Spec changes from 1.0 to 2.0

[pjcozzi]

This issue will be updated as we work on the 1.1 spec (also see the spec pull request, #784)

1.0.1/1.1 was renumbered to 2.0

JSON format changes:

• Removed indirection from animation: sampler now refers directly to accessors, Tightening up animations #712
• material.parameter.value and technique.parameter.value must be an array, Make technique.parameter.value always array #690
• Node can have only one mesh node.meshes vs mesh.primitives #821
• Top-level JSON objects are arrays now
• Added reqs on JSON encoding
• Added reqs on binary data alignment Accessor alignment requirements #802 (comment)

Additions:

• Added accessor.normalized, Add accessor.normalized #691, Сlarify normalized usage #706
• Added "STEP" as valid animation.sampler.interpolation value, Tightening up animations #712
• Added glExtensionsUsed property and 5125 (UNSIGNED_INT) accessor.componentType value, Draft of uint32 support #619
• Added extensionsRequired property, extensionsRequired property #720, Add more extensions docs #721

Removals:

• Removed revision number from profile.version, Drop revision number from profile.version #709
• Removed buffer.type, glTF 2.0 buffer: does it need a type? #786, Little clarification on buffer.type #629
• Removed technique.functions.scissor and removed 3089 (SCISSOR_TEST) as a valid value for technique.states.enable, Remove scissor, see #668 #681
• Techniques, programs, and shaders were moved out to KHR_technique_webgl extension.

Other edits:

• asset is now required, Mandatory dictionaries #642
• accessor.min and accessor.max are now required, Require accessor min/max for POSITION? #593, and clarified that the JSON value and binary data must be the same, Clarification on min/max float precision #628.
• Clarified animation.sampler and animation.channel restrictions, Tightening up animations #712
• buffer.byteLength and bufferView.byteLength are now required, .glTF / .bin mismatches? #560.
• skin.inverseBindMatrices is now optional, Make inverseBindMatrices optional #461.
• Attribute parameters can't have a value defined in the technique or parameter, Thoughts on tightening up the spec #563 (comment).
• Only TEXCOORD and COLOR attribute semantics can be written in the form [semantic]_[set_index], Thoughts on tightening up the spec #563 (comment).
• TEXCOORD and COLOR attribute semantics must be written in the form [semantic]_[set_index], e.g., just TEXCOORD should be TEXCOORD_0, and just COLOR should be COLOR_0, Enforce array form for TEXCOORD and COLOR #649
• camera.perspective.aspectRatio and camera.perspective.yfov must now be > 0, not >= 0, Thoughts on tightening up the spec #563 (comment).
• Application-specific parameter semantics must start with an underscore, e.g., _TEMPERATURE and _SIMULATION_TIME, Thoughts on tightening up the spec #563 (comment).
• Properties in technique.parameters must be defined in technique.uniforms or technique.attributes,
  Thoughts on tightening up the spec #563 (comment).
• technique.parameter.count can only be defined when the semantic is JOINTMATRIX or an application-specific semantic is used. It can never be defined for attribute parameters; only uniforms, d2f6945
• technique.parameter.semantic is required when the parameter is an attribute, 28e113d
• Mesh-only models are allowed, e.g., without materials, Mandatory dictionaries #642
• Skeleton hierarchies (nodes containing jointName) must be separated from non-skeleton hierarchies., Skinning fixes for 1.0.1 #647
• technique.states.functions.blendColor and technique.states.functions.depthRange parameters now must match WebGL function min/max, Align blendColor and depthRange parameters with WebGL spec #707

For upcoming 2.0 changes, see issues labeled "2.0" and not labeled "resolved".

For all 2.0 issues, see issues labeled "2.0".

[pjcozzi]

@javagl moving this discussion to #806 to keep this issue focused on changes from 1.0 to 1.1.

[pjcozzi]

@lexaknyazev can you please make sure this is up-to-date for GDC next week...including as we make changes?

[pjcozzi]

@lexaknyazev @bghgary @sbtron can you please update this 1.0 to 2.0 summary once the spec stabilizes?

[javagl]

Here are some points that I stumbled over while running a code generator (and 1.0-to-2.0 converter) against the 2.0 schema. I'm not sure whether they should be mentioned here (some of them may already be covered by links to issues where they are discussed), but I'll just drop this list here, and remove this comment later, if desired.

