[darcs-devel] the straw that broke the camel's back

Ian Lynagh igloo at earth.li
Fri Jul 13 05:51:14 PDT 2007


On Fri, Jul 13, 2007 at 01:55:26AM +0200, Juliusz Chroboczek wrote:
> 
> On the other hand, Darcs has four very serious and one moderately
> serious issue:
> 
>   - Darcs is not complete (some patches fail to commute);
>   - Darcs is not correct (will sometimes write an incorrect repo);
>   - Darcs is subject to combinatorial explosion;
>   - Darcs is coded inefficiently in some areas.

(Your numbers don't match the length of your list)

> It is not entirely clear to me what the right thing to do is -- work
> on improving Darcs, or throw away everything and start again from
> scratch now that we all know so much more.

I think improving darcs incrementally is just as easy, on the assumption
that you would want to provide a migration route for old repos
regardless. Darcs already has the repo-format stuff to make it easy to
migrate to new patch types over 2 or 3 darcs releases.

The first issues you list are all part of the same thing: We need to
replace mergers with something else, and as far as I know we're still
looking for an algorithm, not at the coding stage (I've yet to read
Jason's PDF).

The fourth one isn't particularly difficult, it just needs time to be
spent on it (and once the merger replacement is done I expect to find
time to do some of it). We already know what more efficient patch
formats, datatypes and algorithms for, e.g., hunk and binary patches
look like, and we already have the repo format stuff; it's just a SMOP
now.

> I really wish I had the
> opportunity to buy a few pints to David, Erik, Tommy, Ian and whoever
> else and discuss all of this.

Who'll be able to be in the Freiburg area between
http://www.cse.unsw.edu.au/~keller/haskellws/HaskellWorkshop.html
and
http://www.haskell.org/haskellwiki/Hac_2007_II
? There's the possibility of a "darcs corner" at the Hackathon if people
are interested.


Thanks
Ian



More information about the darcs-devel mailing list