Slashdot Mirror
FreeBSD Begins Switch to Subversion
An anonymous reader writes "The FreeBSD Project has begun the switch of its source code management system from CVS to Subversion. At this point in time, FreeBSD's developers are making changes to the base system in the Subversion repository. We have a replication system in place that exports our work to the legacy CVS tree on a continuous basis.
People who are using our extensive CVS based distribution network (including anoncvs, CVSup, cvsweb, ftp) will not be interrupted by our work-in-progress. We are committed to maintaining the existing CVS based distribution system for at least the support lifetime of all existing 'stable' branches. Security and errata patches will continue to be made available in their usual CVS locations."
Comment removed based on user account deletion
SVN really is dead. While I could understand the move a few years ago, SVN is antiquated technology now. I'm not sure why anyone would want to use it in a new rollout when there are newer and significantly better options available now.
They don't use git because FreeBSD development has traditionally been "centralized". They use a model where patches are fed to a group of core developers with "commit permission" to the tree, and all source changes are vetted and fed through that funnel. Subversion's centralized source control methodology works well with the FreeBSD development process, and the decentralized aspects of git is not needed.
However, of course, there is still some distributed coding going on at the edges, but they tend to be peripheral and experimental. The developers working on these experimental branches can choose to use whatever source control system they wish. Many FreeBSD developers prefer perforce for their experimental work, but they can use git or mercurial if they wish.
Subversion has an Apache/BSD type license. GIT does not.
Emacs is also GPLed, but that doesn't mean everything I write with it is also automatically GPLed. Unless the BSD people are planning to become Git developers, the source license makes no difference. Try reading it sometime.
Ita erat quando hic adveni.
GIT lost the version control war early on. Its focus on Linux development with little to no support for Windows and Mac made it unpopular. That's a situation that has changed (somewhat), but the stigma is still attached to it. Which is not really a problem. GIT was developed to meet the needs of the Linux Kernel Project. If it happens to meet the needs of other projects, great. If it doesn't, that's just as fine.
In any case, Mercurial ended up being the "best of breed" solution. It offered all the features of the competing version control systems, was portable across platforms, had a significant toolchain appear practically overnight, and is used by HUGE OSS companies like Sun and Mozilla. I've used it in my own projects and have found that it is much easier and more dynamic than the classic, monolithic model of CVS.
Javascript + Nintendo DSi = DSiCade
Path of least resistance; it works much like CVS, it fits in with existing infrastructure, and everyone knows how to use it.
git isn't terribly well suited to very large monolithic projects; you need to split into multiple smaller projects since it tracks entire trees rather than single files. When your tree is 1.3GB+ and has upwards of quarter of a million files that's rather painful either way.
It also isn't well suited to rewriting history, e.g. in the case when you have to remove a changeset because it violates someone's patent or copyright; you can rewrite the repository to remove it, but you end up renaming every commit afterwards, since their names are SHA1's dependent on every previous commit, generating tonnes of churn in many different places as the whole of history basically disappears and reappears elsewhere.
Many of git's advantages can still be leveraged with SVN; git-svn works pretty well, and it doesn't require massive upheavals in all areas of the project.
Thanks for promptly settling the SCM dispute! Now I'd love to hear your ideas on which text editor is the best.
Cretin - a powerful and flexible CD reencoder
Git. It's Linus Torvalds fix for his own software management nightmare with the Linux Kernel.
A talk he did on it
http://en.wikipedia.org/wiki/Git_(software)
You can use a GPL licensed product in any way you wish.
GPL imposes limits on you only if you distribute GPL code (or the result of its compilation)
Well, if you believe Linus, it must be because they are ugly and stupid :)
Mercurial is the solution currently in use at Mozilla, Sun, and quite a few Linux projects. (Though not the main kernel branch. That's GIT. Think something more along the lines of ALSA.)
Mercurial is interesting because it encourages teams to work together, pushing and pulling source code from each other. When the source reaches a stable point, you can push that to a central repository for building and archiving.
The interesting aspect about this design is that it actively encourages branching! Rather than treating branches as a special thing that needs to be done under a certain set of circumstances, it treats every copy of the repository as a branch. So developers can work independently. When they come back together, the tool is able to auto-merge most projects back into a single whole.
Mercurial is able to do this because it tracks the point of divergence. With that information, it can see if any of the changes truly conflict. 95% of the time, there is no conflict and Mercurial is able to merge the files auto-magically. The other 5% of the time, Mercurial will launch a merge tool and make you answer YES/NO to each difference. This process is amazingly smooth.
