test-report-2026-05-17.md — Run summary for the 2026-05-17 session: 64 passed, 3 flaky, 2 skipped, 61 failed (all in pre-existing specs with legacy selectors). Includes per-spec results, root causes for every failure category, and recommended fixes. e2e-test-framework.md — Developer reference covering stack, directory layout, playwright.config.ts rationale (workers: 2, why bcrypt floods the server), shared helpers, selector conventions (PO form has no htmlFor bindings), mobile viewport pattern, and future improvements including auth state sharing. e2e-test-plan.md — Feature coverage matrix mapping all 21 user story groups to their spec files and roles, individual test case tables, regression trigger checklist by code area, gap analysis, and planned CI configuration. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
9.8 KiB
PPMS — E2E Test Framework Reference
This document describes the Playwright-based end-to-end test framework for the PPMS portal: its stack, directory layout, configuration, shared utilities, and the conventions every spec must follow.
Stack
| Layer | Tool | Version |
|---|---|---|
| Test runner | @playwright/test |
1.60 |
| Browser | Chromium (headless) | bundled with Playwright |
| Language | TypeScript | inherits from app tsconfig.json |
| Package manager | pnpm | same as portal app |
| App server | Next.js 15 dev server (pnpm dev) |
auto-started by Playwright config |
Directory Layout
App/pelagia-portal/
├── playwright.config.ts # Root config — workers, retries, baseURL, webServer
└── tests/
├── e2e/
│ ├── helpers/
│ │ ├── login.ts # Shared login(), createDraftPo(), submitPo(), USERS
│ │ └── auth.js # Legacy plain-JS login helper (pre-existing)
│ ├── dashboard/
│ │ └── po-status-badges.js
│ ├── inventory/
│ │ ├── items-tags.spec.ts
│ │ └── cart-icon.spec.ts
│ ├── mobile/
│ │ ├── desktop-required.spec.ts
│ │ ├── manager-approvals.spec.ts
│ │ ├── accounts-payments.spec.ts
│ │ └── bottom-nav.spec.ts
│ ├── admin-bordered-buttons.spec.ts
│ ├── approvals-edit-highlight.spec.ts
│ ├── export-gate.spec.ts
│ ├── notification-bell.spec.ts
│ ├── partial-receipt.spec.ts
│ ├── payment-history.spec.ts
│ ├── po-submit-button.spec.ts
│ ├── profile.spec.ts
│ ├── rebrand.spec.ts
│ └── vendor-auto-verify.spec.ts
├── integration/ # Vitest integration tests (separate suite)
└── unit/ # Vitest unit tests (separate suite)
Configuration (playwright.config.ts)
export default defineConfig({
testDir: "./tests/e2e",
fullyParallel: true,
forbidOnly: !!process.env.CI,
retries: process.env.CI ? 2 : 1, // 1 local retry reduces flakiness from auth concurrency
workers: process.env.CI ? 1 : 2, // 2 local workers — more causes NextAuth bcrypt flooding
reporter: "html",
use: {
baseURL: "http://localhost:3000",
trace: "on-first-retry",
},
webServer: {
command: "pnpm dev",
url: "http://localhost:3000",
reuseExistingServer: !process.env.CI, // reuse running dev server locally
},
});
Why workers: 2
The app uses NextAuth v5 with bcrypt password hashing for every login. Under high parallelism (the default of ~50% CPU cores) all workers attempt to authenticate simultaneously, overwhelming the dev server and causing login redirects to time out. Two workers provide enough parallelism to keep the suite fast without triggering the concurrency limit.
Shared Helpers (tests/e2e/helpers/login.ts)
USERS — seed credentials
export const USERS = {
TECH: { email: "tech@pelagia.local", password: "tech1234" },
MANNING: { email: "manning@pelagia.local", password: "manning1234" },
ACCOUNTS: { email: "accounts@pelagia.local", password: "accounts1234" },
MANAGER: { email: "manager@pelagia.local", password: "manager1234" },
SUPERUSER: { email: "superuser@pelagia.local", password: "super1234" },
AUDITOR: { email: "auditor@pelagia.local", password: "audit1234" },
ADMIN: { email: "admin@pelagia.local", password: "admin1234" },
};
login(page, creds)
Navigates to /login, fills credentials, and waits up to 20 s for the
redirect away from /login. The 20 s timeout is intentional — the bcrypt hash
check plus DB round-trip can exceed the Playwright default 5 s under any load.
await login(page, USERS.MANAGER);
createDraftPo(page, title)
Creates a minimal PO as DRAFT and returns the absolute PO URL. Uses
name-attribute selectors because the PO form labels have no htmlFor/id
binding — getByLabel() will not resolve.
const poUrl = await createDraftPo(page, "Test PO - boiler parts");
submitPo(page, title)
Same as createDraftPo but clicks the Submit for Approval button instead of
Save as Draft. Returns the PO URL after redirect.
Selector Conventions
Critical: PO form has no accessible label bindings
The new-PO form (/po/new) and the edit form use <label> elements that are
visual only — they have no for attribute and the inputs have no id.
page.getByLabel() will not find them.
