This tutorial helps you with:
- step-by-step instructions and screenshots for getting Gerrit access
- setting up Git on your machine
- configuring Git
- installing Git-review
- submitting a patch
- understanding the MediaWiki code review process
What is Git?
Git is a distributed version control system (dvcs) written in C originally developed by Linus Torvalds and others to manage the Linux kernel. In the past couple of years, it has taken off as a very robust and well-supported code repository. “Distributed” means that there is no central copy of the repository. With Subversion, Wikimedia’s servers host the repository and users commit their changes to it. In contrast, with Git, once you’ve cloned the repository, you have a fully functioning copy of the source code, with all the branches and tagged releases at your disposal.
Git is a free and open source distributed version control system designed to handle everything from small to very large projects with speed and efficiency. Git is easy to learn and has a tiny footprint with lightning fast performance. It outclasses SCM tools like Subversion, CVS, Perforce, and ClearCase with features like cheap local branching, convenient staging areas, and multiple workflows.
What is Gerrit?
Gerrit is a free, web-based collaborative code review tool that integrates with Git. It has been developed at Google by Shawn Pearce (co-author of Git, founder of JGit) for the development of the Android project.
Why did Wikimedia engineering move from Subversion to Git
Three major reasons:
- To encourage participation: Since Git is distributed, it allows people to contribute with a much lower barrier to entry. Anyone will be able to clone the repository and make their own changes to keep track of them. And if you’ve got an account in our code review tool (Gerrit), you’ll be able to push changes for the wider community to review.
- To fix our technical process: Subversion has technical flaws that make life difficult for developers. Notably, the implementation of branching is not very easy to use, and makes it hard to use “feature branches”. Our community is very distributed, with many parallel efforts and needs to integrate many different feature efforts, so we’d like to use feature branches more. Git branches are very easy to work with and merge between, which should make things easier for our development community. (Several other large projects, such as Drupal and PostgreSQL, have made the same switch for similar reasons, and we’ve done our best to learn from their experiences.)
- To get improvements to users faster: With better branching and a more granular code review workflow that suits our needs better, plus our ongoing improvements to our automated testing infrastructure, we won’t have to wait months before deploying already-written features and bugfixes to Wikimedia sites.
Set Up SSH Keys in Gerrit
We use SSH keys to establish a secure connection between your computer and Gerrit. Setting them up is fairly easy, but does involve a number of steps. Run the following commands in a terminal.
To make sure you generate a brand new key, you need to check if one already exists.
Generate a new SSH key
To generate a new SSH key, enter the code below. We want the default settings so when asked to enter a file in which to save the key, just press enter.
ssh-keygen -t rsa -C "email@example.com"
Assign the pass phrase (press [enter] key twice if you don't want a passphrase).
Why do passphrases matter?
Passwords aren’t very secure, you already know this. If you use one that’s easy to remember, it’s easier to guess or brute-force (try many options until one works). If you use one that’s random it’s hard to remember, and thus you’re more inclined to write the password down. Both of these are Very Bad Things™. This is why you’re using ssh keys.
But using a key without a passphrase is basically the same as writing down that random password in a file on your computer. Anyone who gains access to your drive has gained access to every system you use that key with. This is also a Very Bad Thing™. The solution is obvious: add a passphrase.
But I don’t want to enter a long passphrase every time I use the key!
Neither do we! Thankfully, there’s a nifty little tool called ssh-agent that can save your passphrase securely so you don’t have to re-enter it. If you’re on OSX Leopard or later your keys can be saved in the system’s keychain to make your life even easier.
It will create 2 files in ~/.ssh directory as follows:
~/.ssh/id_rsa : identification (private) key ~/.ssh/id_rsa.pub : public key
Add your SSH key
Open the id_rsa.pub file with a text editor (Notepad, TextEdit, or gedit will do just fine). This is your public SSH key. You may need to turn on “view hidden files” to find it because the .ssh directory is hidden. It’s important you copy your SSH key exactly as it is written without adding any newlines or whitespace. Now paste it into the “Key” field.
