Skip to content

Playwright Testing

Playwright is a modern browser automation framework by Microsoft. Key advantage over Selenium: auto-waiting for elements, built-in assertions with retry, trace viewer for debugging, and codegen for recording tests. Supports Chromium, Firefox, and WebKit from a single API.

Key Facts

  • Auto-wait: every action waits for element to be actionable (visible, stable, enabled)
  • Built-in assertions: expect(locator).to_have_text("...") retries automatically with timeout
  • Codegen: playwright codegen URL records actions as test code
  • Trace viewer: records screenshots, DOM snapshots, network, console per action
  • Supports Chromium, Firefox, WebKit; headless by default
  • Multiple browser contexts = parallel independent sessions (different users, cookies)

Patterns

Pytest Integration

@pytest.fixture(scope="session")
def browser():
    with sync_playwright() as p:
        browser = p.chromium.launch()
        yield browser
        browser.close()

@pytest.fixture
def page(browser):
    context = browser.new_context()
    page = context.new_page()
    yield page
    context.close()

Locators (Prefer Over CSS/XPath)

# Role-based (most stable, accessibility-first)
page.get_by_role("button", name="Submit")
page.get_by_role("textbox", name="Email")

# Text and label
page.get_by_text("Welcome")
page.get_by_label("Password")
page.get_by_placeholder("Enter email")

# Test ID
page.get_by_test_id("submit-btn")

# CSS/XPath still available
page.locator("css=#username")

Trace Viewer

context.tracing.start(screenshots=True, snapshots=True, sources=True)
# ... test actions ...
context.tracing.stop(path="trace.zip")
# View: playwright show-trace trace.zip

Multiple Contexts (Parallel Users)

admin_ctx = browser.new_context()
user_ctx = browser.new_context()
admin_page = admin_ctx.new_page()
user_page = user_ctx.new_page()
# Independent sessions (separate cookies, storage)

Gotchas

  • Auto-wait is per-action, not per-assertion - use expect() for assertions (retries automatically)
  • page.wait_for_timeout(ms) defeats the purpose - use expect() with custom timeout
  • Codegen output is a starting point - refactor to use role/label locators and POM
  • Browser contexts are isolated - state doesn't persist between contexts

See Also