From Hall A Wiki
A Cheat Sheat for How To Use Git
I, Bob M., got this from Brad Sawatzky in response to an e-mail.
The main thing to remember about git is that it is local (ie. on your machine). Checkouts are from your local copy, and commits are to your local copy -- you get to choose when and in what way you push your changes upstream (more on this later). Since you are only committing to your local copy, you can commit incomplete and/or non-working code with impunity. Once you have a feature that is working, you can build and push the associated set of changes up to Ole. It's impossible to lock people out of a repo the way you can with cvs/svn -- git is specifically designed to remove that problem. You always have to authenticate before pushing changes upstream. Uploading a key just means you don't have to type in your password every time you push upstream. (Note that you do not, and should not, push every change you make upstream right away.) https://help.github.com/articles/generating-ssh-keys git is *way* more lightweight than CVS, but it does require a change in how you think about version control. I created and used 8 independent (local) git repos last week alone while I was debugging a DAQ.
Here are the command I use 99% of the time:
1) Create your repo git init : create empty repo from scratch git add . : add everything in this directory git commit -m'initial commit' : initialize the repository -- OR -- git clone <URL> : copy a repo from elsewhere -- Useful commands: git log : see the log file (can do this anytime) 2) Work on your code inside a new branch: git branch my-working-topic : create a branch git checkout my-working-topic : change into that branch <hack away until I have something that I want to store as a 'change'> - Branches are nice because the isolate your present work from other (possibly conflicting) changes in your repo. They - Useful commands at this stage: git checkout foo.c : retrieve a 'fresh' copy of foo.c, replacing my borked version 3) Build a 'commit' -- Repeat these as often as you want: git add -p : 'interactively' ask about what I changes I want to reco git add <file> : just add all changes in a given file (or add a new file - Misc commands that are useful at this stage: git status : summary of changed files git diff --cached : review the commit you've been working on with 'git add' git diff : see diff of changes made since the last commit that are *not* in commit you're working on git reset HEAD : blow away the commit you've worked on with the 'git add's. Nothing changes in your directory -- you just start at the beginning of step 3 again. I think 'git undo' is a shortcut for this. 4) Save the commit git commit : save the change set with a notation in the log git commit --interactive : gives you a chance to edit the changeset you built in (3) without having to 'git reset HEAD' and start from scratch 5) goto (2) and repeat until you have something that you want to send upstream git fetch : this pulls the changes to your machine, but does NOT change your working directory at all git merge : this will attempt to merge your present state with the upstream <HEAD>. You will need to resolve conflicts at this point -- OR -- git pull : basically shorthand for a 'git pull' followed by a 'git merge'rd as part of this change 6) Merge upstream - Talk to Ole the first time you do this -- there are many ways to handle this. One method (this is the most flexible way, but you guys might be pushing to a personal branch instead of a personal repo): git push : this will transfer the changes from your local working copy up to the github repo you cloned from. - You can then send a 'pull request' to Ole, OR ask that he merge your branch to master. 7) goto (2)
Recommended quick references:
A compilation of links I've found useful here: https://hallcweb.jlab.org/wiki/index.php/Git_Howto