Git doesn't have a central server like Subversion. All of the commands so far have been done locally, just updating a local database. To collaborate with other developers in Git, you have to put all that data on a server that the other developers have access to. The way Git does this is to synchronize your data with another repository. There is no real difference between a server and a client - a Git repository is a Git repository and you can synchronize between any two easily.
Once you have a Git repository, either one that you set up on your own server, or one hosted someplace like GitHub, you can tell Git to either push any data that you have that is not in the remote repository up, or you can ask Git to fetch differences down from the other repo.
You can do this any time you are online, it does not have to correspond
with a commit or anything else. Generally you will do a
number of commits locally, then fetch data from the online shared repository
you cloned the project from to get up to date, merge any new work into the
stuff you did, then push your changes back up.
In a nutshell you can update your project with git fetch
and share your changes with git push. You can manage your
remote repositories with git remote.
Unlike centralized version control systems that have a client that is very different from a server, Git repositories are all basically equal and you simply synchronize between them. This makes it easy to have more than one remote repository - you can have some that you have read-only access to and others that you can write to as well.
So that you don't have to use the full URL of a remote repository every
time you want to synchronize with it, Git stores an alias or nickname for
each remote repository URL you are interested in. You use the
git remote command to manage this list of remote repos that
you care about.
Without any arguments, Git will simply show you the remote repository
aliases that it has stored. By default, if you cloned the project (as
opposed to creating a new one locally), Git will automatically add the
URL of the repository that you cloned from under the name 'origin'. If
you run the command with the -v option, you can see the
actual URL for each alias.
$ git remote origin $ git remote -v origin git@github.com:github/git-reference.git (fetch) origin git@github.com:github/git-reference.git (push)
You see the URL there twice because Git allows you to have different push and fetch URLs for each remote in case you want to use different protocols for reads and writes.
If you want to share a locally created repository, or you want to take
contributions from someone else's repository - if you want to interact in
any way with a new repository, it's generally easiest to add it as a remote.
You do that by running git remote add [alias] [url]. That
adds [url] under a local remote named [alias].
For example, if we want to share our Hello World program with the world, we can create a new repository on a server (I'll use GitHub as an example), which should give you a URL, in this case "git@github.com:schacon/hw.git". To add that to our project so we can push to it and fetch updates from it we would do this:
$ git remote $ git remote add github git@github.com:schacon/hw.git $ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push)
Like the branch naming, remote alias names are arbitrary - just as 'master'
has no special meaning but is widely used because git init
sets it up by default, 'origin' is often used as a remote name because
git clone sets it up by default as the cloned-from URL. In
this case I've decided to name my remote 'github', but I could have really
named it just about anything.
Git addeth and Git taketh away. If you need to remove a remote - you are
not using it anymore, the project is gone, etc - you can remove it with
git remote rm [alias].
$ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) $ git remote add origin git://github.com/pjhyett/hw.git $ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/pjhyett/hw.git (fetch) origin git://github.com/pjhyett/hw.git (push) $ git remote rm origin $ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push)
In a nutshell with git remote you can list our
remote repositories and whatever URL
that repository is using. You can use git remote add to
add new remotes and git remote rm to delete existing ones.
Should you ever need to update a remote's URL, you can do so with
the git remote set-url command.
$ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/pjhyett/hw.git (fetch) origin git://github.com/pjhyett/hw.git (push) $ git remote set-url origin git://github.com/github/git-reference.git $ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/github/git-reference.git (fetch) origin git://github.com/github/git-reference.git (push)
In addition to this, you can set a different push URL when you
include the --push flag. This allows you to fetch from
one repo while pushing to another and yet both use the same remote alias.
$ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/github/git-reference.git (fetch) origin git://github.com/github/git-reference.git (push) $ git remote set-url --push origin git://github.com/pjhyett/hw.git $ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/github/git-reference.git (fetch) origin git://github.com/pjhyett/hw.git (push)
Internally, the git remote set-url command calls
git config remote, but has the added benefit of reporting
back any errors. git config remote on the other hand, will
silently fail if you mistype an argument or option and not actually set
anything.
For example, we'll update the github remote but
instead reference it as guhflub in both invocations.
$ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/github/git-reference.git (fetch) origin git://github.com/github/git-reference.git (push) $ git config remote.guhflub git://github.com/mojombo/hw.git $ git remote -v github git@github.com:schacon/hw.git (fetch) github git@github.com:schacon/hw.git (push) origin git://github.com/github/git-reference.git (fetch) origin git://github.com/github/git-reference.git (push) $ git remote set-url guhflub git://github.com/mojombo/hw.git fatal: No such remote 'guhflub'
In a nutshell, you can update the locations of your remotes
with git remote set-url. You can also set different push
and fetch URLs under the same remote alias.
Git has two commands to update itself from a remote repository.
git fetch will synchronize you with another repo, pulling down any data
that you do not have locally and giving you bookmarks to where each branch on
that remote was when you synchronized. These are called "remote branches" and are
identical to local branches except that Git will not allow you to check them out -
however, you can merge from them, diff them to other branches, run history logs on
them, etc. You do all of that stuff locally after you synchronize.
