There are a couple key factors in this decision that seem to be getting confused with each other. I'm new to PD, but I've been using Git for over a year now with my own repositories. In various consulting projects over the last few years I've developed with CVS, SVN, Mercurial, Darcs, Arch and Bazaar-ng. There are a couple facets to the decision.
1) Distributed Development
The first, and I think most important, is the development model. Distributed development supports a different kind of collaboration, that I think is showing itself to be quite advantageous for open source software. It lets groups of people organize themselves, get a piece of code working, and then it can be shared with the world. Permissions and management of a central repository is no longer an issue. It doesn't exist. Instead Hans-Christoph, for example, would manage an official pd-extended branch that had the fully tested features, while everyone else could still share the latest and greatest stuff without having a centralized bottle-neck.
2) Personal development advantages
In git you have the whole repository locally. This takes more storage space, which is incredibly cheap, but saves LOTS of time. SVN can never merge as well as GIT because you don't have past versions of the repository to merge against. Additionally, with git you get local versioning, local branching, and tons of distributed back-ups of the repository. I'm always working on the train, for example, and with Git I can maintain a much better log of my development because I commit all the time without needing internet access. Branches in git, even for huge projects, are basically free. In space and time. This changes the way you develop. You can make a branch for any little idea you want to mess with or any tricky feature you are working on. It's quite easy to then merge these branches, share them with friends, update them from other's repositories etc. This is a level of power and control that you can not get with svn.
3) User Interface
Git was originally designed to function as a back-end suite that would support easy to use front-end utilities. That has changed over time, and it now includes a nice set of commands and tools that make it work just like you would expect. For sure Git doesn't have the GUI plugins like SVN or CVS and it will take some learning, but I think the benefits in the long term will far, far outweight the initial investment.
There is a Tk interface, GitK, which is incredibly useful to visually look at the history of a repository, merges, branches etc. There is also git-web, which lets you view the repository online. Git also lets you publish a repository just by copying a directory to an http accessible location. This means anyone can share their ideas, features, abstractions etc, without setting up a server, getting permission for a centralized location or anything. Also relevant to the user experience is speed. Git blows away all the other systems in terms of speed. It does everything faster, and you really notice this because it changes what you do. In some of the other systems, like Bazaar-ng, committing, pulling and pushing took so long I wouldn't do it that often, but with git it's all so cheap I do it every time I get a new unit test to pass.
4) Technical
Git writes the repository into a highly compressed format, and then it does not mess with the files. Append only is the standard operation, except for when you occasionally compress the whole thing, in which case it can verify sanity. It is also much cleaner to work with in terms of permissions and access, just because of the whole usage model. I've had many experiences with subversion where the repository had to be recovered because of permissions issues and/or corruption. This is very bad, especially since it is a centralized server that everyone is counting on. It takes a root user to go in and run "svnadmin recover", which in my opinion is a command that shouldn't need to exist in something as important as a source repository.
Keith Packard who runs the X.org project wrote a very good post detailing his research and opinions into source control. It's a bit old now, but worth the read: http://keithp.com/blog/Repository_Formats_Matter/index.html
Sorry for the long post, but this is something I have dealt with a lot recently. In my mind distributed versioning is the only proper way to run a modern open source project. PD is a great piece of software with a seemingly cool and diverse group of developers, which seems perfect for a decentralized model. Either way, I think the decision should be made first between a centralized or distributed versioning system, and then the decision can be simplified to figuring out compatibility, usability etc. It's great news that Git now runs on windows though, because I think it is far better than the others. To quote one of Hans-Christoph's recent signature lines:
Mistrust authority - promote decentralization. - the hacker ethic
:-)
Ciao, Jeff
David Plans Casal wrote:
Speaking from experience, having suffered several migrations to prcs from cvs, to svn from cvs, and to darcs from svn, in commercial projects, I have to say the only one that worked well, made sense, and was least painful was cvs-->svn, and I'd vote for that.
David
On 23 Oct 2007, at 19:41, Stephen Sinclair wrote:
are used ... beside that, i am not sure, how access to specific parts of the repository can be restricted ...
You don't have to restrict anything. If someone messes with your work, you simply don't merge what they did. They're free to do whatever they want in _their_ repository, but you're not forced to accept it into _your_ "official" repo for your project.
Of course, other people would still be free to check out what the other guy did, even if it's not officially part of the main tree. The trick here is to think in terms of "pulling" rather than "pushing". Since no one is pushing, no one can step on anyone else's toes.
Anyways, I think under a system like git, the Pd kernel would have its own repo, and other subprojects (abstraction collections, externals, libraries) would have their own repos. A project like "Pd extended" could then simply be a "super-repo" collecting specific repos as submodules. Each submodule would be tagged to a specific version/branch.
I totally understand the hesitation, git does seem somewhat complicated at first. Once I finally understood the whole distributed way of thinking though, I couldn't help but dive in, and now I like the idea so much I think I can't go back... and actually, I discovered that the interface is really not that hard. Mostly you just use "clone", "pull", and "commit".
Another nice thing about it: you don't even have to "officially" switch all at once. Someone can just start a git repo based off the CVS, and you can go from there... whoever does that is responsible for keeping his repo up-to-date with the CVS and vice-versa. He can check his changes back into the CVS whenever he wants but still work with git on his own computer. It allows for a gradual weening away from the central repository instead of requiring everyone to switch at once.
I won't try any more to push the idea, but I think it's worth considering.
Steve
PD-dev mailing list PD-dev@iem.at http://lists.puredata.info/listinfo/pd-dev
PD-dev mailing list PD-dev@iem.at http://lists.puredata.info/listinfo/pd-dev