presumably for side-effects
a blog about clojure &c.

Search and Replace with git

Using git xargs and rpl

published Jan 13, 2017

If you are like me and you see the command line as the ultimate developer environment even as we're moving into 2017, chances are you also treat Git as a trusty old friend. With most of my work under source control, one thing I find myself reaching for frequently is a way to search and replace strings in the entire project.

What could be simpler than going through all the text files under source control and replacing one string with another? And yet the task turns out to be surprisingly difficult and error-prone. Although editors or IDEs typically have this feature, it's often unwiedly and, much worse, unpredictable.

So following the Unix philosophy, let's decompose the task into discrete steps. The challenges usually involves

  • selecting the right files and
  • replacing the string without fuss.

Most of the time I just want to iterate over every file in my Git repository, though sometimes I want to exclude some of files, either by path or extension.

Just use regular expressions, you say? Actually, no. Most of the time I don't really need the flexibility they offer, and I don't want to remember the confusing and inconsistent regex syntax and esacaping tricks required by the different Unix tools. Mostly I work with fixed strings.

In fact some Unix tools are not Unix-y enough for my taste. A case in point, I don't particularly like using sed or perl for string replacement, as regex support is baked in, and I haven't found a way to opt out - believe me, I've tried.

And then there's the fact that, by default, sed (and perl) only replace a single occurrence of a search string per line. I often forget the s/foo/bar/g modifier.

Fortunately, there is a replacement for sed that operates on fixed strings called rpl(1). It's widely available (brew install rpl, apt-get install rpl) and pretty much does one job, and does it well.

What about selecting files to work on? Git and the popular xargs Unix tool help here, but to compose the two I needed a new tool. So I built git xargs.

Tiny enough to fit in a gist, git xargs only does one thing: it executes a shell command on all files in your Git repository. Here's how you use it (and remember to start with a clean git status):

git xargs rpl foo bar
# replaces "foo" with "bar" in all files in the current repository

git xargs '*.c' '*.cpp' -- rpl foo bar
# restrict to only files ending in .c and .cpp (note the double dash)

git xargs rpl '*hello*' '*goodbye*'
# no regex no problem (but you can specify -w to match only at word boundaries)

git xargs -n1 rpl foo bar
# call rpl invidiually on each file

git xargs sed -i '' s/foo/bar/g
# works with sed too

Of course, the beauty of revision control is that you can easily see what has changed (git diff) and selectively apply or reject the changes (git add -p). If something went wrong, just revert the working directory to its previous state (git checkout --).

Thanks to composability, there are many potential use cases for git xargs besides string replacments. Hopefully it will prove useful to you too!


This is presumably for side-effects, a blog by Paulus Esterhazy. Don't forget to say hello on twitter or by email