$ pbcopy < ~/.ssh/id_rsa.pub # Copies the contents of the id_rsa.pub file to your clipboard
Windows You can open Git Gui, go to Help > Show Key, and then press Copy To Clipboard to copy your public key to your clipboard
$ sudo apt-get install xclip # Downloads and installs xclip $ xclip -sel clip < ~/.ssh/id_rsa.pub
Log into the web interface for gerrit. Click on Settings then SSH Public Keys then add your key.
Setting up Git
These instructions explain how to install Git as a command-line tool. If you also want a GUI client check the list maintained by the Git project.
Mac OS X
- Homebrew package manager:
brew install git– Recommended, simple updating and easy installing of other packages
- Standalone: Git for Mac
- Standalone: Git for Windows – Requires adding the "git" directory to your system PATH.
- MSYS (Minimal Bash for Windows): mSysGit – Recommended, no manual configuration
Linux & UNIX
sudo apt-get install git-core
yum install git
zypper install git
emerge --ask --verbose dev-vcs/git
cd /usr/ports/devel/git; make install
- Solaris 11 Express:
pkg install developer/versioning/git
cd cd /usr/pkgsrc/devel/scmgit/; make install
For other Linux distributions please check your vendor documentation.
Now that you have Git set up and your SSH keys entered into Gerrit, it’s time to configure your personal info.
Now that you have Git on your system, you’ll want to do a few things to customize your Git environment. You should have to do these things only once; they’ll stick around between upgrades. You can also change them at any time by running through the commands again.
Git comes with a tool called git config that lets you get and set configuration variables that control all aspects of how Git looks and operates.
git config -l
Set your username and email
Git tracks who makes each commit by checking the user’s name and email. In addition, we use this info to associate your commits with your gerrit account. To set these, enter the code below, replacing the name and email with your own.
git config --global user.email "firstname.lastname@example.org" git config --global user.name "preilly"
Download the Examples extension using Git
Let's practice downloading the Examples extension. Simply run the following on your command line:
git clone https://gerrit.wikimedia.org/r/p/test/mediawiki/extensions/examples.git
This will copy the entire history of the Examples extension repository. You will have a working directory of the extension's main branch, so you can look at the code and start editing it.
If you change into the new directory, you can see the .git subdirectory. That is where all the project data is.
By default, Git will create a directory that has the same name as the project in the URL you give it - basically whatever is after the last slash of the URL. If you want something different, you can just put it at the end of the command, after the URL. So, in this example you will have a "examples" directory.
Prepare to work with gerrit
In order to work with gerrit, you need to add a "change id" to your commit summary (you can see these if you browse changes on gerrit, they look like Change-Id: Ibd3be19ed1a23c8638144b4a1d32f544ca1b5f97 starting with an I). Each time you amend a commit in response to gerrit feedback git gives it a new commit ID, but because this change ID stays the same gerrit will keep track of it as a new "patch set" addressing the same change.
There's a git add-on called git-review that manages this and other aspects of using gerrit.
First, install the release version of git-review.
Generally, the easiest way to get last version is to install it using the python package installer
$ sudo pip install git-review
If that worked, skip to the next section. If you need to install pip, then:
In Debian/Ubuntu, to install
$ sudo apt-get install python-pip
On OpenSuse, install via YaST
python-setuptools, then run
$ sudo easy_install pip
$ sudo pip install git-review
On FreeBSD, you directly have a kept up to date devel/git-review port:
$ cd /usr/ports/devel/git-review $ make install
Note: if you don't have
apt-get but have
python installed, you can use this:
$ sudo easy_install pip $ sudo pip install git-review
Or, if you already have git-review installed, you can upgrade it using
$ sudo pip install --upgrade git-review
git is a distributed version control system; it's possible to fetch code from one host, push your changes to another, and submit them to a third for gerrit code review. It's a lot simpler to work with one "remote" for all three. Since git's default remote name is "origin" and most guides on the web use that name, let's tell git-review to use this as well. In your home directory, edit .config/git-review/git-review.conf and add the two lines
[gerrit] defaultremote = origin
Setting up git-review
After cloning a repository, you need to set it up for git-review. This will automatically happen the first time you try to submit a commit, but it's generally better to do it right after cloning. In your project's directory ("examples"), enter
$ git review -s
If you see "the authenticity of host gerrit.wikimedia.org can't be established..." Don’t worry, this is supposed to happen. Type “yes”.
