Fix fork divergence: add clean upstream PR workflow, fix rebase merge-base bug, generalize sync backup

[Copilot]

Fork-specific commits were leaking into upstream PRs because the rebase workflow counted commits via the GitHub API (which includes all fork-deviation commits), and the sync workflow only preserved .github/workflows/ files, missing README.md, dependabot.yml, etc.

Changes

• New: .github/workflows/create-upstream-pr.yml
workflow_dispatch workflow to create clean upstream PR branches. Accepts issue_number, source_branch, commit_shas (comma-separated), and optional pr_title. Cherry-picks only the specified SHAs onto a fresh upstream/master-based branch (upstream-pr/issue-{N}), runs verification (git log + git diff --stat), force-pushes, and outputs the direct PR comparison URL:

  https://github.com/eclipse-jdt/eclipse.jdt.ui/compare/master...carstenartur:eclipse.jdt.ui:upstream-pr/issue-{N}
  

• Fix: .github/workflows/rebase-upstream.yml — rebase step
  Replaced HEAD~${PR_COMMIT_COUNT} (broken: count from GitHub API includes fork-specific commits) with git merge-base to find the true fork point:

  MERGE_BASE=$(git merge-base HEAD upstream/master)
  ACTUAL_COMMITS=$(git rev-list --count $MERGE_BASE..HEAD)
  git rebase --onto upstream/master $MERGE_BASE

  Also removed the now-unused commit_count output from the "Get PR branch info" step.

• Fix: .github/workflows/sync-upstream.yml — backup/restore steps (both scheduled and manual jobs)
  Replaced the dynamic upstream-diff scan (only caught .github/workflows/ files) with a manifest-driven approach: reads .github/fork-specific-files.txt line by line, backs up each listed file before reset, restores with cp -r + git add -A to handle any path. Uses POSIX-compatible case syntax instead of [[.

• New: .github/fork-specific-files.txt
  Explicit manifest of fork-specific files to preserve across upstream syncs:

  .github/dependabot.yml
  .github/workflows/codacy.yml
  .github/workflows/codeql.yml
  .github/workflows/maven.yml
  .github/workflows/rebase.yml
  .github/workflows/rebase-upstream.yml
  .github/workflows/sync-upstream.yml
  .github/workflows/create-upstream-pr.yml
  .github/fork-specific-files.txt
  README.md
  

  The manifest includes itself so it survives sync resets.

Original prompt

Problem

The fork carstenartur/eclipse.jdt.ui has a persistent divergence from eclipse-jdt/eclipse.jdt.ui. When PRs are opened from fork branches against upstream, fork-specific commits leak into the PR. This happened with PR #2779 which was supposed to only contain a fix for issue eclipse-jdt#1865 (overload resolution detection in Move Instance Method), but instead included 38 commits and 16 changed files — 10 of which were fork-specific (workflows, badges, pom changes, etc.).

Two Things to Fix

1. Add a create-upstream-pr.yml workflow

Create .github/workflows/create-upstream-pr.yml — a workflow_dispatch workflow that automates creating clean upstream PR branches. It should:

• Accept inputs: issue_number, source_branch, commit_shas (comma-separated SHAs to cherry-pick), and pr_title
• Fetch upstream/master from https://github.com/eclipse-jdt/eclipse.jdt.ui.git
• Create a branch upstream-pr/issue-{issue_number} based on upstream/master
• Cherry-pick only the specified commit SHAs onto this clean branch
• Force-push the branch to the fork
• Output a direct link to create the PR on upstream: https://github.com/eclipse-jdt/eclipse.jdt.ui/compare/master...carstenartur:eclipse.jdt.ui:upstream-pr/issue-{issue_number}
• Add a verification step that runs git log upstream/master..HEAD --oneline and git diff upstream/master..HEAD --stat so the user can see exactly what will be in the PR

2. Fix the existing rebase-upstream.yml workflow

The current rebase-upstream.yml has a bug: it uses HEAD~${PR_COMMIT_COUNT} to determine which commits to rebase, but PR_COMMIT_COUNT comes from the GitHub API which counts ALL commits that differ between the PR branch and upstream — including fork-deviation commits. This means it replays fork-specific commits onto upstream.

Fix the rebase step in .github/workflows/rebase-upstream.yml to use git merge-base instead:

Replace the current rebase logic (lines 142-165):

git remote add upstream https://github.com/eclipse-jdt/eclipse.jdt.ui.git
git fetch upstream master

PR_COMMIT_COUNT=${{ steps.pr.outputs.commit_count }}
...
git rebase --onto upstream/master HEAD~${PR_COMMIT_COUNT}

With:

git remote add upstream https://github.com/eclipse-jdt/eclipse.jdt.ui.git
git fetch upstream master

# Find the actual fork point with upstream
MERGE_BASE=$(git merge-base HEAD upstream/master)
echo "Merge base with upstream: $MERGE_BASE"

# Count actual commits to rebase (between merge-base and HEAD)
ACTUAL_COMMITS=$(git rev-list --count $MERGE_BASE..HEAD)
echo "Actual commits to rebase: $ACTUAL_COMMITS"

if [ "$ACTUAL_COMMITS" -eq 0 ]; then
  echo "Error: No commits to rebase"
  echo "success=false" >> $GITHUB_OUTPUT
  exit 0
fi

# Rebase using merge-base (correct: only replays commits after the fork point)
if git rebase --onto upstream/master $MERGE_BASE; then
  echo "success=true" >> $GITHUB_OUTPUT
else
  echo "success=false" >> $GITHUB_OUTPUT
  git rebase --abort
fi

Also remove the commit_count output from the "Get PR branch info" step since it's no longer needed by the rebase logic (but keep the rest of that step intact).

3. Improve sync-upstream.yml to also backup/restore non-workflow fork-specific files

The current sync-upstream.yml only backs up fork-specific files under .github/workflows/. It should also handle:

• .github/dependabot.yml
• README.md (if it differs from upstream, i.e., contains fork badges)

Update the backup step to scan for ALL files that differ between the fork and upstream (not just .github/workflows/), using a configurable list or by diffing. A simple approach: maintain a file called .github/fork-specific-files.txt that lists the paths of fork-specific files (one per line). The sync workflow reads this file to know what to back up and restore.

Create .github/fork-specific-files.txt with:

.github/dependabot.yml
.github/workflows/codacy.yml
.github/workflows/codeql.yml
.github/workflows/maven.yml
.github/workflows/rebase.yml
.github/workflows/rebase-upstream.yml
.github/workflows/sync-upstream.yml
.github/workflows/create-upstream-pr.yml
.github/fork-specific-files.txt
README.md

Update sync-upstream.yml to use this file list instead of only scanning .github/workflows/.

Summary of Files to Create/Modify

1. CREATE .github/workflows/create-upstream-pr.yml — new workflow
2. MODIFY .github/workflows/rebase-upstream.yml — fix the rebase logic to use merge-base
3. MODIFY .github/workflows/sync-upstream.yml — use fork-specific-files.txt for backup/restore
4. CREATE .github/fork-specific-files.txt — list of fork-only files

This pull request was created from Copilot chat.

---

💬 We'd love your input! Share your thoughts on Copilot coding agent in our 2 minute survey.