Development:Git repository

This page provides some information about our switch to git (which, as of Nov 14th, has been completed). As an initial comforting note, git really isn't so complicated as some people would have you believe. A few months ago I became familiar with distributed version control by using Mercurial, and since a few weeks we're using git at work. I think we'll benefit a lot from switching to it on the long run.

The primary repository

Initial setup

With Git, we'll have one repository for each project. The central repositories through which we cooperate are hosted on gitorious.org. Gitorious is a friendly website that is also open source. On Gitorious the main repository for each project is called mainline. Once you click to the mainline repository, you can see several ways to clone it (the new svn checkout).

We've categorized all projects related to The Mana World, so you can easily see the complete list of The Mana World projects on Gitorious. The projects have different clone URLs for read-only or developer access. The URL for developer access is called the "push URL", since it allows you to push commits into the repository via ssh. The list below is for your convenience.

Project	Read-only URL	Push URL (Developers only)	Atom feed
TMW client	git://gitorious.org/tmw/mainline.git	git@gitorious.org:tmw/mainline.git	Atom
This repository contains the client for The Mana World
TMW client data	git://gitorious.org/tmwdata/mainline.git	git@gitorious.org:tmwdata/mainline.git	Atom
This repository has the client data.
TMW server	git://gitorious.org/tmwserv/mainline.git	git@gitorious.org:tmwserv/mainline.git	Atom
This repository contains the our own server.
TMW server data	git://gitorious.org/tmwserv-data/mainline.git	git@gitorious.org:tmwserv-data/mainline.git	Atom
This repository contains the data for TMWServ.
TMW web frontend	git://gitorious.org/tmwweb/mainline.git	git@gitorious.org:tmwweb/mainline.git	Atom
This repository has the web module for TMWServ.
TMW website	git://gitorious.org/tmw/website.git	git@gitorious.org:tmw/website.git	Atom
This repository has our website.
eAthena server	git://gitorious.org/tmw-eathena/mainline.git	git@gitorious.org:tmw-eathena/mainline.git	Atom
This repository contains our hacked up eAthena.
eAthena data	git://gitorious.org/tmw-eathena-data/mainline.git	git@gitorious.org:tmw-eathena-data/mainline.git	Atom
This repository has the data for our eAthena.
TMW Art	git://gitorious.org/tmwart/mainline.git	git@gitorious.org:tmwart/mainline.git	Atom
This repository has some sources for our artwork.

Git uses ssh's private/public key authentication for identifying committers. For development purposes you should clone the "push url", but this requires that you have:

1. Signed up to gitorious.org
2. Generated a private/public ssh keypair (if you haven't got this already)
3. Filled in your public key in your account details on Gitorious

Cloning

If you simply git clone the URL without any additional arguments, it will create the repository in a directory called "mainline". This is generally not what you want. Hence, after the clone url, you should pass the name of the directory you want to have created (just like with Subversion):

$ git clone <clone_url> project

If you want to have all projects in one place, you probably want to do something like this:

$ mkdir tmw
$ cd tmw
$ git clone git@gitorious.org:tmwserv/mainline.git tmwserv
$ git clone git@gitorious.org:tmw-eathena-data/mainline.git eathena-data
etc.
or for all of them in one go (after the cd tmw step):
$ for repo in  tmw tmwdata tmwserv tmwweb tmw-eathena tmw-eathena-data ; do git clone git://gitorious.org/${repo}/mainline.git $repo  ; done

The way Gitorious works, we can't have one top-level "tmw" project under which we put all these repositories, since they're really separate projects. Gitorious allows multiple repositories for each project, but they are clones of each other (you can only create new ones by cloning existing ones). This allows anybody (not just the development team) to make clones and start hacking on them. Changes can easily be merged from one one repository to another. So instead of all the inconvenience with TMW forks we had in the past, now comes the time to encourage people to clone!

