I don't know, that's the point! Implementations frequently have unexpected behavior that can be abused for computational purposes. For example, you can perform computations on an x86 without actually executing any instructions, by setting up the page tables such that the MMU functions as a rudimentary one-instruction computer. https://github.com/jbangert/trapcc
Real, asymptotically-efficient computations in CSS would require some kind of layer of indirection -- a way to make the styling of one element depend on another one. I'm far from a CSS expert and I don't know if such a thing is at all possible, but it's not out of the realm of possibility.
Off the top of my head, you might be able to get somewhere by requiring the user to put their mouse pointer in a certain ___location, and using box sizes to trigger the :hover pseudo-class. If the pointer covers multiple overlapping elements, is :hover applied to all of them, or just the topmost?
Real, asymptotically-efficient computations in CSS would require some kind of layer of indirection -- a way to make the styling of one element depend on another one. I'm far from a CSS expert and I don't know if such a thing is at all possible, but it's not out of the realm of possibility.
Off the top of my head, you might be able to get somewhere by requiring the user to put their mouse pointer in a certain ___location, and using box sizes to trigger the :hover pseudo-class. If the pointer covers multiple overlapping elements, is :hover applied to all of them, or just the topmost?