Around the end of March 2013, BioJava will move from SVN to GIT for version control. This page should be viewed as a proposal for how the migration process will progress. It is a work in progress. Please share your comments on the talk page or on the mailing list.
Benefits of the move
- Use branches in your daily workflow without merge terror
- Track buggy/untested code without breaking the build for everyone else
- Collaborate within a small team before sharing the finished feature
- Simple, well integrated bug tracking
- Accept contributions from the public via pull requests
The primary BioJava SVN repository requires an ssh account for each developer. As described in Developer Code Access, the repository url is
svn co svn+ssh://dev.open-bio.org/home/svn-repositories/biojava/biojava-live/trunk/
There are also two read-only mirrors which permit anonymous access. The recommended public access point for SVN is
A commit hook is used to update a github mirror after each commit. This repository is read-only (by policy). Although it is possible to develop from a GIT clone & submit push-requests with new code, none such requests are accepted. Github repositories can be accessed using either GIT (recommended) or SVN
svn co --depth empty
svn up biojava/trunk
Neither of these methods are recommended, although they have been suggested to work around open-bio.org server problems.
Linking SVN to GIT
As an open source project we want to make sure that contributors are fairly acknowledged. Moving to github requires using github accounts to identify developers. This means linking old SVN account names to a github account so that past and future contributions are linked to the same person.
- Wherever possible, map SVN accounts to Github accounts. Most active developers have already added themselves to the google spreadsheet to enable this (see mailing list)
- If no Github account is known, associate SVN account with their email address. Developers can link this email to a github account at a future time, which will automatically tie them to past commits
- If no email address is known, or the email is no longer accessible, commits will be credited to their SVN account name only.
We currently have two classes of SVN permissions:
- Developers. Full write access to any Biojava project through the developer SVN.
- Users. Read-only access to the anonymous mirrors.
The decentralized nature of GIT will allow a third class of users:
- Developers. Members of the Biojava Github organization. Can configure push access on a project-specific level.
- Contributors. Github users who are not part of the biojava organization, but maintain a personal fork of the project & submit push requests
- Users. General users who clone the code & don’t submit push requests.
Current developers may find it easier to submit push requests rather than join the Biojava organization. However, frequent contributors should join so that the burden of accepting pull requests does not become too high for the administrators.
On Sat, April 29, 2013:
- Generate authors file for svn2git based on known github accounts or email addresses
- Set SVN to read-only
- Delete the existing git repository. All forks of the current repo will become incompatible.
- Use svn2git tool to create a new git repository, converting author information along the way
- All developers checkout git repositories & begin pushing changes to github
It is unfortunate that the 12 existing forks (plus local clones) will be incompatible, but none of the public forks have diverged significantly from the trunk so this should be acceptable.
Unlike SVN, each project will have a separate repository under the BioJava organization. Only actively developed projects will be moved to github, with inactive projects (eg DASRepository) remaining on the read-only SVN server. Specifically, the following projects will be moved to github:
- biojava-live, renamed to biojava
- RCSB_SequenceViewer, renamed to rcsb-sequenceviewer
- RCSB Viewers, renamed to rcsb-viewers
The github repository for each project will contain the following branches, based on a variant of the git-flow model:
- master Current development branch, corresponding to SVN trunk. All active developers are free to make commits directly to this branch, as well as merging feature branches and pull requests
- release By definition, any commit to this branch counts as a release. Each commit should be tagged with a new version number, eg ‘v3.1.0’, and should result in a new set of jar files getting uploaded to the wiki. Only project leads should commit to this branch.
Additionally, any number of feature branches may be present should developers wish to collaborate on specific features. The master branch will be marked as the default branch. This is analogous to checking out trunk from svn to get the most recent code:
Specific branches for each release (eg a release-3.1.0 branch containing only bugfixes) could also be used. However, past releases have mostly not had significant bug fixes during testing, so this may be unnecessary complexity.
The SVN repository used SVN 1.4, which does not store information about merges. This made it extremely difficult to deal with the major refactoring that occurred as part of the Maven Migration. Another problem was that a number of branches don’t contain the full trunk (only a single project), and git lacks a mechanism to deal with partial checkouts. As a result, no history prior to September 2009 (r7227) was migrated to Git. This is unfortunate, but it would have been extremely difficult and time-consuming to document all the large moves and refactors from BioJava’s history in a way consistent with git.
The anonymous SVN will continue to be accessible as documentation of the full history in SVN.
 Github blog. Announcing SVN support. Accessed 2013-03-18
 Github blog. Collaborating on Github with Subversion. Accessed 2013-03-18