Limitiations of GitHub Forks

Assumption:

  1. git is pretty awesome, but it's conceptually complex. As a result using git demands a preexisting familiarity with git itself or some sort of wrapper to minimize the conceptual overhead.
  2. The collaboration methods (i.e. hosting) provided by git, which are simple by design to allow maximum flexibility, do not provide enough structure to be practically useful. As a result providers like GitHub (and BitBucket and gitorious) offer a valuable service that makes it easier--or even possible--for people to use git.

Caveats:

  • there are problems with using centralized repository services controlled by third parties, particularly for open source/free software projects.

    There are ways that GitHub succeeds an fails in this regard. but this dynamic is too complex to fully investigate within the scope of this post.

  • If you use GitHub as designed, and the way that most projects use nGitHub, then you have a very specific and particular view of how Git works.

    While this isn't a bad thing, it's less easy to use git in some more distributed workflows as a result. This isn't GitHub's fault so much as it is an artifact of people not really knowing how git itself works.

Assertions:

  1. GitHub's "fork" model[^fork] disincentives people from working in "topic" branches.

  2. By making it really easy for people to publish their branches, GitHub disincentives the most productive use of the "git rebase" command that leads to clean and clear histories.

  3. There's no distinction between a "soft fork" where you create a fork for the purpose of submitting a patch (i.e. a "pull request") and a "hard fork," where you actually want to break the relationship with the original project.

    This is mostly meaningful in context of the other features that GitHub provides, notably the "Network" chart, and the issue tracker. In a soft-fork that I would intend to merge back in, I'd like the issues to "come with," the repository, or at least connect in some way to the "parent." For hard forks, it might make sense to leave the old issues behind. The same with the network chart, which is incredibly powerful, but it's not great at guessing how your repository relates to the rest of its "social network."

The solution: keep innovating, keep fighting lock-in, and don't let GitHub dictate how you work.

comments powered by Disqus