A test suite that had never passed, repaired and extended

The challenge

An industrial manufacturer had a browser test suite that had never been green, 0 of 103 passing, and no unit or integration tests at all. A red suite on every run had stopped telling anyone anything, so defects kept coming back.

The approach

1. Drove the application directly, outside the suite, to settle whether the application was broken or the test harness was. The application was healthy.
2. Fixed the harness itself: setup steps without a usable timeout, fixed waits longer than the test limit, unstable element ids, and nothing that started the app before the tests ran.
3. Audited whether a pass would mean anything. Tests that skipped when data was missing, or swallowed their own errors, were repaired or removed.
4. Added a unit and integration layer where none existed, on the code paths the browser tests miss or reach differently. Job, part and subtask validation alone has 129 unit tests behind it, and it runs on both the server and the screens, so a rule is covered whichever route a user took to save.
5. Kept browser tests for the paths that matter, aiming both layers at the areas with the most reported defects.
6. Broke the code on purpose under every new test, watched the test catch it, and restored the code.

The result

The browser suite was repaired from 0 of 103 passing to 111, and a unit and integration layer that did not exist was added: 458 tests. With all four pull requests merged that is 569 tests, 111 browser, 207 client and 251 server, 530 of them added. The work also surfaced production defects, each with its fix written in its own pull request.