Working with Git

This document contains information about working with the ​Git version control system, tailored to developers familiar with ​Subversion.

Initial setup

Git commits contain your name and email address. You should set them in your git configuration using the following commands:

git config --global user.name "Your Name"
git config --global user.email YOUR_MACPORTS_HANDLE@macports.org

If you work on multiple Git projects and do not want to modify your email address for those, these commands can also be run without --global in a clone of MacPorts' repositories to only change the option for these repositories. If you are not a MacPorts committer, use any email address.

Common git tasks & notes about MacPorts' Subversion export

Getting a working copy

To start working with git, you need a copy of the sources. Follow the sections for getting the ports tree or MacPorts base, depending on where you want to make changes, then continue with the next section.

Committing changes in your working copy

A fundamental difference between Subversion and Git working copies is that svn commit by default commits all changes in your working copy, but git commit by default commits none. Git uses a staging area called "index" that allows you to mark changes for inclusion in the next commit. To add changes to the next commit, use

git add <filename>...

git status gives you an overview of the current index and your working copy. Additionally, it lists the commands to revert local uncommitted modifications (git checkout -- <filename>) and to remove files from the next commit, but preserve the modifications in your working copy (git reset HEAD <filename>).

Once you have chosen which files to include in your next commit using git add, it is a good practice to review this list using

git status

and show the diff to be committed using

git diff --cached

If you are not satisfied with your changes, you can keep changing your files. Note that you will have to add any new modifications to the index using git add again. Once you are satisfied with your change run

git commit

which prompts you for the commit message. See the section on commit messages in git for more information on git conventions and expectations in commit messages.

Because of Git's distributed nature, a commit on your local machine is not immediately available on the central server, like it was the case with Subversion. This means that you can continue to prepare further changes in additional commits before you publish your changes as a set. In fact, it is a very common practice in Git to do many small changes that are logically consistent in themselves and then publish them in one step.

If you have commit access, you can publish your commits using git push <remote-name> <branch-name>. <remote-name> is the name of the repository to which you want to push. The most common push target is the location you initially cloned, which is automatically named origin. <branch-name> is the name of the branch you want to push. The Git equivalent to Subversion's trunk is called master. It is considered best practice to always specify your push target and the branch you are pushing, since git's default is pushing all branches that have a remote equivalent when you run git push, which might publish changes that you do not consider final yet (you can disable this behavior by changing the push.default git-config setting to nothing, see ​git-config(1)).

git push origin master

Note that the push will fail if the remote repository has new changes. Contrary to Subversion, it does not matter whether your changes conflict with the remote ones. If this happens, you must update your local working copy as described in the section on fetch the latest changes and re-try the push.

Commit messages

There are a number of conventions to writing Git commit messages. For a detailed explanation, see ​http://chris.beams.io/posts/git-commit/. As a tl;dr, here are seven short rules:

1. Separate subject from body with a blank line
2. Limit the subject line to 50 characters
3. Capitalize the subject line
4. Do not end the subject line with a period
5. Use the imperative mood in the subject line
6. Wrap the body at 72 characters
7. Use the body to explain what and why vs. how

If you don't want to remember these rules, you can configure your git client to load a template whenever it prompts you for a commit message by setting the commit.template ​git-config(1) option. The KDE developers ​have a nice example.

Fetching the latest changes

Git's equivalent to svn update is a little more complicated due to Git's distributed nature. Most of the complexity is not visible if you do not have commits in your working copy that have not been pushed yet. If both the local and the remote repository have changes (git calls them "diverged"), you will run into one of Git's core principles: Every commit has (at least) one parent commit, i.e. the commit history forms a directed acyclic graph.