Always use name-attribute selectors for PO form fields:
// CORRECT
page.locator('input[name="title"]')
page.locator('select[name="vesselId"]')
page.locator('select[name="accountId"]')
page.locator('input[name="projectCode"]')
// WRONG — will time out
page.getByLabel(/title/i)
page.getByLabel(/vessel/i)
Role-badge selectors (profile page)
The user's role appears in both the desktop sidebar/header and the profile page.
getByText("Technical") will fail with a strict-mode violation. Scope to <dd>:
// CORRECT — scoped to the profile <dd> role badge
await expect(page.locator("dd span").filter({ hasText: "Technical" })).toBeVisible();
// WRONG — strict-mode violation (role appears in header too)
await expect(page.getByText("Technical")).toBeVisible();
Mobile viewport
Mobile tests must set the viewport explicitly before login():
const MOBILE_VIEWPORT = { width: 375, height: 812 };
test("...", async ({ page }) => {
await page.setViewportSize(MOBILE_VIEWPORT);
await login(page, USERS.MANAGER);
// ...
});
Checking CSS classes (not computed styles)
For visual-only assertions (bordered buttons, colored badges), check the element's
class attribute rather than computed CSS, since Tailwind classes are the source
of truth:
const cls = await page.locator("button", { hasText: "Edit" }).first().getAttribute("class");
expect(cls).toMatch(/border/);
Writing a New Spec
File naming
- Feature specs:
tests/e2e/<section>/<feature>.spec.ts - Pre-
@playwright/testscripts:tests/e2e/<section>/<feature>.js(legacy format)
New specs must use @playwright/test format (import { test, expect } from "@playwright/test").
Header comment
Every new spec file must open with a JSDoc block:
/**
* User stories covered: Feature N — <Name>
* - Story 1
* - Story 2
*
* Created: YYYY-MM-DD
*/
Step logging
Log every meaningful assertion with a ✓ prefix so CI output is scannable:
console.log("✓ Export buttons hidden on DRAFT PO");
console.log(`✓ Logged in as ${creds.email}`);
Graceful skips for seed-dependent tests
Tests that require specific seed data (e.g., an item with multiple vendors, or a PO in a particular status) should skip rather than fail hard if the precondition is absent:
const poRow = page.locator("[data-status='MGR_APPROVED']").first();
if ((await poRow.count()) === 0) {
test.skip(true, "No MGR_APPROVED PO in seed data");
return;
}
HTTP-level assertions (no browser needed)
Use request.get() for API-level checks (status codes, headers) rather than
driving the full browser:
test("export returns 403 for DRAFT PO", async ({ request }) => {
// Must first obtain a session cookie — use page-based login or apiRequestContext
const resp = await request.get(`/api/po/${draftPoId}/export?format=pdf`);
expect(resp.status()).toBe(403);
});
Running the Suite
cd App/pelagia-portal
pnpm test:e2e # headless, 2 workers
pnpm test:e2e:ui # Playwright interactive UI
pnpm test:e2e -- --headed # watch the browser
# Single file
pnpm test:e2e -- tests/e2e/mobile/bottom-nav.spec.ts
# Name filter
pnpm test:e2e -- --grep "Feature 20"
# All tests with trace on every run (debugging)
pnpm test:e2e -- --trace on
HTML report at playwright-report/index.html after every run.
Known Gotchas
| Situation | Symptom | Fix |
|---|---|---|
| Many workers, all trying to log in at once | Login times out; page stays on /login |
Keep workers ≤ 2 locally; use storageState for auth |
getByLabel(/title/i) on PO form |
Locator times out — no htmlFor binding |
Use locator('input[name="title"]') |
getByText("Technical") on profile page |
Strict-mode violation — appears in header AND profile | Scope to page.locator("dd span").filter(...) |
| Multi-role flow in one test (submit → approve → pay) | Flaky under 2 workers; competing for same user | Use beforeAll + a dedicated seed PO or run single-threaded |
Viewport-dependent md:hidden elements |
Element found at desktop viewport but not mobile, or vice versa | Always set viewport before login for mobile tests |
router.push soft navigation |
waitForLoadState('networkidle') still sees old URL |
Use page.waitForURL(pattern) concurrently with the click/select |
Future Improvements
-
Auth state sharing — Save one
storageStateper role in a global setup file. This eliminates ~100 login round-trips and should cut suite time from 25 min to under 5 min. -
Fix pre-existing specs — Update
submitter-journey.spec.tsandpo-export.spec.tsto use the shared helper's name-based selectors (FIX-1 in test report). -
data-testidattributes — Add sparsedata-testidattributes to ambiguous elements (unit price input, line-item rows) so specs don't depend on implementation details like placeholder text or CSS class names. -
CI integration — Run
pnpm test:e2ein GitHub Actions on every PR. Useworkers: 1andretries: 2(already wired forprocess.env.CI). -
Visual regression — Add Percy or Playwright's built-in screenshot comparison for the status badge colors and mobile card layout.