Pull request workflow

Pull request workflow

The so-called “PR workflow” used by Godot is common to many projects using Git, and should be familiar to veteran free software contributors. The idea is that only a small number (if any) commit directly to the master branch. Instead, contributors fork the project (i.e. create a copy of it, which they can modify as they wish), and then use the GitHub interface to request a pull from one of their fork’s branches to one branch of the original (often named upstream) repository.

The resulting pull request (PR) can then be reviewed by other contributors, which might approve it, reject it, or most often request that modifications be done. Once approved, the PR can then be merged by one of the core developers, and its commit(s) will become part of the target branch (usually the master branch).

We will go together through an example to show the typical workflow and associated Git commands. But first, let’s have a quick look at the organization of Godot’s Git repository.

Git source repository

The repository on GitHub is a Git code repository together with an embedded issue tracker and PR system.

Note

If you are contributing to the documentation, its repository can be found here.

The Git version control system is the tool used to keep track of successive edits to the source code - to contribute efficiently to Godot, learning the basics of the Git command line is highly recommended. There exist some graphical interfaces for Git, but they usually encourage users to take bad habits regarding the Git and PR workflow, and we therefore recommend not to use them. In particular, we advise not to use GitHub’s online editor for code contributions (although it’s tolerated for small fixes or documentation changes) as it enforces one commit per file and per modification, which quickly leads to PRs with an unreadable Git history (especially after peer review).

Forking and cloning

The first step is to fork the godotengine/godot repository on GitHub. To do so, you will need to have a GitHub account and to be logged in. In the top right corner of the repository’s GitHub page, you should see the “Fork” button as shown below:

Click it, and after a while you should be redirected to your own fork of the Godot repo, with your GitHub username as namespace:

You can then clone your fork, i.e. create a local copy of the online repository (in Git speak, the origin remote). If you haven’t already, download Git from its website if you’re using Windows or macOS, or install it through your package manager if you’re using Linux.

Note

If you are on Windows, open Git Bash to type commands. macOS and Linux users can use their respective terminals.

To clone your fork from GitHub, use the following command:

$ git clone https://github.com/USERNAME/godot

Note

In our examples, the “$” character denotes the command line prompt on typical UNIX shells. It is not part of the command and should not be typed.

After a little while, you should have a godot directory in your current working directory. Move into it using the cd command:

$ cd godot

We will start by setting up a reference to the original repository that we forked:

$ git remote add upstream https://github.com/godotengine/godot
$ git fetch upstream

This will create a reference named upstream pointing to the original godotengine/godot repository. This will be useful when you want to pull new commits from its master branch to update your fork. You have another remote reference named origin, which points to your fork (USERNAME/godot).

You only need to do the above steps once, as long as you keep that local godot folder (which you can move around if you want, the relevant metadata is hidden in its .git subfolder).

Note

Branch it, pull it, code it, stage it, commit, push it, rebase it… technologic.

This bad take on Daft Punk’s Technologic shows the general conception Git beginners have of its workflow: lots of strange commands to learn by copy and paste, hoping they will work as expected. And that’s actually not a bad way to learn, as long as you’re curious and don’t hesitate to question your search engine when lost, so we will give you the basic commands to know when working in Git.

In the following, we will assume as an example that you want to implement a feature in Godot’s project manager, which is coded in the editor/project_manager.cpp file.

Branching

By default, the git clone should have put you on the master branch of your fork (origin). To start your own feature development, we will create a feature branch:

# Create the branch based on the current branch (master)
$ git branch better-project-manager
# Change the current branch to the new one
$ git checkout better-project-manager

This command is equivalent:

# Change the current branch to a new named one, based on the current branch
$ git checkout -b better-project-manager

If you want to go back to the master branch, you’d use:

$ git checkout master

You can see which branch you are currently on with the git branch command:

$ git branch
  2.1
* better-project-manager
  master

Be sure to always go back to the master branch before creating a new branch, as your current branch will be used as the base for the new one. Alternatively, you can specify a custom base branch after the new branch’s name:

$ git checkout -b my-new-feature master

Updating your branch

This would not be needed the first time (just after you forked the upstream repository). However, the next time you want to work on something, you will notice that your fork’s master is several commits behind the upstream master branch: pull requests from other contributors would have been merged in the meantime.

To ensure there won’t be conflicts between the feature you develop and the current upstream master branch, you will have to update your branch by pulling the upstream branch.

$ git pull --rebase upstream master

The --rebase argument will ensure that any local changes that you committed will be re-applied on top of the pulled branch, which is usually what we want in our PR workflow. This way, when you open a pull request, your own commits will be the only difference with the upstream master branch.

While rebasing, conflicts may arise if your commits modified code that has been changed in the upstream branch in the meantime. If that happens, Git will stop at the conflicting commit and will ask you to resolve the conflicts. You can do so with any text editor, then stage the changes (more on that later), and proceed with git rebase --continue. Repeat the operation if later commits have conflicts too, until the rebase operation completes.

If you’re unsure about what is going on during a rebase and you panic (no worry, we all do the first few times), you can abort the rebase with git rebase --abort. You will then be back to the original state of your branch before calling git pull --rebase.

