Overview

Every backend maintains a per-pixel uint32_t object-ID buffer alongside color and depth. Picking reads this buffer to identify the object at a given screen coordinate.

Object IDs

Multiple meshes may share the same ID if the application treats them as one selectable entity.

Backend Contract

Every backend must:

1. Allocate an object-ID buffer during framebuffer_create
2. Clear it to 0 during frame_begin
3. Write draw_call->object_id for every rasterized pixel during draw
4. Implement pick_read_id(device, fb, x, y) → uint32_t
5. Implement pick_read_depth(device, fb, x, y) → float

Result Struct

typedef struct MopPickResult {
    bool     hit;         /* true if object_id != 0 at (x,y) */
    uint32_t object_id;   /* valid only when hit == true      */
    float    depth;        /* [0,1] normalized depth           */
} MopPickResult;

Usage

mop_viewport_render(viewport);
MopPickResult r = mop_viewport_pick(viewport, mouse_x, mouse_y);
if (r.hit) {
    select_object(r.object_id);
}

Picking reads the most recently rendered frame. Render before picking.