The key thing to keep in mind with Mercurial is that you won't want to keep all your source in one repository. (Like most companies do with CVS.) Keep a separate repository for each project or module. You can keep the repositories all in the same path, but it's much easier to work with only the code you need rather than copying around a 10GB source tree from developer to developer.
If you do decide to try Mercurial and are given a Windows development machine, I highly recommend TortoiseHG. You'll occasionally have to run 'hg update' from the command line (the tool will prompt you), but it's otherwise a very slick way of working with Mercurial repositories.
Oh, and don't use the CVS->Mercurial conversion tools. It leaves CVS-style droppings all over creation. Just import the latest codebase and keep CVS running in read-only mode for as long as you need historical data.
Javascript + Nintendo DSi = DSiCade
Huh? Git didn't have windows support very early on but very soon you could compile it with Cygwin. As for Mac support... it got it very quickly. In fact, I'm willing to claim that not having very early Win support didn't do anything to adoption rate. Target audience were Linux hackers so having support for various other systems wouldn't have done much at all.
As for "changed (somewhat)", what do you mean somewhat? I use Git daily on my Mac laptop and have even used it under Vista without any issues.
Git is HUGE these days. Rails project, for example, has completely switched to Git. Same is happening to Pythin community as more and more things are moving to places like GitHub (which is amazing btw). Also, Google Code now provides Git repos for almost ALL of the projects.
As for Hg, it's lost the war. Git has won. If you want proof, try some searches for "git tutorial" and "mercurial tutorial" and see who's winning. Searches with " tutorial" appeneded are great because they indicate adoption rate and indicate that there are people out there who are trying to get started.
in short, Git has already won and expect it to be the biggest source code versioning system in less than two years from now.
I've never used Mercurial before, so I don't know if the comparison fits exactly, but the branch/merge style of development you've described can be done very easily within CVS and SVN. That sort of thing is much cleaner in SVN as you're branching from a base repository revision rather than from a tag that can be moved/reapplied, but it's possible if you're careful in CVS as well. From there, it comes down to personal preference with regard to whatever merge tool you care to use.
I greatly prefer that style of development myself, as it tends to keep the trunk much cleaner, and allows a code reviewer to concentrate on specific changes in a given branch rather than worrying about the stability of the codebase as a whole. Of course, old branches can be a bitch to merge if you've got really active development, but there's probably no way around that no matter what tools you're using.
I think any distributed version control system (ie, pretty much all the new free ones except SVN) will encourage branching, lack of a central authority just about requires this.
Debian's popcon agrees with this synopsis.
Basically, in the first year or so of the "distributed SCM revolution" in 2005, Mercurial was in the lead. However in mid 2006 git passed it by, and took off exponentially. At the moment Mercurial usage is about half that of git. It looks like within a year or so git will pass rcs and soon after that cvs, to become the second most popular scm after svn.
It seems that in general, Mercurial is chosen by comities (Solaris, Mozilla etc.) as it offends the least number of people due to its earlier better windows support. However, in projects run by a single benevolent dictator (like Wine, linux kernel and X.org) git tends to win due to its better technology and the lesser importance of political issues. (The fact that Linus originally wrote git will definitely hold back its usage in non-linux OS's due to politics. I can't see Microsoft using it, for example.)
Googlefight!
git tutorial: 512,000 results
mercurial tutorial: 1,100,000 results
Winner: Mercurial!
Google doesn't provide JACK for GIT. GIT uses SVN. In order to use GIT with Google, you need to have a GIT->SVN translator:
http://nigel.mcnie.name/blog/using-git-for-your-sourceforgegoogle-code-project Installing a Cygwin environment is not a supportable solution for most corporations. They needed native solutions. Something which has begun to appear.I agree with you wholeheartedly. The target audience is Linux hackers. They are the ones using GIT. The business world, OTOH, has chosen Mercurial. Such is the way of things.
Javascript + Nintendo DSi = DSiCade
If you go by Debian's popcon, Subversion has already kicked both our asses and we should all go home:
http://people.debian.org/~igloo/popcon-graphs/index.php?packages=darcs%2Cgit-core%2Cmercurial%2Cbzr%2Csubversion%2C+cvs&show_installed=on&want_legend=on&want_ticks=on&from_date=2003-10-01&to_date=&hlght_date=&date_fmt=%25Y-%25m&beenhere=1
Javascript + Nintendo DSi = DSiCade
If you are used to CVS then Subversion is definitely a step up, and it will be very familiar to your users. What's more it is well supported by IDEs and has piles of other tools like Tortoise that makes it easy for non-developers to use. Heck, if push comes to shove it can even be used as a WebDAV share with the advantage that it will automagically version your files.
