What problem does this solve or what need does it fill?

A full 3D affine transform is a poor fit for UIs and often overkill for 2D uses. Transform and it's propagation system is a major bottleneck in PostUpdate, with a large number of potentially separable systems dependent on them.

What solution would you like?

Split out Transform2D and RectTransform from Transform, explicitly for their respective use cases.

Pros

• Both can be smaller in memory than Transform and GlobalTransform.
• Math on both will be simpler (and faster)
• Would likely fix any oddities when dealing with Z-positions in both UI and 2D by strictly defining inheritance behavior of depth.
• UI and 2D propagation and dependent systems parallelize more readily with 3D systems.
• UI probably doesn't need a dedicated propagation or GlobalTransform equivalent, relying on the UI layouting algorithm instead.

Cons

• Higher complexity. Recreating propagation outside of PostUpdate is now more involved.
• Different components adds mental overhead to users.

What alternative(s) have you considered?

Leave it as is for now.