Proper handling of app contexts during tests

[torotil]

The documentation recommends to use a test-wide app context for convenience when many interactions with the DB are needed:

@pytest.mark.fixture
def app_ctx(app):
    with app.app_context():
        yield

According to a discussion in the flask issue queue this will create problems for test methods that also emit test requests to the flask app. In this case the test-wide app context and with it the scoped session is used both in the test method (and its fixtures) and inside the test requests. This means the state of the session is leaked between the test method and the test request which makes the tests behave subtly different than the live environment.

An example is:

@app.route("/test")
def test():
    page = model.Page.query.get(1)
    page.visit_count += 1
    # Forgot to commit here so nothing is saved to the database.
 
def test_visit_count(app_ctx):
    page = model.Page(visit_count=0)
    db.session.add(page)
    db.session.commit()
    client = app.test_client()
    client.get("/test")
    assert page.visit_count == 1

The recommendation of avoiding test-wide app contexts (flask) contradicts the recommendation to use one (flask-sqlalchemy).

• How would a good (aka DRY) way of working with models in tests / fixtures look that doesn’t need a test-wide app context?
• Does the flask-sqlalchemy documentation need updating?