Positioned layout in RenderBox is tangled up in ways that make it hard to build on: * Setting and calculating values from containing block geometry is scattered throughout the code. * Offset calculations for converting to/from container coordinates are unnecessarily confusing and not well separated from layout rule implementations. * We repeat some calculations for no other reason that someone happened to put them in a function that's called multiple times. * We pass around long lists of parameters in order to pass cached information from one function to another, which makes it hard to notice the parameters that actually changed between function calls. To have a better foundation for adding position-area and alignment code, we should: * Encapsulate containing block geometry logic and data. * Package up cached values into a single struct so they can be passed as a single parameter. * Better organize calculations into phases so that position-area adjustments to the containing block and alignment logic can be more easily and correctly integrated.