Playwright, written down

Choosing a framework

• What Playwright does not do

  Every limit with its source, and not one of them is a compliment wearing a disadvantage's clothes.

• Playwright vs Selenium

  Not which is better — what breaks when you move a suite you already have, and the four places Selenium still wins.

• Playwright vs Cypress

  The differences are architectural, and each one has a price. Where Cypress is better, said plainly.

• Playwright vs Puppeteer

  A library against a test framework, which is not a fair fight and not the question — so here is the one worth asking.

• Playwright vs WebdriverIO

  One arrives whole, one is assembled from services and reporters. And one of them reaches a native app.

• TypeScript or Python?

  Same browser API in both. The test runner is Node-only, and that is the whole decision.

Getting started

• Playwright, after the quickstart

  The four decisions that are cheap on day one and a project to undo later — the hour the documentation has no reason to write.

• What Playwright is, and what it is not for

  Three engines, one API, four languages, and the four things it will not do.

• How Playwright actually works

  One connection instead of a request per command, and what that changes about the tests you write on top of it.

Writing tests

• Best practices, ranked

  Ordered by what they save you, with four marked contested and left unresolved.

• Why your Playwright test is flaky

  Four causes, how to tell which one you have, and the one fix that is not a fix.

• Which wait actually fired

  Seven timeouts, three defaults, and the nesting that makes one of them unreachable.

• What to assert, and which assertions wait

  The ones that poll and the ones that judge once — and why the second kind is how a suite acquires a race.

• Iframes and shadow DOM

  Two different boundaries that readers arrive confusing, and the one nothing gets into.

• Which locator to use

  The seven built-in locators ranked for a real application, and what to do when the recommended one does not exist.

• Staying logged in

  Four login patterns, and the expiry whose failure message never says the word session.

• What to mock, and what it costs

  A mocked test and a real one answer different questions — so the page says which is worth what.

• When to write a fixture

  Setup that survives a failing test body, and the teardown a discarded worker still runs.

• Do you need a page object model?

  Twelve arguments, each with a verdict attached — including the two weakest, one from each side.

• Visual regression, without the noise

  A baseline is keyed to the platform that made it. What belongs in the frame, and which tolerance option applies.

• Where test data comes from

  Three ways to get a record in front of a test, what each costs, and who deletes the row when the run dies halfway.

• API calls from a browser test

  The request fixture sends real HTTP from a test file. Where that beats a dedicated API tool, and where it stops.

• Uploads and downloads

  The drop zone with no input in it, and the export nobody opens — the cases setInputFiles does not cover.

• Accessibility checks in the suite

  What a scan decides on its own, what it leaves to a person, and why a green run is not a conformance claim.

• What the generator hands you

  Codegen, three test agents and a bundled MCP server in 1.62, and the regression a generated test stayed green through.

• Component testing in a real browser

  What a browser catches that jsdom cannot, what the gallery costs to keep, and where the line against end-to-end sits.

Running them at scale

• Workers, shards, and where each stops paying

  A scaling curve measured on one machine, and the arithmetic for the shard count from two numbers you time yourself.

• Playwright in GitHub Actions

  The docs’ workflow goes green on the first push. Week two is the bill: pull request against nightly, shards, and who reads a red job.

• Reading a failing CI run

  The build is red, the machine is gone, and the trace is the only witness left.

• Which reporter, and who reads it

  Four readers, one reporter array that serves all four, and what the HTML report does not render off the file system.

• Playwright in Docker

  The official image is usually the whole job, until its tag and the version in your project disagree.

• Playwright vs TestCafe

  TestCafe stopped defaulting to its proxy in 3.0, and the browser string still decides which mechanism the whole run gets.

• The Playwright MCP server

  Twenty-four tools out of the box, 68 with capabilities on, and a --help that names three of the eight.

• Playwright cheat sheet

  The locators, assertions and CLI flags worth keeping open in a tab, every one of them run against 1.62.1.

• Playwright interview questions

  Eight questions an interviewer can actually score, and what a good answer sounds like when the candidate has done the work.

• Mobile emulation

  A viewport, a user agent and a touch flag in a desktop engine. What that catches, and the things it cannot.

• The library or the runner

  Two packages, and a working `playwright test` command is no evidence you have the one you think you have.

• Onboarding a team

  The first weeks, the conventions worth fixing early, and the two checks that do not cover for each other.