The problem is that you're trying to use the tool in a way it can't be used. Consider the following process instruction:

<?
    echo 'This is from PHP.';
?>
[download]

That runs over three lines. However, if you're reading things in line-by-line, you lose all context. When HTS gets to an individual line, it can't parse it as a process instruction because it doesn't know that it is one. I'm afraid you'll have to take a different approach to solving your problem.

I'm not sure there is a way around this problem, frankly, because the tokens will never have a direct correlation with line numbers. One possibility would be to (forgive me!) track the number of line endings in all tokens as a rough heuristic of which line a token is on.