[darcs-users] Darcs and Operational Transformations (Team ECOO)
willu at cse.unsw.edu.au
Tue Sep 15 00:46:18 UTC 2009
Sounds very interesting. If you let me know when the next meeting
is, if there is one, I might see if I can make it. I went as far as
starting to develop my own VCS, but time constraints overtook me (this
is a side-interest of mine). I have a few things to note:
- Google Wave also appears to use OT (and was developed by some
guys I know at Google Sydney I believe).
- My understanding is that instead of enforcing TP2, both SO6 and
Google Wave enforce tree-structured patch sharing. Is that still true
- Thanks for the links to FoCAL. Looks interesting.
My thoughts on making an OT based VCS went in a different direction
from 'conflict free', but rather moving to a representation where the
conflicts were an integral part of the representation. This means
moving away from linear strings. The OT internals all worked with
that more expressive representation, but it would be linearised and
conflict markers introduced when you built a working copy. The patch
creation code would handle mapping back from the linearised workspace
to the more expressive representation.
The best analogy I can think of is wanting to introduce square root
in a number system. The reals are great, but sometimes the square
root operation takes you outside the reals, e.g. sqrt(-1). So you
expand your representation space to the complex numbers. In that
space you can take square roots freely and everything works. I guess
you could say that in this expanded space my system is conflict free,
or you could look at is as the 'conflicts' corresponding to the
'imaginary' part of the representation.
When you need to feed a complex number into a system that cannot
handle them, you need to force things back to plain reals - in the VCS
case this consists of building a conflict marked version in a plain
On 15/09/2009, at 1:47 AM, Eric Kow wrote:
> Dear Darcs-Users, Team ECOO and William Uther,
> Last week I had the great opportunity to meet and talk about Darcs
> members of Team ECOO in Nancy.
> Team ECOO are an INRIA research group that working on lots of
> topics related to computer-mediated cooperation <http://
> They have a special interest in (distributed) revision control
> wikis and other such software. The team has a lot of experience
> in distributed revision control; in fact they have even put together a
> patch-based system known as 'so6' ("saucisse") with some intriguing
> similarities to Darcs.
> Overall, the purpose of this meeting was for Darcs to start
> a long-term relationship with the research world. There is a lot of
> work out there on advanced revision control. If we could only make
> that we're all talking to each other on a regular basis, perhaps we
> could pool our efforts somehow. Maybe one day, we can put revision
> control back on the map as a serious research topic. Maybe one day,
> can give these Gits a run for their money.
> The meeting was long and interesting. I presented Darcs a little bit;
> we talked about the formal properties that Darcs tries to ensure and
> also about the potential relationship between Darcs and Operational
> Transformations. I hope to put together a more detailed write-up
> the things we talked about. Here I just want to give a very quick
> 1. I talked about some key Darcs features: dependency inference
> and pervasive cherry-picking (even in our undo operations).
> 2. Pascal Molli brought up a standard test from the
> Operational Transformations literature (TP2;
> 3. I discussed how Darcs patch theory can be seen as a sort of
> multi-layer onion:
> a. properties of the commutation/inverse operations
> b. conflicts (eg darcs-1 vs darcs-2)
> c. definition of primitive patches and commutation
> d. conflict marking
> Here I was trying to convey the idea that while the bug above was on
> the level of (c-d), what the Darcs community is most pre-occupied
> at the moment are (a-b), which we think we may be able solve
> independently of each other.
> 4. We talked about the relationship between Darcs and Operational
> Transformations. If I can give a caricature of the state of
> * OT has definitions and proofs of formal properties
> * Darcs has a story for conflicts (ie. parts a-b above).
> Our story may not be perfect (exponential merges, open
> but it's out there in the wild and mostly works in practice.
> 5. We suggested some interesting steps forward such as
> * Within the Darcs community? Attempt to formulate Darcs (3a) in
> terms of Operational Transformations
> * Within the ECOO team? Study the Darcs (3b) story for conflicts
> and compare with previous work on OT 'conflict objects'. I
> still think that the Darcs idea that 'conflicting patches cancel
> each other out' is potentially new and interesting for ECOO
> * Within the Darcs community? Read up on OT literature (in
> separate mail) on formal properities and proofs thereof
> (Could be very interesting for Ian and all his Coq-writing
> * Darcs and ECOO: Bring a Patch Theory Expert (ahem Ian?) to Nancy
> they cane talk to Team ECOO!
> 6. Team ECOO talked a bit about some new directions they are
> exploring in conflict-free revision control. It sounds rather
> familiar to Jean-Phillipe Benardy's work on Focal.
> Apologies if I've glossed over or misunderstood anything important!
> I hope to follow up one this with a few more concrete details.
> Many thanks to Claudia Ignat, Gerald Oster and the rest of team ECOO
> this opportunity. I hope we can continue this discussion and keep on
> exchanging ideas :-)
> Eric Kow <http://www.nltg.brighton.ac.uk/home/Eric.Kow>
> PGP Key ID: 08AC04F9
More information about the darcs-users