GNOME considers DVCS choices

THE GNOME project has completed a survey of active contributors about which distributed version control system (DVCS) they would prefer to switch to in 2009. The project is now in the process of interpreting the results and deciding on the next steps. As the process unfolds, it provides a vivid snapshot of how free and open source software (FOSS) contributors regard DVCS applications.

DVCSes are an idea that have taken hold strongly in the FOSS community in the last few years, and that is starting to edge out older version control systems such as CVS and Subversion. Unlike such older, centralized version control systems, DVCSes do not require a single repository. Instead, on a DVCS, all contributors have their own repositories, and decision-makers decide which ones to merge for a release. In general, DVCSes are seen as more flexible, although critics argue that they can be too decentralized, and that they cause more conflicts than traditional version control systems during merges. All the same, they continue to be popular, partly because of the high-profile of Git, the DVCS originally written by Linus Torvalds for Linux kernel development after the switch from BitKeeper.

The GNOME survey was privately distributed by Behdad Esfahbod, one of the directors of the GNOME Foundation, in December 2008 to gather background information for a possible move away from Subversion, the version control system used by most of GNOME. The survey was distributed to GNOME contributors with Subversion accounts and SSH keys — a total of 1083 people in all, of which 579 replied.

The survey asked those who replied about their current use of Subversion, as well as their role within GNOME, what DVCSes they used or with which they were familiar, and how they felt about switching from Subversion. They were then asked to rank their preferences for Git, Mercurial (often abbreviated to Hg, after the abbreviation for the element Mercury), Bazaar (Bzr), and Subversion.

Analyzing the results

On 3 January, 2009, Esfahbod published the raw results of the survey. Over the next few days, the results were analyzed by a number of people, including Shaun McCance, Andy Wingo and Elijah Newren, all of whom charted the results. Newren's analysis in particular has become a center for discussion about the survey, undoubtedly because it was the most exhaustive analysis.

Looking at Newren's charts, viewers can quickly see approximate but definite results (those who want more exactness can refer to the raw data). The result is a thorough picture of GNOME contributor's views on DVCS.

Newren cross-correlates survey results in every possible way, but here are some highlights:

• About 60% of respondents claimed familiarity with Git, and 25% with Bzr and 20% with Hg.
• 37% of respondents preferred to switch from Subversion, 34% were indifferent either way, and the rest either support Subversion or did not want to switch.
• 48% chose Git as their first choice, and 25% preferred Subversion. Bzr was favored by about 12%, and Hg by 7%, while 5% expressed no preference. Among second choices, Subversion, GIT and Bzr were all between 20-25%, and possibly within any margin of error. However, in the third through fifth choices, Bzr and Hg were favored as much or more than Subversion or Git.
• Preferences for different roles in the project were clearly defined: Package maintainers and coders steadily preferred Git and Subversion. Translators and testers held the same preferences, but less strongly. Surprisingly, documenters preferred Bzr, but, since only four documenters replied, the validity of that result is questionable.
• Those who wanted to switch, or were indifferent both strongly favored Git.

Newren's conclusion was that "there's a strong preference in the community toward switching, and that git has a strong lead in preference among the community."

As might be expected in an online discussion of FOSS, replies to both Esfahbod's publication of results and Newren's results were not long in coming. Still other comments were posted below LWN's brief mention of Newren's analysis.

Replies to the survey and analyses

Probably the most common criticism was that the survey was as much a popularity contest as anything else. Several commenters also wondered why other DVCS software such as Monotone and Darcs were not included in the survey (a question that, so far, no one has answered). In addition, some commenters were quick to talk about Git's shortcomings, while others — obviously unfamiliar with Git — asked questions about its features.

On the GNOME desktop-devel list, Esfahbod's announcement produced dozens of replies. One of the most articulate was by Andrew Cowie, who maintained that "The way the whole survey exercise was conducted it was impossible for Git to lose", and that the desired results were plain beforehand from discussion on the #gnome-hackers channel. Cowie also complained about the fact that GNOME contributors to projects that do not use Subversion were not invited to participate, and that the survey required listing preferences for all choices when he would have preferred not to vote for Git, Mercurial, or Subversion at all.

Much of the discussion below Esfahbod's announcement of the results, did quickly come to center on the assumption that a move from Subversion was now inevitable. Some questioned whether GNOME had the resources to devote to such a move, while others volunteered to be part of a task force to plan and implement the move. Another possibility raised was hiring someone to coordinate the change. Still others attempted to rough out the steps needed to move away from Subversion.

However, while most assumed — however reluctantly — that a change to Git would happen, others tried to raise alternatives. Some suggested that each separate GNOME module should be allowed to use its own DVCS, which others argued would discourage new contributors.

Others championed a suggestion raised on John Carr's blog that GNOME use its bzr-playground server and create plug-ins so that developers could use the DVCS software they personally preferred. However, this idea was dismissed by Esfahbod, who in a reply to the discussion about his announcement condemned such "hacks [developed] in house" because of the potentially high-maintenance they might require in the future.

In the end, however, the usefulness of these criticisms and alternate suggestions is limited. As Esfahbod states: "[...] This thread is not about making decisions. This thread is about giving those making the decision input they need to consider. Who makes the decision? Those who actually have to implement, oversee, and maintain any change" — in other words, the GNOME Foundation directors, and the project's release team and system administrators, the ones who commissioned the survey in the first place. The unspoken assumption in the survey appears to be that the move from Subversion is inevitable, and only the details are up for discussion. So far, though, those details have not been finalized, either by consensus or an announcement.

Meanwhile, for those unaffected by the decision, the survey and the resulting commentary provides a seldom-seen insight into the mindset of active members of the GNOME project — and, very likely, of active FOSS contributors in general.

---

(Log in to post comments)

       familiar with
bzr&git    bzr&hg     git&hg  
54  git    16  bzr    33  git 
27  bzr    14  git    18  hg  
 7  hg      9  hg     10  bzr 
 5  svn     2  svn     7  svn 
 3  any                1  any 

       use regularly
git&bzr    bzr&hg   git&hg
10  git    3  bzr   7  hg  
 6  bzr    2  hg    5  git 
 2  any    1  svn   2  svn 
 1  svn             1  bzr 
 1  hg