refactor(ci): scope PR jobs to packages changed by the PR

[xsahil03x]

Replaces the previous diff-aware-CI attempt on this branch with a version that addresses the bugs we found in the original — silent test failures from the bash &&/|| trap, and example packages leaking back into the affected set via melos's dependent expansion.

What changes

Four new diff-aware melos scripts that scope each job to only the packages a PR touches (plus their transitive dependents):

• analyze:changes — dart analyze on changed packages and their dependents (includes examples; analyzing them catches API drift in the sample code)
• format:changes — dart format --set-exit-if-changed on the same set
• test:changes — flutter test --no-pub --coverage, with a wrapper script that skips packages without a test/ directory
• lint:pub:changes — flutter pub publish --dry-run on publishable changed packages

The workflow's analyze/format/test jobs now call the :changes variants directly. Push events to master use the same scripts — HEAD~1 resolves to the previous tip there, scoping to the merge commit's changes.

Why HEAD~1...HEAD (no env var, no github.base_ref)

GitHub Actions's default checkout for pull_request events checks out an auto-generated merge commit at refs/pull/N/merge. On that commit:

• HEAD^1 (first parent) = base branch tip — master, v10.0.0, feat/design-refresh, whatever the PR targets
• HEAD^2 (second parent) = PR head tip

So HEAD~1...HEAD is always exactly the PR's effective diff, regardless of which branch it targets, with zero workflow plumbing. On push events, HEAD~1 is the previous tip — same diff shape.

Verified empirically:

$ git merge --no-ff feature -m "Merge feature into main"
$ git diff HEAD~1...HEAD --name-only
c
d

Pre-existing files (a, b) correctly excluded; only the feature branch's added files appear.

Why test:changes needs a wrapper script

The melos filter pipeline (packages/melos/lib/src/package.dart:800) runs in this order:

1. applyIgnore, applyDirExists, applyFileExists
2. filterPrivatePackages, applyScope, applyCategories
3. applyDependsOn / applyNoDependsOn
4. filterNullSafe, filterPublishedPackages
5. applyDiff
6. applyIncludeDependentsOrDependencies ← runs LAST, walks _packageMap (the unfiltered workspace)

So --ignore=\"*example*\" and packageFilters: { dirExists: test } correctly filter examples out of the initial set — but --include-dependents then pulls them straight back in from the full workspace. Confirmed locally: a single change to stream_chat with --include-dependents selects 11 packages, 5 of them *_example.

We can't drop --include-dependents (we need to test that a change to stream_chat doesn't break tests in stream_chat_flutter), and we don't want to remove examples from the workspace (they need to participate in analyze/format). So the skip decision happens per-package, after melos has computed the affected set:

# .github/workflows/scripts/test-package.sh
set -e
if [[ \"\$MELOS_PACKAGE_NAME\" == *_example ]]; then exit 0; fi
if [ ! -d test ]; then exit 0; fi
exec flutter test --no-pub --coverage

set -e plus explicit exec ensures real test failures propagate — the original `[ -d test ] && flutter test ... || echo skip` pattern would silently swallow non-zero exits from flutter test (classic bash &&/|| precedence bug).

lint:pub:changes doesn't need a wrapper because pub publish --dry-run only validates this package's own pubspec/files/metadata — never reads the parent's code — so we drop --include-dependents entirely and --no-private filters examples cleanly.

Other workflow cleanups

• Coverage threshold checks gain if: hashFiles('packages/X/coverage/lcov.info') != '' guards so they skip cleanly when their package wasn't touched (no "file not found" errors)
• Codecov upload moves to last so we never push coverage that failed local gates
• Drop the legacy ACTIONS_ALLOW_UNSECURE_COMMANDS: 'true' env var (security smell, deprecated)
• Kept everything still useful from master: gate job, fastlane Android/iOS build matrix, SQLite3 install for persistence tests, validate-formatting.sh

Local usage

# Default — diffs against HEAD~1 (your last commit)
melos run test:changes
melos run analyze:changes
melos run format:changes
melos run lint:pub:changes

# For \"what does my whole branch touch vs master\", invoke melos exec directly:
melos exec --diff=\"origin/master...HEAD\" --include-dependents -- \"dart analyze --fatal-infos .\"

The full analyze / format / test:all / lint:pub scripts are kept untouched for the local "check everything" use case.

Caveats worth knowing

• Requires actions/checkout defaults — if anyone later sets ref: \${{ github.event.pull_request.head.sha }}, the HEAD~1 trick stops resolving to the base branch.
• Requires fetch-depth: 0 (already set everywhere in this workflow) so HEAD~1 resolves on the runner.

Summary by CodeRabbit

• Chores
  • Optimized CI pipeline to run checks and tests only on changed packages, reducing build time and providing faster feedback on pull requests.
  • Enhanced test coverage validation with improved handling of packages without coverage data.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 60.57%. Comparing base (07e1a35) to head (a68cdfa).
Report is 22 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2109   +/-   ##
=======================================
  Coverage   60.57%   60.57%           
=======================================
  Files         383      383           
  Lines       23927    23927           
=======================================
  Hits        14493    14493           
  Misses       9434     9434           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 06d04e8d-6a4a-4240-9c17-fdef8ac98f69

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch feat/melos-diff

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}