This will make it easier for the parser to walk through the token sequence
with one token of lookahead, maintaining state of where we've got to and
also allowing us to switch into and out of the newline-sensitive mode
as we transition to and from newline-sensitive constructs.