This may ask you for your git username, if it's different from the shell username you're using.
How to submit a patch
The main avenue for submitting changes to MediaWiki is to first join the MediaWiki development community so you can submit changes to Gerrit, our code review tool. Getting developer access is relatively easy.
Make sure that your master branch (the branch created when you initially cloned the repository) is up to date:
git pull origin master
However, note that a few repositories use different terms (for example operations/puppet has a "production" instead of a "master" branch).
Create a branch
First, create a local branch for your new change. Give the branch a short but reasonably descriptive name.
git checkout -b BRANCHNAME master
This will create a new branch (BRANCHNAME) from 'master' and check it out for you. This is equivalent to doing
git branch BRANCHNAME master git checkout BRANCHNAME
Use a descriptive branch name. If you are working on a bug, include its id, for example "2012/bug12345".
Make and commit your change
Modify your local code in some fashion.
Then check the changes you've made, within the file(s) and within the directory:
Without any extra arguments, a simple git diff will display in unified diff format (a patch) what code or content you've changed in your project since the last commit that are not yet staged for the next commit snapshot.
Then check the changes you've made, within the file(s) and within the directory:
You run git status to see if anything has been modified and/or staged since your last commit so you can decide if you want to commit a new snapshot and what will be recorded in it.
This will show all modified files. To prepare submitting a file, you should add your changes to the index (the staging area between your working copy and your local repository), which is done by using the git add command. (Note that git add newly stages files. But git -a stages any files previously committed.)
git add Example/Example.body.php
You run git add on a file when you want to include whatever changes you've made to it in your next commit snapshot. Anything you've changed that is not added will not be included - this means you can craft your snapshots with a bit more precision.
While doing so you can always review the change already added to the staging area by invoking git status and look at the diff with
git diff --cached
The git diff --cached command will show you what contents have been staged. That is, this will show you the changes that will currently go into the next commit snapshot.
Once you are happy with the change list, you can add them to your local repository by using
Now that you have staged the content you want to snapshot with the git add command, you run git commit to actually record the snapshot.
You will then be prompted with your favorite editor to add a descriptive message for this commit. This is what other people will see when you will later push that commit to another repository. If you do not manually add a Change-ID line to your commit message, it will be automatically generated and added for you.
You can repeat this step over and over until you have a set of changes that you want to have pushed to the master branch. One of the cool things about git is that when you git commit, you are committing to your local copy. This means you can commit as often as you like without potentially screwing things up for another developer on the project, unlike in SVN where you would want to be very careful that the changes you commit would not cause things to break.
Hence the workflow is something like:
# Add change: $ git add <some file> # Verify list of files added to the staging area $ git status # Review diff of changes staged: $ git diff --cached # repeat until you are happy with your changes $ git commit <edit commit message>
Prepare to push your change set to Gerrit
Before your changes can be merged into master, they must undergo review in Gerrit.
But first, it's a good idea to synchronize your change set with any changes that may have occurred in master while you've been working. From within the branch you've been working on, execute the following command:
git pull origin master git rebase master
git pull will update the code in your local copy of the master branch.