Background knowledge

A picture is worth a thousand words:

 A --- B --- C ---- R1 ---- R2 ---- R3  <= origin/master
              \
               +--- L1 ---- L2          <= master

A, B and C are commits that are both in your local and in the remote repository. R1-3 are commits that have been pushed into the remote repository "origin"'s master branch while you were working. L1 and L2 are commits you prepared locally on your master branch. Git offers two different ways to bring R1-3 into your local branch:

Merging

A merge commit, created by git merge, is a commit that has multiple parents. If no conflict occurs, merge commits do not usually have a diff attached (i.e. they do not modify files). On conflict, merge commits contain the diff that resolves the conflict. In pictures:

 A --- B --- C ---- R1 ---- R2 ---- R3   <= origin/master
              \                      \
               +--- L1 ---- L2 ------ M  <= master

The new commit M is the merge commit and can be pushed back to origin. This preserves the information that work was done in parallel, but unfortunately tends to mess up the history graph. See the attached screenshot of a commit history that always merges. To avoid this, you can instead rebase your changes.

Rebasing

Rebasing commits rewrites their parent commit IDs and avoids the need for a merge commit. Running git rebase origin/master will take all commits in your local working copy that are not yet pushed and attach them after the end of origin/master, which yields this picture:

 A --- B --- C ---- R1 ---- R2 ---- R3   <= origin/master
                                      \
                                       L1' ---- L2'  <= master

Note that L1 and L2 have been modified by this operation; their commit IDs changed because of that. This new state can be pushed back to origin without the need for a merge commit, and the history graph will stay linear. We recommend that all developers rebase their changes rather than merge when conflicts occur during pushing.

Putting the background knowledge into production

First, get all new commits from the remote repository using git fetch <remote-name>, where <remote-name> identifies the repository from which you want to fetch and defaults to "origin":

git fetch

Then, rebase your local changes (if any) on top of any new changes in the remote repository and fix any conflicts that occur:

git rebase origin/master

Because these two operations are very common, Git offers a shorthand for them:

git pull --rebase

Note: git rebase requires that you do not have uncommitted modifications in your working copy. If you have modifications, you can temporarily save them using git stash and restore them after the rebase using git stash pop.

Warning: git pull without the --rebase flag is a shorthand for git fetch && git merge origin/master, which will automatically create a merge commit if it thinks that's necessary.

If you do not want to remember passing --rebase to git pull every time you run it, you can set a couple of ​git-config(1) options to make it the default:

• Setting pull.rebase to true will change the default to always rebase when calling git pull. Note that this will also flatten any local merge commits you might have committed on purpose with git merge, which might be undesirable when merging development branches for MacPorts base. Consider using the preserve setting, which avoids this.
• Rebasing can be enabled on a per-branch basis using the branch.<name>.rebase setting, which accepts the same values as pull.rebase.
• You can make branch.<name>.rebase true the default for all branches that you clone by setting branch.autoSetupRebase to always. This allows you to change the setting back to a different value for specific branches but still keep the default to rebase. Note that this setting will not affect branches that you have already created.

Reverting changes

Subversion has two methods for reverting changes: svn revert, which drops uncommitted local modifications and restores the committed state and svn merge -c -12345 to undo committed changes.

Due to Git's distributed nature, there are three stages that can be reverted:

• To drop uncommitted modifications, use git checkout -- <filename>. If you had already added the file to the index using git add, you have to unstage it first using git reset HEAD <filename>. git status prints these commands, so you don't have to remember them.
• To undo a change that has already been committed and pushed, use git revert <commitID>. This will create a new commit that applies the inverse diff. Note that you still have to push this commit to publish it.
• To throw away all changes that you have locally committed but not yet pushed, use git reset --hard origin/master. You will loose all your uncommitted and committed modifications. If that is not what you want, Git provides a variety of tools that allow you to change commits that you have not pushed yet (and theoretically also commits that have already been pushed, which will prevent you from pushing any changes again). Since this is an advanced topic it will not be covered here. As a pointer for further research, look for git commit --amend to change the topmost commit and git rebase --interactive, the so-called "interactive rebase", to change older commits.

