-- Most of the time, we update the screen_top of nodes from Viewport.y.
			-- But here we went the other way.
			-- It's very important to avoid creating a recurrence, to avoid running
			-- both sides of this feedback loop in a single frame. These apps are
			-- not very numerically precise (e.g. we force text lines to start at
			-- integer pixels regardless of zoom, because that keeps text crisp),
			-- and the computations of Viewport.y and node.top will almost certainly
			-- not converge. The resulting bugs are extremely difficult to chase
			-- down.
			-- The optional skip_updating_screen_top_for arg ensures we don't run
			-- the other side of the feedback loop.

			-- Most of the time, we update the screen_top of nodes from Viewport.y.
			-- But here we went the other way.
			-- It's very important to avoid creating a recurrence, to avoid running
			-- both sides of this feedback loop in a single frame. These apps are
			-- not very numerically precise (e.g. we force text lines to start at
			-- integer pixels regardless of zoom, because that keeps text crisp),
			-- and the computations of Viewport.y and node.top will almost certainly
			-- not converge. The resulting bugs are extremely difficult to chase
			-- down.
			-- The optional skip_updating_screen_top_for arg ensures we don't run
			-- the other side of the feedback loop.