This page provides some GIT "best practices" based upon the experiences of the page's author. I welcome suggestions from other GIT users to help improve this page and to learn other ways to use GIT (which is a very useful, albeit complicated, tool.) Some helpful links to GIT resources:
and to lighten things up:
- a GIT man page generator which creates fake man pages which seem scarily real.
During development, a project's state changes: files are added or removed, registered files get modified, etc. At certain points you decide the current state should be saved, which is when you commit the changes to the repository. This document uses a diagram notation to describe the contents of a repository. An example is:
[A]-[B]-[C] ^ master
The diagram above shows a repository containing three states of a project, A, B, and C. The diagram doesn't show differences between the states; it could be a change of a single line in a file or the addition of a hundred files in a subdirectory. The diagram just shows that the project's state has changed and has been recorded in the repository. The direction of the diagram always goes left to right. For instance, state B happened after state A. Branch tags (like "master", in the example) will point to the state with which they're associated.
When you commit changes to a
git repository, a 160-bit hash value is computed and associated with that commit. The diagrams would be difficult to read if the states were labelled using these hash values so, in this document, we'll just use single letters to label a state.
Creating a Repository¶
Creating a Local Repository¶
Sometimes you have a project that's not ready to share, but you still want to use
git to save stable points of your development. In this case, you go to the base directory of your project and run
Your local project is now associated with a
git repository! You can use the commands discussed later to add files and commit changes to this local repo.
Creating a Redmine remote repository¶
Go to a project's settings and click the "repository" tab. Select Git as the repo format. There can only be one main repository, but there can be many secondaries, each requires a unique identifier. You can use an identifier for the main repository, but it is usually left blank. When blank, the main repository's identifier is set to the project's identifier. Keep the identifier field simple; it gets appended to the project's identifier, and you run into problems when the combined name is over 30 characters. For the path to the repo, use "create_me". Enable the "Report last commit” feature.
Wait 1/2 hour, or so, for the system to create the empty repository. If you refresh the "Settings/Repository" page, you’ll eventually see "create_me" change to an actual path to the repository.
Cloning a Repository¶
git is a distributed SCM system which means, in
git's view, no repository is more important than any other. It is a project's agreed upon conventions that a particular repository is the "official" one, not
git's. Redmine projects can have a
git repository associated with them. In these cases, the Redmine copy is the "official" copy and every remote copy tries to keep synchronized with it.
Project repositories in Redmine have a unique identifier. To allow all members of a project to make changes, Redmine creates a user account and adds the project's members to its
.k5login file. An active valid Kerberos ticket is required to interact with Redmine. The user account is the same as the identifier with "p-" prepended. If, for example, your project's identifier was
my-project, you could clone its repository with this command:
$ git clone ssh://firstname.lastname@example.org/cvs/projects/my-project
This command creates a subdirectory called
my-project and places the contents of the repository in it. This directory is called the working directory and contains not only source code you can edit, but has a copy of the repository itself (found in the
.git directory -- don't mess with the contents of this directory!)
git remembers, in its local configuration, from which repository your copy came. By default, it is given the label
origin (see the
git remote command below.)
Saving Your Changes¶
After a development/debugging cycle, you'll have some changes that need to be saved to your local repository. In the next section, we'll show how to forward these commits to the remote repository.
Saving to the Local Repository¶
Committing to the local repository takes two steps in
git. The first step is to specify what needs to be committed. The second step is to actually commit it. This may seem like extra work, and it is, but it provides some powerful abilities that other SCMs don't allow.
$ git add file1 [file2 ...]
This command adds all the changes in the specified files to the commit staging area. Incidentally, this is the same command when adding a new file to
git (the changes being staged, in this case, are the entire contents of the file.)
$ git commit
All the commits in the staging area will be committed. An editor will be started so the commit message can be entered.
$ git add -p
This is a powerful version of the
add command. The
-p option puts the
add command in an interactive "patch mode":
git presents each group of changes in each file and asks whether you want to add them to the commit. For instance, while fixing a bug you discover a typo in an unrelated comment. Rather than making one commit with both of your fixes, you can use "
git add -p" to add the typo patch, commit it, then stage the rest of the changes and commit them. Keeping each commit as small as possible helps when you get the inevitable conflict when merging, so the
-p option is a fantastic feature.
For those of you that don't subscribe to formal processes, the following command adds all changes to the staging area and commits with one command. This essentially does the same as
cvs commit. I don't recommend using it, though.
