I very much like the idea of these prims not actually being inside the node graph. Keeping them as a pure UI concept like backdrops seems a lot cleaner to me. I find it hard to mentally map out all the cases and determine if there is some important feature that is going to be excessively complicated by this approach. But if the various use cases can be covered by writing a more complex UI layer on top of the combined (node graph, bundles) set of prims, I think this is the superior design.
I’ll ask Rafal to have a look at this thread, as he’s a lot deeper into the technical details in this area than I am.
I aim to write up a full proposal post-siggraph. To two of James’s later questions:
layering opinions - since a bundles-Dot is almost entirely specified by a relationship, we get to use full list-editing of targets in stronger layers.
identifying Material splice points - I’m not positive they will want the exact some treatment, but this very much reminds me of the some-years-old forum discussions (possibly the old forum) about “publishing” aov’s for consumption by RenderVars, and there seemed to be consensus that aov’s should be a sub-namespace of outputs on a Material/NodeGraph prim that connect to the shader output that produces a result.