The second command that will fetch down new data from a remote server is
git pull. This command will basically run a git fetch
immediately followed by a git merge of the branch on that remote
that is tracked by whatever branch you are currently in. I personally don't much
like this command - I prefer running fetch and merge
separately. Less magic, less problems. However, if you like this idea, you
can read about it in more detail in the
official docs.
Assuming you have a remote all set up and you want to pull in updates, you
would first run git fetch [alias] to tell Git to fetch down all the
data it has that you do not, then you would run git merge [alias]/[branch]
to merge into your current branch anything new you see on the server
(like if someone else has pushed in the meantime). So, if I were working on my
Hello World project with several other people and I wanted to bring in any changes
that had been pushed since I last connected, I would do something like this:
$ git fetch github remote: Counting objects: 4006, done. remote: Compressing objects: 100% (1322/1322), done. remote: Total 2783 (delta 1526), reused 2587 (delta 1387) Receiving objects: 100% (2783/2783), 1.23 MiB | 10 KiB/s, done. Resolving deltas: 100% (1526/1526), completed with 387 local objects. From github.com:schacon/hw 8e29b09..c7c5a10 master -> github/master 0709fdc..d4ccf73 c-langs -> github/c-langs 6684f82..ae06d2b java -> github/java * [new branch] ada -> github/ada * [new branch] lisp -> github/lisp
I can see that since the last time I synchronized with this remote, five branches have been added or updated. The 'ada' and 'lisp' branches are new, where the 'master', 'c-langs' and 'java' branches have been updated. In this case, my team is pushing proposed updates to remote branches for review before they're merged into 'master'.
You can see the mapping that Git makes. The 'master' branch on the remote
repository becomes a branch named 'github/master' locally. That way now I can
merge the 'master' branch on that remote into my local 'master' branch by running
git merge github/master. Or, I can see what new commits are on that
branch by running git log github/master ^master. If your remote
is named 'origin' it would be origin/master instead. Almost any
command you would run using local branches you can use remote branches with too.
If you have more than one remote repository, you can either fetch from specific
ones by running git fetch [alias] or you can tell Git to synchronize
with all of your remotes by running git fetch --all.
In a nutshell you run git fetch [alias] to synchronize your
repository with a remote repository, fetching all the data it has that you do
not into branch references locally for merging and whatnot.
To share the cool commits you've done with others, you need to push your
changes to the remote repository. To do this, you run
git push [alias] [branch] which will attempt to make your [branch]
the new [branch] on the [alias] remote. Let's try it by initially pushing
our 'master' branch to the new 'github' remote we created earlier.
$ git push github master Counting objects: 25, done. Delta compression using up to 2 threads. Compressing objects: 100% (25/25), done. Writing objects: 100% (25/25), 2.43 KiB, done. Total 25 (delta 4), reused 0 (delta 0) To git@github.com:schacon/hw.git * [new branch] master -> master
Pretty easy. Now if someone clones that repository they will get exactly what I have committed and all of its history.
What if I have a topic branch like the 'erlang' branch we created earlier and I just want to share that? You can just push that branch instead.
$ git push github erlang Counting objects: 7, done. Delta compression using up to 2 threads. Compressing objects: 100% (6/6), done. Writing objects: 100% (6/6), 652 bytes, done. Total 6 (delta 1), reused 0 (delta 0) To git@github.com:schacon/hw.git * [new branch] erlang -> erlang
Now when people clone or fetch from that repository, they'll get an 'erlang' branch they can look at and merge from. You can push any branch to any remote repository that you have write access to in this way. If your branch is already on the server, it will try to update it, if it is not, Git will add it.
The last major issue you run into with pushing to remote branches is the
case of someone pushing in the meantime. If you and another developer clone
at the same time, you both do commits, then she pushes and then you try to
push, Git will by default not allow you to overwrite her changes. Instead,
it basically runs git log on the branch you're trying to push and
makes sure it can see the current tip of the server's branch in your push's
history. If it can't see what is on the server in your history, it concludes
that you are out of date and will reject your push. You will rightly have to
fetch, merge then push again - which makes sure you take her changes into
account.
This is what happens when you try to push a branch to a remote branch that has been updated in the meantime:
$ git push github master To git@github.com:schacon/hw.git ! [rejected] master -> master (non-fast-forward) error: failed to push some refs to 'git@github.com:schacon/hw.git' To prevent you from losing history, non-fast-forward updates were rejected Merge the remote changes before pushing again. See the 'Note about fast-forwards' section of 'git push --help' for details.
You can fix this by running git fetch github; git merge github/master
and then pushing again.
In a nutshell you run git push [alias] [branch] to update a
remote repository with the changes you've made locally. It will take what your
[branch] looks like and push it to be [branch] on the remote, if possible. If
someone else has pushed since you last fetched and merged, the Git server will
deny your push until you are up to date.