Note

If you omit the --rebase argument, you will instead create a merge commit which tells Git what to make of the two distinct branches. If any conflicts arise, they would be resolved all at once via this merge commit.

While this is a valid workflow and the default behavior of git pull, merge commits within PRs are frowned upon in our PR workflow. We only use them when merging PRs into the upstream branch.

The philosophy is that a PR should represent the final stage of the changes made to the codebase, and we are not interested in mistakes and fixes that would have been done in intermediate stages before merging. Git gives us great tools to “rewrite the history” and make it as if we got things right the first time, and we’re happy to use it to ensure that changes are easy to review and understand long after they have been merged.

If you have already created a merge commit without using rebase, or have made any other changes that have resulted in undesired history, the best option is to use an interactive rebase on the upstream branch. See the dedicated section for instructions.

Tip

If at any time you want to reset a local branch to a given commit or branch, you can do so with git reset --hard <commit ID> or git reset --hard <remote>/<branch> (e.g. git reset --hard upstream/master).

Be warned that this will remove any changes that you might have committed in this branch. If you ever lose commits by mistake, use the git reflog command to find the commit ID of the previous state that you would like to restore, and use it as argument of git reset --hard to go back to that state.

Making changes

You would then do your changes to our example’s editor/project_manager.cpp file with your usual development environment (text editor, IDE, etc.).

By default, those changes are unstaged. The staging area is a layer between your working directory (where you make your modifications) and the local Git repository (the commits and all the metadata in the .git folder). To bring changes from the working directory to the Git repository, you need to stage them with the git add command, and then to commit them with the git commit command.

There are various commands you should know to review your current work, before staging it, while it is staged, and after it has been committed.

git diff will show you the current unstaged changes, i.e. the differences between your working directory and the staging area.
git checkout -- <files> will undo the unstaged changes to the given files.
git add <files> will stage the changes on the listed files.
git diff --staged will show the current staged changes, i.e. the differences between the staging area and the last commit.
git reset HEAD <files> will unstage changes to the listed files.
git status will show you what are the currently staged and unstaged modifications.
git commit will commit the staged files. It will open a text editor (you can define the one you want to use with the GIT_EDITOR environment variable or the core.editor setting in your Git configuration) to let you write a commit log. You can use git commit -m "Cool commit log" to write the log directly.
git commit --amend lets you amend the last commit with your currently staged changes (added with git add). This is the best option if you want to fix a mistake in the last commit (bug, typo, style issue, etc.).
git log will show you the last commits of your current branch. If you did local commits, they should be shown at the top.
git show will show you the changes of the last commit. You can also specify a commit hash to see the changes for that commit.

That’s a lot to memorize! Don’t worry, just check this cheat sheet when you need to make changes, and learn by doing.

Here’s how the shell history could look like on our example:

# It's nice to know where you're starting from
$ git log
# Do changes to the project manager with the nano text editor
$ nano editor/project_manager.cpp
# Find an unrelated bug in Control and fix it
$ nano scene/gui/control.cpp
# Review changes
$ git status
$ git diff
# We'll do two commits for our unrelated changes,
# starting by the Control changes necessary for the PM enhancements
$ git add scene/gui/control.cpp
$ git commit -m "Fix handling of margins in Control"
# Check we did good
$ git log
$ git show
$ git status
# Make our second commit
$ git add editor/project_manager.cpp
$ git commit -m "Add a pretty banner to the project manager"
$ git log

With this, we should have two new commits in our better-project-manager branch which were not in the master branch. They are still only local though, the remote fork does not know about them, nor does the upstream repo.

Pushing changes to a remote

That’s where git push will come into play. In Git, a commit is always done in the local repository (unlike Subversion where a commit will modify the remote repository directly). You need to push the new commits to a remote branch to share them with the world. The syntax for this is:

$ git push <remote> <local branch>[:<remote branch>]

The part about the remote branch can be omitted if you want it to have the same name as the local branch, which is our case in this example, so we will do:

$ git push origin better-project-manager

Git will ask you for your username and password, and the changes will be sent to your remote. If you check the fork’s page on GitHub, you should see a new branch with your added commits.

Issuing a pull request

When you load your fork’s branch on GitHub, you should see a line saying “This branch is 2 commits ahead of godotengine:master.” (and potentially some commits behind, if your master branch was out of sync with the upstream master branch).

On that line, there is a “Pull request” link. Clicking it will open a form that will let you issue a pull request on the godotengine/godot upstream repository. It should show you your two commits, and state “Able to merge”. If not (e.g. it has way more commits, or says there are merge conflicts), don’t create the PR, something went wrong. Go to IRC and ask for support :)

Use an explicit title for the PR and put the necessary details in the comment area. You can drag and drop screenshots, GIFs or zipped projects if relevant, to showcase what your work implements. Click “Create a pull request”, and tadaa!

Modifying a pull request

While it is reviewed by other contributors, you will often need to make changes to your yet-unmerged PR, either because contributors requested them, or because you found issues yourself while testing.

