At CodeScene, we used to not delete merged git branches, at all. The practice came from Branch Analysis. To get that data, we didn’t want to remove a merged branch immediately. However, we ended up, never deleting them which also causes a problem: over time you accumulate thousands of branches. That creates a mess and makes it hard (or impossible) to select proper branch in CodeScene itself (the max number of branches we load and display is 1000).
Recently, we decided that it’s time to start removing older merged branches We still want to keep recent merged branches for ocassional review (via CodeScene’s branch analyses) but we do not need anything older than 1 month.
The question arises, how to do it: There’s setting in GitHub to do this automatically (see Managing the automatic deletion of branches). But that’s pretty much instantaneous - it deletes the merged branch immediately, which isn’t what we want.
So we have to resort to a more manual approach - we decided to write a script that’s periodically run via GitHub Actions.
It’s quite simple - I ended up writing this "one-liner"[1]:
MAIN_BRANCH=master
MAX_AGE_DAYS=30
git branch -r --merged origin/${MAIN_BRANCH} --no-contains ${MAIN_BRANCH} --format='%(committerdate:raw)%09%(refname:short)' | awk -v max_age=$(( $MAX_AGE_DAYS * 86400 )) -v now=$(date +%s) '{ diff = now - $1; if (diff > max_age) print $3 }' | sed 's/origin\///' | xargs git push --delete originIt works like this:
Find all the branches merged into the main branch (excluding the main)
git branch -r --merged origin/${MAIN_BRANCH} --no-contains ${MAIN_BRANCH} ...… while adding committerdate at the same time - using the raw format to get Unix Epoch seconds:
... --format='%(committerdate:raw)%09%(refname:short)'Use awk to calculate the difference between current time and print branch names only for refs with commiterdate older than the threshold (30 days)
awk -v max_age=$(( $MAX_AGE_DAYS * 86400 )) -v now=$(date +%s) '{ diff = now - $1; if (diff > max_age) print $3 }'Notice how I use awk -v to define awk variables max_age and now because I cannot reference shell variables easily inside the awk script
Remove the 'origin/' prefix
sed 's/origin\///'Finally, delete all the matching branches
xargs git push --delete originThe output of the first two steps may look like this:
1683723780 +0200 origin/branch-a
1683190390 +0200 origin/branch-bRunning this via GH Actions is relatively easy, but there’s one critical piece you need to be aware of
- the checkout step has be configured with fetch-depth: 0 [2]:
...
- uses: actions/checkout@v3
with:
# Fetch all the branches to be able to list them later
fetch-depth: 0Otherwise, the script won’t be able to find anything and fails with a rather cryptic error: "fatal: --delete doesn’t make sense without any refs"
name: Delete old merged git branches
on:
schedule:
- cron: "0 6 * * 1-5"
workflow_dispatch:
inputs:
max-age-days:
description: "Branches older than 'age' (days) will be deleted. Default: 30"
required: true
default: 30
type: number
main-branch:
description: "The main branch that shouldn't be deleted. Default: 'master'"
required: true
default: master
type: string
env:
MAX_AGE_DAYS: ${{ github.event.inputs.max-age-days }}
MAIN_BRANCH: ${{ github.event.inputs.main-branch }}
jobs:
build:
timeout-minutes: 5
runs-on: ubuntu-latest
steps:
- name: "Print info"
run: echo "The script will remove all git branches older than ${{ env.MAX_AGE_DAYS }} days and merged into the ${{ env.MAIN_BRANCH }} branch."
- uses: actions/checkout@v3
with:
# Fetch all the branches to be able to list them later
fetch-depth: 0
- name: "Delete old merged branches"
# The best way to understand this is to actually run it piece-by-piece, perhaps skipping the last command (actual delete)
# ----
# 1. The "git branch" command shows all the merged branches with the date (unix epoch seconds) of the latest commit, e.g.
# 1636550813 +0100 origin/1095-improve-trello-provider
# 2. The "awk" command filter out those that are older then the threshold, printing only the branch names ($3)
# 3. Finally, we delete the branches one by one via `xargs git push --delete ...`
# ----
# NOTE: the 'origin/' prefix for the branch name is needed later, otherwise it fails on GH actions with this error: "fatal: --delete doesn't make sense without any refs"
run: git branch -r --merged origin/${{ env.MAIN_BRANCH }} --no-contains origin/${{ env.MAIN_BRANCH }} --format='%(committerdate:raw)%09%(refname:short)' | awk -v max_age=$(( ${{ env.MAX_AGE_DAYS }} * 86400 )) -v now=$(date +%s) '{ diff = now - $1; if (diff > max_age) print $3 }' | sed 's/origin\///' | xargs git push --delete origin