Branching Model

Pulp lives on GitHub. The “pulp” repository is for the platform, and then each supported content family (like “rpm” and “puppet”) has its own repository.

Pulp uses a version scheme x.y.z. Pulp’s branching strategy is designed for bugfix development for older x.y release streams without interfering with development or contribution of new features to a future, unreleased x or x.y release. This strategy encourages a clear separation of bugfixes and features as encouraged by Semantic Versioning.

Note

Pulp’s branching model is inspired by the strategy described by Vincent Driessen in this article, but is not identical to it.

master

This is the latest bleeding-edge code. All new feature work should be done out of this branch. Typically this is the development branch for future, unreleased x or x.y release.

Version-Specific Branches

Each x.y release will have one corresponding branch called x.y-dev. For example, all work for the 2.7.z series of releases gets merged into 2.7-dev.

Protected Branches

Within each repository, the master branch, any branch ending in -dev, and any branch ending in -release should be marked as protected on GitHub. The basic protection that disallows force-push and deletion is the only option that should be enabled. There should be no restrictions on required status checks or who can push. There is a script at devel/scripts/protected-branches.py that will mark all appropriate branches as protected. Any time new branches are created that should be protected, that script can be run to do the work.

Build Tags

Builds will be represented only as tags.

Note

In the past, the latest beta and GA release of an x.y stream would be represented additionally by branches, but that is no longer the case as of pulp 2.7.

Build Lifecycle

Alpha and Beta releases will be built from the tip of an x.y-dev branch. If the beta fails testing, blocking issues will have fixes merged to the x.y-dev branch like any other bug fix, and then a new build will be made. Other changes unrelated to the blocking issues may get merged to the x.y-dev branch between builds, and no effort will be made to “freeze” the branch. Any such unrelated changes will be included in the next beta build.

Release candidates will be built from the most recent beta tag. GA releases will be built from the most recent release candidate tag.

Hotfix

When a hotfix needs to be made, a branch will be created from the most recent x.y.z release tag. The fix will be made (via pull request from a personal fork to the hotfix branch), a new tag will be built from the tip of the hotfix branch, and the hotfix branch can be merged to x.y-dev.

Bug Fix Branches

When creating a Pull Request (PR) that fixes a specific bug, title the PR as you would a git commit message with a short, human-readable description. Bug fixes should always be made against the latest available x.y-dev branch.

Feature Branches

Similar to bug fix branches, the name of a feature branch and its associated Pull Request should be a short, human-readable description of the feature being added. For example, a branch to add persistent named searches might be named feature/named-searches. Also new features should go into latest x.y-dev branch which does not have corresponding x.y-release branch. In case there is no such branch then the master branch is the right one. If you are not sure master branch is always the correct one.

Choosing an Upstream Branch

When creating a bug fix or feature branch, it is very important to choose the right upstream branch. The general rule is to always choose the oldest supported upstream branch that will need to contain your work. For more info see above Feature Branches or Bug Fix Branches.

Naming of the new Branch

It is advised to use the number of your issue or story when you are creating your new branch name. Some examples of naming:

  • Issue #2524 - Vagrant enviroment is not starting properly => 2524-vagrant-init-fix
  • Story #2523 - Implement regex upload of packages => 2523-regex-upload

Commit Messages

Commit messages in Pulp should contain a human readable explanation of what was fixed in the commit. They should also follow the standard git message format of starting with a subject line or title (usually wrapped at about 50 chars) and optionally, a longer message (usually wrapped at 72 characters) broken up into paragraphs. For more on what constitutes a good commit message, we recommend Tim Pope’s blog post on the subject.

It’s also recommended that every commit message in Pulp reference an issue in Pulp’s Redmine issue tracker. To do this you should use both a keyword and a link to the issue.

To reference the issue (but not change its state), use re or ref:

re #123
ref #123

To update the issue’s state to MODIFIED and set the %done to 100, use fixes or closes:

fixes #123
closes #123

You can also reference multiple issues in a commit:

fixes #123, #124

Putting this altogether, the following is an example of a good commit message:

Update node install and quickstart

The nodes install and quickstart was leaving out an important step on
the child node to configure the server.conf on the child node.

closes #1392
https://pulp.plan.io/issues/1392

Note

In case you have multiple commits use re or ref in all of them and fixes or close only in the last one to avoid closing the issue before it’s completely done.

Cherry-picking and Rebasing

Don’t do it! Seriously though, this should not happen between release branches. It is a good idea (but not required) for a developer to rebase his or her development branch before merging a pull request. Cherry-picking may also be valuable among development branches. However, master and release branches should not be involved in either.

The reason is that both of these operations generate new and unique commits from the same changes. We do not want pulp-x.y and master to have the same bug fix applied by two different commits. By merging the same commit into both, we can easily verify months later that a critical bug fix is present in every appropriate release branch and build tag.

Note

If you are not sure what “rebasing” and “cherry-picking” mean, Pro Git by Scott Chacon is an excellent resource for learning about git, including advanced topics such as these.