What I said is the idea of my first proposal.
In the revised version, I have adopted your comments and suggestions
including:
(1) treate binary and text files equally
(2) sending full text as deltas
(3) if time permitted, to implement rsync.
On 5/15/06, Peter N. Lundblad <peter@famlundblad.se> wrote:
>
> Qi Fred writes:
> > The reason is NOT text files are small to enable only the binary files
> > working without text-base. The basic idea is that text files have
> > many special properties, such as svn:eol-style, svn:keywords, etc.,
> > which need special considerations. The process of text files would be
> > more complex than binary ones to achieve a same performance.
>
> I think you are mistaken here. What does properties have to do with
> this? We can't drop the propsbases, because they are needed for a lot
> of things. And you can have keyword expansion enabled in binary files
> as well.
>
> > Further more, small files may waste a lot of disk spaces in certain
> > systems. Working without text base is a charming feature.
> > In my mind, a rsync-like algorithm is a way to achieve this goal.
> > This is suggested in my revised proposal submitted to SoC.
>
> Yes, but as I've pointed out before, you don't *need* an rsync
> algorithm.
> It will just require more network bandwidth.
> Note that I'm opposed to a simplified "rsync-like" algorithm as you've
> proposed, that only works if blocks are exactly aligned, because we
> will be stuck with supporting that even when the have the real thing.
>
> Best Regards,
> //Peter
>
--
Best Regards,
Fred Qi
Received on Mon May 15 05:13:46 2006