Great to see this problem tackled :)

• I think it's definitely a good idea to ignore faces for object in bounding box mode. Those are most of the time complex meshes which are instanced a lot like trees. They would occlude the geometry the user really see and give unexpected result for the user (particularly when the bounding box is not visible in the current view, because the zoom level is too high). But for wireframe, I don't see a clear reason to do it. Maybe it could be an option in the snapping part of the 3DView header.
• I'm not a specialist in this area, so excuse me if the question is stupid, but why not use the selection method (Automatic/OpenGL Occlusion Queries/OpenGL select) for snapping to face? It works good to select and the goal is the same, get the first face under the mouse cursor.
• Maybe for a later version, it would be great to make it offer the same options/behavior like for selection: have a key to display a list of objects under the mouse cursor that could be snapped to (shift+alt+select mouse in the default keymap, it's the enumerate option). It would also solve the problem of deciding whether the draw method of objects should influence the snapping point and let the user decide on a per case basis.

The idea behind ​​ignoring faces in wireframe is that, as in BounBox, we don't see the faces. But it may not be ideal to change the current behavior of Blender, since users are already accustomed to it.

In D2527#59067, @mathieu menuet (bliblubli) wrote:

• I'm not a specialist in this area, so excuse me if the question is stupid, but why not use the selection method (Automatic/OpenGL Occlusion Queries/OpenGL select) for snapping to face? It works good to select and the goal is the same, get the first face under the mouse cursor.

In fact we could use OpenGL to snap. For example, to test occlusion, we can use the unproject function and compare the Z-depth with Z_depth = dot_m4_v3_row_z(pmat, co)/mul_project_m4_v3_zfac(pmat, co);
But as we see in some bug reports, different GPUs sometimes have a problem with OpenGL.
The entire snap system could be done with the help of OpenGL, but in that case the functions would need to redraw the entire View3d and test each element individually (as is already done with the selection system). This is not as efficient as using CPU and BVHs.

In D2527#59067, @mathieu menuet (bliblubli) wrote:

• Maybe for a later version, it would be great to make it offer the same options/behavior like for selection: have a key to display a list of objects under the mouse cursor that could be snapped to (shift+alt+select mouse in the default keymap, it's the enumerate option). It would also solve the problem of deciding whether the draw method of objects should influence the snapping point and let the user decide on a per case basis.

This is an interesting idea that would really help users ;)
A more urgent idea would be to expose the snap functions for Python. Thus, addons focused on architecture could add these features and enhance Blender

In D2527#59083, @Germano Cavalcante (mano-wii) wrote:

A more urgent idea would be to expose the snap functions for Python. Thus, addons focused on architecture could add these features and enhance Blender

Fully agree :) If it can be done for 2.79, it would be awesome.