

Basically an instanceXX - as in instanceProgram - enable referencing a shared/read-only object and augmenting or specializing it with additional parameters. so far we have made node reference specific objects, and not use instances. This may create a bit of duplication, but its a simpler design, and so far the duplications have been minor, and it IMHO it fits better a run-time format.

So, could we change a bit the skin design, so that node can reference a specific skin, rather than using a instanceSkin?

Other questions: can we have both skins and meshes in a node?
Also noticed: a node can have many meshes (meshes: []) but only one instanceSkin ?

[RemiArnaud]

The proposal would be to move skeletons and sources in skin.
(which means that some skins may have to be duplicated in some cases, but the size is quite small)
One advantage is that it make it easier to parse gltf json, as all the nodes can be parsed first, and then the skins - so the skins references to skeletons and jointID can be resolved before parsing skins.
Otherwise, as it is now, the instanceSkin does forward references into the nodes.

[pjcozzi]

@RemiArnaud can you provide a concrete before/after JSON example? @fabrobinet will be the best person to provide feedback, but this is the most confusing part of glTF to me (perhaps necessarily so) so I am interested in anything that will simplify it.

Other questions: can we have both skins and meshes in a node?

From the schema:

"A node will have either the camera, light, meshes, or instanceSkin property defined"

[pjcozzi]

@fabrobinet for 1.0?

[fabrobinet]

I am not comfortable agreeing or punting this without a concrete schema proposal, @RemiArnaud can you write it down so that there is no ambiguity ?

[pjcozzi]

If we want to consider this for 1.0, please provide an concrete JSON schema proposal.

[pjcozzi]

Seeing no proposals, closing.

[tparisi]

Hold on, wasn't this really just a naming thing? Why not just go with meshes and skin as the property names on the node (vs instanceSkin)

In general I think we could ditch the 'instance' naming and simplify things. A material could have a technique property instead of instanceTechnique-- you and I just discussed this earlier today.

[pjcozzi]

I'm not sure it is as simple as a rename, I think there is a semantic difference. @fabrobinet any chance you recall?

If this is just a rename, then, yes, I am all for dropping the instance prefix.

[tparisi]

Beyond the rename, I don't think we want to contemplate a redesign right now.

Do you like the technique rename idea too?

[pjcozzi]

Do you like the technique rename idea too?

Yes, but let's see the schema proposal.

[tparisi]

For which, the new materials/techniques/common design?

[pjcozzi]

I'm not sure what you are talking about anymore, but let's keep all material discussion in #84.

[pjcozzi]

For 1.0 for this issue I suggest we either

(1) do nothing, or
(2) "flatten" instanceSkin into node (like we did with the other instance properties for program and technique), e.g.,

Before

        "Geometry-mesh019Node": {
            "children": [],
            "instanceSkin": {
                "meshes": [
                    "Geometry-mesh019"
                ],
                "skeletons": [
                    "locator047Node",
                    "locator051Node"
                ],
                "skin": "Controller-mesh019"
            },
            "matrix": [...],
            "name": "Cesium_Man"
        }

After

        "Geometry-mesh019Node": {
            "children": [],
            "meshes": [
                "Geometry-mesh019"
             ],
            "skeletons": [
                "locator047Node",
                "locator051Node"
            ],
            "skin": "Controller-mesh019",
            "matrix": [...],
            "name": "Cesium_Man"
        }

This explicitly implies that a node can't have both a skin and meshes (for rendering), which we already say in the schema (#255 (comment)).

@tparisi @tfili what do you think? I'm OK with either option. Let's try to converge ASAP.

[tparisi]

In general I am for simplifying

Tony Parisi
tparisi@gmail.com
415.902.8002

On Oct 9, 2015, at 10:14 AM, Patrick Cozzi notifications@github.com wrote:

For 1.0 for this issue I suggest we either

(1) do nothing, or
(2) "flatten" instanceSkin into node (like we did with the other instance properties for program and technique), e.g.,

Before

    "Geometry-mesh019Node": {
        "children": [],
        "instanceSkin": {
            "meshes": [
                "Geometry-mesh019"
            ],
            "skeletons": [
                "locator047Node",
                "locator051Node"
            ],
            "skin": "Controller-mesh019"
        },
        "matrix": [...],
        "name": "Cesium_Man"
    }

After

    "Geometry-mesh019Node": {
        "children": [],
        "meshes": [
            "Geometry-mesh019"
         ],
        "skeletons": [
            "locator047Node",
            "locator051Node"
        ],
        "skin": "Controller-mesh019",
        "matrix": [...],
        "name": "Cesium_Man"
    }

This explicitly implies that a node can't have both a skin and meshes (for rendering), which we already say in the schema (#255 (comment)).

@tparisi @tfili what do you think? I'm OK with either option. Let's try to converge ASAP.

—
Reply to this email directly or view it on GitHub.

[tfili]

I'm good with simplifying as well.

[pjcozzi]

OK, let's do it.

• @tfili - update converter
• @pjcozzi - update schema
• @tparisi - update spec - if needed.

[tfili]

Updated converter in dev-1.0 at fb40b4c

[pjcozzi]

@tparisi please close this when the spec is updated.

[tparisi]

@pjcozzi please review the language thanks

[pjcozzi]

Looks good.