Re: [Monotone-devel] netsync over ssh revisited

[Richard Levitte]

In message <address@hidden> on Thu, 06 Sep 2012 04:56:45 -0400, Stephen Leake 
<address@hidden> said:

stephen_leake> Richard Levitte <address@hidden> writes:
stephen_leake> 
stephen_leake> > In message <address@hidden> on Thu, 06
stephen_leake> > Sep 2012 05:56:12 +0200 (CEST), Richard Levitte 
<address@hidden>
stephen_leake> > said:
stephen_leake> >
stephen_leake> > richard> I had a look in std_hooks.lua and noticed the ssh+ux: 
schema, which
stephen_leake> > richard> basically does what I'm after...
stephen_leake> >
stephen_leake> > Speaking of this, I'm thinking about that scheme name, and it 
comes
stephen_leake> > out as odd to me...  there seems to be a concensus out there 
that a
stephen_leake> > protocol transported through another protocol should be named
stephen_leake> > {protocol}+{transport}.  ssh+ux: does the exact opposite, and 
ux+ssh:
stephen_leake> > would be more appropriate.  Furthermore, since that's 
basically UNIX
stephen_leake> > domain sockets piped through SSH, and mtn supports local UNIX 
domain
stephen_leake> > sockets through the scheme local: (because netxx does that, 
that's
stephen_leake> > why), I'm pondering that the ssh+ux: scheme should really be 
renamed
stephen_leake> > to local+ssh: (of course, we can keep ssh+ux: as an alias).
stephen_leake> >
stephen_leake> > Thoughts?
stephen_leake> 
stephen_leake> local+ssh sounds more like your second scheme
stephen_leake> 
stephen_leake> Attempting to generalize/abstract a naming convention:
stephen_leake> 
stephen_leake>     ssh {host} socat - UNIX-CONNECT:{path}
stephen_leake> 
stephen_leake> I'm not clear what's a protocol and what's a transport here; it 
seems to
stephen_leake> me the actual data exchange protocol is always mtn (that defines 
how to
stephen_leake> exchange certs, keys, revisions, and possibly authentication), 
and both
stephen_leake> ssh and unix domain sockets are transports.

You're right, it was poorly expressed.  Say {transpport}+{tunnel}
then.  My inspiration is really git+ssh:, and I believe I've seen
similar constructs elsewhere, but memory fails me for the moment.

And you're absolutely right, we're not entirely consistent either (but
neither is anyone else)...  "ssh:" should really be "mtn+ssh:" if we
were to be entirely consistent, for example.

stephen_leake> > Either way, I realised that the possible uris aren't fully 
documented
stephen_leake> > in the manual, so I'm doing so now.  
stephen_leake> 
stephen_leake> That's good.
stephen_leake> 
stephen_leake> > And I think it would be good to have a consensus on local+ssh: 
vs
stephen_leake> > ux+ssh: vs ssh+ux: before the added documentation gets 
published.
stephen_leake> 
stephen_leake> Now is the time to rename it, if we are going to. But so far, I 
don't
stephen_leake> see a clear naming convention. 
stephen_leake> 
stephen_leake> 
stephen_leake> This discussion suggests a way to make file: work on Windows 
native; use
stephen_leake> TCP sockets to connect between the client and server instances 
of mtn.

Not sure that's a good idea from a security point of view.

Cheers,
Richard

-- 
Richard Levitte                         address@hidden
                                        http://richard.levitte.org/

"Life is a tremendous celebration - and I'm invited!"
-- from a friend's blog, translated from Swedish