kandev/docs/plans/fork-pr-comparison-targets/task-05-cross-fork-e2e.md

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

id	title	status	wave	depends_on	plan	spec
05-cross-fork-e2e	Prove cross-fork comparison end to end	done	5	04-comparison-target-ui-and-docs	plan.md	../../specs/platform/requirements/workspace-git-status.md

Task 05: Cross-fork desktop/mobile regression

Add deterministic browser coverage for the reported stale-fork failure and its manual recovery path. Use real temporary Git repositories plus the mock provider; do not assert only mocked numeric payloads.

Red tests first

Create an upstream repository, an outdated fork with the same main branch name, and a feature branch whose single commit changes three files. Associate a mock provider PR whose head repository is the fork and target repository is upstream. Before production changes, the fixture must reproduce inflated commits/diff from the fork base.

Desktop scenario

Add apps/web/e2e/tests/git/fork-pr-comparison-target.spec.ts:

• create/start the fork-attached task and wait for its initial Git observation;
• associate through the real task-PR association route with complete provider head/base identity;
• assert target label upstream/widget:main;
• assert exactly one local contribution commit and the expected three-file +/- scope in Changes/Review;
• assert the task-card totals refresh without backend/session restart;
• reload and prove the same target and totals survive;
• choose fork main in Compare against, assert the provider target clears, and confirm this action does not change the PR base, origin, checkout branch, or push remote; and
• cover an unreadable/colliding comparison target with the explicit unavailable state and no inflated totals.

Mobile scenario

Add apps/web/e2e/tests/git/mobile-fork-pr-comparison-target.spec.ts using the mobile project:

• open Changes through the normal mobile control;
• open the existing branch details touch drawer and assert the full target identity is accessible;
• assert the one-commit/three-file scope and unavailable-state recovery copy;
• perform the same-name manual override through touch input;
• verify one Changes scroll owner, touch-sized controls, no clipped drawer content, and no horizontal document overflow.

Fixture and cleanup rules

• Extend shared Git/provider helpers only where both specs reuse them.
• Use the managed E2E backend and API client; do not start a second backend.
• Keep repositories under the test-owned temporary directory and restore/reset shared seed state in cleanup.
• Use causal API/WS waits rather than arbitrary sleeps. Keep screenshots optional and deterministic.

Acceptance

• The exact public-repository fork topology that produced the bug is covered on desktop and mobile.
• The test fails against branch-only origin resolution and passes only when association, persistence, materialization, cache refresh, and UI are all connected.
• Existing PR drift/history and multi-PR E2E suites remain green.

Likely files

• apps/web/e2e/tests/git/fork-pr-comparison-target.spec.ts (new)
• apps/web/e2e/tests/git/mobile-fork-pr-comparison-target.spec.ts (new)
• apps/web/e2e/helpers/git-helper.ts
• apps/web/e2e/helpers/api-client.ts
• apps/web/e2e/fixtures/* only if the existing fixture cannot express two remotes
• apps/backend/internal/backendapp/e2e_reset.go only for provider identity fixture support not completed in Task 03

Verification

cd apps/web && pnpm e2e:run --host --no-build --project chromium -- e2e/tests/git/fork-pr-comparison-target.spec.ts
cd apps/web && pnpm e2e:run --host --no-build --project mobile-chrome -- e2e/tests/git/mobile-fork-pr-comparison-target.spec.ts
cd apps/web && pnpm e2e:run --host --no-build --project chromium -- e2e/tests/git/git-changes-panel.spec.ts e2e/tests/pr/pr-detection.spec.ts
cd apps/web && pnpm e2e:run --host --no-build --project mobile-chrome -- e2e/tests/git/mobile-pr-checkout-drift.spec.ts

Run cd apps && pnpm install --frozen-lockfile first in a fresh worktree.

Parallelism

sequential. Final integration gate after Tasks 01-04.

Output contract

Record the reproduced before-values, corrected after-values, desktop/mobile command results, cleanup behavior, and any remaining executor limitation. Mark Task 05 and the plan complete only after both new projects pass.

Completion record

• The fixture keeps the local fork's origin/main at the old commit, advances the disposable upstream, and bases feature/fork on that upstream commit before adding one three-file contribution commit.
• Corrected results are one local contribution commit, three PR files, and upstream/widget:main in both desktop and mobile Changes surfaces. The branch-only baseline would include the upstream commit and is therefore distinct from the asserted result.
• Green commands:
  • pnpm e2e:run --host --no-build --project chromium -- e2e/tests/git/fork-pr-comparison-target.spec.ts
  • pnpm e2e:run --host --no-build --project mobile-chrome -- e2e/tests/git/mobile-fork-pr-comparison-target.spec.ts
• The host runner cleaned each disposable repository/backend fixture after the run; no executor limitation remains for this regression.