Test failure from installation page

[rmccole]

Expected Behavior

Tests should complete. Tests are run using command
py.test --ignore cwltool/schemas/ --pyarg cwltool

Actual Behavior

Test failed. Output:

`==================================================== FAILURES ====================================================
__________________________________________ TestOverride.test_overrides ___________________________________________

self = <cwltool.tests.test_override.TestOverride testMethod=test_overrides>

@pytest.mark.skipif(onWindows(),
                    reason="Instance of Cwltool is used, On windows that invoke a default docker Container")
def test_overrides(self):
    sio = StringIO()

    self.assertEquals(main([get_data('tests/override/echo.cwl'),
                            get_data('tests/override/echo-job.yml')],
                           stdout=sio), 0)

  self.assertEquals({"out": "zing hello1"}, json.loads(sio.getvalue()))

E AssertionError: {'out': 'zing hello1'} != {u'out': u'-n zing hello1\n'}
E - {'out': 'zing hello1'}
E + {u'out': u'-n zing hello1\n'}
E ? + + +++ ++

anaconda/envs/CWLtest/lib/python2.7/site-packages/cwltool/tests/test_override.py:26: AssertionError`

Workflow Code

`py.test --ignore cwltool/schemas/ --pyarg cwltool`

Full Traceback

Paste the full traceback in case there is an exception
Run the workflow with ``--debug`` flag for more verbose logging 

Your Environment

• cwltool version:
  cwltool 1.0.20180116213856
  Check using cwltool --version
  OS: Mac OS 10.12.6

[anton-khodak]

The problem is that the standard shell for executing commands from a string in cwltool is /bin/sh which does not support -n argument to echo on MacOS. We can either change the default shell to /bin/bash for MacOS specifically or for cwltool in general. Are there specific reasons to use /bin/sh by default, @tetron ?

[mr-c]

@anton-khodak Good catch. I'm guessing the assumption of the availability of -n was an oversight. The shell needs to be a POSIX shell, so the use of /bin/sh is correct.

[mr-c]

@anton-khodak I think it would be fine to drop the -n and amend the tests to expect the newline character