knoopx/jj-hunk

jj-hunk

Programmatic hunk selection for Jujutsu. Split, commit, or squash specific hunks without interactive prompts.

Basic Workflow

List available hunks, create a spec file, then apply:

# See what hunks exist
jj-hunk list
jj-hunk list --format text    # Human-readable output
jj-hunk list --files          # Files with hunk counts only

# Split changes using a spec file
jj-hunk split --spec-file /tmp/spec.yaml "feat: add feature A"

# Commit specific files, leave rest uncommitted
jj-hunk commit --spec-file /tmp/commit-spec.yaml "fix: handle edge case"

# Squash into parent
jj-hunk squash --spec-file /tmp/squash-spec.yaml

Targets any revision with -r <rev> (default is @ for working copy). Read specs from stdin with -.

Spec Format

YAML files control which hunks to keep or reset:

files:
  src/main.rs:
    hunks: [0, "hunk-7c3d..."] # Select by index or stable id
  path/to/skip:
    action: reset # Discard all changes
default: reset # Unlisted files get this action

• hunks: [indices|ids] — select by 0-based index or stable id from jj-hunk list
• ids: ["hunk-..."] — select by id string (sha256-based, stable across runs)
• action: keep / action: reset — keep or discard all changes in a file
• default — action for unlisted files

Use jj-hunk list --spec-template to generate an ID-based starting spec. IDs are more reliable than indices since they're stable across runs.

Example: Split by Hunk

# 1. List hunks
jj-hunk list --format text
# Output: M src/main.rs — hunk 0 insert hunk-abc123, hunk 1 insert hunk-def456

# 2. Create spec for first hunk only
cat > /tmp/split-spec.yaml << 'EOF'
files:
  src/main.rs:
    hunks: [0]
default: reset
EOF

# 3. Split — first hunk becomes commit A, rest stays in working copy
jj-hunk split --spec-file /tmp/split-spec.yaml "feat: add feature A"

Example: Commit Subset of Changes

cat > /tmp/commit-spec.yaml << 'EOF'
files:
  src/fix.rs:
    action: keep
default: reset
EOF

jj-hunk commit --spec-file /tmp/commit-spec.yaml "fix: handle edge case"

Tips

• IDs are stable (sha256-based), indices are not — prefer ids for reproducibility
• Use --files for a quick summary without full hunk details
• Read spec from stdin: cat spec.yaml | jj-hunk commit - "message"
• Use -r <rev> with split/squash to target any revision
• When verifying after operations, always use the original change ID — see jj-core skill for template syntax (concat(), description field)