How to handle an optional package dependency in tests?

[rimey]

#1691 adds gcloud.monitoring for accessing the Google Monitoring API. The initial focus is on querying of metric data.

The primary way in which users will want to get metric data is as a pandas DataFrame, and the Query class offers an as_dataframe() method for this purpose.

However, pandas is a large package with dependencies on yet other packages. I would not recommend gcloud-python taking a dependency on it.

We have taken the approach of importing pandas dynamically in as_dataframe(). The documentation explains that use of this method requires that you have pandas installed. This works perfectly.

Tests are still an issue, however. The unit tests for as_dataframe() naturally require pandas, which is not currently present in the test environment.

How should we handle this?

[tseaver]

Maybe something like:

try:
     import pandas
except ImportError:
    HAVE_PANDAS = False
else:
    HAVE_PANDAS = True
from unittest2 import TestCase, skipUnless

class TestWithMaybePandas(TestCase):
    @skipUnless(HAVE_PANDAS, 'No pandas')
    def test_something_using_pandas(self):
          # whatever

Ideally you would add a tox environment (but maybe not one of the defaults) which added pandas to the dependencies, so that those tests could be exercised.

[rimey]

Done. PTAL.

What do we do about the test coverage analysis?

[dhermes]

We may just have you add a no-cover pragma.

[rimey]

I added a py27-pandas tox environment, and I added no-cover pragmas.

Thank you, @tseaver and @dhermes, for the advice.