These days, I’m a developer on a project managed with a central CVS server. And there’s no way I could convince the team to switch to a VCS that makes more sense (imho). I’m not really allowed to do branches either, and merging probably sucks anyway with CVS, so that’s perfectly fine with me if I can’t try. But I do miss branching, it’s just so helpful to develop big new feature or to try really experimental stuff.
My main experience is with Subversion, but I don’t think that mixing Subversion and CVS is a good idea. I’m also coding a lot with my laptop lately, which most of the time don’t have internet access. Buses in Gatineau and Ottawa don’t provide free wireless internet for their users 
. So, a decentralized version control system like Git or Mercurial seemed like good solution for me. It’s not the best scenario, but using a DVCS on top of cvs checkout seemed is a possible solution.
I’ve tried Mercurial and Git for some time, and Git was a winner in my case. Honestly, they seems to have the same capabilities, I don’t think one is better then the other. Coming from SVN, Mercurial is the easier to learn/work with. But strangely, I have decided to use Git for the single reason that there’s some really nice looking GUI tools for it on OS X. Who knows, maybe a cool app could convince the other developers to try it (I can’t believe they love the look of MacCvs)
Here’s some notes on my setup. Mainly to act as a reminder for my own personal use. This is not intended to be a tutorial for anybody, but It may be useful to some.
there’s some nice tutorial on learn.github.com too
Git config
First thing to do is to present myself to my Git installation
git config --global user.name 'John Doe' git config --global user.email 'johndoe@example.com'
and then activate the colors on the console
git config --global color.diff auto git config --global color.status auto git config --global color.branch auto
Creating the git repos
It’s as easy as running git init at the root of the codebase. This will create a .git folder the the Git data.
Ignoring useless files
The codebase I’m working on is a CVS checkout, and I’m using Eclipse on OS X so I have created a .gitignore file at the root of the project that contains:
CVS
.DS_Store
.cache
.settings
.buildpath
.project
GUI for OS X
Some gui are bundled with Git on OS X, they are gitk and git-gui (once opened, pressing command-+ a couple of time increase the font to a reasonable size). They’re really useful, but they don’t look that great. I’m mainly using GitX in the hope to impress my collegues, but GitNub looks cool too, I want to try it someday.
Backup (aka Dropbox)
I’m backing up my repository on a shared drive
git clone --bare /dev/coolproject /Volumes/SharedHome/gitrepos/coolproject.git
The ‘–bare’ argument is used to create only the repository without a checkout. I also use the same technique to backup and share some project with Dropbox. This way I can share with people I know. They can git pull my project without me having to create a public repository on github or setting up the necessary stuff on my web server.
Now I’m ready to push to this repos with
git push /Volumes/SharedHome/gitrepos/coolproject.git
I can also set this remote as the origin (or any other name)
git remote add origin /Volumes/SharedHome/gitrepos/coolproject.git
And I will be able to push using a shorter command
git push origin
this idea have been stealed from this great blog post titled Version Control Makes You A Better Programmer
Git and CVS
Git have a nice feature that let you commit with different author information. When I update from CVS to get new code from my coworkers, I can commit stuff to Git with
git commit -a --author="cvs <dev@cie.com>"
after it have been done once, I don’t need to type the full author info, name will be enough
git commit -a --author=cvs
I have an alias in my .profile (.bash_profile) to make it faster
alias gitcvs='git commit -a --author="cvs" -m "cvs"'
This way, everything coming from CVS will be easily identifiable. Obviously, every commits in the CVS don’t translate in a commit in Git, but I don’t really care, as I will use cvs log if I really need to digg something. I’ve heard about git-cvsimport, but that sound too complicated for my simple needs for now, will take a look at it later. Now I have to get some work done.
Before running cvs update, I usually commit all my stuff to Git and CVS. Even better, with Git I can just ’stash’ my modifs (git stash) run cvs update and then re-apply the stash (git stash apply). The stash is like a temporary placeholder for your code. I need to read more on it.
To be continued
Let’s see if this setup will stay usable and useful for me in the near future. Stay tuned
