Quick Git for Bioinformatics

Introduction

This demo has the parts:

1. Create a git repository, both from a directory and from cloning. Add and commit files.
2. Working with remotes.
3. Branch and merge.

Unless you're working with collaborators, you'll mostly use the first and third parts.

Think this is just useful for code? Think again! You can should version control papers, and small datasets (e.g. an RNA-seq counts table, not a 5GB SAM file).

But first, tell Git who you are with:

git config --global user.name "Vince Buffalo"
git config --global user.email vsbuffalo@ucdavis.edu

Creating repositories, adding files, and committing

Creating a repository from a directory, adding files, and committing

mkdir vince
cd vince
echo "This is a test repository." > test.txt
git init
git add test.txt # this is called "staging"
git commit -m "initial import" # this commits all staged files only. # use a commit message
cd .. # get to where we were

Note something very important about Git: when you add a new file, you're not just telling Git it should be under version control, but you're staging it in the next commit. When a file already exists, you've made changes to it, and you want it to be in the next commit you need to add it with git add, or tell git commit to commit all files using -a. This allows AWESOME incremental commits. You can also unstage a file for commit using git reset (that's advanced, we can talk about that later, but to unstage a file, you do git reset FILENAME).

Cloning a repository

If you have an existing repository you want to clone, you use git clone. This essentially initializes a empty local git repository, and pulls a clone from the remote repository.

git clone ./vince joe
ls
cd joe
git log

Working with remotes

Let's mimic the workflow we use most often: a remote "bare" repository. The issue with the current workflow is that if Joe commits changes, and pushes them to my remote, he would be asked if really wants to push to my working repository. In fact, he probably doesn't: this is where I am working and I want to be able to pull in his changes whenever I like, not have them pushed to me.

Consequently, a very effective git workflow is to have a remote "bare" reposistory. The "bare" means it's not a real working directories, which we have to explicitly state by default all git repositories are working directories (i.e. you can edit files). This is in contrast to CVS and SVN.

We set up a server that is frequently backed up that has a /git directory full of our project's bare reposistories. In this example, we'll create a bare repository named "remote". Remember, Git works with SSH!

rm -rf joe # sorry Joe.

mkdir remote # this is our pretend remote; in real life, put this on a server everyone has access to
cd remote
git init --bare # older versions of Git require this to be git --bare init, but use the newest version!
ls # see all the Git internal stuff? Not a place to do work!
cd ..

Now, we're going to tell my local repository about this remote one.

cd vince
git remote add origin ../remote
git remote ## or cat .git/config, SVN users, note that there is only ONE, YES ONE directory!

Now, let's push our commits to remote.

git log # what commit are we at?
git push origin master

Why origin master? We want to tell git push (push is the subcommand) where we want to push (origin, which is the name of the remote we added; "origin" is used by convention) and which branch we're on (master is the default).

Now, let's pretend we're Joe again. Joe wants to get my work, add something to it, and check it in.

cd .. # get to where we were
git clone remote joe
cd joe
ls 
git log
echo "new gibberish" >> test.txt
git commit -a -m "some gibberish added"
git push origin master

Ok, so Joe added something to test.txt, commited it, and pushed it to the remote. Now, Vince wants to pull in Joe's changes:

cd ../vince git pull origin master git diff HEAD^

This means take the HEAD (basically a pointer in Git's history to where your current working directory is), go back one commit in history (the ^), and diff the current working directory with the last commit.

We can also look at specific files from older commits using git show:

git show HEAD^:test.txt

Git branching

So suppose you're writing an analysis in Python, but someone says it would be 10000x easier in R. You want to keep your data organized in the existing directories. You think to create a repo_in_R copy of your directory, but Keith tells you that this could lead to the innocent drowning of a kitten (see http://bitsandbites.posterous.com/details-of-next-meeting-12th-january-2012).

What do you do? If you're from the land of SVN and CVS, you're pulling out hair, flipping through pages of documentation, and practicing branching on fake repositories. If you're using Git, you just sip some tea, lean back, and branch. Branches in Git are super easy, because they're virtual, not physical.

git branch # see, we're in master now
git checkout -b new-branch
git branch

echo "more gibberish" >> test.txt
git commit -a -m "more gibberish added on a dev branch"

git checkout master # go back to master
echo "different gibberish added. Uh oh." >> test.txt
git commit -a -m "different gibberish added, clash anyone? "

We can look at this situation with:

gitk --all &
# or 
git log --graph --all

Now, let's merge that new-branch into master:

git branch # make sure you're in the branch you want to merge the other into. git merge new-branch

Uh oh, a merge conflict!! It's easy to resolve. Look what's up with git status, then manually resolve the conflict. When you're down add the files, and commit. That easy! And normally you won't find merge conflicts.

emacs test.txt
git add test.txt
git commit -m "merge conflict resolved"
git push origin master # might as well?

What's next?

• Following remote branches
• git refspecs
• git stash
• Github
• git blame