Re: Howto read line from a stream

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Sun, 31 May 2009 05:56:41 -0700 (PDT), Tomek Wa�kuski wrote:

> Thank you Dimitry. I have implemented something like this:
> 
> ---------- CODE -----------
>    function Get_Line (Channel : in Stream_Access) return String is
>       type String_Access is access String;
>       Buffer      : Character;
>       Result      : String_Access;
>       Quit        : Boolean := False;
>       Line_Length : Natural := 0;
>       procedure Free is new Ada.Unchecked_Deallocation (String,
> String_Access);
>    begin
>       loop
>          Character'Read (Channel, Buffer);
>          if Result = null then
>             Result := new String (1 .. 1);
>          end if;
>          if Result'Length = Line_Length then
>             declare
>                Old : String_Access := Result;
>             begin
>                Result := new String (1 .. Old'Length +1);
>                Result (1 .. Old'Length) := Old.all;
>                Free (Old);
>             end;
>          end if;
>          Line_Length := Line_Length + 1;
>          Result (Line_Length) := Buffer;
>          if Quit = True and then Buffer = ASCII.LF then
>             exit;
>          elsif Buffer = ASCII.CR then
>             Quit := True;
>          else
>             Quit := False;
>          end if;
>       end loop;
>       return Result.all;
>    end Get_Line;
> ---------- CODE -----------

The problem with above is that it is at least as inefficient as 
Unbounded_String:

1. You allocate a new string per each input byte;
2. You do not keep the allocated buffer between calls to Get_Line;
3. You copy the buffer content once read;
4. It leaks memory.

The idea I tried to convey by my solution was to address the issues 1-3:

1. The buffer is allocated in big pieces, so that it would adjust to the 
maximal line length in few iterations.
2. The buffer is kept between calls. 1 & 2 mean that there would be only 
one or two allocations per session.
3. The buffer content is not copied. Get_Line returns of an access to 
String, of which slice is then renamed.

> My protocol says that every line ends with CR & LF, only length can
> vary.

The best practice of dealing with texts (streams or not) is to use LF as a 
line terminator and remove training (or any) CR's from the buffer. This 
makes it working under both Linux and Windows.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de