Bagisto Playwright Testing

Playwright Testing

Do not write the spec first

"Add Playwright tests for X" is not an instruction to start typing a spec. It is an instruction to find out what X actually does, what already covers it, and which of its behaviours a browser can prove that a Pest test cannot.

A test written before that investigation is usually a duplicate, a test of the wrong layer, or a test that passes whether or not the feature works. All three cost more than they return: someone maintains them, and nobody can trust them.

Work the pipeline in order — understand the feature and its rules, map what already covers it, decide the scenarios and which belong in Pest, design the data and locators, implement, then classify every failure before changing anything:

Scale the process to the change. A one-line locator fix needs the investigation step (why did the locator break?) and the proof step (does it still fail when the feature does?), not a coverage matrix. A new feature area needs all of it. What never gets skipped: reading the application code, and proving the test can fail.

How to think about this work

• Maximise meaningful behavioural coverage, not test count. Ten tests that differ only in input value are one test with a data table.
• A passing test is not evidence of a good test. Ask separately whether it could have failed — if the feature were deleted, would this go red?
• Do not copy an existing test without deciding what it proves. These suites contain tests that assert only that a toast appeared.
• A business prerequisite is not an execution dependency. Establish the state you need; never rely on another spec having run.
• Do not add a wait without naming the thing you are waiting for.
• Classify a failure before changing anything. If the application is wrong, the red test is doing its job — never edit an assertion to get green.
• Choose a locator because it is stable and uniquely identifies the intended element, never because it is short or made strict mode quiet.
• Search before you build. A new fixture, helper or page object is the last resort, after reuse, extension and refactoring.
• A regression test is not finished until it has failed against the bug.
• Prefer the strongest behavioural verification with the least machinery, and stop when the risks are covered — test-design.md says when.

Architecture

Three independent Playwright projects — packages/Webkul/{Admin,Shop,Installer}/tests/e2e-pw/, each with its own playwright.config.ts, setup.ts, pages/, tests/, utils/ and data/ — plus a root-level tests/e2e-pw/helpers/ layer of dependency-free helpers all three import through @shared/*.

Two rules follow, and both are load-bearing:

• The shared layer takes no dependencies. CI installs node_modules per package only, so nothing under helpers/ may import @playwright/test, dotenv or any third-party package. A helper that needs one belongs in that package's utils/.
• The suites share no page objects, fixtures or specs. Each carries page objects for whatever screens it drives, including the other side's. Never import across package boundaries.

Per-test timeouts are 60 s in Admin, 240 s in Shop and 300 s in Installer; expect waits 30 s in all three. Raise a single flow with test.setTimeout(...), never the config.

Full layout, the where-does-code-go table and the wrapper pattern: architecture.md.

Running

Run from the package directory, never the repo root — that is where node_modules lives. All three packages expose the same scripts:

cd packages/Webkul/Admin          # or Shop, or Installer
npm install && npm run install:browsers
npm run test:e2e
npm run test:e2e -- tests/catalog/categories.spec.ts -g "create a category"
npm run typecheck                 # tsc --noEmit over the suite

test:e2e:headed, :ui, :debug and :report mirror the Playwright flags. Admin and Shop need a running, installed, seeded app; Installer needs an uninstalled one. troubleshooting.md has the environment contract.

Non-negotiables

• The suite shares one database and does not roll back. An E2E run leaves every record it creates. Assertions must survive that: scope to the row you created, never to a global count or a list position.
• A test owns its state. It creates what it needs, restores any global setting it changed, and deletes only what it made. Never operate on "the first row" or an arbitrary existing record because it was convenient — test-data.md.
• Tests are independently runnable, across suites too. workers: 1 and file order are not permission to depend on another test having run, and a Shop spec never assumes an Admin spec configured anything — admin-and-shop.md.
• Admin auth is cached to .state/admin-auth.json by the adminPage fixture. Do not log in by hand in a spec; the Shop promotion specs that call loginAsAdmin(page) are legacy — see authoring.md.
• Rebuild assets before running after any frontend change, or the browser loads the previous bundle and the failure will look like a test bug.
• No comments anywhere under tests/e2e-pw/ — no //, no /* */, no docblock. This is a Bagisto convention, held without exception across all three suites today, and it diverges from generic Playwright advice deliberately. Put the reason in a method name instead; authoring.md shows how.
• Generate test data inside beforeEach, never at module scope. One module-scope uniqueStamp() gives every test in the file the same SKU, and the file only passes while cleanup never misses.
• A test that depends on the clock is a timezone question first. The app runs on APP_TIMEZONE (Asia/Kolkata here) and the configs pin Node and the browser to the same zone. Anything involving today, tomorrow, slots, availability or expiry is investigated with time-and-timezone.md before anything else.
• "Flaky" is a conclusion, not a starting hypothesis. One shard failing, one database driver failing, or a local pass are not evidence of flakiness. Get the trace, the timestamps and the server log first — troubleshooting.md.
• Filenames are lower-kebab and a file that declares a class is named exactly for it — CatalogAclPage.ts, not catalog.ts. Describe and test titles are lower case with single spaces. No numeric ordering prefixes.

The spec names the intent and composes the workflow; the page object owns every locator and asserts through named expect… methods. A spec that contains a CSS selector belongs in a page object instead — authoring.md.

REQUIRED SUB-SKILL: Use bagisto-change-verification before calling any change done.