Refactor tests

[sandello]

Hi there.

I am looking towards using this library so I decided to look through and (possibly) adapt to my needs. As the first step I decided to run tests (on Mac OS X) and I have encountered some problems (Kafka was not starting, unittest module is not so convenient, etc).

So I have refactored tests a little bit. :)

Here is the list of major changes:

• Configured tests to spawn a Zookeeper instance. I am not a Java developer, so I don't have Zookeeper running locally. :)
• Extracted fixture code to the separate file, just not to mix test logic and fixture logic.
• Updated CLASSPATH settings to reflect kafka-run-class.sh from Kafka distribution.
• Configured tox and py.test to work with all that stuff.
• Marked unit tests as "xfail", just to silent them.
• Added has_gzip() / has_snappy() methods to not to fail on Snappy tests.
• Did some PEP8 styling by the way.

[mumrah]

@sandello, this is a pretty big patch so it will take me some time to review :)

[sandello]

No worries.

BTW, why 'kafka-python', and not 'python-kafka'?

[mumrah]

There is another python client called pykafka, and I didn't want to cause any confusion.

[mumrah]

After a quick glance through the commits a few thoughts:

• Overall, I like the changes. The tests are one area I know needs improvement, so a lot of this is much needed
• I don't like introducing tox as a dependency for testing. It's fine to include a tox descriptor that works with our tests, but I always want to be able to run the tests with just "python"
• The classpath in kafka-run-class.sh is pretty messed up. As you saw, it has dependencies on local ivy cache, and it also includes a bunch of stuff that's not really needed. After the 0.8 release, it will hopefully improve
• +1 for ZK fixture
• +1 for splitting out fixtures from tests

More detailed review to follow

[sandello]

On (2): Tox is not a dependency, actually. It is just a convenient way to run tests under py26 and py27.
I personally run tests as py.test, because it provides nice filtering facilities and also separates stdout/stderr for different tests. After I am done running the test I do a "pre-push" run with tox to be sure that py26 also works. That's all.

Still, nothing prevents you from doing just python ./test/test_integration.py. It is more a matter of how you write tests (e.g. you don't use pytest magic fixtures and stick with plain ol' unittest library).

On (3): Yeah... I can think only of invoking kafka-run-class.sh directly to not to mess with ivy cache directly.

[mumrah]

1. It seems to be a dependency in that you've integrated it into setup.py as the default test runner
2. The only other option would be to check-in the Kafka libs into source, which is not appealing. I think your patch is a good solution until Kafka sorts out their build

[sandello]

1. Yeah, but this is a development dependency, right? Not a runtime one.

[mumrah]

Sure, but what I'm saying is, I don't want it as a development dependency :)

[sandello]

Okay, I don't mind about dropping it. I am not quite familiar on how to use python setup.py test when you don't use neither py.test nor tox. Shall I just drop the extra code and write a few lines in README saying that we are tox & py.test-friendly?