$ git commit -a
Saving to the Remote Repository¶
After a "
git commit", your changes are only in your local copy of the repository.
git is a DVCS with no central, official repository so committing to the local repo is considered enough. This allows developers to work on projects and commit their work without a network even being present. It's a project's decision to specify which repository eventually needs to receive everyone's efforts.
In most cases, you "push" your changes back to the original repo with the
$ git push
When a project is cloned,
git automatically records from where the source came. It assumes that eventually you're going to send your changes back to it.
git gives it the easier-to-remember label, "
When a branch is first pushed to a remote repository, you need to specify the target repo and the branch name. In fact, if you clone an empty repository (like when you make a new project on Redmine), you have to use this extended command to push the first commit(s) of the
$ git push origin master
Commands Providing Information¶
When in the working directory,
git provides several commands that provide useful information.
$ git status On branch master Your branch is up-to-date with 'origin/master'. nothing to commit, working directory clean
This command shows the current status of the working directory. The top of the output indicates the branch checked-out. The command also shows which files are ready to commit; which files have changes, but haven't been staged for commit; and which files aren't tracked by
git at all.
git status doesn't use the network so it doesn't contact the remote repository and only uses information that was cached during the last
git pull and
git push commands. This means if you've committed two times since the last
git pull, the output from
git status will mention you're "2 commits ahead of 'origin/master'", or something like that.
$ git log
Displays the history of commit messages along with the commit hash, date of commit, and who committed it.
$ git diff
diff command shows all the differences between the files in the working directory and the staging area (not the repository!) In other words, it shows the changes that haven't been staged to be committed. The
diff command has many, many options. All
git commands recognize a set of special symbols.
HEAD, for instance, refers to the head commit of the current branch. So "
git diff HEAD" shows the differences between the working directory and your local repository, bypassing the staging area. You can also compare against a remote repo with "
git diff origin/master" which compares the working directory against the
master branch on the remote repository,
$ git branch -v
Shows all the branches in your local repository and marks the one currently checked-out in your working directory (with an asterisk.)
$ git remote -v
Shows all remote repositories the local repository interacts with. 99% of the time, this will only display the default remote repository,
origin, with its network path.
Getting Updates from Remote¶
The main purpose of SCMs is to keep software in sync with all its developers. This means occasionally pulling in changes made by colleagues.
pull command is the simplest way to do this.
$ git pull
This command retrieves, from the remote repository, changes not present in the local repository and applies them. Then it attempts to merge the changes to the checked out copy. Usually the merge succeeds but, once in a while, the remote change modifies a section of code that has local changes. This is called a "conflict".
git will mark the conflicting areas of the file and won't finish the merge until you resolve all conflicts. Using an editor, review the marked up areas and decide how the final version should look. When all conflicts have been resolved, use "
git add" to add the resolved files and "
git commit" to finish the merge.
git pull" and "
git push" can get most of the job done. However, when two developers have their own changes at the head of the branch, merging them creates a slight mess. Part of
git's philosophy is to make sure the development history is clear and easy to understand. This section will discuss how to use branches to make development more robust and keep the project's history organized. The following is not the only way to use branches, nor is it the way. It is, however, a useful method the author uses that works well.
git names the main branch of development "
master". Released code (i.e. code to be used by other projects) is typically taken from the end of the
master branch. Following this convention requires anything committed to
master to be reasonably debugged and compiles cleanly. This may be impossible while debugging or implementing new features so branches should be used to isolate unfinished work from the
For the following discussion, let's assume a project called "Project" exists on a remote machine and has three commits, A, B, and C, on the
Remote Repository: [A]-[B]-[C] ^ master
Creating a Branch¶
When you clone a repository, unless otherwise specified, you'll be using the
$ git clone PATH_TO_PROJECT project $ cd project
Your project's environment is now:
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C] ^ master Working Directory (on master branch): C
The local copy of the repository is stored in a subdirectory named
.git. Note that the local repository matches the remote's and your working directory matches the state of the project that was last committed.
