Git allows a wide variety of branching strategies and workflows.
Because of this, many organizations end up with workflows that are too complicated, not clearly defined, or not integrated with issue tracking systems.
Therefore, we propose GitLab flow as a clearly defined set of best practices.
It combines [feature-driven development](https://en.wikipedia.org/wiki/Feature-driven_development) and [feature branches](https://martinfowler.com/bliki/FeatureBranch.html) with issue tracking.
Since many organizations new to Git have no conventions for how to work with it, their repositories can quickly become messy.
The biggest problem is that many long-running branches emerge that all contain part of the changes.
People have a hard time figuring out which branch has the latest code, or which branch to deploy to production.
Frequently, the reaction to this problem is to adopt a standardized pattern such as [Git flow](https://nvie.com/posts/a-successful-git-branching-model/) and [GitHub flow](http://scottchacon.com/2011/08/31/github-flow.html).
We think there is still room for improvement. In this document, we describe a set of practices we call GitLab flow.
Merging everything into the `master` branch and frequently deploying means you minimize the amount of unreleased code, which is in line with lean and continuous delivery best practices.
However, this flow still leaves a lot of questions unanswered regarding deployments, environments, releases, and integrations with issues.
With GitLab flow, we offer additional guidance for these questions.
GitHub flow assumes you can deploy to production every time you merge a feature branch.
While this is possible in some cases, such as SaaS applications, there are many cases where this is not possible.
One case is where you don't control the timing of a release, for example, an iOS application that is released when it passes App Store validation.
Another case is when you have deployment windows — for example, workdays from 10 AM to 4 PM when the operations team is at full capacity — but you also merge code at other times.
In these cases, you can make a production branch that reflects the deployed code.
You can deploy a new version by merging `master` into the production branch.
If you need to know what code is in production, you can just checkout the production branch to see.
You only need to work with release branches if you need to release software to the outside world.
In this case, each branch contains a minor version, for example, 2-3-stable, 2-4-stable, etc.
Create stable branches using `master` as a starting point, and branch as late as possible.
By doing this, you minimize the length of time during which you have to apply bug fixes to multiple branches.
After announcing a release branch, only add serious bug fixes to the branch.
If possible, first merge these bug fixes into `master`, and then cherry-pick them into the release branch.
If you start by merging into the release branch, you might forget to cherry-pick them into `master`, and then you'd encounter the same bug in subsequent releases.
Merging into `master` and then cherry-picking into release is called an "upstream first" policy, which is also practiced by [Google](https://www.chromium.org/chromium-os/chromiumos-design-docs/upstream-first) and [Red Hat](https://www.redhat.com/en/blog/a-community-for-using-openstack-with-red-hat-rdo).
Every time you include a bug fix in a release branch, increase the patch version (to comply with [Semantic Versioning](https://semver.org/)) by setting a new tag.
The merge request serves as a code review tool, and no separate code review tools should be needed.
If the review reveals shortcomings, anyone can commit and push a fix.
Usually, the person to do this is the creator of the merge request.
The diff in the merge request automatically updates when new commits are pushed to the branch.
When you are ready for your feature branch to be merged, assign the merge request to the person who knows most about the codebase you are changing.
Also, mention any other people from whom you would like feedback.
After the assigned person feels comfortable with the result, they can merge the branch.
If the assigned person does not feel comfortable, they can request more changes or close the merge request without merging.
In GitLab, it is common to protect the long-lived branches, e.g., the `master` branch, so that [most developers can't modify them](../permissions/permissions.md).
So, if you want to merge into a protected branch, assign your merge request to someone with maintainer permissions.
Link to issues by mentioning them in commit messages or the description of a merge request, for example, "Fixes #16" or "Duck typing is preferred. See #12."
GitLab then creates links to the mentioned issues and creates comments in the issues linking back to the merge request.
To automatically close linked issues, mention them with the words "fixes" or "closes," for example, "fixes #14" or "closes #67." GitLab closes these issues when the code is merged into the default branch.
With Git, you can use an interactive rebase (`rebase -i`) to squash multiple commits into one or reorder them.
This functionality is useful if you want to replace a couple of small commits with a single commit, or if you want to make the order more logical.
However, you should never rebase commits you have pushed to a remote server.
Rebasing creates new commits for all your changes, which can cause confusion because the same change would have multiple identifiers.
It also causes merge errors for anyone working on the same branch because their history would not match with yours.
Also, if someone has already reviewed your code, rebasing makes it hard to tell what changed since the last review.
You should also never rebase commits authored by other people.
Not only does this rewrite history, but it also loses authorship information.
