Handle CRLF files correctly in let-syntax

Previously a CRLF source file would be re-emitted with \r\n translated
to \r\r\n which affected the fragments in src/preprocessing.ml. The
lexer is now altered so that \r is never written to the output channel
so on Windows the result will always be CRLF encoded and on Unix the
result will always be LF encoded, regardless of the input.

Signed-off-by: David Allsopp <[email protected]>

src/let-syntax/lexer.mll

@@ -28,12 +28,21 @@ let fail lb msg =
   exit 1
 
 let ps t s =
-  String.iter s ~f:(function
-    | '\n' -> t.line <- t.line + 1
-    | _    -> ());
-  output_string t.oc s
+  let len = String.length s in
+  if len > 0 then
+    if s.[len - 1] = '\n' then begin
+      assert (String.index s '\n' = len - 1);
+      t.line <- t.line + 1;
+      if len > 1 && s.[len - 2] = '\r' then begin
+        output_substring t.oc s 0 (len - 2);
+        output_char t.oc '\n'
+      end else
+        output_string t.oc s
+    end else
+      output_string t.oc s
 let pc t = function
   | '\n' -> t.line <- t.line + 1; output_char t.oc '\n'
+  | '\r' -> assert false
   | c    -> output_char t.oc c
 let npc t n c = for _ = 1 to n do pc t c done
 let pf t fmt = ksprintf (ps t) fmt
@@ -179,8 +188,8 @@ and lhs t = parse
   | eof
     { ()
     }
-  | newline as s
-    { Buffer.add_string t.buf s;
+  | newline
+    { Buffer.add_char t.buf '\n';
       Lexing.new_line lexbuf;
       lhs t lexbuf
     }