Question

My organization is using git branching extensively. As a result, we have produced over 2000 branches in the past year. We are now trying to adopt a strategy for cleaning up all the old branches that are of some given age. I know how to delete branches, but I can’t find a straightforward way to list all of the branches with heads of a given age. The plan is to set up a cron that periodically deletes all branches of a given age, except those that are on some list.

Has anyone tried anything like this before?

Answer 1

The answers using committer dates are a good direction… if you want to delete branches that point to old commits. But you might want to delete branches which are actually old; if you create a branch today pointing to a commit from last year, you don’t want it wiped!

So, you want to examine the reflog dates instead.

You can get a human-readable form with git reflog show --date=local mybranch:

8b733bc mybranch@{Tue Mar 22 13:21:49 2011}: commit: foo
7e36e81 mybranch@{Tue Mar 22 13:21:25 2011}: commit: bar
99803da mybranch@{Tue Mar 22 13:20:45 2011}: branch: Created from otherbranch

(You might also like --date=relative)

The entry on the top is the most recent thing that happened on that branch, so that’s the bit we care about. Unfortunately, there’s no log format placeholder for just the date, so to grab out just the date, we do a little work:

git log -g -n 1 --date=local --pretty=%gd mybranch | sed 's/.*{\(.*\)}/\1/'
# Prints "Mon Mar 21 13:23:26 2011"

Of course, for scripting, that’s not very useful, so let’s go ahead and get the epoch time instead:

git log -g -n 1 --date=raw --pretty=%gd mybranch | sed 's/.*{\(.*\) .*/\1/'
# Prints 1300731806

Now we’re getting somewhere!

#!/bin/bash
cutoff_date=$(date --date="July 23, 2010" +%s)
git for-each-ref refs/heads --format='%(refname)' | while read branch; do
    reflog_date=$(git log -g -n 1 --date=raw --pretty=%gd $branch -- | sed 's/.*{\(.*\) .*/\1/')
    if [ -n "$reflog_date" && "$reflog_date" -lt "$cutoff_date" ]; then
        git branch -D ${branch#refs/heads/}
    fi
done

An example script! I used date to convert a human-readable date for the cutoff, then for each branch, I checked if the reflog’s last date was before the cutoff, and if so, deleted the branch. You could add in a check against a whitelist there, to save yourself from accidentally deleting something you care about. (Edit: if the branches are older than 90 days, this won’t delete them, because their reflogs will already be empty… up to you what you want to do in that case, really. You could fall back to checking the committer date, which ought to be pretty safe at that point.)

Edit: Here’s another approach. Expire the reflogs at the cutoff time, then delete the branches whose reflogs are empty. The problem here is that if the cutoff time is older than the time when your reflogs already expire (90 days) it’ll really just be deleting branches older than 90 days instead. You could work around that, of course.

#!/bin/bash

# Git lets you use very readable time formats!
cutoff_time="1 year ago"
# other examples:
# cutoff_time="July 23, 2010"
# cutoff_time="yesterday"

git for-each-ref refs/heads --format='%(refname)' | egrep -v 'master|other-whitelisted-branch' |
while read branch; do
    git reflog expire --expire="$cutoff_time" $branch
    if [ "$(git reflog show -1 $branch | wc -l)" -eq 0 ]; then
        git branch -D ${branch#refs/heads/}
    fi
done