You want to do some development on this version of the code without affecting the
master branch until you're through, so you create a branch. You can name it anything you wish but in this example we'll call it
$ git checkout -b devel
Now your project's environment has slightly changed:
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C] ^ master devel Working Directory (on devel branch): C
A new branch called
devel is created which refers to the same changeset as
master and your working directory is now tracking it. But now we can commit changes to the
devel branch without affecting
master. Assume we've made changes to the files in the project. The source code is in a new, unique state which we'll call "D". If these changes are committed, the new environment becomes:
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C]-[D] ^ ^ | devel master Working Directory (on devel branch): D
master hasn't changed. .
devel is your local development and, once you're done with your changes (and you tested them!), you'll merge them back into
master and push to the remote server.
You can see what branches are available in your local repository:
$ git branch -v master 0123456 Commit message for C * devel 6543210 Commit message for D
The output shows the names of the branches (the asterisk shows which branch is current checked-out in the working directory.) The next field is the commit hash followed by the commit message at the head of the branch.
checkout command to switch between branches. Using the example environment from the previous section,
$ git checkout master
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C]-[D] ^ ^ | devel master Working Directory (on master branch): C
Switching back to
$ git checkout devel
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C]-[D] ^ ^ | devel master Working Directory (on devel branch): D
Note that switching branches doesn't affect the repositories. It only affects the contents of the working directory and which branch the working directory is tracking.
Let's assume you've made a few more commits, states E and F, and you're ready to merge them back into the main repository. At this point, the project environment looks like this:
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C]-[D]-[E]-[F] ^ ^ master devel Working Directory (on devel branch): F
To merge it back properly, we switch to the
master branch, make sure no one has added to it, merge
devel, and push
master back to the remote repository. You'll note that in the above diagram, the remote repository matches the local's
master branch. Later, we'll cover the case when some one has pushed commits to the remote repository.
First, switch to the
$ git checkout master
Make sure no one added to master:
$ git pull Already up-to-date.
The "Already up-to-date." message means you're in sync! This is best situation to have since it means everything else will go smoothly. It's safe to merge and push the changes in
$ git merge devel
Remote Repository: [A]-[B]-[C] ^ master Local Repository: [A]-[B]-[C]-[D]-[E]-[F] ^ devel master Working Directory (on master branch): F
By the way, in the output generated by the merge, you may see the term "Fast-forward". This means
git was able to simply move the head of the branch to a new commit position without performing any
diff calculations. Fast-forward merges are the cleanest merges since they don't generate any new artifacts in the project's history.
Now we can push the
master branch to the remote repository:
$ git push
Remote Repository: [A]-[B]-[C]-[D]-[E]-[F] ^ master Local Repository: [A]-[B]-[C]-[D]-[E]-[F] ^ devel master Working Directory (on master branch): F
Congratulations! Your changes are incorporated in the project and are available to other developers! Also, the project's history is easy to follow1.
At Fermilab, most projects are worked on by one person at a time which means the previous section's example is what normally happens. However, if you are working on an active project that has multiple developers, you'll occasionally have to pull in their changes before pushing yours. Let's reset the environment back to the previous section's starting point, except let's assume the remote repository has been committed to while we were developing our code:
Remote Repository: [A]-[B]-[C]-[G]-[H] ^ master Local Repository: [A]-[B]-[C]-[D]-[E]-[F] ^ ^ master devel Working Directory (on devel branch): F
When we switch to the
master branch and pull changes from the remote repository, instead of getting "Already up-to-date.", you see output showing which files have been updated (due to the remote changes.) The environment has now diverged:
Remote Repository: [A]-[B]-[C]-[G]-[H] ^ master Local Repository: [G]-[H] / ^ [A]-[B]-[C] master \ [D]-[E]-[F] ^ devel Working Directory (on master branch): H
master branch matches the remote's history, which is good. However
devel is based off the state C and doesn't incorporate the changes of G or H. Before we can confidently merge our changes, we need to make sure G and H didn't break our code (because G and H are on
master we have to assume H is the new, stable version of our project.)
git has a powerful command,
rebase which helps us in this situation. Go back to the
devel branch and tell
git to re-base our branch to the new
$ git checkout devel $ git rebase master
If there were no conflicts, the new environment state is:
Remote Repository: [A]-[B]-[C]-[G]-[H] ^ master Local Repository: [A]-[B]-[C]-[G]-[H]-[D']-[E']-[F'] ^ ^ master devel Working Directory (on devel branch): F'
rebase command saves all the commits on
devel from where
master split and then applies them, one by one, to the new head of
master. The diagram shows the applied commits as D', E', and F' because they may not be the exact same patches as before since they may be applied to a file changed by G or H. If a conflict occurs,
git will stop to let you resolve it. Once resolved, use "
git add file" to add the file and then "
git rebase --continue" to apply the rest of the commits.
