gitYou can use git as a full-fledged, commit-driven VCS, with all its awesome features. I wrote about that (and probably will again) before. But often you just have to track some things over the course of a few minutes/hours and don’t even want a “repository” or a commit-history or branches.

Often you have some code/text in it’s original form and need to change a few lines until something works (again). Sometimes you’d like to keep track of these changes (for example to apply them on a different codebase/host). Before git, I just copied the files (or directory) I wanted to edit to some or destination and later did a diff with the new files. That sucks for obvious reasons.

With git it can be much more convenient: You go into the directory with the vanilla files in it. Just do a git init and git add .. Thats pretty much it, now you just go on to edit whatever you want. To see the changes you made since the first git add . just type git diff (I don’t have to tell you that you can pipe this output into a patch).

If you happen to add new files, I usually commit the original version using git commit -m ‘initial’ (even later on, since you haven’t staged any of your new changes for committing yet), and then do a git add -u and git add <new -files> and then diff with git diff –cached. If there is a better way please let me know!

That, imho, is one of the examples where git beats svn hands-down. The notion of a local repository (or in this case the index) is just great once you understood the concept somewhat. I wouldn’t use svn for little changes since it would mean setting up an svn-repo (in a different dir), checking it out, commit the vanilla version, changing stuff and then diffing it. And then you’d have all those .svn directories cluttering your working tree, whereas with git you can just do a rm -rf .git at the end of the day and forget that you even used git.

Something else I missed in my svn days was committing only some changes (called “hunks” in git) to a file. In git you can use git add –patch or git-gui to do just that and avoid the problem of having multiple unrelated changes in a single commit (which sucks when debbuging). I read a really good article (and comparison with other SCMs) today at