python nose --with-doctest adding extra test

Question

I am needing to defer the import of some libraries unless they are needed so I have created have an object to defer the import.

class Importer(object):
    def __init__(self, name):
        self.name = name

    def __bool__(self):
        try:
            mod = importlib.import_module(self.name)
            return True
        except ImportError:
            return False

    def __getattr__(self, name):
        try:
            mod = importlib.import_module(self.name)
            attr = getattr(mod, name)
            setattr(self, name, attr)
            return attr
        except ImportError as e:
            msg = self.name + " must be installed"
            raise ImportError(msg)

I have run nose with my tests and everything passes. However, when I run --with-doctest it runs an extra test (that I have not defined) that generates an error in any environment where a given library isn't installed.

Testing the code below recreates the issue I am experiencing. In an environment without numpy, nosetests --with-doctest is running a second test which I have not defined.

numpy = Importer('numpy')

def mean(array):
    """
    Mean of an array.

    Examples
    --------
    >>> mean([1, 2, 3])
    2.0
    >>> mean([1, -1])
    0.0
    """
    if numpy:
        return numpy.mean(array)
    return float(sum(array) / len(array))

Clearly, there is only one doctest available to test. Also if numpy is installed only 1 test is run. So why then when numpy is not installed am I getting a second test? Here is the output.

.E
======================================================================
ERROR: Failure: ImportError (numpy must be installed)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/devin/Dropbox/Devin_S_+_Spark_Wave/nimble/zissue/issue.py", line 16, in __getattr__
    mod = importlib.import_module(self.name)
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 994, in _gcd_import
  File "<frozen importlib._bootstrap>", line 971, in _find_and_load
  File "<frozen importlib._bootstrap>", line 953, in _find_and_load_unlocked
ModuleNotFoundError: No module named 'numpy'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/site-packages/nose/failure.py", line 39, in runTest
    raise self.exc_val.with_traceback(self.tb)
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/site-packages/nose/plugins/manager.py", line 154, in generate
    for r in result:
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/site-packages/nose/plugins/doctests.py", line 228, in loadTestsFromModule
    tests = self.finder.find(module)
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/doctest.py", line 933, in find
    self._find(tests, obj, name, module, source_lines, globs, {})
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/doctest.py", line 992, in _find
    if ((inspect.isroutine(inspect.unwrap(val))
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/inspect.py", line 512, in unwrap
    while _is_wrapper(func):
  File "/Users/devin/anaconda3/envs/issueEnv/lib/python3.6/inspect.py", line 503, in _is_wrapper
    return hasattr(f, '__wrapped__')
  File "/Users/devin/Dropbox/testing/issue/issue.py", line 22, in __getattr__
    raise ImportError(msg)
ImportError: numpy must be installed

----------------------------------------------------------------------
Ran 2 tests in 0.003s

Answer 1

Well, this is fun!

The core difficulty is that the doctest module "tricked" you into attempting a numpy import. That is, it probed for numpy.__wrapped__, and you took the bait. You need to distinguish between a probe call, for which returning a dummy value is acceptable, and a real call, for which you must import or raise.

Perhaps a heuristic of testing for __wrapped__ suffices. But you may find it necessary to inspect the call stack to find the caller, notice it is e.g. autodoc, and change behavior based on that.