Added tracking of order by which tokens are processed by lexical analyser.
This helps in correctly formating error messages when constructs begin in one file, and end in another.