We are in the process of deprecating this API. Neither SRP uses this API, and it has never worked as intended, there is not an easy way to support this, and there isn't a proper workflow specified for how this data would be used.

Unity culls all lights by frustum and occlusion and put into the VisibleLightList. If the redner pipeline requests per-object lighting, then Unity does a PerObjectCulling that tests each light against each object and it builds a lightIndices list as follows:

[Obj0Light0, Obj0Light1, ... Obj0LightN, Obj1Light0, Obj1Light1, ... Obj1LightK, ... ObjZLightZ]
Per-object reflection probes would also have such a list. So when we setup state for drawing each object, we setup a variable that tells the offset and count into the visible light index list and visible reflection probe index list. So if you want to draw Obj0, you set offset as 0 and count as N+1. If you want to draw object 1, you set offset as N + 1 and count as K+1.

The buffer of visible light indices is based on the VisibleLight array. If the client side sorts the VisibleLight array, then we either you have to go back to the light index light and update all of it to the new sorted lights. Or you create an indirection to the light index map to tell how to map from the old light indices to the new light indices.

The identity of that is {0, 1, 2, ... N}. But let’s say you sort and swap VisibleLight[0] and VisibleLight[1], so you have to update the light index map to be {1, 0, 2, ... N }. The light index map is a remap list for both light indices and reflection probe indices, but they need to be seperate.

Please let us know if there is a real world scenario where this functionality would be useful or critical to performance.