johnhooks

Tag: git

  • Managing Parallel Work

    Git Worktrees Part 2

    Once you start using worktrees, the immediate problem becomes obvious: you have multiple branches evolving independently, and they all need to stay synchronized with your main branch. When a hotfix lands in production or someone merges a feature that touches the same code you’re working on, every active worktree potentially needs updating. Do this wrong and you’ll spend your afternoon resolving merge conflicts. Do it right and the synchronization becomes almost invisible.

    The key insight is that worktrees share a single git database, which means operations in one worktree affect all the others instantly. When you fetch new commits, they’re immediately available everywhere. When you create a branch, it shows up in every worktree. This shared state is what makes synchronization practical, but it also requires understanding which operations are local to a worktree and which are global to the repository.

    Keeping Your Base Branches Fresh

    The most common synchronization task is pulling updates from your remote repository into your local branches. The naive approach—running git pull in every worktree—works but creates unnecessary complexity. A better pattern treats one worktree as the source of truth for your base branches.

    Start by designating your main worktree as the place where you update shared branches. When you want to sync with the remote, navigate there and pull:

    Because all worktrees share the same git database, this updates the main branch everywhere. Your other worktrees don’t automatically check out the new commits, but the branch itself is updated. When you’re ready to integrate those changes into a feature branch, switch to that worktree and rebase:

    cd ../feature
git rebase main

    This rebases your current branch onto the newly updated main. If you’ve been working on the new-auth-system branch in this worktree, the rebase incorporates all the latest main branch commits beneath your feature work.

    The reason this works cleanly is that rebasing is a per-branch operation, not a per-worktree operation. You’re not moving the worktree itself—you’re updating the branch that happens to be checked out there. Other worktrees tracking different branches are unaffected until you explicitly rebase them too.

    Handling Conflicts Across Worktrees

    When you rebase a branch in one worktree and hit conflicts, git stops and waits for you to resolve them. This happens entirely within that worktree’s context. Your other worktrees continue to work normally because they’re tracking different branches, and git’s conflict state is worktree-specific.

    Here’s where the isolation becomes valuable. If you’re mid-rebase in your feature worktree and get pulled into an urgent code review, you can just cd ../review and handle the review in a completely clean environment. The conflicted rebase in the feature worktree sits there waiting for you, exactly as you left it. When you return to resolve the conflicts, the process is standard git conflict resolution—fix the conflicting files, stage them, and continue the rebase:

    cd feature
# Fix conflicts in the files git identified
git add .
git rebase --continue

    Once the rebase completes, that branch is synchronized with main. Your other feature branches in other worktrees still need their own rebases when you’re ready to update them.

    Integration Testing Across Features

    A powerful pattern enabled by worktrees is testing how multiple in-progress features work together before merging any of them. Create a dedicated integration worktree that exists purely for combining and testing branches:

    git worktree add integration -b integration main

    This creates a branch called integration based on main. Now you can merge multiple feature branches into it:

    cd integration
git merge feature-auth
git merge feature-billing
git merge feature-notifications
npm test

    This tests how all three features interact without touching any of the individual feature branches. If tests pass, you know the features are compatible. If tests fail, you know there’s an integration problem to solve before merging to main.

    The integration branch is disposable. After testing, you can delete it and recreate it fresh the next time you need integration testing. The pattern works because your feature branches remain untouched—you’re only testing a temporary combination.

    When you’re satisfied that features work together, merge them to main individually:

    cd ../main
git merge feature-auth
git push origin main
git merge feature-billing
git push origin main

    Each merge is a deliberate, tested step. This approach catches integration issues early while keeping your main branch clean and your feature branches focused.

    Cleaning Up Finished Work

    As you complete features and merge branches, worktrees accumulate. Some track branches that no longer exist. Others were created for one-time tasks and are no longer needed. Regular cleanup prevents your workspace from becoming cluttered with obsolete directories.

    When you’re done with a worktree, remove it:

    git worktree remove feature

    This deletes the working directory and unregisters the worktree. The branch itself remains in your repository—you’ve only removed the working directory where it was checked out. If the branch is also finished and merged, delete it separately:

    git branch -d new-auth-system

    The -d flag is safe because git prevents deleting unmerged branches. If you’re certain you want to delete an unmerged branch, use -D instead.

    Sometimes you’ll manually delete a worktree directory without using git worktree remove. Maybe you cleaned up your filesystem and forgot to tell git. When this happens, git still thinks the worktree exists. Clean up the stale metadata:

    git worktree prune

    This removes references to worktrees whose directories no longer exist. Running this periodically keeps your worktree list accurate.

    The Daily Rhythm

    After using worktrees for a while, a natural workflow emerges. Start your day by updating main in the main worktree. Rebase your active feature branches to incorporate those changes. When new work comes in, create feature branches in dedicated worktrees. When you need to review code, use your review worktree. When you finish work, merge to main and clean up the worktree.

    This rhythm eliminates the constant branch switching that fragments your attention. Each workspace maintains its own context. You move between them by changing directories, and your mental context switches cleanly because the filesystem itself shows you where you are.

    The mechanics are simple, but the effect is profound. You stop thinking about git as a sequence of checkouts and stashes and start thinking about it as a set of parallel workspaces. Each workspace evolves independently until you decide to synchronize them. The synchronization itself becomes a deliberate action rather than an automatic side effect of switching branches.

