Rebase

Mandatory Pre-Rebase Analysis

Complete all steps below before running git rebase.

Step 1 — Identify the merge base and branch files

MERGE_BASE=$(git merge-base <branch> <target>)
git diff <target>...<branch> --name-only        # files touched by the branch
git diff "${MERGE_BASE}..<target>" --name-only  # files changed on target since divergence

Step 2 — Diff overlapping files

For each file appearing in both lists above, read both sides:

git diff "${MERGE_BASE}..<target>" -- <file>    # what target changed
git diff "${MERGE_BASE}..<branch>" -- <file>    # what the branch changed

Determine what each side changed and in which regions.

Step 3 — Assign a disposition to every overlapping file

Disposition	When to use
KEEP	Branch version wins; target change is irrelevant or already superseded
MERGE	Both sides changed different regions — list which regions each side owns
DROP	Branch change superseded by what target already landed; discard it
REWRITE	Branch intent survives, but implementation must change to account for target's changes
NO_CONFLICT	File touched only by the branch — no overlap with target

Step 4 — State the plan before executing

Output the full plan in this format before any git rebase command:

Pre-rebase plan — <branch> onto <target>

Overlapping files:
  path/to/file.py: MERGE — branch adds X in foo(); target rewrites bar(); no region overlap
  path/to/other.py: DROP — target already landed the same change
  path/to/third.py: REWRITE — branch intent survives; must account for renamed parameter on target

No-conflict files (branch-only): path/a.py, path/b.py

Step 5 — Execute the rebase

git rebase <target> <branch>

Pass <branch> — the same one Steps 1-4 analysed. git rebase's positional form is [<upstream> [<branch>]]; omitting <branch> rebases whatever is currently checked out, which is a different branch whenever the analysis targeted one you are not standing on. That rewrites commits the plan never looked at and leaves the requested branch untouched.

Naming <branch> makes git check it out first, which fails with fatal: '<branch>' is already used by worktree at ... when another worktree holds it. Run the rebase from the worktree that owns the branch:

git worktree list          # find the worktree whose HEAD is <branch>

Then run the git rebase <target> <branch> above from that directory. Do not free the branch by detaching or switching the other worktree — another session may be mid-task in it.

On each conflict:

1. Resolve according to the plan.
2. When a conflict deviates from the plan (e.g., a region marked NO_CONFLICT has an unexpected conflict): stop, explain the deviation, update the plan entry, then resolve.

After completing, verify with git log --oneline and run the test suite.

Rules

• Write the Step 4 plan before running git rebase.
• Name <branch> in the git rebase invocation — never rely on the current checkout matching it.
• Resolve each conflict according to the plan — check before accepting either side.
• Read file-level diffs (Step 2); commit message titles are not a substitute.