git branchstack

Motivation

When I am working on multiple changes to a Git repository, I usually want to combine all of my changes in a single branch, but send them upstream in small, reviewable chunks. As stated in the related articles one advantage is that you can base new work on previous changes, and test them in combination.

Git already supports this workflow via git format-patch and git send-email, however, many projects prefer to receive patches via pull requests. To make proposed changes easy to review, you'll want to submit a separate pull request for each independent change on your worktree's branch. This means that you want to create branches containing those independent changes and nothing else.

git branchstack creates the desired branches without requiring you to switch back and forth between branches (and invalidating builds). This allows you to submit small pull requests, while enjoying the benefits of a branchless workflow. After making any changes to your worktree's branch you can easily update the generated branches: just re-run git branchstack.

Installation

Via pip

$ pip install --user git-branchstack

Instead of the last command you can also run ./git-branchstack directly, provided you have git-revise>=0.7.0.

Via pipx

Use this instead to avoid breakage when your Python installation is upgraded.

$ pipx install git-branchstack

Usage

Create some commits with commit messages starting with a topic tag [...]. The topic name, <topic>, inside the square bracket tag markers [<topic>], must be an unused valid branch name. Then run git branchstack to create the branch <topic> with the given commits.

For example, if you have created a commit history like

$ git log --graph --oneline
* 9629a6c (HEAD -> local-branch) [some-unrelated-fix] Unrelated fix
* e764f47 [my-awesome-feature] Some more work on feature
* a9a811f [my-awesome-feature] Initial support for feature
* 28fcf9c Local commit without topic tag
* 5fb0776 (master) Initial commit

Then this command will (re)create two branches:

$ git branchstack
$ git log --graph --oneline --all
* 9629a6c (HEAD -> local-branch) [some-unrelated-fix] Unrelated fix
* e764f47 [my-awesome-feature] Some more work on feature
* a9a811f [my-awesome-feature] Initial support for feature
* 28fcf9c Local commit without topic tag
| * 7d4d166 (my-awesome-feature) Some more work on feature
| * fb0941f Initial support for feature
|/
| * 1a37fd0 (some-unrelated-fix) Unrelated fix
|/
* 5fb0776 (master) Initial commit

By default, git branchstack looks only at commits in the range @{upstream}..HEAD. It ignores commits whose subject does not start with a topic tag.

Created branches are based on the common ancestor of your branch and the upstream branch, that is, git merge-base @{upstream} HEAD.

To avoid conflicts, you can specify dependencies between branches. For example use [child:parent1:parent2] to base child off both parent1 and parent2. The order of parents does not matter: the one that occurs first in the commit log will be added first.

Pass --keep-tags to mark dependency commits by keeping the commits' topic tags. Use keep-tags=all to keep all topic tags. To only keep topic tags of select dependencies, prefix them with the + character (like [child:+parent]).

If a commit cannot be applied cleanly, git branchstack will show topics that would avoid the conflict if added as dependencies. You can either add the missing dependencies, or resolve the conflict in your editor. You can tell Git to remember your conflict resolution by enabling git rerere (use git config rerere.enabled true; git config rerere.autoUpdate true).

Instead of the default topic tag delimiters ([ and ]), you can set Git configuration values branchstack.subjectPrefixPrefix and branchstack.subjectPrefixSuffix, respectively.

Integrating Commits from Other Branches

You can use git-branchstack-pick to integrate other commit ranges into your branch:

$ git branchstack-pick ..some-branch

This behaves like git rebase -i except it prefills the rebase-todo list to cherry-pick all missing commits from some-branch, prefixing their commit subjects with [some-branch]. Old commits with such a subject are dropped, so this allows you to quickly update to the latest upstream version of a ref that has been force-pushed.

Here's how you would use this to cherry-pick GitHub pull requests:

$ git config --add remote.origin.fetch '+refs/pull/*/head:refs/remotes/origin/pr-*'
$ git fetch origin
$ git branchstack-pick ..origin/pr-123

Tips

You can use git revise to efficiently modify your commit messages to contain the [<topic>] tags. This command lets you edit all commit messages in @{upstream}..HEAD.
```
$ git revise --interactive --edit
```
Like git revise, you can use git branchstack during an interactive rebase.
git-autofixup can eliminate some of the busywork involved in creating fixup commits.

In Stacked Diffs Versus Pull Requests, Jackson Gabbard describes the advantages of a patch-based workflow (using Phabricator) over the one-branch-per-reviewable-change model; git branchstack can be used to implement the first workflow, even when you have to use pull-requests.
In My unorthodox, branchless git workflow, Drew DeVault explains some advantages of a similar workflow.

Peer Projects

While git branchstack only offers one command and relies on standard Git tools for everything else, there are some tools that offer a more comprehensive set of commands to achieve a similar workflow:

Unlike its peers, git branchstack never modifies any worktree files, since it uses git revise internally. This makes it faster, and avoids invalidating builds.

Contributing

Submit feedback at https://github.com/krobelus/git-branchstack/ or to the public mailing list by sending email to mailto:~krobelus/git-branchless@lists.sr.ht.

krobelus / git-branchstack