Thanks for all the suggestions. Actually, this is a one-off program for which I want a correct solution in the minimal lines of code. So paramount for me is just sticking together that which is known to work. Investigating this further, here is what I've found. The spirit multi-pass iterator is almost ideal. It is a one liner - I like that. One can specify a different storage policy - fixed_queue<size_t N> so that as long as I know that my pattern doesn't exceed a certain size, it will handle and unbounded file while still keeping a bound on the amount of storage used. It even will throw an exception if make an error and the internal buffer overflows. The authors even made a typedef for this version of multi-pass which they call lookahead<input_iterator, N>. Could this be more perfect for my application? I don't think so. But alas, there's a rub. multi-pass hardcodes "forward_iterator_tag" rather than inheriting the trasversal_tag from it's storage policy - in this case "random_access_tag". So even though all the code is there and it would work (I believe) I'm stuck. It's so close, that I'm considering making a tweaked version of multi-pass to fix this problem. I'm loath to do this as the implementation is sufficiently general that its a whole learning curve to understand enough to mess with it with confidence. And of course the time invested has to be weighed against using a cruder solution. This is one of the most frustrating things about working with libraries and computers. The curse of things that "almost work". It's maddening. Note that I'm not blaming anyone. Its the two-edge sword of importing code. Robert Ramey