User:Timothee Flutre/Notebook/Postdoc/2012/08/14

From OpenWetWare
Owwnotebook icon.png Project name Report.pngMain project page
Resultset previous.pngPrevious entry      Next entryResultset next.png

About Git

  • Motivation: nowadays, it's pretty common to use a computer for a project in which one wants to: keep history of the changes, access them on different machines with different operating systems, share our work with someone else, etc. In such cases, it's very useful to use a distributed versioning system, such as Git.

  • Conflicts: when updating one branch with the content of another one (git checkout branch1; git merge branch2), some conflicts can happen, and it is usually hard to know how to solve them properly (but see a concrete example here). In the following, branch1 can be master and branch2 can be origin/master, or branch1 can be master and branch2 can be dev.
    • The first solution is to edit each conflicted files by hand, then run git add fileX.txt (staging indicates to git that the conflict is resolved) and finally run git commit -m "merge branch2 and solve conflicts" fileX.txt.
    • The second solution is to ignore the conflicts and overwrite the files of branch1 with the content of branch2, one file at a time: git checkout --patch branch2 fileX.txt.
    • The third solution, even more radical, is to "overwrite" all of branch1 with the content of the branch2, all files at once: git reset --hard branch2.

  • Tips:
    • undo uncommitted changes: git checkout myfile.txt; (much) more on "undo"
    • split a big commit in several smaller commits: git add -p myfile.txt
    • usual config: git config --global 'Timothée Flutre'; git config --global ''; git config --global color.diff auto; git config --global color.status auto; git config --global color.branch auto; git config --global core.editor emacs; git config --global i18n.commitEncoding 'utf8'; git config --global i18n.logOutputEncoding 'utf8'; git config --global core.autocrlf input
    • remote via ssh tunnel: first open the tunnel ssh -l tflutre -Nf -L 20400:maincluster:22, then add the remote git remote add mcl ssh://tflutre@localhost:20400/home/tflutre/myproject/.git
    • create release on github: first create tags, and then create the release (automatic via tag name) curl --user "timflutre" --data '{"tag_name":"v1.0","target_commitish":"master","name":"v1.0","body":"first release"}'
    • get download count of release: curl -u "timflutre" -i where the release id can be obtained via curl -u "timflutre" -i
    • avoid being rejected by github because SSL certificates are not installed on your machine: add export GIT_SSL_NO_VERIFY=true to your ~/.bash_profile (source)
    • make the ChangeLog file: git log --no-merges --format="%an: %s" > ChangeLog
    • collaborate with some using Windows: ask him to do git config --global core.autocrlf true

  • Writing a paper: in this example, I am writing a paper with two colleagues. We decide to do it as a centralized workflow, the shared repository being hosted by github.
    • Setting up the infrastructure:
      • Each of us needs to create a free account.
      • I need to upgrade my account in order to have the right to manage private repositories ($7/month).
      • I create a private repository named "paper" and add my colleagues as collaborators to it.
      • I retrieve the repository on my local machine: git clone git://
      • I create my first file, for instance "paper_main.tex", and add it to git in my local repository: git add paper_main.tex followed by git commit -m "first commit" paper_main.tex.
      • I create one branch per collaborator (the default branch being "master"): git branch tim, then git branch colleague1 and finally git branch colleague2. I can list the local branches with git branch and I can switch to my branch with git checkout tim for instance.
      • I push the changes I made from my local repo onto github: git push origin master, this for each branch I created.
      • I send an email to my colleagues telling them that they can retrieve the content of the repository from github into their local machine(s): git clone
    • Typical working cycle:
      • Each of us can make modifications on its own branch, and push them on github in order to allow the others to access the changes: git push origin colleague1 for instance.
      • From time to time, one of us has the responsibility to merge the changes and update the "master" branch with the latest version.
      • Once this is done, the others need to retrieve the new content of "master" in their local repo: git checkout master, git fetch origin, git diff master origin/master, git merge origin/master.
      • Then, they need to update their local branch with the new content of "master": git checkout colleague1, git diff --name-status colleague1..master. This will list the files having differences between their local branch and the new content of "master".
      • One can look at the differences file by file: git diff --color-words colleague1:paper_main.tex master:paper_main.tex. The options "--color-words" is especially useful in LaTeX.
      • To merge the content of the recently-updated local "master" into his own local branch, we do: git merge master.
    • Tips: don't version the output pdf in the repository because, as it is binary, git can't merge it properly. But you can add a Makefile (see below) and, by entering make main -i on the command-line, it will compile your pdf document when you need it
all: main supp

	latex paper_main.tex
	bibtex paper_main
	latex paper_main.tex
	latex paper_main.tex
	pdflatex paper_main

	latex paper_supplements.tex
	bibtex paper_supplements
	latex paper_supplements.tex
	latex paper_supplements.tex
	pdflatex paper_supplements

	rm -f *~ *.aux *.dvi *.log *.pdf *.bbl *.blg *.toc

  • Two remotes: let's imagine that on cluster1 I have 2 branches, "master" and "dev", on github I only have "master", and I want to work with "dev" on cluster2.
  • Personal repositories: for some small project, I am the only one working on them and I want to keep them confidential, so I have my own server in my home on a computer cluster (~/opt/git)
    • mkdir ~/smallproject; cd ~/smallproject; echo "hello" > README; git init; git add README; git commit -m "first commit"; cd ..; git clone --bare smallproject/.git
    • edit smallproject.git/config to only keep section [core] and, inside, set bare = true
    • mkdir -p ~/cluster1; sshfs -oworkaround=rename me@cluster1: ~/cluster1; cp -r smallproject.git ~/cluster1/opt/git
    • cd ~/smallproject; git remote add origin ~/cluster1/opt/git/smallproject.git
    • I can now push and pull from this minimal server