Just a few days ago the curl project turned twelve years old, and I decided that it was time for us to ditch our trusty old CVS setup and switch over to use git instead for source code control.
Why Switch at All
I’ve been very content with CVS over the years and in our small project we don’t really have any particularly weird or high demands on the version control software.
Lately (like in recent years) I’ve dipped my toes into various projects that have been using git, and more and more over time I’ve learned to appreciate the little goodies that git does that CVS simply cannot. I’m then not even speaking about branches or merges etc that git does a whole lot better and easier than CVS, I’m in fact even more in love with git’s way to ease handling with diffs sent by email and its great way of keeping track of authors separately from the committer etc. git am and git commit –author are simply two very handy tools missing in CVS.
So if we want to switch from CVS to another tool what would we chose? That wasn’t really the question in my case so I didn’t answer it. In my case, it was rather that I’ve been using git in several projects and it is used in some of the biggest projects I work with so it was some git’s features I wanted. I didn’t consider any of the other distributed version tools as quite frankly: they wouldn’t be much better for me than what CVS already is. I want to reduce the number of different tools I need, and I’m quite sure anyway that git is one of the top contenders even if I would do an actual comparison.
So the choice to go git was quite selfish and done by me, but I felt that quite a few guys in the curl community supported this decision and very few actually believe remaining with CVS was a better idea.
The fact that git itself uses libcurl for its HTTP access of course also proves its good taste! 🙂
How did the conversion go
Very easy and swiftly. First, as I mentioned above we never used branches much so we basically had a linear development with a set of tags. I did an rsync of the full repo to get a local copy to work with, then I ran ‘git cvsimport’ on that to created a new repo. I did run it a couple of times to make sure I had done a correct mapping of all CVS user names to their git equivalents. Converting >10 years of CVS commits took roughly 10 minutes on my desktop machine so it wasn’t that tedious even.
Once I had a local repo created with all authors looking good, I simply followed the instructions on github.com on how to add a remote origin to a local branch and when I pushed to that, git sent off all commits ever made to curl to the remote repo now exposed to the world from github.com.
When that part was done, I did a quick read on the ‘git help daemon’ docs and 30 seconds later I had a local repo setup that is a mirror of the github one, so that users can still opt to get the code from haxx.se.
Unchanged work flow
Git allows different ways of working with the code, but I’ve decided that at least as a start we won’t change the way we work. I’ll offer all committers push rights to the master branch on the repository and we will simply all push to that, as our head development branch.
We will prefer patches made with git format-patch sent to the mailing list, but as before you can still produce patches by diffing source code using extracted tarballs or whatever approach you prefer.
All details on how to get the code for curl using git is available online.
3 thoughts on “curl goes git”
On http://curl.haxx.se/source.html, you still mention “Browse the CVS Repository”, even though it links to github, I guess you’d want that updated 🙂
Ah, thanks fixed now. There are still some traces of CVS all over but I’m slowly tracking them down and updating them accordingly…
Thanks for making this decision. It is so great so see how many projects moved to Git in recent times. Currently, the most frequented tools for checking out code from repositories are Git, svn and mercurial. I really hope mercurial and svn will slowly disappear because Git offers everything one needs without lots of dependencies (*thinking about mercurial”). I myself only uses the basic commands Git offers. That’s why svn also worked fine for me but it would be really nice to see if some day you don’t need 10 different tools just to check out the latest code and keep it updated. Conversion tools are IMO only appropriate for conversion but my CPU is too slow for complex things like cvs conversions. Converting a public CVS repository which dates back about 6 years ago with lots of commits took about half an hour. Updating also takes much time. So, we’d all be better off to use only one single solution that matches all your problems and help our users to more accessibility to our source code. In the end of the day, it’s for our own good. 🙂
Comments are closed.