Then, git rebase will temporarily set aside the changes you've made in your branch, apply all of the changes that have happend in master to your working branch, then merge all of the changes you've made back into the branch. Doing this will help avoid future merge conflicts. Plus, it gives you an opportunity to test your changes against the latest code in master.
Once you are satisfied with your change set and you've rebased against master, you are ready to push your code to Gerrit for review.
Push your change set to Gerrit
If you installed
git review and you ran
git review -s to set it up for this repository, the command to push changes to Gerrit is very simple:
git review -R
Upon success, you'll get a confirmation and a link to the changeset in Gerrit. New patchset: preilly; "Added get version method to extension" [test/mediawiki/extensions/examples] (master) - https://gerrit.wikimedia.org/r/9332
(If you forgot to run
git review -s, "remote" will complain about "missing Change-id in commit message". But it will also suggest a commit message with a Change-Id: INNNXXXNNN... line. Copy that line starting with "Change-Id", run
git commit --amend, and paste the Change-Id line at the end. Then repeat
git review -R and it should complete.)
You can view this change in the Gerrit Web UI:
If you want to see your changed files in their context then click on the
(gitweb) links and the
tree link (you then will see).
Amending a change
Sometimes, you might need to amend a submitted change. You can amend your own changes as well as changes submitted by someone else, as long as the change hasn't been merged yet.
If you have git-review, checkout the change like this:
git review -d <change number>
Note, if you already have the change in a branch on your local repository, you can just check it out instead:
git checkout <branch name>
git review -d 9332
Or, if you already have the branch,
git checkout review/preilly/2012/bug12345
should accomplish the same thing.
Next, make some changes.
git add the files as needed, then commit the change (ensuring you are amending the commit):
git add Example/Example.body.php git commit --amend
NOTE: DO NOT use the
-m flag to specify a commit summary: that will override the previous summary and regenerate the Change-Id. Instead, use your text editor to change the commit summary if needed, and keep the Change-Id line intact.
Push the change
git review -R
- The -R is important here. It tells git-review to not rebase your change against master, which clutters diffs between patch set 1 and 2.
New patchset: preilly; "Added get version method to extension" [test/mediawiki/extensions/examples] (master) - https://gerrit.wikimedia.org/r/9332
git review complains about multiple commits
If git review asks you if you really want to submit multiple commits, and lists a bunch of unrelated commits from different branches, try either of this:
git fetch --all
git remote update
Both commands do exactly the same thing, they fetch objects from all remote repositories set. So just pick the command you remember easily and forget about the other one.
git complains "you are in the middle of a merge -- cannot amend"
When after rebasing and merging your
git commit --amend
message: fatal: You are in the middle of a merge -- cannot amend.
apply these steps and reapply your changes
git stash git reset --hard git checkout master git review -d <change number> git stash pop git commit -a --amend
How we review code
Code review is an essential part of our contribution workflow. The principle is basic: any patch must be reviewed by others before being merged.
This means that you code will need reviewers. Check our advice for getting reviews.
Review before merge
It's important to us to have a review-before-merge workflow for MediaWiki core and also for any extension we deploy. We will also offer that option to any extension author who wants it for their extension. The one exception is localisation and internationalisation commits, which will be able to be pushed without review.
Who can review? Gerrit project owners
Who has the ability to do code review?
We use gerrit.wikimedia.org to manage code review. Anyone can ask for a Gerrit account (Get an account!). Within Gerrit, anyone can comment on commits and signal their criticisms and approvals. Anyone can give a nonbinding "+1" to any commit. However, for any given repository ("Gerrit project"), only a small group of people will have the ability to approve code within Gerrit and merge it into the repository. (Within gerrit, this superapproval is a "+2" even though that's a misleading name, because two +1 approvals DO NOT add up to a +2.) These people are "Gerrit project owners". To learn about becoming a Gerrit project owner, see Gerrit/Project ownership.
Even within a Gerrit project, we can also specify particular branches that only specific people can pull into.
