---
name: jj
description: >
  The jj mental model: working copy as commit, change IDs, anonymous branches,
  revsets, merges, and conflicts as data. Use when jj behavior is confusing or
  you are translating git habits; vcs holds this repo's commit rules.
---

# jj concepts

Commands operate on repository data, not on the working copy. The working copy is itself a commit, so most commands edit the database and leave `@` in place. Rebases always succeed because a conflict is recorded as data in the commit, never raised as an error.

- The working copy is a commit. Nothing is uncommitted, and editing files amends `@`.
- Change IDs name the logical change across rewrites; commit IDs name content. Refer to changes by change-ID prefix.
- Describe intent first. Create the change, describe what it will do, refine the description as you work. Git commits after the fact; jj describes before.
- Two changes sharing a parent are a branch. Branches are anonymous by default; name one only when the name adds value. `jj log -r 'heads(all())'` lists every head.
- A merge is a change with multiple parents, so `jj new A B C` merges any number of changes. There is no `jj merge`.
- Conflicted commits store conflict markers (`+++++++` starts the snapshot, `%%%%%%%` starts the diff) and descendants keep working. Resolving the conflict rebases descendants automatically; the fix propagates.

## Revsets

`-r` on almost every command takes a revset and defaults to `@`. Symbols name single commits: `@`, change IDs, commit IDs. Operators: `@-` parent, `@+` child, `::x` ancestors, `x::` descendants, `&` intersection, `|` union. Functions: `heads(x)`, `description(substring:x)`, `mine()`, `trunk()` (the remote main, master, or trunk bookmark). A useful log for larger repositories: `jj log -r '@ | ancestors(remote_bookmarks().., 2) | trunk()'`.

## Moves

- Squash workflow, the git-index equivalent: describe the change, run `jj new` for an empty scratch change on top, work in the scratch, then `jj squash` moves it into the described change. `jj squash PATH` stages one path; `jj squash -i` selects hunks interactively. `jj abandon` discards the scratch change. `jj squash` is short for `--from @ --into @-` and works between any change and its parent.
- Edit workflow: `jj new -B @ -m "desc"` inserts a change before `@` and rebases descendants; make the fix there, then `jj next --edit` returns to the following change.
- Rebase scope: `-r REV` moves one revision, `-b REV` its whole branch, `-s REV` the revision plus descendants; `-o DEST` sets the destination.

Source: https://steveklabnik.github.io/jujutsu-tutorial/