Shallow cloning for non-developers

One of our repositories, tmwdata, has grown quite large cause of its long history filled with relatively large binary files. If you are only interested in getting the latest version, and have no need to be able to push back changes, then you can make a shallow clone:

$ git clone --depth 1 git://gitorious.org/tmwdata/mainline.git tmwdata

Working with git

Commit

From now on, a commit is something you do locally. Others won't see your change on Gitorious unless you push it there. You'll notice committing is very fast, and you can commit multiple times before you decide to push. You can also make corrections to your last commit.

Before you start committing, it is important to identify yourself to git, so that it can include the correct authorship information with your commit. You are no longer identified with a username, as was the case with Subversion. You can read exactly how to do this, as well as other useful information geared towards people switching from Subversion, on this page:

• Git - SVN Crash Course: http://git.or.cz/course/svn.html

Pushing

Once you have committed some stuff, you can push these to the repository on Gitorious using git push. This works since by default the push command pushes to a remote called origin, and this remote is automatically set up when you clone. However, the push will fail if there have been new commits on the remote repository. In that case, you'll first have to pull in these changes (just like with Subversion, however Subversion allowed this as long as the same files weren't touched, git doesn't).

Pulling

When you want to get the latest changes from the repository on Gitorious, you generally use git pull. However, note that this command does not work when you have local changes. Also, when you have local commits, the pull command will generate a merge commit (and before that you may have to resolve some conflicts).

If you don't want to create merge commits, but would rather stack your local commits on top of any incoming commits, you should use git pull --rebase. This rebases your local commits on top of the incoming ones. You should never do this when you have pushed these commits elsewhere, so only do it when you are sure the commits are only on your machine.

If you have local changes and want to update your checkout, then there are several options:

• You commit your local changes, and do a pull, optionally with --rebase.
• Or you use git stash to place your local changes on a "hidden" stash. Then, after pulling, you apply your changes again with git stash apply.
• Or you create a patch of your local changes that you apply again after the pull. This approach sometimes makes sense, but I would say in general it's the more clumsy way to go. There are git commands that help you with this though.

Resolving conflicts

Rather similar to Subversion. When there are conflicts, a merge or a rebase will add conflict markers into files. Use git status to see which files remain in conflict and use git add on files to mark them as resolved. When you did a merge and you have resolved all conflicts, you commit. When you were doing a rebase of several commits, you do git rebase --continue instead.

Patch making

Git has an easy way to send patches to other people to review and commit for you. After you have made a commit, git format-patch will make a patch out of it. The patch includes your author information the commit message you gave, and all the changes to be done. The recipient can just git am [patch file] to apply the commit. After it has been pushed, you'll need to remove the patch from your local repository, git reset --hard HEAD^ will do that. If you don't do that, you'll get a conflict when your patch is pulled from the central repository.

Good to know

Git has several useful commands to figure out the current state of your repository, your files and what recently changed. Below is a non-exhaustive list of commands that are useful to know:

• git branch: Without any parameters, this command lists your local branches, and indicates which branch you're currently on.
• git whatchanged: This shows a list of all commits on the current branch similar to git log, but with a list of the files that have been touched in each commit as well.
• git status: This shows all kind of things about your current checkout: which files changed, untracked (unknown) files, added or removed files, files that have conflicts (during merge), etc. It also shows the status of your index, which is what git will commit once you do git commit. If you're new to git I would recommend to wait a bit with learning how to use the index, but not to avoid it forever.

There are also additional applications that help you with using git:

• gitk: A simple but effective tool that visualizes the history and some of your current state. Run with --all to have it show all branches, otherwise it will just show stuff relevant to your current branch.
• tig: A textual interface, rather similar to an email reader.
• git gui: A gui tool like gitk which helps you prepare and perform your commits. Also makes it easier to understand the index concept.

git on Windows

When using git on Windows you might use msysgit. If you notice that some files seem to have changed after doing a fresh clone, you may want to disable core.autocrlf using git config core.autocrlf false. However, this is not recommended for contributors since the setting makes sure you don't commit Windows style newlines into the repository. When encountering this problem it is usually best to consult other developers about the affected files.