How to comment on, review, and merge code in Gerrit
Anyone can comment on code in Gerrit.
Viewing and commenting on code
- Make sure you have a https://gerrit.wikimedia.org login (Get an account!). If you don't know, try logging in at https://labsconsole.wikimedia.org; the username and password should be the same. If you can't, ask in for someone to help.
- Log in to Gerrit. If you know the changeset you want to look at (URL will look like https://gerrit.wikimedia.org/r/#change,8939 ), go to that. Otherwise, use the search box and try searching. There is no fulltext search in Gerrit, but you can search by author ("Owner"), Gerrit project, branch, changesets you've starred, etc. The Gerrit search documentation covers all of the different search operators you can use.
- The changeset has a few important fields, links and buttons:
- Reviewers. 'jenkins-bot' is the autoreviewer that auto-verifies anything that passes the Jenkins tests. If it passes, you see a green checkmark. If it fails, a red X. A changeset needs a green checkmark before anyone can merge it.
- Add reviewer (manually ping someone to request their review. It'll show up in their Gerrit stream)
- Side-by-side diff button:
- Opens the diff. You can double-click on a line and comment on that line, then save a draft comment! Then, click "Up to change" to go back to the changeset.
- Abandon Change button (you'll see this if you wrote this diff. This action removes the diff from the merge queue, but leaves it in Gerrit for archival purposes)
- Review button:
- The purpose of this page is to signal your thoughts on the commit. You can leave an overall comment under Cover Message. You can also view inline comments from the diff that are still in draft form and awaiting publication.
- If upon code review, you approve, use "+1" under Code Review; otherwise, use "-1" to disapprove. These numbers are nonbinding, won't cause merges or rejections, and have no formal effect on the code review.
- If you have actually tested the code, and it behaves as intended, use "+1" under Verified; if you tested and it behaves incorrectly, use "-1". If you did not test this change, leave Verified at 0. To publish your draft inline comments plus your comments on the diff as a whole, click Publish.
Comparing patch sets
Every time you amend your commit and submit it for review, a new patch set is created. You can compare the different patch sets like this:
- Select the older patch set in the "Old Version History" list.
- Expand the newer patch set details by clicking the arrow near it.
- Click Side-by-Side. Note that in Gerrit 2.3 this only works if you open the diff in the same tab, so don't open it in a new tab.
Formally reviewing and merging or rejecting code
If you are one of the Gerrit project owners, you'll also see:
- Abandon Change button
- on the Review page, additional Code Review options to +2 (approve) or -2 (veto) a diff, and a Publish And Submit button (publish your comment and merge diff into the branch, in 1 step)
- Submit Patch Set 1 button (merge -- only useful if you or someone else has already given a +2 approval to the diff, but not merged it)
And once you've merged something into the example Gerrit project you'll see it in https://gerrit.wikimedia.org/r/gitweb?p=test/mediawiki/extensions/examples.git;a=summary .
If you merge a commit that references a Bugzilla bug, please go to that bug and mark it RESOLVED: FIXED and reference the merge ID.
Also useful are these pages:
- Git and gerrit in Wikimedia labs
- An intro-to-git "training mission"
- An introduction to git-svn for Subversion/SVK users and deserter by Sam Vilain
- GitMagic A git guide with translations - recommended
- A git branching model
- git: the simple guide
- Git Community Book will take you gently into Git internals. Actually it is very hard to "get" git until knowing something about how it works internally. After this, everything become simple, just hidden without convoluted and anti-productive user interface.
- How we're going to use Git: slides by Brion Vibber, as presented autumn 2011
- Audio from Git presentation by Brion Vibber, autumn 2011
- Notes from Brion's presentation
- Git's website with documentation
- MediaWiki Git Guide (MGG) - selection of relevant Git pages in the MW Virtual Library; for PDF generation or printing
- A very useful Git guide
Also useful (if you subtract away GitHub stuff):