This post is from the CollabNet VersionOne blog and has not been updated since the original publish date.
Subversion or Git? Decisions, decisions ….
Are you facing a difficult choice between version control systems? Are you having trouble sorting out the alternatives? Are you bewildered by all the opinions you find on Google? Let’s see if I can help sort this all out.
CollabNet (the original stewards of the Subversion project) provides both Subversion and Git, either through our enterprise TeamForge product or through our CloudForge Platform. We think there’s a place in the universe for each. Here’s how to find yourself in that universe.
The short form
Here are some specific recommendations. If you want to stop reading at the end of this section, you should make out just fine.
- If you have compelling requirements for a single, certain, master copy of your work, use Subversion. You can do this with Git, so long as there are no slip-ups. But you can’t do anything else with Subversion (slip-ups or no), and “compelling requirements” like Sarbanes-Oxley are happier with guarantees than possibilities.
- If you plan to maintain parallel, largely shared but permanently somewhat different lines of the same product, use Git. One common example: perhaps you have a large product that you customize for each customer. The customizations are permanent, and generally not shared among code lines, but most of the code is common to all. Git was designed for just this case (in Git terms, local customizations to the common core, and occasional feature or bug-fix contributions back up-tree).
- Neither of those? Take your pick, you should be fine with either tool.
What do you want from your version control system? A lot of this is the same for everyone, and both tools accomplish these tasks just fine. If all you care about is the basics, you could easily just flip a coin and get on with your work, confident that your choice would be sound. These “basics” include:
- Store all versions of your files (“version control”)
- Associate versions of each file with appropriate versions of all other files (“configuration management”)
- Allow many people to work on the same files, toward a common goal or release (“concurrency”)
- Allow groups of people to work on substantially the same files, but each group towards its own goal or release (“branching”)
- Recover, at any time, a coherent configuration of file versions that correspond to some goal or release, either for investigation or extension like bug fixing (“release management”)
Both systems do all these things quite well. If these things are truly all you care about … flip that coin! All the discussion and decision has to do with other details, details that are always secondary to the basics, but can become crucial to particular projects or environments.
So, if you haven’t flipped that coin and gone away, let’s look at the differences.
But first, some demythologization
Here are a few things you may have heard that just aren’t true, or aren’t true any longer:
Git hates windows. No, it doesn’t. It used to, but now there are good Git integrations with Windows Explorer (TortoiseGIT, SmartGIT) and most of the major IDEs (Visual Studio, Eclipse, Netbeans), and you no longer need to use the Unix emulation environment Cygwin. Git is only for hackers. As a Unix-centric, command-line only tool, Git was originally anathema for many workers. But with all the GUI integrations, it’s considerably more friendly, within reach of nearly everyone (see Windows integrations above, plus non-Windows tools like GitX, Coda via GitX or Tower, Emacs). Git is hard to learn. Well, it’s a lot easier to learn now, anyway, thanks to those GUIs, and to improvements in the command-line UI and documentation.
On the other hand
Subversion is slow on Windows. The latest Subversion releases, up to the current 1.7, have made great strides in Windows performance. Subversion merging is hard. Well, it’s a lot easier now, anyway, thanks to the continuing progress on “merge tracking.”
So, what’s different?
As you’ve surely heard, the key difference is that Subversion is “centralized,” while Git is “distributed.” You can go many other places to learn about what that means in their implementation, I don’t spend time on that here. But I will talk about why it matters to you.
Git’s distributed model means that you have full version control operations purely locally on your workstation. Of course, the changes you make locally are not visible to anyone else, until you do something else (Git’s “push” or “pull”), but you can check changes in so they’re not lost, create branches, merge among branches, back up to older versions, and browse and compare versions — all without a network. The one cost in all that is, you have to learn some additional commands (push, pull, clone) and workflows, if you ever expect anyone else to see your work. You can do everything locally that any version control system can do … and you also have extra steps required to make your work visible to others, kind of an inescapable trade-off. On balance, local versioning is a significant convenience for the developer, and a big part of the popularity of Git.
Most processes have some reason to want one copy of everything, somewhere, that’s reliably known to be “the official version.” With Subversion, that’s the central repository. With Git, that’s … some repository or other: the tool doesn’t privilege any one repository over another, but users and conventions can. With Git, you have to remember to push or pull all changes into the designated central repository — not hard to remember, especially if you support it with other conventions, such as only building releases from the central, but still an extra step. With Subversion, there’s nothing to remember or agree to: there is only one repository. By definition, anything that’s checked in at all is checked in to the central. One place where this can become particularly important is in “governance,” where external constraints like Sarbanes-Oxley or escrow contracts demand hands-off guarantees. On balance, enterprises with governance constraints value the guarantees of Subversion.
(Governance, traceability, ALM, and DevOps all require broader support than just the code and files, of course: there has to be integration among the code repository, the tests, the deployed product, and the tracking system. Fortunately, CollabNet now integrates both Subversion and Git with these other components.)
Somewhat paradoxically, there are times when you want to throw away (or keep private) some work. Git is better at throwing things away than Subversion is: you keep the potentially throw-away work in its own repository until you decide whether to make it official. If you decide to toss it, there are virtually no left-overs in the repositories you keep. By contrast, in Subversion, everything goes into the one central repository, and you don’t have that same option. Typically in Subversion, you don’t actually “throw away” such work, but only leave it on some branch you never look at. But it’s still there, taking up space, and possibly cluttering the history browser. The possibility of throw-away work is a fairly big consideration in open-source work (and in fact it was one of the key design considerations for Git), but rather less so for enterprise work (where throw-away work is also throw-away salaries, equipment, lighting, and all the other employee expenses), so enterprises typically avoid ever doing it in the first place. I wrote about handling throw-away work at some length a while back.
Some things are possible, but I wouldn’t recommend them — at least, not to someone who has to ask:
Git-SVN. Git is able to pull files out of a Subversion repository, store them in a Git repository (allowing all that local version control, and some inter-git-repository push/pull/merge), and then push the results back into Subversion. It sounds like the best of both worlds, doesn’t it? But you end up having to be an above-average expert in both systems, and parts of the workflow are very slow. Some folks really like it. Maybe you would, too. But it’s not what you’d call “mainstream.” Cygwin.Cygwin is a system for providing a lot of Unix-like capabilities on a Windows system. Early Git Windows implementations ran inside Cygwin. Again, this ends up forcing you to be an above-average expert in two very different systems. If you’re already steeped in Unix and Windows both, it can be handy. But it’s a high price to pay for just adding a little Git. Fortunately, Git Windows installation no longer requires Cygwin.
So, which one’s for you? Try a free, fully functional 30 day trial of Subversion and Git on CloudForge to help make this choice.