The downside of Subversion is that it isn't very good at merging. Merging branches in current versions of Subversion is a manual process that is ridiculously painful. This can be mitigated somewhat using the svnmerge Python script, but even with the script merging still isn't as easy as any of the distributed version control systems. For people like Linus Torvalds that's basically a showstopper for Subversion. To them merging is basically the whole point of a version control system.
It's quite possible, however, that you have different needs than the Linux kernel. For example, none of the distributed version control systems deal well with large files. If you want to store multi-media files along with your source then Subversion is basically your only option. Likewise, if you plan on having designers or random office workers use your repository then you can forget about the distributed tools.
It is definitely possible to do branch/merge work with CVS/SVN, but you need to track the merge points yourself, and be aware not to merge the same changeset twice. Git and Mercurial make the process far easier by tracking merge points for you.
Also, one big thing distributed source control gives you is the ability to "clone" or "fork" your own copy of the repository. This lets you make small, incremental changes that you can roll-back and merge back as you like, and then you can "push" all of your commits back to another fork (presumably a dev trunk somewhere). As an example, check out this visual representation of the many forks of the Rails project on Github: http://github.com/rails/rails/network
You should ALWAYS use quotes for specific phrases. if you've done some SEO work or actually tried paying for AdWords, you'd realize how important phrase searches are on Google.
Everyone we hire is either already experienced with SVN or picks it up in about 5 minutes. I don't know what the OP is smoking.
Often wrong but never in doubt.
I am Jack9.
Everyone knows me.
Contrary to Linux, FreeBSD uses a centralized development methodology which SVN is very well-suited for. (Now, if only they hurry up with the stable release of their merge-tracking, I could say that Subversion is perfect for a centralized development model.)
Then, there's the obvious licensing issue. GIT is released under the GPL, which, I'd guess, is a little too restrictive for BSD-style people. :P
Morality is usually taught by the immoral.
If you are still using CVS, or CVSup, for updating your ports tree, I suggest you have a look at portsnap(8):
PORTSNAP(8) FreeBSD System Manager's Manual PORTSNAP(8)
NAME
portsnap -- fetch and extract compressed snapshots of the ports tree
bash$
Stop Computers/Cars Analogies on S
There are different tools for different purposes:
:-)
CVS is (was) what the central repository uses to store the software.
Perforce is a central repository for internal development. That way the limitations of CVS for this part of the job don't limit the developers.
But Perforce is commercial software and you can't push it on to the community.
Subversion is a free software which has the capabilities which are set as a requirement for the FreeBSD project. It has some capabilities of Perforce, it has some capabilities of CVS and it can be integrated in the current distribution framework.
Oh, and it understood most of the FreeBSD CVS Repository
bash$
For the forseeable future, yes. Commits to SVN are being replicated into CVS, so all the existing CVS infrastructure will continue working.
Tarsnap: Online backups for the truly paranoid
That wasn't any factor in the discussions about the switch as I remember. It was mostly, "What do people like and what will satisfy the most people."
Windows or Unix shop? How large do you expect your repository's history to get (in terms of revisions per year)?
For a UNIX shop with a very deep history, you're going to be best off with Git.
For a UNIX shop without an exceptionally deep history, Bazaar (Canonical's revision control system) is worth considering. It works well on Windows, but the GUIs there are immature -- so if you're not a UNIX shop and your Windows folks are revision control power users who hate the command line (for whom the currently available GUIs aren't adequate but the command line is unsuitable), that's something which will need to be taken into consideration.
For a Windows shop where you need the best available GUIs, Subversion has the stability you need right now -- but you're missing out on functionality which can save a lot of headaches later (when, say, you try to merge between a branch where you did a mass renaming and one where you didn't... though Git doesn't always handle this one well either, Bazaar takes care of it beautifully). SVN 1.5 will be slightly better along those lines, and bzr-svn (allowing Bazaar to integrate with Subversion repositories) can help your UNIX power users take advantage of distributed revision control's advantages while still keeping all the SVN shininess available for your Windows folks.
That said -- SVN set out to be the best replacement for CVS it could possibly be. It's a very nice, very polished replacement for CVS -- but that's not the same thing as "the best revision control system it could possibly be"; distributed workflows have some nontrivial advantages even in centralized environments; look at the PQM the Bazaar folks use to manage their own project for a prime example.