On Thu, 28.11.13 09:19, David Herrmann (dh.herrm...@gmail.com) wrote:
> > Hmm, so if end and start match the buffer is empty? So you can never
> > fill it entirely? Or am I missing something?
> >
> > I'd always maintain start index + fill level instead of a start
> > index + end index, to avoid the confusion regarding the fill-level...
>
> Well, start+end is how ring-buffers were implemented historically. I
> think the reason is that "wrapping over" is easier to calculate for
> indices than for lengths (you cannot do: r->len = (r->len + 1) &
> RING_MASK). Even all examples on wikipedia use two indices
> (http://en.wikipedia.org/wiki/Circular_buffer). And all kernel
> ring-buffers use start/end..
Well, you only need the wrapping over for the reading side, the writing
side never needs that.
int append(..., length, ...) {
...
if (current_length + length > MAX_LENGTH)
return -ENOBUFS;
...
current_length += length;
...
}
int remove(..., length, ...) {
...
if (length > current_length)
return -EAGAIN;
...
rindex = (rindex + length) & RING_MASK;
current_length -= length;
...
}
Lennart
--
Lennart Poettering, Red Hat
_______________________________________________
systemd-devel mailing list
systemd-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/systemd-devel
