Allow passing fixtures that pytest-factory's fixtures should depend on

[underyx]

I have the following fixtures:

@pytest.yield_fixture(scope='session')
def db_url():
    with testing.postgresql.Postgresql() as db:
        yield db.url()


@pytest.yield_fixture
def db(db_url):
    """Create a database session."""
    engine = sqlalchemy.create_engine(db_url)

    models.metadata.create_all(engine)

    factories.Session.configure(bind=engine)
    session = factories.Session()  # creates the thread local session used by factories
    yield session
    session.rollback()
    factories.Session.remove()  # disables the thread local session

    models.metadata.drop_all(engine)


pytest_factoryboy.register(factories.BookFactory)

My problem now is that trying to use the book_factory fixture on its own doesn't work; I need the db fixture to open a session the factory can use, first.

I'd like to be able to specify something like

pytest_factoryboy.register(factories.BookFactory, depends_on=['db'])

which would have the same effect as the manual implementation of

def book_factory(db):
    # …

would.

[bubenkoff]

can't you just use autouse fixture for that?
and then separate db related tests via package so then fixture will be used only for tests using it

[underyx]

@bubenkoff I was thinking of trying it but

1. I'm worried about the order, is it guaranteed to run first?
2. I'm not sure how viable the package separation is, we have a few beastly functions which access up to five databases.

[bubenkoff]

to be sure about the order you do need to explicitly declare that - i agree, so avoiding it is better
but autouse fixtures always go before the explicitly used ones

[olegpidsadnyi]

@underyx What kind of the base class are you using for your factories?
I'm using SQLAlchemy, so it is lazily bound to the same session maker that my test setup is using.

In what way your factory is storing the model? Basically what ORM are you using?

[bubenkoff]

@underyx this doesn't look like very big overhead

@pytest.yield_fixture(scope='session')
def db_url():
    with testing.postgresql.Postgresql() as db:
        yield db.url()

if db is session also scoped, then making it autouse is efficient

[underyx]

@underyx What kind of the base class are you using for your factories?

@olegpidsadnyi: factory.alchemy.SQLAlchemyModelFactory

I'm using SQLAlchemy as well — the factories.Session in the issue description's code is an SQLAlchemy scoped_session.

[bubenkoff]

@underyx doing models.metadata.create_all(engine) on every test if very inefficient
i would invest time in making tests isolated still, but sharing the same database during the whole session.
The way we do it is normally via randomization, using https://pypi.python.org/pypi/Faker
there's a plugin as well for it's integration pytest-faker

[underyx]

@bubenkoff well, alright, we have everything configured now with autouse=True and it works. But it's still not ideal that any test run will automatically create all databases even when I run tests that don't need any database at all. Any chance of this feature making it into pytest-factoryboy?

[olegpidsadnyi]

@underyx At this point I don't see it is coming. It would be yet another magical point of declaring a dependency besides a factory class, fixture registration (partial specialization), fixture propagation via conftest.py and finally the @pytest.mark.parametrize.
This would be just an invitation to start abusing it when there are decent ways to configure the test suite.