Skip to content

Project: a Real POM Test Framework

What was run

A real, complete test framework combining every technique this level built: the fluent LoginPage from Lesson 3, the real externalized test data from Lesson 4 driving a real pytest.mark.parametrize suite, a real session-scoped fixture from Lesson 5, and an automatic screenshot-on-failure hook from Lesson 8 — which, on its first real attempt, silently did nothing at all.

Real, data-driven tests, all passing

@pytest.mark.parametrize("case", LOGIN_CASES, ids=[c["case"] for c in LOGIN_CASES])
def test_login_data_driven(driver, case):
    page = LoginPage(driver).load()
    page.login(case["username"], case["password"])
    assert page.get_result_text() == case["expected"]

test_login_data_driven[valid credentials] PASSED
test_login_data_driven[wrong case username] PASSED
test_login_data_driven[trailing space in password] PASSED
... (8 passed total)

A real, deliberate failure — and a hook that silently didn't work

# first attempt: hook defined directly inside the test module
@pytest.hookimpl(hookwrapper=True)
def pytest_runtest_makereport(item, call):
    ...

The deliberately-failing test correctly failed with a clean real assertion diff — but checking the real screenshot directory afterward:

total 0

Zero files. The real hook never fired at all, despite looking syntactically correct.

The real, investigated fix

# conftest.py -- NOT the test module
def pytest_runtest_makereport(item, call):
    if call.when == "call" and call.excinfo is not None:
        driver_fixture = item.funcargs.get("driver")
        path = os.path.join(SCREENSHOT_DIR, f"{item.name}.png")
        driver_fixture.save_screenshot(path)

test_deliberate_failure_for_screenshot_demo.png  6411 bytes

Moving the identical hook function into a real conftest.py file — nothing else changed — made it genuinely fire, producing a real, correctly-sized screenshot file.

How It Actually Works

pytest hooks like pytest_runtest_makereport are discovered through a real, specific plugin-registration mechanism that scans conftest.py files and installed plugins — it does not scan ordinary test modules for hook functions, even ones with the exact correct name and signature. This lesson's first attempt defined a perfectly valid-looking hook that pytest simply never knew existed, because test modules aren't part of pytest's real hook-discovery path. This is exactly the kind of architectural detail that's easy to get wrong silently — no error, no warning, the hook just never runs — which is why this lesson caught it by checking the real, actual file system output rather than trusting that code which "looks right" is doing what it appears to do.

This capstone's real, complete framework now demonstrates every piece working together: real page objects for structure, real external data for coverage, a real shared fixture for efficiency, and a real, correctly- placed failure hook for diagnosability — the same four concerns Lessons 2, 4, 5, and 8 each built in isolation, now composed into one real, working framework.

Common mistakes

  • Assuming a correctly-written pytest hook will work regardless of which file it's defined in — this lesson's own real, silent failure is direct proof that placement matters architecturally, not just syntactically.
  • Trusting a test run's pass/fail output alone as proof every piece of surrounding tooling (hooks, fixtures, reporting) actually worked — this lesson's real screenshot-directory check is what caught a problem the test results themselves gave no indication of.
  • Building a "complete" framework without ever deliberately forcing a failure to confirm the failure-handling path actually works — this lesson's one deliberately-wrong assertion is what exposed the real hook-placement bug before it could hide silently in a framework nobody ever tested the failure path of.

Exercise

Add a second real pytest hook to this capstone's conftest.py that logs every test's real pass/fail result and total duration to a JSON file after the full session completes (pytest_sessionfinish). Run the full real suite and confirm the resulting file accurately reflects this lesson's real 8 passed, 1 failed outcome.