[pytest-dev] fixtures as context managers

[holger krekel]

On Fri, May 24, 2013 at 16:50 +0200, Harro van der Klauw wrote:
> As long as it throws an error hinting that you might need yielding=True it
> should be obvious on how to fix
> the backwards incompatibility issue as soon as you run your tests.

We cannot easily throw an error with a hint.  Consider this example:

    import pytest

    @pytest.fixture
    def fix():
        yield 1
        yield 2

    def test_fix(fix):
        for x in fix:
            assert x < 3

This runs fine on pytest-2.3.5.  On trunk it gives this error:

    ...

    fix = 1

        def test_fix(fix):
    >       for x in fix:
                assert x < 3
    E           TypeError: 'int' object is not iterable

I've never written or seen somebody writing such a generator fixture, though.
And what you would need to do is rewrite the fixture:

    @pytest.fixture
    def fix():
        def gen():
            yield 1 
            yield 2
        return gen()

Then again, when i first saw the contextlib.contextmanager decorator 
i found it not very intuitive.  Did anyone?  It looks like a hack.
>From that angle i'd rather go for requiring "contextyield=True" or
@pytest.contextfixture because that can be looked up in documentation
and thus is easier to read for people not familiar with yields/contextlib.
   
best,
holger
 
> I don't see a big problem with this, updating of a requirement is something
> that you should never do automatically.
> 
> Cheers,
> Harro
> 
> 
> 
> On 24 May 2013 16:36, Andreas Pelme <andreas at pelme.se> wrote:
> 
> > On Thursday 9 May 2013 at 15:56, holger krekel wrote:
> > > This is probably used by very few people but to be on the safe side,
> > > we probably should introduce a flag like this:
> > >
> > > @pytest.fixture(ctx=True) # signal this is a context manager style
> > fixture
> > > def fix():
> > > yield 1
> > >
> > > What do you think? Any other suggestions for the flag name?
> > >
> > > I'd rather not introduce something like @pytest.contextfixture
> > > because it would be a duplication of the API (scope, params).
> > > But i am open to be convinced otherwise.
> >
> > I agree that another API like contextfixture should be avoided.
> >
> > We had a short discussion on IRC about this, Holger had the idea of doing
> > the opposite if ctx=True - a new default argument "yielding=False" which
> > would make it possible to restore the old behavior by putting yielding=True
> > on fixtures that would be affected by this.
> >
> > A fixture that is a generator that currently looks like this
> >
> > @pytest.fixture
> > def fix():
> >     yield 1
> >     yield 2
> >
> > Would then have to be changed to
> >
> > @pytest.fixture(yielding=True)
> > def fix():
> >     yield 1
> >     yield 2
> >
> > This is backward incompatible, but given that it seems questionable if
> > "generator fixtures" useful/is used, with a note in the release notes and
> > documentation I think this could be a good compromise.
> >
> > I will be happy to give this a try if it is decided this could be a good
> > approach!
> >
> > Cheers
> > Andreas
> >
> >
> > _______________________________________________
> > Pytest-dev mailing list
> > Pytest-dev at python.org
> > http://mail.python.org/mailman/listinfo/pytest-dev
> >

> _______________________________________________
> Pytest-dev mailing list
> Pytest-dev at python.org
> http://mail.python.org/mailman/listinfo/pytest-dev