UI Testing
Two pages to practise web UI testing: a form with elements to find with Selenium or Playwright, and a case study of a button that lost its styling after a stylesheet rebuild.
What a UI test checks
A UI test drives a real browser. It opens a page, finds elements, clicks and types, and then asks what the page shows: is the button visible, does the counter read 3, is the box still in the corner. The thing under test is the rendered document: the HTML after the scripts have run, the box model, and the CSS the browser actually applied, plus the events a user fires.
That is exactly the layer the other two levels cannot see. A unit test calls one function with a fixed input and never builds a document. An API test checks the contract (status code, headers and body) and stops before the renderer. Both are cheap and finish in milliseconds, so you want many of them. A UI test is slower and breaks when a button moves, and it is the only one that fails when a stylesheet is not served, when a cookie banner covers the form, or when a click lands on the wrong element.
The order that works: push the logic down to unit tests, the contract to API tests, and keep UI tests for the few paths a person really walks.
The tools
Two tools cover most browser work. Selenium is the older standard:
WebDriver talks to every major browser, and bindings exist for many
languages. Playwright ships its own browser builds, waits for an
element before acting on it, and keeps a trace of the run for the
failure report.
The locators are the same in both: id, class name, tag name, CSS
selector, link text. The practice page in this folder has one block per
locator type, plus the controls that catch people out — a form, radio
buttons, checkboxes, a drop-down, nested tables, and three blocks that
differ only by id, by class, or by text alone.
From a locator to a regression test
Four steps, in this order, cover most UI defects and keep the test small.
- Find the element by a stable attribute. An id, a name or a data-test attribute survives a redesign; link text works for links. A positional selector such as nth-child is the last resort: the next markup change moves it.
- Check the state. Is it visible, enabled, checked, selected, and does it show the expected text? Assert on the element you found, not on the page title: the title can be right while the form below it is broken.
- Check the CSS when the look is the point. Read the computed style of the element (width, position, background, border), not the class attribute. A class can be present in the markup even though its rule never reached the browser, and that is how a stylesheet regression hides.
- Write one regression test. Open the page, assert the state and the computed value, then revert the fix and watch the test fail. A test that passes on the broken page is checking the wrong property, and a test nobody saw fail is not evidence.