The current version of the spec - while not outride forbidding what we are doing in practice - does not reflect a lot of the aspect that are important for proper optimization. For one, the spec currently focused almost exclusively on the worst-case solutions that we know do not perform well (but allow to meet current needs). However, we have made significant progress that allows us now to use much more optimal structures and representations for a lot/most of the cases, and we have ideas for addressing the remaining ones.