On 1/26/08, David Roundy <[EMAIL PROTECTED]> wrote:
> Yes, that would be nice. It *would* add a requirement for darcs to be
> present on the server (not a huge requirement, and we could always fall
> back on scp). And most folks who get and pull over ssh will also be
> pushing over ssh, which already requires darcs be present on the server.
I agree.
> It also would require that we develop an entire ftp protocol including
> transmition of errors.
The command set required is fairly small. I'm sure I'm missing a few
details, but you need at least get, put, rm, mv, mkdir, stat -- and
they would all be very simple, just the barest minimum needed to serve
Darcs' needs.
You would not need working-directory management or globbing. As far as
I understand the Darcs repo format, you would not need ls, ch{mod,grp}
or ln.
> I think this would definitely be harder than
> learning haskell. Which isn't to say it's not a reasonable problem to
> tackle, but rather that your lack of Haskell knowledge isn't an excuse!
Really? I don't think it sounds difficult at all in any of the
languages I do know. :) If it's designed to be purely sequentially
pipelined, you don't even need to handle much state.
> One might also wonder why we don't use the existing sftp program, which
> itself implements quite a similar protocol to what you're proposing. In
> fact, we do so, but haven't wanted to deal with the headache of maintaining
> the connection, so we only do so during darcs get (when we know in advance
> an entire list of files we'd like to grab). And this itself turns out to
> be somewhat fragile, as not all sftp servers are identical (and of course,
> putty never names any of its commands in a standard manner).
I forgot about the SFTP command mode. I don't know if it's pipelined,
but I can see how it could be fragile.
Alexander.
_______________________________________________
darcs-devel mailing list
darcs-devel@darcs.net
http://lists.osuosl.org/mailman/listinfo/darcs-devel
