[PATCH] Branch and pull-- now with remote

[Martin Pool]

On  6 Jun 2005, duchier at ps.uni-sb.de wrote:
> Martin Pool <mbp at sourcefrog.net> writes:
> 
> >> Getting somewhat off-topic here.  When it comes to remote branches.  I
> >> know you Martin has been talking about having inteligence on both
> >> sides of the pipe. I think this would be a good idea, however, I also
> >> think that it's needed to have support for remote dumb storage.
> >
> > Yes, I certainly want remote dumb storage.  Writing to remote dumb
> > storage is a bit harder because of the need to get the locking and
> > permissions right, but I think it can be done in this model.
> 
> This is getting back to something I mentioned a while back (revlib server
> ideas).  The point to keep in mind is that even when the backend is dumb, you
> are talking to a smart server: it's just that the smart server is then running
> in the client.
> 
> I think that it is essential to not make the same mistake as arch: namely,
> dumbing it all down to operations on a file system.  Let the interface to the
> specific backend dumb things down where necessary, but leave open the
> possibility of smarter backends.

Such a layer already exists in bzr through the Branch class.  I think
Erik just said that he introduced a new subclass that operates over
ssh.  You could easily have another one that e.g. looks things up in
an SQL database.

I wouldn't claim the abstraction is perfectly factored just yet but I
think it's in approximately the right place.

-- 
Martin