Git worktrees

Use per-branch mode to work on several branches at once:

# Initialize once in your primary checkout.
db-git init --database-url postgresql://localhost/myapp --mode per-branch

git worktree add ../myapp-feature -b feature/auth
cd ../myapp-feature
db-git run -- npm run dev

With the hook installed, git worktree add creates the new branch's database from seed if it does not already exist. An existing recorded branch database is reused. The primary checkout and linked worktree can run applications simultaneously with separate databases. Databases belong to branches, so forcing the same branch into two worktrees does not give it two databases. PostgreSQL's existing connection policy still applies to cloning and reset operations; template cloning may need to stop connections to its source. Use on_active_connections = "fail" to refuse those operations instead of terminating sessions.

To start from another branch's data, create the database before adding its worktree:

db-git create feature/review --from feature/auth
git worktree add ../myapp-review -b feature/review

All worktrees use the primary checkout's .db-git.toml. init from a linked worktree also updates that shared file. Relative snapshot_dir settings resolve against the primary checkout, even from nested directories in another worktree. A conflicting .db-git.toml in a linked checkout is rejected; reconcile it with the primary file before continuing. Environment and CLI overrides retain their normal precedence. Keep the primary checkout available while using linked ones.

Resource	Scope
Configuration	Primary checkout's .db-git.toml
Branch ownership	Common Git directory: db-git/state.json
Branch operation journals and lock	Common Git directory: db-git/operations/
Default snapshot directory	Common Git directory: db-git/snapshots/
Disable marker and checkout/rebase state	Current worktree's Git directory

list, create --from, and recover see the same ownership and recovery records from every checkout. An interrupted operation blocks mutations and run across worktrees until recovered. disable and enable affect only the current worktree. Older versions' worktree-local ownership or recovery files are detected and preserved; reconcile those records with the common state before proceeding. Existing database names are never automatically changed.

Hook installation and removal respect core.hooksPath, preserve an existing hook as an adjacent .legacy file, and restore it on removal. Legacy hooks still run when db-git is disabled. With Git's default hooks directory, one installation serves every worktree. An absolute core.hooksPath is also shared. A relative core.hooksPath resolves within each checkout, so run db-git hook install in each worktree; if its initial checkout had no hook, run db-git create there too. Removing a shared hook affects every worktree that uses it.

Shared mode cannot safely represent several checked-out branches in one working database. With multiple worktrees, db-git refuses shared-mode snapshot changes, automatic switching, run, and enable; doctor reports the conflict. Existing snapshots and recovery records remain available. Use per-branch mode or remove extra worktrees before resuming shared mode.