tests/e2e/ uses Sections in headless Chrome the way students and staff do: it clicks the real
pages, served by the app with the React build, and checks what changes on the page and in the database.
The E2E workflow runs these tests on each push and pull request.
| File | Flows |
|---|---|
test_student.py |
A student sees their section, joins a discussion, switches labs, leaves a lab, and needs a section's enrollment code to join it |
test_staff.py |
A TA starts a session and takes attendance, which the student then sees in their history. A teacher adds a student to a section's roster by email and removes another |
test_admin.py |
An admin sets the staff page link that everyone sees in the nav bar, stops students joining discussions, syncs the course's students with bCourses, and merges a student imported under an old email into the same student under their new one. Students can't open the admin page |
test_offerings.py |
A TA imports a course from Canvas. Signing out, then signing in from a course page, returns to that page |
Everyone signs in through /dev_login/ against the mock Canvas, as Oski Bear
(teacher), Yu Long (TA) or Jimmy Xu (student).
The server, browser and seed data are in tests/browser.py, shared with the a11y audit. One
server and one Chrome serve the whole run. Before each test, the site fixture (tests/e2e/conftest.py)
seeds a fresh database and signs out, so each test starts from the same course: a lab, a discussion and a
tutoring section run by the TA, with the student in the lab.
site has helpers for finding things the way a person does, by their visible text:
site.visit(path, user), site.click("Join Section", within), site.card("Wheeler 150"), and
site.wait(condition). The database fixture opens the app's database, to set up what a test needs or
check what the page did.
After an action, wait for its result. The course pages are React, so a click's API call may only
start after the click returns, and the page updates after the response arrives. Wait for what the action
should do (site.wait(lambda: sections_of(database) == [...]), or for text to appear) instead of checking
right away. Checking right away passes or fails depending on timing.
They need the frontend build in server/static/ and Chrome, as the a11y audit does:
yarn build && cp -R build/. server/static/
pytest tests/e2eRebuild after changing the frontend. A failing test saves a screenshot of the page in e2e-screenshots/.
In CI these are uploaded as the e2e-screenshots artifact of a failed run.
Add a test for a flow people depend on, in the file for whose flow it is, or a new test_<subject>.py.
Use the seeded course where you can, and add what a test needs with database, as
test_student_switches_to_another_lab does. Unit tests (tests/unit/) are faster and the better place
for the server's rules and edge cases. The end to end tests check that the pages and the server work
together.