Hi all, several Apache projects have moved parts of their development to GitHub, some more (no more Jira), some less (issues in Jira, code on GitHub).
I have used GitHub more and more lately and find its processes quite convenient. This mostly revolves around the concept of "pull requests". People submit "pull-requests" instead of patches. Based on these, GitHub allows to do: * convenient code reviews * leave comments (line-based as well as general) * approve/request changes * discuss the pull request as it evolves * even exporting a pull request as a patch is easy Pull requests are essentially branches and it is possible for interested developers to: * check out these branches * collaborate on them * finally merge them into the master (trunk) branch when ready. I have also made some experience setting up Jenkins in such a way that it can automatically check pull requests: * an admin developer leaves a comment on a pull request such as "Jenkins, test this please" * Jenkins sees this comment and triggers a build of that particular contribution * Jenkins updates a "build status" that is visible directly in the discussion thread of the pull request on GitHub. Depending on the result of the build, it will give a red or green light. * That is very useful to do an integration test of a contribution after an initial code screening - no need to locally integrate a patch, locally test it, etc. The initial screening can happen by looking at a side-by-side div on the GitHub website - again no local effort. Finally, GitHub allows to enforce certain processes if desired. E.g. one can *optionally* enforce that: * limit access to specific branches (e.g. master/trunk or maintenance branches) * no changes can go into master (trunk) without being reviewed first as pull requests (a very strong 4 eyes principle, has ups and downs) * pull requests must have passed a Jenkins check before they can be merged All in all, I believe our project would benefit from moving to GitHub. After an initial transition and adaption phase, our processes should become smoother, easier and as a result more interactive. The main hurdle for us developers would in my view be to learn how do to source code management using a distributed approach. The way one has to think about code management in git (or probably any other DSCM tool) is different from a centralized approach like SVN. To take full advantage, branches move into the center of attention and branching/merging becomes a much more common operation than in SVN. Fortunately, git supports such operations very well. The hurdle for contributions would also be lower (more and more people are familiar with GitHub processes, and already have GitHub accounts. Juggling with patch files is (at least for me) a great annoyance). Of course, there is also the draw back that GitHub may shut down at some point (like Google Code did) or start charging or introduce terms-of-service that are incompatible with our goals. But due to the strong interest of various Apache projects on using the GitHub infrastructure, INFRA has been working on various integrations e.g. mirroring GitHub repositories (not sure how they handle it when Apache projects use GitHub issues though). Also at present GitHub appears to be very community-oriented when it comes to the terms-of-service, e.g. making text updates available to the community early, soliciting and incorporating community feedback, etc. On the bottom line, my personal judgement at this point in time is, that the benefits of smoother process and easier community involvement seriously outweigh such risks and that many of the risks have been mitigated by INFRA anyway. What do you think? Best, -- Richard
