Dear list,
I have the funniest problem.

I use Scan to scan a file, and output an edited variant, when edits are done
I just need to copy the remainder of the file. 

This ludicrious task is proving more elusive to handle than the whole project.

The problem is that when I finish editing, the scanning buffer is active and not empty and
my reading position in the input channel is not where I am currently at in scanning.


I don't want to use scanner to consume the remaining 3 terrabytes digesting 
it line by line. I can't seem to be able to scan bigger chunks than lines either unless I can name a character
that certainly does not appear in the text I am reading.
If there is a character like this, say @, then I am ok with

1)
scan "%s@@%!" (fun s -> Printf.fprintf corpus_out "%s" s)

this one reads till the next @ character which is ignored or the end of the input, which is checked with putting %! explicitly.
This passes my tests, but horribly ugly, since there is no character that I can guarantee this way.
Plus I might not have memory for passing this whole chunk as one string if the file is large.

So as an alternative I did this:

2)
(* we set the input channel reading position to where we are in scanning *)
let _ = scan "%n" (fun x -> seek_in corpus_in (x - 1)) in
(* and then dump the rest trivially in chunks of buf_size chars *)
let buf = String.create buf_size in
let rec dump () =
let len = input corpus_in buf 0 buf_size in
if len > 0 then (output corpus_out buf 0 len; dump () )
in

or in one go

3)
let end_pos = in_channel_length corpus_in
let len = end_pos - pos_in corpus_in in
let s = String.create len in
let _ = really_input corpus_in s 0 len in
Printf.fprintf corpus_out "%s" s;

On my mac and linux, all works smoothly, till I used it on windows.
These do not work on windows, since in_channel_length and seek do not take into account the newline translations that take
place at reading.
Or in other words, the scan module reports character positions incorrectly since CRLF=\013\010 is counted as one character and matched by \n.
I suspect 2 does not work either, since seek is probably the same as in_channel_length when it comes to counting chars.

So there is no way to combine scan positions and in_channel/seek type positions.

If there was a way to dump and empty the scanning buffer, then I could then just use (2), since
then scanning pos and pos_in would align, but I found no way of doing that.

I have no idea how to solve this. Well, I guess I am missing something trivial.

Thanks for help

Viktor