Commit Management for GitHub Pages

When GitHub Pages is served from a dedicated branch (e.g. gh-pages), its history can be rewritten to manage repository size. One way to do this is to simply amend the most recent commit and force push.

git commit --amend
git push --force

The upside is that previous updates no longer contribute to the repository size. The downside is that a bad commit can’t be easily undone.

To have the best of both worlds, a more sophisticated approach can be used. Instead of squashing every change into a single commit, preserve the N most recent commits and only the remaining commits.

The git-squash-range script provided here does exactly this.

Algorithm

The history can be divided into two commit ranges:

The oldest commits to be squashed: [root, mid]
The newest commits to be preserved: (mid, head]

The commit IDs for the root, mid, and head commits are determined as follows:

root - git rev-list --max-parents=0 HEAD
mid - git rev-parse HEAD~$N where N is the number of commits we want to preserve
head - git rev-parse HEAD

The current branch also needs to be recorded:

branch - git branch --show-current

The algorithm is then:

Squash the oldest commits
1. Checkout root
  git checkout $root
2. Squash merge from mid
  git merge --squash $mid
3. Amend root with the merge result
  git commit --amend -m squashed
Preserve the newest N commits
```
git cherry-pick $mid..$head
```
Move the branch to the new HEAD
1. Move the branch
  git branch --force $branch HEAD
2. Checkout the branch
  git checkout $branch

Development

Run the tests.

rake

rfdonnelly/git-squash-range

Commit Management for GitHub Pages

Algorithm

Development