Follow the Git flow Paradigm

The term git flow refers to several distinct but related concepts.

Most importantly, it refers to a strategy or paradigm for managing code branches in a way that best promotes both innovation and continuous delivery of functional software. These are both core principles in agile software development with proven results.

Git flow may also refer to an application; an extension to the git command line tool that helps developers to follow the git flow paradigm. The git-flow application and its use is described in a separate document.

JEDI developers are not required to use the git-flow application. However, all JEDI developers are expected to comply with the git-flow paradigm as described below.

The git flow paradigm was introduced in 2010 in a brief but compelling blog post by the software engineer and project manager Vincent Driessen. We strongly recommend that all JEDI developers take a few moments to read this article.

Furthermore, we recommend that all JEDI developers keep the git flow diagram handy as a (virtual) desktop reference.

Note

Where in the above references master branch is used, in JEDI we use main branch, as many other Git communities do.

In the git flow model, the code is organized into the following branches that are maintained using the git version control system:

  • main branch
    • One permanent main branch that is used for releases only

    • Must pass all tests at all times

    • Direct code changes (commits and pushes) not allowed: all changes are made through GitHub pull requests and subject to code reviews

    • All changes are tagged with a release number

  • develop branch
    • One permanent develop branch that is used for development

    • Must pass all tests at all times

    • Direct code changes (commits and pushes) not allowed: all changes are made through GitHub pull requests and subject to code reviews

  • feature/* branches
    • multiple branches implementing specific code changes

    • where most development work happens

    • Branch off of develop

    • Merge back into develop

    • Temporary: each branch is deleted after it is merged

  • bugfix/* branches
    • For correcting errors or omissions

    • Branch off of develop (or a feature branch)

    • Merge back into develop (or a feature branch)

    • Temporary: each branch is deleted after it is merged

  • hotfix/* branches
    • For correcting errors or omissions

    • Branch off of main

    • Merge back into main and develop

    • Temporary: each branch is deleted after it is merged

  • release/* branches
    • For refinement, bug fixes, and documentation leading up to a release

    • Branch off of develop

    • Merge into main and develop

    • Temporary: each branch is deleted after it is merged

One of the most important principles of agile software development is to:

Keep your feature branches as small and focused as possible

Ideally, feature branches should exist for no more than a week or two. You should break large changes into small parts that can be implemented sequentially, tested, and readily reviewed by your peers. Then, after those changes are reviewed and merged, you can proceed to the next stage of changes.

Large feature branches that exist for weeks and change dozens of files become too cumbersome to review and will likely diverge from the develop branch, leading to multiple conflicts when it finally comes time to merge.

Life Cycle of a Feature Branch

Under the git flow paradigm, and using the git flow application, the typical life cycle of a feature branch is as described below. Other git-flow branches are handled in a similar way. If you do not wish to use the git flow application, you can achieve the same steps just with standard git. For example, git flow feature start newstuff is equivalent to:

git checkout develop
git branch feature/newstuff
git checkout feature/newstuff

Note

If you haven’t previously, you may need to initialize git flow for the repository by running git flow init -d

For further tips on working with git and git-flow, see our accompanying document on the git-flow application.

Step 1: Start the feature branch

git flow feature start newstuff

This creates a new branch called feature/newstuff that branches off of develop. Then you can edit files and commit them as you would with any other git repository:

git add *
git commit

Step 2. Push your branch to GitHub for the first time

After making one or more commits, you can push your branch to GitHub as follows:

git flow feature publish newstuff

Now there is a copy of your branch on the web, within GitHub, in addition to the copy on your computer.

Step 3. Additional commits and pushes as needed

Now typically you will make multiple commits as you add a feature and repeatedly recompile the code and test your changes. Don’t forget to add a test that specifically checks the code you have added.

git commit -a
git push

Each time you do a git push, this will transfer your changes from your computer to the copy of your branch that exists on GitHub.

If someone else is working on the same branch, you can do a git pull to retrieve the latest code from GitHub and merge it with the version that is on your computer. Note that this may occasionally lead to code conflicts that must be resolved. See the GitHub Guides for tutorials and examples on how to work with git and GitHub.

Step 4: Keep your branch up to date with develop

Step 4 does not really come after Step 3 - it should accompany it - they should be executed together.

As you make changes to the code, you don’t want your feature branch to diverge too much from the develop branch. If it does, then when you try to merge it you may find many conflicts. Furthermore, as noted above, feature branches with multiple changes are difficult to review by your peers. You want to make it easier on them by making sure that the changes you intend to merge into develop are only the changes you’ve added, not previous code that is left over from past versions of develop.

So, every day or two, you should execute these commands to merge in the latest changes from the develop branch on GitHub:

git checkout develop
git pull
git checkout feature/newstuff
git merge develop

Step 5: Finish the feature branch with a GitHub Pull Request

When your feature branch is finished, it should be merged into the develop branch. Finished means that the feature is implemented, the code compiles and all tests pass.

Though the git flow application has a finish function to do this, you should not use it. Instead, as noted above, all changes to the develop branch must be reviewed by other developers through GitHub pull requests.

For tips on properly issuing a GitHub pull request, see the next item in our list of Best Practices for Developers.

After your feature branch is triumphantly merged into develop, the remote branch (on GitHub) will be deleted. But, it will still exist on your computer. To bring your computer up to date, you can issue the following commands:

git remote update -p
git checkout develop
git pull origin develop
git branch -D feature/newstuff

The first command synchronizes the metadata that describe the changes that have been made on the remote repository (i.e. GitHub). The -p option prunes branches that have been deleted on the remote repository, including your feature/perf-enhance branch (if it was merged). Then the next two commands switch to the develop branch and synchronize it with GitHub. If your pull request was merged, your local copy of develop should now include your changes.

The last command deletes your local copy of the feature branch. You don’t need it any more since those changes are now included in the develop branch.