Rebasing prevents the other authors from being attributed and sharing part of the [`git blame`](https://git-scm.com/docs/git-blame).
If a merge involves many commits, it may seem more difficult to undo.
You might think to solve this by squashing all the changes into one commit before merging, but as discussed earlier, it is a bad idea to rebase commits that you have already pushed.
Fortunately, there is an easy way to undo a merge with all its commits.
The way to do this is by reverting the merge commit.
Preserving this ability to revert a merge is a good reason to always use the "no fast-forward" (`--no-ff`) strategy when you merge manually.
NOTE: **Note:**
If you revert a merge commit and then change your mind, revert the revert commit to redo the merge.
Git does not allow you to merge the code again otherwise.
Having lots of merge commits can make your repository history messy.
Therefore, you should try to avoid merge commits in feature branches.
Often, people avoid merge commits by just using rebase to reorder their commits after the commits on the `master` branch.
Using rebase prevents a merge commit when merging `master` into your feature branch, and it creates a neat linear history.
However, as discussed in [the section about rebasing](#squashing-commits-with-rebase), you should never rebase commits you have pushed to a remote server.
This restriction makes it impossible to rebase work in progress that you already shared with your team, which is something we recommend.
Rebasing also creates more work, since every time you rebase, you have to resolve similar conflicts.
Sometimes you can reuse recorded resolutions (`rerere`), but merging is better since you only have to resolve conflicts once.
Atlassian has a more thorough explanation of the tradeoffs between merging and rebasing [on their blog](https://www.atlassian.com/blog/git/git-team-workflows-merge-or-rebase).
If you need to utilize some code that was introduced in `master` after you created the feature branch, you can often solve this by just cherry-picking a commit.
Sometimes you can use .gitattributes to reduce merge conflicts.
For example, you can set your changelog file to use the [union merge driver](https://git-scm.com/docs/gitattributes#gitattributes-union) so that multiple new entries don't conflict with each other.
The last reason for creating merge commits is to keep long-running feature branches up-to-date with the latest state of the project.
The solution here is to keep your feature branches short-lived.
Most feature branches should take less than one day of work.
If your feature branches often take more than a day of work, try to split your features into smaller units of work.
If you need to keep a feature branch open for more than a day, there are a few strategies to keep it up-to-date.
One option is to use continuous integration (CI) to merge in `master` at the start of the day.
Another option is to only merge in from well-defined points in time, for example, a tagged release.
You could also use [feature toggles](https://martinfowler.com/bliki/FeatureToggle.html) to hide incomplete features so you can still merge back into `master` every day.
> **Note:** Don't confuse automatic branch testing with continuous integration.
> Martin Fowler makes this distinction in [his article about feature branches](https://martinfowler.com/bliki/FeatureBranch.html):
>
> "I've heard people say they are doing CI because they are running builds, perhaps using a CI server, on every branch with every commit.
> That's continuous building, and a Good Thing, but there's no *integration*, so it's not CI."
In conclusion, you should try to prevent merge commits, but not eliminate them.
Your codebase should be clean, but your history should represent what actually happened.
Developing software happens in small, messy steps, and it is OK to have your history reflect this.
You can use tools to view the network graphs of commits and understand the messy history that created your code.
If you rebase code, the history is incorrect, and there is no way for tools to remedy this because they can't deal with changing commit identifiers.
## Commit often and push frequently
Another way to make your development work easier is to commit often.
Every time you have a working set of tests and code, you should make a commit.
Splitting up work into individual commits provides context for developers looking at your code later.
Smaller commits make it clear how a feature was developed, and they make it easy to roll back to a specific good point in time or to revert one code change without reverting several unrelated changes.
Committing often also makes it easy to share your work, which is important so that everyone is aware of what you are working on.
You should push your feature branch frequently, even when it is not yet ready for review.
By sharing your work in a feature branch or [a merge request](#mergepull-requests-with-gitlab-flow), you prevent your team members from duplicating work.
Sharing your work before it's complete also allows for discussion and feedback about the changes, which can help improve the code before it gets to review.
A commit message should reflect your intention, not just the contents of the commit.
It is easy to see the changes in a commit, so the commit message should explain why you made those changes.
An example of a good commit message is: "Combine templates to reduce duplicate code in the user views."
The words "change," "improve," "fix," and "refactor" don't add much information to a commit message.
For example, "Improve XML generation" could be better written as "Properly escape special characters in XML generation."
For more information about formatting commit messages, please see this excellent [blog post by Tim Pope](https://tbaggery.com/2008/04/19/a-note-about-git-commit-messages.html).
