Clarify TRS properties in the Object Model

[lexaknyazev]

/cc @aaronfranke @hybridherbst

[javagl]

Fixes #2547 ?

"Since the glTF 2.0 specification allows..."

It sounds like a reasonable change for me.

I cannot be sure that I fully understand the implications for consumers of such assets, and even less for producers. It may imply that, as a general rule, individual TRS properties should be preferred over a matrix. But to my understanding, that would perfectly make sense for anything that is related to interactivity, and have no drawbacks.

(The difficulties for consumers to track and handle a mix of TRS and matrix already exist, independently of the object model)

Tag @emackey because you also had some thoughts in the issue.

[lexaknyazev]

Implications for consumers (i.e., engines or loaders):

• Animation pointer & Interactivity: No need to support the ambiguous case when rotation or scale are updated on a node that defines a static matrix.
• Interactivity only: Requirement to reject graph access (both read & write) to rotation and scale properties of such nodes. We should make test assets to assert that.

Implications for producers (i.e., exporters):

• Do not write static matrices for nodes used for rotations or scale animations via animation pointers (it's already the case for base glTF 2.0 animations).
• In case of interactivity, prefer to not export static matrices at all just to be safe.

[aaronfranke]

My opinion is that translation should always be allowed, and it's good to specify this.

[lexaknyazev]

My opinion is that translation should always be allowed

That's what the PR says.

[bghgary]

lgtm

[emackey]

It's unfortunate that the spec needs to reference what exists "in JSON" as opposed to some parameter setting or value. But it's not clear how to get around this without creating more problems, so, I won't object to this.

[hybridherbst]

One question for something that's not entirely clear to me, what happens when matrix and translation are both defined? Does translation "win" and override the translation of the matrix? Can I in that case have the matrix and translation and then animate or make interactive just the translation property?

[lexaknyazev]

what happens when matrix and translation are both defined

That's disallowed by the base spec.

[hybridherbst]

Thanks, then I'm not sure if the section is clear enough: https://github.com/KhronosGroup/glTF/pull/2559/changes#diff-8f228d937d6ff200a2972f8164aa5f5738bf0377eab6023a9eba4f2b106e6fe5.

For me it sounds like "translation" can still be utilized with both interactivity and animation data, only rotation and scale are ruled out. But I might be misreading that.

[lexaknyazev]

@hybridherbst

• Static JSON use of any TRS property with matrix is disallowed in the base spec. That remains unchanged.
• The Object Model now says that rotation and scale pointers are undefined if the static matrix property is defined.
• The statement about rotation and scale being undefined does not include translation, and the note below explicitly highlights why matrix presence does not affect translation validity.

Do you think more clarifications are needed?