kandev/docs/plans/remote-contribution-head-drift/task-04-responsive-force-push-regression-e2e.md

id, title, status, wave, depends_on, plan, spec

id	title	status	wave	depends_on	plan	spec
04-responsive-force-push-regression-e2e	Prove rewritten PR behavior end to end	done	4	03-diverged-changes-ui-and-mobile-parity	plan.md	../../specs/tasks/system-design/remote-contribution-tasks.md

Task 04: Prove Rewritten PR Behavior End to End

Recreate the reported stale-checkout/current-PR mismatch in browser tests and prove that the behavior is correct on desktop and mobile.

Desktop scenario

Extend apps/web/e2e/tests/git/git-changes-panel.spec.ts with a deterministic local Git graph and mocked current GitHub PR commits:

1. Create a task checkout with several commits representing the original contributor history and record its local/upstream head.
2. Associate a mocked open PR whose current commit list uses different SHAs, simulating a force-pushed rewrite. At least one rewritten commit may intentionally reuse a message and file totals to prove metadata is not used for deduplication.
3. Open Changes and assert:
   • the drift warning is visible;
   • “Current PR commits” and “Local checkout commits” are separate;
   • each list contains only its own SHAs/messages;
   • the old local rows have the neutral checkout marker, not the green unpushed arrow;
   • Push and Pull are unavailable with the reconciliation reason;
   • no action mutated local HEAD or the worktree.
4. Add a safe local-ahead control case: provider/upstream head match and one local commit exists. Assert one Push commit even when base ahead is larger.

Reuse mockGitHubAddPRs, mockGitHubAddPRCommits, and mockGitHubAssociateTaskPR. Extend the E2E helper only if the test cannot establish the required upstream ref with its real temporary repository.

Mobile scenario

Add apps/web/e2e/tests/git/mobile-pr-checkout-drift.spec.ts under the mobile-chrome project:

• Open the same kind of linked task on Pixel 5 dimensions.
• Enter the shared Changes panel through the normal mobile navigation.
• Assert the warning and both histories are readable without horizontal overflow.
• Open the 44px Git actions trigger and assert Pull/Push are disabled while Commit remains available.
• Assert the panel keeps one vertical scroll owner and the warning/headings can be reached by ordinary touch scrolling.

Use stable roles, translated accessible names, and data-testid only where no semantic locator exists. Do not use fixed sleeps.

Acceptance

• The original visual failure is reproduced by test data and fails before Tasks 01–03.
• Desktop and mobile tests pass after the fix.
• Tests prove both containment of a rewritten history and preservation of an ordinary local-ahead flow.
• The test does not depend on a live GitHub account or network timing.

Files likely touched

• apps/web/e2e/tests/git/git-changes-panel.spec.ts
• apps/web/e2e/tests/git/mobile-pr-checkout-drift.spec.ts (new)
• apps/web/e2e/helpers/api-client.ts only if a narrow mock/status helper is required
• relevant Changes/Session page object only if semantic locators cannot express the new state

Verification

Run the focused specs first, then the related browser project if time permits:

cd apps/web && pnpm e2e:raw --project=chromium tests/git/git-changes-panel.spec.ts --grep 'rewritten PR|upstream push count'
cd apps/web && pnpm e2e:raw --project=mobile-chrome tests/git/mobile-pr-checkout-drift.spec.ts

Dependencies and parallelism

Depends on Task 03. Run after the UI contract is stable; this task may expose wiring defects that must be fixed in the owning files rather than hidden in test helpers.

Output contract

Record the red failure signature, green desktop/mobile command output, screenshots or traces for any failure investigated, viewport/scroll assertions, and blockers. Update this task and the plan checkbox when complete.

Completion evidence

• The browser fixture initially exposed a missing local origin/main tracking ref. The regression setup now restores the temporary remote URL, fetches the ref, and establishes the graph without relying on fixture history.
• Desktop rewrite and local-ahead scenarios passed with Chromium; the rewritten-history scenario passed on Pixel 5. Assertions cover separate histories, accurate provenance, disabled remote actions, unchanged local HEAD/worktree, one scroll owner, and no horizontal overflow.
• Screenshot capture hooks are present for the desktop and mobile regression states; fresh PR assets will be captured during final delivery verification.