MeshMerge node

[johnhaddon]

We would like the ability to instance meshes on to a point cloud as the Instancer does, but merging all the meshes into a single object ideal for raytracing (think leaves on a tree or banknotes on the wind). One approach would be to have an instancer style node (or an extension to the existing node) which simply outputs a single mesh. Another would be to use the instancer as-is, and have a subsequent MeshMerge node which collapses a hierarchy of meshes into a single object.

[carstenkolve]

I principle I do like the idea of the merging node (because of its flexibility - one might not want to do any raytracing etc); however with gaffer expanding all bounding boxes, this approach would still lead to a very slow graph in the ui?

[johnhaddon]

I like the idea of the merging node too. I would expect it to be a bit slower but I think it'd be worth prototyping it to see rather than first writing the more specialised solution first.

[carstenkolve]

agreed, especially if the instancer does get an option to create an approximate bbox based on the incoming point data as per #957 the speed issue might not be so bad

[johnhaddon]

Cool - that sounds like a plan then. Adding an approximate bound is easy - much easier than doing the accurate bound, so we can get that in quickly whenever it gets prioritised.

[johnhaddon]

I've updated the title on this one to be a MeshMerge node, as discussed previously. I'd propose that it works as follows :

• A parent plug specifies the location where the merged mesh will be created
• A name plug specifies the name of the merged mesh
• A filter input specifies the meshes to be merged
• A pruneOriginalMeshes toggle causes the original meshes to be pruned from the scene

In terms of implementation, I think we can derive from BranchCreator to do the work of making the merged mesh, and use a Prune node to do the pruning work. Those would be child implementation nodes of the main MeshMerge node.

There's code for merging two meshes in Cortex, which I think would provide a useful reference. But since we want to merge an arbitrary number of meshes, I think it's important that we do that in one step for speed. We might still want to put the "merge multiple meshes" code in Cortex somewhere anyway, perhaps in a new MeshAlgo.h file, since we've been discussing making Cortex functionality available directly in functions as well as via the higher level Ops.

How does that sound?