• The asset.profile and asset.premultipliedAlpha properties have been removed.
• The byteStride property was moved from accessor to bufferView (Move byteStride from accessor to bufferView? #827 , Issue 820 #833 )
• The animation.channel.target.id was renamed to animation.channel.target.node ( de1023f - I think this was raised as an issue originally, but have to look closer)
• The skin.bindShapeMatrix has been removed. It is now assumed to be pre-multiplied with the mesh or the inverse bind matrices (Culling animated models. #723, 6c41899)

[pjcozzi]

@javagl just curious, what 1.0 to 2.0 converter are you using? We would welcome your use and contributions to CesiumGS/gltf-pipeline#191

[vsantosu]

Hello @pjcozzi , is the AnalyticalGraphicsInc/gltf-pipeline tool generating gltf 2.0 already? I can't find clear information in the repo of which version is using. Thanks in advance!

[pjcozzi]

@mastayoda there is initial work for a 1.0 to 2.0 conversion stage in gltf-pipeline in CesiumGS/gltf-pipeline#191, but it needs to be updated to include the latest 2.0 changes. Contributions are welcome!

[vsantosu]

Thanks so much @pjcozzi ! I will contribute for sure! Our company is trying to move away from the 3D mesh format mess and we are loving gltf so far. We will be jumping into the Blender export/import project soon to support 2.0

[javagl]

Referring to the comment above (i.e. to keep this issue focussed on the changes), only the short answer: I internally created a small converter in JglTF, for "1.0" -> "2.0+Technique" conversion, but there are still many open issues, and it's probably worthwhile to focus on the official glTF tools for contributions.

[pjcozzi]

This all sounds great, @mastayoda and @javagl!

[pjcozzi]

@javagl or @sbtron any chance one or both of you want to update this? Someone asked on gitter and this would also be useful for the COLLADA2GLTF/gltf-pipeline/obj2gltf/Cesium updates.

[javagl]

I can run the code generator on the latest (final) schema, and see whether the compiler complains at a certain point. This may not cover all changes (particularly not necessarily the semantics of changes), but could be a starter (in addition to my comment above #605 (comment) )

[pjcozzi]

Sounds like a great start, @javagl! You are right that there will be many more subtle changes as well.

[javagl]

In the initial comment above, I don't see things like morph targets, PBR or the CATMULLROMSPLINE and CUBICSPLINE interpolation methods for animation.sampler being mentioned... (some of them may be "too big to be called a 'spec change'", though...)

After re-generating the code based on the latest schema, the following changes appeared in the diff (in addition to the points that I mentioned above... )

• The animation.channel.target.node is now optional (to allow other targets in the future), as of 3d3db04

• The point from the initial comment

  accessor.min and accessor.max are now required, Require accessor min/max for POSITION? #593, and clarified that the JSON value and binary data must be the same, Clarification on min/max float precision #628.

  is no longer true: They are optional now, as of 9de934a (I found some related discussions, but not where the decision was made, @lexaknyazev )

• The maximum value of bufferView.byteStride has been changed from 255 to 252 in 6e6ae37 (not sure why...)

• For camera.orthographic, the zfar value must be greater than znear, as of 0b3dbdd

• node.children, scene.nodes, and skin.joints can't be empty when defined, as of f44c3bd

• The components of the node.rotation quaternion must be in the range [-1,1], as of 75be81f

[lexaknyazev]

@javagl

• Accessors bounds are required only for positions (incl. morphs) and animation inputs. They don't make much sense for other cases.
• byteStride must be divisible by 4. So it can't be 253, 254, or 255.

[pjcozzi]

@javagl thanks for the input, you are very welcome to create a comprehensive 1.0 to 2.0 changelog as this one is clearly out-of-date.

[javagl]

• bufferView.byteOffset and accessor.byteOffset are now optional, with 0 being the default.
• texture.sampler is now optional

(Yes, the TODO of creating a change log is still implicitly pending)

[pjcozzi]

I think it is OK to close this since there is no action here.

[pjcozzi]

@javagl a complete 1.0 to 2.0 list is still welcome if you have the bandwidth.