Darcs

[Manuel M T Chakravarty]

David,

Thanks for that summary!

Given that a first version of the New Darcs seems to be just around  
the corner, I am very much in favour of giving that a shot before  
moving over to a new vcs.  Switching to a new vcs is a significant  
investment, especially for those who haven't used that new vcs  
seriously before.  Maybe we should just invest that effort into  
helping David improve his code instead.

Last, but not least, we have made a lot of noise about what we want to  
see improved in darcs.  Now, it seems David has invested much time to  
address these issues.  So, I think we owe him at least to seriously  
try his new code before moving somewhere else.

Manuel

David Roundy:
> Hi all,
>
> Don just pointed out this discussion, and suggested I drop by with a  
> status
> update.
>
> First off, you're right that we're very short on developers for darcs.
> None of our experienced developers have much time just for darcs
> development just now (with the exception of myself), and we've had a
> distinct drop in the number of new developers joining the project.   
> I'm not
> quite sure why that is, but I think in part it is because darcs is  
> Good
> Enough for most developers.  It's almost exclusively large projects  
> (like
> ghc) that run into the major infelicities of darcs, and most  
> developers who
> work on large projects don't have much time to contribute to darcs,  
> while
> haskellers who work on small projects are happy with darcs as it is.
>
> I've been working on the conflicts problem (as some of you are aware)
> pretty much exclusively (apart from time spent with my Physics PhD  
> student)
> for the past few months (since August).  As of last week, I've got an
> implementation of the internals that passes all tests I've been able  
> to
> throw at it, and I am now in the process of adjusting the high-level  
> code
> to use the new patch semantics.  If I didn't have a grant proposal  
> due next
> Monday, I'd expect to have a working version of darcs in the next  
> week,
> it's moving that fast.  It helps that the new code implements  
> semantics are
> very similar to the existing semantics.
>
> Darcs 2 will use the new "hashed inventory" repository format that  
> some of
> you are familiar with.  The format cleans up the concept of a  
> "partial"
> repository basically by allowing lazy downloading of patches as  
> needed.  I
> believe this will address most of the problems ghc folks have  
> reported with
> --partial repositories, which is why I developed this format.  It  
> also has
> advantages in terms of atomicity and in that it enables easy signing  
> of
> repositories to verify the complete contents.
>
> I suspect there will be major performance regressions, however, with  
> the
> new code, because much of it was written with scaling and  
> correctness in
> mind, but without special care to ensure laziness to keep down  
> memory use.
> So hopefully once the test version is out, we can get some serious
> performance testing.  I plan on devoting what time I can to fixing
> performance (and correctness) issues over the next semester,  
> although I'll
> also need to devote more time to Physics than I have this semester.
>
> I do plan on writing a paper on darcs' semantics and submitting it  
> to POPL.
>
> Oh yes, and also in the plans is a couple of new formats for darcs  
> patches,
> particularly hunk patches, which should greatly improve parsing and
> file-writing times as well as memory consumption.  Basically, the  
> idea is
> to store the new and old version as contiguous blocks of bytes  
> (without the
> '+' and '-' at the beginning of each line), so that we can skip  
> right over
> hunk patches when parsing.
>
> I'm sure that I haven't addressed all of the concerns that have been
> expressed here, but hopefully you've got a somewhat clearer idea as to
> what's been going on in darcs-land.  To reiterate, we *are* very  
> short of
> developers.  Most of us (former and current darcs maintainers) seem  
> to in
> academic jobs which don't leave us much time for hobbies.   
> Contributions to
> the development effort would be very much appreciated!
>
> Incidentally, at this point, darcs development is also provides an
> opportunity to play with GADTs and existential type witnesses, as  
> we've
> gotten that going in the code.  I expect I'll end up writing a  
> separate
> paper on that subject, since it's a neat set of tricks to statically
> eliminate rather a large class of bugs.
> -- 
> David Roundy
> Department of Physics
> Oregon State University
>
> _______________________________________________
> Cvs-ghc mailing list
> Cvs-ghc at haskell.org
> http://www.haskell.org/mailman/listinfo/cvs-ghc