The good news is that you can modify a pull request simply by acting on the branch you made the pull request from. You can e.g. make a new commit on that branch, push it to your fork, and the PR will be updated automatically:

# Check out your branch again if you had changed in the meantime
$ git checkout better-project-manager
# Fix a mistake
$ nano editor/project_manager.cpp
$ git add editor/project_manager.cpp
$ git commit -m "Fix a typo in the banner's title"
$ git push origin better-project-manager

However, be aware that in our PR workflow, we favor commits that bring the codebase from one functional state to another functional state, without having intermediate commits fixing up bugs in your own code or style issues. Most of the time, we will prefer a single commit in a given PR (unless there’s a good reason to keep the changes separate), so instead of authoring a new commit, considering using git commit --amend to amend the previous commit with your fixes. The above example would then become:

# Check out your branch again if you had changed in the meantime
$ git checkout better-project-manager
# Fix a mistake
$ nano editor/project_manager.cpp
$ git add editor/project_manager.cpp
# --amend will change the previous commit, so you will have the opportunity
# to edit its commit message if relevant.
$ git commit --amend
# As we modified the last commit, it no longer matches the one from your
# remote branch, so we need to force push to overwrite that branch.
$ git push --force origin better-project-manager

The interactive rebase

If you didn’t follow the above steps closely to amend changes into a commit instead of creating fixup commits, or if you authored your changes without being aware of our workflow and Git usage tips, reviewers might request of your to rebase your branch to squash some or all of the commits into one.

Indeed, if some commits have been made following reviews to fix bugs, typos, etc. in the original commit, they are not relevant to a future changelog reader who would want to know what happened in the Godot codebase, or when and how a given file was last modified.

To squash those extraneous commits into the main one, we will have to rewrite history. Right, we have that power. You may read that it’s a bad practice, and it’s true when it comes to branches of the upstream repo. But in your fork, you can do whatever you want, and everything is allowed to get neat PRs :)

We will use the interactive rebase git rebase -i to do this. This command takes a commit ID or a branch name as argument, and will let you modify all commits between that commit/branch and the last one in your working branch, the so-called HEAD.

While you can give any commit ID to git rebase -i and review everything in between, the most common and convenient workflow involves rebasing on the upstream ``master`` branch, which you can do with:

$ git rebase -i upstream/master

Note

Referencing branches in Git is a bit tricky due to the distinction between remote and local branches. Here, upstream/master (with a /) is a local branch which has been pulled from the upstream remote’s master branch.

Interactive rebases can only be done on local branches, so the / is important here. As the upstream remote changes frequently, your local upstream/master branch may become outdated, so you can update it with git fetch upstream master. Contrarily to git pull --rebase upstream master which would update your currently checked out branch, fetch will only update the upstream/master reference (which is distinct from your local master branch… yes it’s confusing, but you’ll become familiar with this little by little).

This will open a text editor (vi by default, see Git docs <https://git-scm.com/book/en/v2/Customizing-Git-Git-Configuration#\_core\_editor>\_\_ to configure your favorite one) with something which may look like this:

pick 1b4aad7 Add a pretty banner to the project manager
pick e07077e Fix a typo in the banner's title

The editor will also show instructions regarding how you can act on those commits. In particular, it should tell you that “pick” means to use that commit (do nothing), and that “squash” and “fixup” can be used to meld the commit in its parent commit. The difference between “squash” and “fixup” is that “fixup” will discard the commit log from the squashed commit. In our example, we are not interested in keeping the log of the “Fix a typo” commit, so we use:

pick 1b4aad7 Add a pretty banner to the project manager
fixup e07077e Fix a typo in the banner's title

Upon saving and quitting the editor, the rebase will occur. The second commit will be melded into the first one, and git log and git show should now confirm that you have only one commit with the changes from both previous commits.

But! You rewrote the history, and now your local and remote branches have diverged. Indeed, commit 1b4aad7 in the above example will have changed, and therefore got a new commit hash. If you try to push to your remote branch, it will raise an error:

$ git push origin better-project-manager
To https://github.com/akien-mga/godot
 ! [rejected]        better-project-manager -> better-project-manager (non-fast-forward)
error: failed to push some refs to 'https://akien-mga@github.com/akien-mga/godot'
hint: Updates were rejected because the tip of your current branch is behind
hint: its remote counterpart.

This is a sane behavior, Git will not let you push changes that would override remote content. But that’s actually what we want to do here, so we will have to force it:

$ git push --force origin better-project-manager

And tadaa! Git will happily replace your remote branch with what you had locally (so make sure that’s what you wanted, using git log). This will also update the PR accordingly.

Deleting a Git branch

After your pull request gets merged, there’s one last thing you should do: delete your Git branch for the PR. There won’t be issues if you don’t delete your branch, but it’s good practice to do so. You’ll need to do this twice, once for the local branch and another for the remote branch on GitHub.

To delete our better project manager branch locally, use this command:

$ git branch -d better-project-manager

Alternatively, if the branch hadn’t been merged yet and we wanted to delete it anyway, instead of -d you would use -D.

Next, to delete the remote branch on GitHub use this command:

$ git push origin -d better-project-manager

You can also delete the remote branch from the GitHub PR itself, a button should appear once it has been merged or closed.