This documentation was written to describe the 1.7.x series of Apache™ Subversion®. If you are running a different version of Subversion, you are strongly encouraged to visit and instead consult the version of this documentation appropriate for your version of Subversion.

To Branch or Not to Branch?

To branch or not to branch—that is an interesting question. This chapter has provided thus far a pretty deep dive into the waters of branching and merging, topics which have historically been the premier source of Subversion user confusion. As if the rote actions involved in branching and branch management aren't sometimes tricky enough, some users get hung up on deciding whether they need to branch at all. As you've learned, Subversion can handle common branching and branch management scenarios. So, the decision of whether or not to branch a project's history is rarely a technical one. Rather, the social impact of the decision often carries more weight. Let's examine some of the benefits and costs of using branches in a software project.

The most obvious benefit of working on a branch is isolation. Changes made to the branch don't affect the other lines of development in the project; changes made to those other lines don't affect the branch. In this way, a branch can serve as a great place to experiment with new features, complex bug fixes, major code rewrites, and so on. No matter how much stuff Sally breaks on her branch, Harry and the rest of the team can continue with their work unhindered outside the branch.

Branches also provide a great way to organize related changes into readily identifiable collections. For example, the changes which comprise the complete solution to a particular bug might be a list of non-sequential revision numbers. You might describe them in human language as revisions 1534, 1543, 1587 and 1588. You'd probably reproduce those numbers manually (or otherwise) in the issue tracker artifact which tracks the bug. When porting the bug fix to other product versions, you'd need to make sure to port all those revisions. But had those changes been made on a unique branch, you'd find yourself referring only to that branch by its name in conversation, in issue tracker comments, and when porting changes.

The unfortunate downside of branches, though, is that the very isolation that makes them so useful can be at odds with the collaborative needs of the project team. Depending on the work habits of your project peers, changes made to branches might not get the kind of constructive review, criticism, and testing that changes made to the main line of development do. The isolation of a branch can encourage users to forsake certain version control best practices, leading to version history which is difficult to review post facto. Developers on long-lived branches sometimes need to work extra hard to ensure that the evolutionary direction of their isolated copy of the codebase is in harmony with the direction their peers are steering the main code lines. Now, these drawbacks might be less of an issue for true exploratory branches aimed at experimenting with the future of a codebase with no expectation of reintegrating the results back into the main development lines—mere policy needn't be a vision-killer! But the simple fact remains that projects generally benefit from an orderly approach to version control where code and code changes enjoy the review and comprehension of more than one team member.

That's not to say that there are no technical penalties to branching. Pardon us while we go meta for a bit here. If you think about it, every time you checkout a Subversion working copy, you're creating a branch of sorts of your project. It's a special sort of branch. It lives only on your client machine; not in the repository. You synchronize this branch with changes made in the repository using svn update—which acts almost like a special-cased, simplified form of an svn merge command.[36] You effectively reintegrate your branch each time you run svn commit. So, in that special sense, Subversion users deal with branches and merges all the time. Given the similarities between updating and merging, it's no surprise, then, that the areas in which Subversion seems to have the most shortcomings—namely, handling file and directory renames and dealing with tree conflicts in general—are problematic for both the svn update and svn merge operations. Unfortunately, svn merge has a harder time of it precisely because of the fact that, for every way in which svn update is a special-cased, simplified kind of generic merge operation, a true Subversion merge is neither special-cased nor simplified. For this reason, merges perform much more slowly than updates, require explicit tracking (via the svn:mergeinfo property we've discussed in this chapter) and history-crunching arithmetic, and generally offer more opportunities for something to go awry.

To branch or not to branch? Ultimately, that depends on what your team needs in order to find that sweet balance of collaboration and isolation.

[36] Actually, you could use svn merge -rLAST_UPDATED_REV:HEAD . in your working copy to quite literally merge in all the repository changes since your last update if really wanted to!