Such an action depends from a specific conditions that influence specific method realization, I guess.
I believe that snapping to a 3d cursor as to a vertex (being able to use it as snap source or snap target), including its orientation could be useful for different operations, for example, for self-snapping.
Self-snapping is a separate iceberg though.
An example - putting a cube from its bottom to its top.
There was proposed ghosting (like in most CAD programs), but it could be performance expensive when you work with heavyweight selections (like precisely placing houses that consists from up to 50k objects). Ghosting can be optional, but it is also challenging to make I guess.
At the moment self-snapping can be solved with building additional geometry (temporary constructions aka helpers), like drawing a mesh edge from snap source to snap target, and removing it after operation (or keeping it for multiple similar operations, which is typical for working with instances for example), so it is a question of a QOL-feature rather than critical missing functionality, so it can be postponed.