At this point, you should re-test your changes because the other developer's commits may have affected the features you added. This may require you to commit more on the
devel branch. Once you've tested the code again, start over at the top of this section: switch to
master; pull from remote; if no changes from remote, merge from
devel; if remote has changes, pull and rebase. Repeat, ad nauseum.
git supports tagging a state of the project with a label. The command is
$ git tag TAG ITEM
ITEM can be a commit hash, a branch name, or another tag. Using the last example's environment, if we wanted to tag state C of the project with the label "v1.0", we would find the hash of state C using "
git log" and then tag it with
git tag v1.0 STATE_C_HASH
Remote Repository: [A]-[B]-[C]-[G]-[H] ^ master Local Repository: [A]-[B]-[C]-[G]-[H]-[D']-[E']-[F'] ^ ^ ^ v1.0 master devel Working Directory (on devel branch): F'
To push the tag to the remote repo:
$ git push --tags
Branches and tags are very similar in git; they both mark a state in the history of a project. The big difference is that when your working directory is attached to a branch "tag" and you commit, the branch is updated to refer to the new commit.
Branching Used for Formal Processes¶
This section describes a way to use
git branches for organizing deployment. A project wishing to use this approach must follow these conventions:
- All major versions have their own branch named "
#is the major version number. For instance, all 1.x releases are found in the
masterbranch is typically deleted. If you choose to keep
master, there's extra steps to keep it synchronized with its corresponding release branch. It's simpler to remove it.
- All development should be done on private, local branches that forked from the associated release branch.
- When merging back to the release branch, pull changes before trying to merge (as described in the previous section on "Branching").
- When merging, use "
git merge --no-ff devel", where devel is the name of your development branch. The merge commit's message should summarize the changes done on the branch.
- Tag the merge with a version number (for instance, the 6th merge on the
v1.xbranch should be tagged
A new project is started, which will take a while before it has a stable API. As the API grows, we want to release stable points for people to use and test. This will be the
v0.x branch. When you create a new
git project, your environment is like this:
Local Repository: [A] ^ master Working Directory (on master branch): A
Now we create the
v0.x branch and delete the master:
$ git checkout -b v0.x $ git branch -d master $ git push origin v0.x
Remote Repository: [A] ^ v0.x Local Repository: [A] ^ v0.x Working Directory (on v0.x branch): A
Now you want to add features to the
v0.x branch. You create a
devel branch, which you will never push to the remote repo, and make several commits. After testing and debugging, you're ready to release the first usable version,
v0.1. Make sure any code in your project that needs to report the version number has been updated and committed, too. The environment looks like this:
Remote Repository: [A] ^ v0.x Local Repository: [A]-[B]-[C]-[D] ^ ^ v0.x devel Working Directory (on devel branch): D
After making sure the remote and local repos are in sync, we can merge in the changes:
$ git checkout v0.x $ git merge --no-ff devel $ git tag v0.1 $ git push origin v0.x $ git push --tags
We use the
--no-ff option to force a commit on the branch, even if a fast-forward would work. The repo now looks like this:
Remote Repository: [A]-------------[E] \ / ^ [B]-[C]-[D] v0.x v0.1 (tag) Local Repository: [A]-------------[E] \ / ^ [B]-[C]-[D] v0.x ^ v0.1 (tag) devel Working Directory (on v0.x branch): E
Node E is called a "merge commit". It consists of a merge between its two parent nodes. This approach keeps the individual (possibly broken) commits out of the release branch. After another session to add a feature,
v0.2 is released:
Remote Repository: [A]-------------[E]---------[H] \ / ^ \ / ^ [B]-[C]-[D] | [F]-[G] | | | v0.1 (tag) v0.x v0.2 (tag) Local Repository: [A]-------------[E]---------[H] \ / ^ \ / ^ [B]-[C]-[D] | [F]-[G] | | | v0.1 (tag) v0.x v0.2 (tag) Working Directory (on v0.x branch): H
As you can see, the backbone of the release branch are the stable points of the project. A "
git log --merge v0.x" will display only the merge commits with their associated summary messages.
1 One might argue that the
master branch contains the "broken" states of D and E. It is possible, using
git , to merge all the changes in
devel into one commit on the
master branch so no one could gain access to the D and E states. This article doesn't cover that usage.