    That deliberateness is the real benefit. You control when contexts merge, when branches update, and when work moves between worktrees. Git stops interrupting your flow and starts supporting it.

  • Escape Branch Switching

    Escape Branch Switching

    Git Worktrees Part 1

    If you’ve ever been deep in debugging a feature branch when someone asks you to quickly review a pull request or fix a production bug, you know the pain. You can’t just switch branches—you’ve got uncommitted changes, half-finished work, and a mental context that will take ten minutes to rebuild when you come back. So you either commit incomplete work with a message like “WIP – will fix later” or you stash everything and hope you remember what you were doing.

    There’s a better way. Git worktrees let you check out multiple branches simultaneously, each in its own directory. No more branch switching. No more stashing. No more losing your place.

    What Worktrees Actually Are

    A worktree is just a working directory connected to your repository. When you clone a repository normally, you get one worktree—the directory where your files live and where git status shows what’s changed. Git worktrees let you create additional working directories, each with its own checked-out branch, all sharing the same underlying repository data.

    Think of it this way: your repository is a database of commits, branches, and history. A worktree is a view into that database, showing you one particular branch’s files. With multiple worktrees, you can have multiple views open at the same time.

    The key insight is that worktrees share everything except the working directory itself. Commits made in one worktree are immediately visible in all others. Branch updates propagate instantly. But each worktree has its own set of files, its own staging area, and its own checked-out branch.

    Why This Changes Everything

    The most immediate benefit is eliminating context switching. When you’re working on a feature in one worktree and need to review a PR, you don’t stop what you’re doing. You just cd ../review-worktree and check out the PR branch there. Your feature branch work sits untouched in its directory, exactly as you left it. When you’re done with the review, you cd back and continue where you left off.

    This is particularly powerful when you’re working with AI agents or automation. You can have one worktree where an agent is running tests, another where you’re actively developing, and a third where you’re reviewing someone else’s changes. Each workspace operates independently without the chaos of switching branches or managing multiple clones of the repository.

    The second major benefit is that worktrees prevent common git footguns. Ever accidentally committed to the wrong branch? Run a destructive rebase when you meant to be on a different branch? Worktrees make it physically obvious which branch you’re on because you’re literally in a different directory. The file path in your terminal shows you exactly where you are.

    Getting Started

    Here’s what the basic setup looks like. Instead of cloning normally, you start with a bare repository:

    git clone --bare git@github.com:user/repo.git repo.git
cd repo.git

    A bare repository has no working directory—it’s just the git database. This positions all your worktrees as equals rather than having one “main” directory and several “linked” ones. Now create your first worktree:

    git worktree add main main

    This creates a directory called main and checks out the main branch there. The first argument is the directory name, the second is the branch to check out. Add a few more:

    git worktree add -b review review
git worktree add -b hotfix hotfix

    Now you have three directories, all tracking the main branch. Wait—didn’t we say you can’t check out the same branch in multiple worktrees? That’s correct, and that’s where the pattern gets interesting. You don’t actually work directly on main in these worktrees. Instead, you create feature branches:

    cd review
git checkout pr/123

cd ../hotfix
git checkout -b fix/urgent-bug hotfix

    The worktree directories are just namespaces. What matters is which branch is checked out inside them.

    The Core Workflow

    Once you have worktrees set up, your daily workflow changes in a subtle but significant way. Instead of using git checkout to switch contexts, you use cd to change directories. This feels weird at first because cd seems too simple for something as important as switching branches. But that simplicity is the point.

    When you want to work on a feature, you navigate to its worktree and create a branch. When you want to review something, you navigate to your review worktree and check out that branch. When you need to make a hotfix, you navigate to your hotfix worktree and branch from there. Each context stays isolated and preserved. The mental model shift is from “I have one workspace and I change what’s in it” to “I have multiple workspaces and I move between them.”

    What You Need to Know

    Each worktree requires its own project setup. If your project has dependencies that need installing, you’ll run npm install or the equivalent in each worktree. This sounds like overhead, but it’s actually a feature—each worktree can have different dependencies installed, matching the branch you’re working on. When you switch between worktrees, you’re not wondering if your node_modules match your current branch state.

    Disk space is not a major concern. Worktrees share the git object database, so you’re not duplicating your entire repository history multiple times. You’re just duplicating working directories and the specific files checked out on each branch.

    You also need to be aware that git commands run in a worktree context. When you run git status in a worktree, it shows the status of that worktree only. When you commit, you’re committing in that worktree’s context. But branch operations—creating branches, merging, rebasing—affect the shared repository and are visible across all worktrees.

    When Worktrees Make Sense

    Worktrees aren’t for everyone or every project. They make the most sense when you frequently switch contexts—reviewing PRs, jumping to urgent bugs, managing multiple features simultaneously. If you typically work on one thing at a time until it’s done, traditional branch switching might be simpler.

    The sweet spot is when you need to maintain multiple parallel workstreams in a single repository. This could be because you’re collaborating with others and need to review their work frequently. It could be because you’re managing both development and production hotfixes. Or it could be because you’re orchestrating multiple AI agents, each working on different aspects of the codebase simultaneously.

    In part two, we’ll dive into the practical patterns for managing these parallel workstreams, keeping them synchronized with your main branch, and merging work back together cleanly.