mtn_setup()

-- This test is a bug report.
xfail_if(true, false)

-- Todo:
-- 1) Write monotone-agent
-- 2) Write tests for it here
--
-- Desired user experience:
--   - start the agent at login time
--     (it would be nice if we could spawn it automatically the first
--     time monotone is run, but then how would we communicate the
--     agent's address to independent processes?  and cause it to exit
--     when the session ends?)
--   - first time monotone needs to sign something, it prompts for a
--     passphrase
--   - second and later times monotone needs to sign something, no
--     passphrase is necessary (even if the second time is in a
--     different monotone process entirely)
--   - probably have some command that does nothing except prompt for a
--     passphrase, so people who want to type it once at the beginning
--     of a session can do that.
--
-- Internal design:
--   - agent opens an access controlled unix-domain socket (what's the
--     windows equivalent?  cf. pageant below)
--   - an environment variable saves the socket (and pid, I suppose,
--     for cleanup), like ssh-agent (again, what's the windows
--     equivalent?)
--   - when monotone wants to sign something, it checks to see if the
--     agent has the private key.
--     - if it does, then monotone sends the agent the data to be
--       signed, and waits for the response
--     - if it doesn't, then monotone requests the passphrase, decrypts
--       the key, sends the key to the agent, and then either goes to
--       the above step (maybe simpler) or just signs the data itself
--       (maybe faster)
--   - so the agent needs protocol packets:
--     - do you have key <hash>?
--       - response
--     - sign data <data> (or just hash of data <hash>) with key <hash>
--       - response
--     - add key <hash> <key pieces>
--       - response
--     - for debugging, might want 'list keys' and 'remove key'; 'list
--       keys' probably means agent should store key ids too?  depends
--       on how debuggish it is, I guess.
--   - note that this design assumes that monotone will have
--     independent access to the list of keys, and so doesn't need to
--     get the ids and hashes from the agent.  this seems reasonable,
--     at least for a first pass; it should be simpler to code, and the
--     cases where it fails (e.g., agent forwarding) don't seem very
--     compelling to me.
--
-- Other similar projects to steal ideas from:
--   - ssh-agent in openssh
--     their protocol is documented in
--     draft-ylonen-ssh-protocol-00.txt.  It's simple enough that this
--     isn't that useful, though.  (Unfortunately, ssh-agent provides
--     challenge-responses, not signatures, so we can't use it
--     directly.  Using it directly would have been much nicer.  I
--     suppose in the long run we could try and convince the ssh-agent
--     people that providing signing services for some keys might be
--     generally useful...)
--     It's not clear that signing services are something you
--     want to forward to other machines, though, even if
--     challenge-response services are...
--   - "pageant" in putty is a ssh-agent that works on Windows
--     - how does this work in windows, with no SSH_AUTH_SOCK and no
--       unix-domain sockets?
--   - codeville has an agent implementation too