Repository split

Unlike Subversion, Git does not permit selective checkout of individual subdirectories. As part of the GitHub migration, the Subversion repository has been restructured into several smaller Git repositories so that contributors don't have to clone the entire project.

Here is a summary of the restructuring; the full set of ​svn2git conversion rules is available on ​Clemens Lang's GitHub.

Subversion (Mac OS Forge)	Git (GitHub)
trunk/base	macports/base.git
trunk/dports	macports/ports.git
trunk/dports/{PortIndex,PortIndex.quick}	deleted
trunk/doc-new	macports/guide.git
trunk/doc	macports/guide.git, branch "doc-old"
trunk/www	macports/www.git
trunk/base/portmgr/jobs	macports/infrastructure.git, subfolder "jobs"
contrib	macports/contrib.git
contrib/{MacPorts_Framework,Pallet}	macports/pallet.git
contrib/mpab	macports/mp-buildbot.git, branch "mpab"
contrib/mp-buildbot	macports/mp-buildbot.git
contrib/buildbot	macports/infrastructure.git, subfolder "buildbot-old"
contrib/buildbot-test	macports/infrastructure.git, subfolder "buildbot"
users/{dluke,jberry,pipping}	deleted (exceeds GitHub size limit, only contains old files)
users/*	separate repository in macports/users/*
distfiles	deleted
downloads	deleted
branches	branches of the appropriate repository/-ies
tags	annotated tags in the appropriate repository/-ies

Common git tasks while working with ports

Checking out a working copy of the ports tree

To get a working copy of the MacPorts ports tree to start changing ports, clone of copy of the repository:

git clone git@github.com:macports/ports.git # or
git clone https://github.com/macports/ports.git # if SSH does not work on your network

This will give you the entire history of the ports tree, with the latest version being checked out in the filesystem.

Note: If you intend to submit a pull request on GitHub to get your changes included in MacPorts, you may want to create a fork of the repository first. To do that, go to ​https://github.com/macports/ports/ and click the fork button at the top right. Then, run the command above, but use <yourusername>/ports.git instead of macports/ports.git.

See the section on committing changes to find out how to get your changes into the repository.

Common git tasks while working with MacPorts base

Checking out a working copy of MacPorts base

The source code of MacPorts itself is no longer managed in the same repository as all ports. Contrary to Subversion, checking out a sub-directory of a repository is not possible with Git. In order to avoid that all port maintainers have to clone the complete history of MacPorts base as well, the Subversion repository has been split into multiple separate repositories. MacPorts base is now available using

git clone git@github.com:macports/base.git # or
git clone https://github.com/macports/base.git # if SSH does not work on your network

See the section on repository splitting during the export to get an overview of where a path in the old Subversion history is now available in Git.

Merge a single change from master into a release branch

The equivalent to Subversion's svn merge -c <revision> . is git cherry-pick. Use git cherry-pick to apply a single change from master to a release branch. To do this, look up the commit ID of the commit you want to pick:

git log
# copy the commit ID

Switch to the target branch of the cherry pick:

git checkout release_2_3

Cherry-pick the commit. It is good practice to pass -x to git cherry-pick, which will automatically add a "Cherry picked from commit <commmitID>" line to the commit message of your cherry pick. You will have the option to modify the commit message, e.g. to describe why the backport was necessary.

git cherry-pick -x <commitID>

Finally, push the new commit using

git push origin <branchname>

Tools

• ​gitk is a Tk-based graphical repository browser included with Git itself.
• ​GitX is a graphical repository browser designed for macOS. Unfortunately, our GitX port is currently nonfunctional.
• ​Tig is an ncurses-based interface, available from the tig port.

Learning Resources

• ​Git Documentation
• ​GitHub Help
• ​Git's list of external resources
• ​GitHub's list of external resources
• ​“Infrastructure/Git/Configuration” at the KDE Community Wiki