Those elements need to be attached to something on the page. Actually placing them next to
the element in the DOM causes problems though, if any parent element is anything
overflow: visible, the element gets cut off. So you need absolute positioning
in the body.
Some of the time absolute positioning is right, but what about if the thing we're attached to is fixed to the center of the screen? We'll have to move it every time the user scrolls. What about if the element is in a scrollable container, if the overlay is inside of it (so no clipping), it would be cool if the code were smart enough to move it inside when that area is scrolled. That way we need to reposition it even less.
It would also be nice if the code could somehow figure out whether positioning it from the top, bottom, left, or right would result in the fewest repositionings as the user scrolls or resizes.
Most of the time you're building these elements it would be nice for the element to flip to the other side of the element if it hits the edge of the screen, or a scrollable container it might be in. It would be nice if we could confine the element to within some area, or even hide it when it leaves.
It would be nice for the element to be repositioned with CSS transforms rather than top and left when possible, to allow the positioning to be done entirely in the GPU.
Now that the positioning is so fancy, you're going to use it for more and more elements. It would be cool if the library could optimize all of their repositioning into a single repaint.
All of that is baked into Tether.