An experiment in using context-managers in unittest fixtures.
with_fixture is an experiment in using context-managers naturally with Python unittest fixtures.
Test fixtures and context-managers have very similar structures. In both cases, they define bits of code to be run before and after other bits of a priori unknown code. They are the bread in little code sandwiches.
Yet, as far as I know, there is no natural way to use them together. That is, there is no way to use context-managers in - or perhaps as - test fixtures with the standard unittest module. This is doubly egregious because context-managers are a central element in modern Python design, and we should be encouraging their use widely.
with_fixture, then, is an experiment in making them work together. I want to start to explore the options we have for using context-managers in test fixtures, with the long-term goal of getting something into the standard library.
Experiment 1: with_fixture.TestCase.withFixture()
with_fixture currently implements a single, simple approach to combining context-managers and fixtures. It’s a kind of minimum viable product. The idea is to subclass unittest.TestCase and override the setUp() and tearDown() methods to drive a new generator method, withFixture().
withFixture() must call yield once. Everything before the yield is the equivalent of setUp() in unittest.TestCase, and everything after the yield is equivalent to tearDown(). By placing the yield in a with-block, it becomes very natural to use context-managers in test fixtures.
Example: Temporary file
Here we use the tempfile.TemporaryFile context-manager in withFixture():
class TestBindingToMembers(TestCase): TEST_DATA = b'1234567890' def withFixture(self): # This creates a tempfile and binds it to self.f with tempfile.TemporaryFile() as self.f: # This yield sends control to the test yield # The test has now been executed. Check for the data we # expect in the file. self.f.seek(0) assert(self.f.read() == self.TEST_DATA) def test_nothing(self): self.f.write(self.TEST_DATA)
This is clearly a bit contrived (it’s taken from the somewhat awkward test suite for with_fixture itself), but you can see what’s going on.
The current implementation is small and simple, and there’s plenty of room for further experimentation. Areas to investigate include:
- Allowing users to use setUp() and tearDown() along with withFixture(). Currently we make that difficult. There are a number of interesting problems that arise when you start to think about this.
- Integrating an implementation directly into unittest.
- Consider alternative names. I thing withFixture is pretty good and communicates the design intent, but others may find it confusing.
- Consider equivalent support for setUp/tearDownClass().
- Do some deeper thinking on exceptions. I think the currently implementation is pretty sound WRT exceptions (i.e. that setup and teardown will get executed correctly in the face of exceptions) but I’m not testing it and haven’t given it great throught yet.
- Completely different designs. I can imagine completely different ways to approach this, and I’m sure others can too.