This is necessary for incremental parsing correctness, as the incremental parser will attempt to reuse these types of nodes in both contexts, and we much ensure it creates the same trees you would get if you were parsing normally.