Probably missing something wildly obvious here.
Context: Maya (2017) modeler, Arnold renderder.
We want to display an object as one type of geometry in the viewport, and have a different geometry at rendertime. Without any user interaction required (no toggling layers, visibilities, etc).
In ye-olde-days, a pair of pre/post-renderMELs would get most of the way, flipping the visibility of objects pre and post export of a frame render. This is not as meaningful in an IPR context, given they monitor changes to the scene state, such as visibility.
I think what we need is a pre-translation/export callback hook? So we can mask what objects are translated into the .ass being used.
To give a more explicit example:
we want to render tonnes of cylinders, which we do with an instancer-fed setup (> ginstances)
its too heavy to display this instancer set in the viewport with full geometry (and the bounding box instancer display options arent sufficient for interactive evaluation of the cylinder placements in the scene [they clutter])
for proxy representation of the cylinders, we use a particle system with 'streak' display type for OpenGl line drawing of the cylinder axes (or just a custom MPxLocator drawer)
we want to always keep our proxy representation in the viewport, but see the full geo rep in renders (eg in Arnold's new renderView in IPR mode). [or sometimes just render the proxy, and ignore the full geo, and occasionally disp and render geo, hide proxy. Flexibility.]
exporting for Arnold stand-ins doesnt (seem) to fit, as there are constant edits being made to the 'asset' [unless we can at-translation-time export an .ass for a procedural node to use?... a pre-translation callback slot somewhere?]
Any suggestions greatly appreciated!