Code review comment for lp://staging/~vanvugt/mir/parenting-server-api

I'm also wondering about type morphing, which design wants us to support... What if a surface morphs from type A to type B and only type B requires parenting?

In that case you'd have to reject a simple set_type() request and add more client functions that do an atomic set_type_B(surface, newparent). So there's the scalability issue still, and you do gain atomicity, but I'm not sure that atomicity is more important than the scalability advantage.