Interview: David Roundy of Darcs Revision Control

comforteagle writes "In the aftermath of our last interview with Tom Lord, regardless of personalities, it became apparent that the idea of decentralizing CVS is a big deal. Many mentioned darcs as an alternative to Arch. Mark Stosberg has interviewed project head-hancho David Roundy about darcs, his 'theory of patches,' what's next, and on using Haskell for the project."

Comments

Re:Haskell just won't cut it

[PhYrE2k2]

I did read the article, however I do DISAGREE with that comment in the article. People won't learn a language for one program, and there is not a large enough body who know the language to really truly UNDERSTAND the program and enough about it to make modifications and additions to it. Compare that to a bunch of C/C++/Java/Perl developers with a massive community body, it's a lot easier to get people to contribute. -M

Re:Haskell just won't cut it

[Pseudonym]

The problem with it is that because it's functional you often end up restructuring half the program for what would have been a trivial change in an imperative language.

While I don't disagree with this, there are some counter-arguments too:

• If you find yourself doing that, you may have written your original program in an imperative style in the first place. Alan Holub's argument about getter/setter methods applies to declarative programming too. If you wrote in a more language-ideomatic style, you might not be facing a huge restructure at all.
• Much the same problem can happen in imperative languages, only the class of changes which would trigger such a restructure are different. For example, in a non-GC'd language, you may end up restructuring your program if some critical data lifetime changes. Or, instead of restructuring your program, you might prefer to hack it up instead, making it less maintainable. (It might be argued that languages like Haskell, which discourage this kind of hackery, might be a good thing in the hands of a certain kind of programmer.)
• Even if you do have to restructure half the program, tools like Haskell's type system make this a less painful task than it would otherwise be.

Knowing a language also means knowing what kinds of changes are painful and what kinds of changes are not. Knowing this in advance helps you write your programs to be more future-proof.

Re:Haskell just won't cut it

[pnot]

me: Why on earth would the users of a piece of software care about the language it's written in?

you: If your users are FOSS developers, they quite likely care about the ability to modify the tool, which includes caring about the languate in which it is written.

Interesting point. Certainly FOSS developers care about being legally allowed to modify code, but I'm not sure that they care, on the whole, about the language.

emacs, for example, is largely written in elisp -- hardly a mainstream language. Yet it's extremely popular, even among people who don't know any lisp. People who find the need to extend it get a good excuse to learn lisp in a well-motivated, incremental way.

Speaking personally, I'd be ''more'' inclined to hack on a project written in something interesting like Haskell, ML, Smalltalk, or Lisp. (In fact I chose my current main project partly as an excuse to learn Lisp.) A lot of people like having a motivation for learning a new language, or a practical use for an "academic" language they happen to know. (I learned Haskell in university, so I'd be quite keen to get to use it in the "real world".)

I think the only time I'd care about the language of a program I'm using would be if it were written in something particularly horrible -- "urgh, if I ever want to modify this I'll have to learn befunge!" But perhaps that's the way some people view Haskell ;-).

It's a matter of taste, I suppose. I do acknowledge that I'm a bit of a language nut.

I still maintain that quality is more important than quantity, though. I've been teaching C and Java to second-year undergraduates this year -- having seen some of their code, I can safely say that if I were starting an OSS project, I'd rather have one seasoned Haskell hacker on board than the entire lot of 'em :-).

Needs wider adoption

[haeger]

While darcs is nice it needs wider adoption. When it comes to a project that people are working on, you have almost as many boxes as you have developers and for a revision control program to be adopted and used there has to be binaries for all those devels. AFAIR there are some issues with the win32 binary? One of our devels had major problems with it and now we're living with both a cvs and a darcs repository, and noone really knows where to send patches. I think it's safe to say that our project is dying, if not dead already.

Not that I blame darcs or anything, just that one need to be sure that darcs work for everyone before commiting to it. CVS works on all platforms and is well tested. Darcs will hopefully get there.

And yes, I did my part and created a package for my platform. It's linked from the binary download page.

.haeger

Re:Darcs is KISS

[Earlybird]

• One of the nice things about subversion (recently converted user, very happy so far) is the support for multiple url formats and communications methods.

Darcs and Arch both have this. (Arch undoubtedly has the most extensive protocol support of any revision control system.)

• Another notable thing (for windows users) is TortoiseSVN, (an explorer shell extension) which is just great.

Tortoise is quite nice indeed -- I used TortoiseCVS for years.

• I can see how the distributed, multi-repo model of bitkeeper/darcs/arch is superior but svn looks good if you only need single-repo.

Need is just one aspect of the development process; right now CVS gives most people what they need, despite the cracks in the lacquer. Darcs doesn't just erase the cracks, but improves the process.

For example, I occasionally submit patches to certain open-source projects. The easiest way to do this is to check out the CVS repository, make my changes, and do "cvs diff -u" to get the patches in that format, which I tend post to some Bugzilla server or email to somebody. But I can't commit them. I don't mean to the master repository -- I mean locally. There's no way I can bundle my file patches in a changeset and keep its history. I'm basically managing a CVS working directory where my changes are never checked in.

With Darcs, I just do "darcs get" to get the master repository, make my changes, commit them locally. I can use "darcs send" to submit my changes to the project maintainer. Anyone else can grab my patches with "darcs get" or "darcs pull". I can be Alan Cox to some Linus without breaking my back over patch management.

Re:Here:

[boa13]

One important feature is missing from the page:

Support for signing patches and archives

Allows to verify who created/commited the patches. Allows to verify the integrity of a repository in case of compromise.

* Arch: Excellent. Each patch can be signed, repositories can be fully verified.
* Darcs: Incomplete. Patches sent by email can be signed so the recipient can verify the identity of the submitter. No support for verifying repository integrity. [1]
* Subversion: N/A

[1] Problem is: You can only sign something that will not vary once distributed, Darcs patches vary once distributed.