One useful feature in 2.7x Blender is the zoom-dependent (adaptive) grid in the viewport. Zooming in always produced a new one tenth subdivision in the grid. And the snap to grid also snapped to the proper increments in top/side/front views. So I could zoom in and model with millimeter-precision snapping, and then zoom out and position the objects with 1 meter-precision.
Now this seems to be gone in 2.8. It always uses the same snapping increments that are defined under scene properties.
What happened here? Or am I simply missing something?