[Monotone-devel] Re: Future of monotone

[Graydon Hoare]

Thomas Keller wrote:

> You've probably read Graydons recent message; I did and I have to say 
> that it made me a little sad. Not because Graydon "officially" stopped 
> working on the project - he wasn't doing many things lately anyways 
> beside the recent attempt of redoing netsync with something smarter in 
> nvm.nuskool (Whats the status of this Graydon? Do you think it is worth 
> that somebody else picks up your work there?) 

I think it's a good branch as they go in this sort of thing. I just ran 
out of interest. It contains a variety of interrelated new code. Anyone 
else is welcome to pick it up of course.

The changes are, primarily:

  - a JSON printer/parser and querying system
  - an SCGI i/o facility such that monotone can serve SCGI requests
    from a webserver or a raw socket
  - a sketch of the long-planned upgrade to certificates
  - a micro HTTP client
  - a much simpler synchronization system that works over HTTP+JSON+SCGI

My plan here was to rip out netsync, packets, basic_io, netio, and the 
old cert system. Put in the new cert system and the new sync system, and 
 replace the automate commands with json-speaking equivalents. Do away 
with all the legacy i/o stuff, and provide one that's actually easy to 
speak.

The idea was that this would both make hosting and deploying much easier 
-- HTTP is more friendly -- remove some of the nonsense that makes 
debugging netsync difficult, and make it easier to write browser-based 
frontends since you could just code them in JS and have them do XHR to 
the server to pluck out JSON objects describing database structures 
directly. Plus, lots of script languages can just slurp in JSON, so it 
might make it a little easier to script.

The one key point of this was that it required a small bit of 
policy-branch work to connect each branch to a corresponding 
certificate-storage lineage. It's really just a branch-ID -> 
lineage-root-revid mapping you need to store, nothing fancy. This is 
because the new sync system, being compeltely DAG-based, was to transmit 
certs by transferring and then trivially auto-merging a "filesystem 
representation" of the set of certs associated with a branch. The 
details actually make sense, and it's way simpler than using merkle 
tries; if anyone wants to care I'm happy to explain further. Early 
experiments with the new sync system suggest it would work well and be 
simpler than netsync.

-Graydon