Travis CI fails intermittently due to file not found

[dohmatob]

It says, IOError: [Errno 2] No such file or directory: '/tmp/titi/func/3.hdr'

[KamalakerDadi]

I am not sure I get this issue. Can you a bit more explicit ?

For me file exists, /tmp/titi/func/3.hdr

[arthurmensch]

I think this is because one of the test function is badly written and writes twice the same data, which creeps travis out. I tried to address it in next PR

[mrahim]

PR Merged. Closing.
Thanks !

[dohmatob]

The issue still persists (moments, a go I've had to rebuild a PR, because is failed thus). Re-opening.

[lesteve]

I can reproduce locally but you have to clean the files in /tmp/titi (maybe by sending a grosminet):

rm -rf /tmp/titi && nosetests pypreprocess/tests/test_subject_data.py -m test_not_uniq

Doing that in a loop I get 10-20 failures for 100 attempts:

(for i in $(seq 1 100); do
    rm -rf /tmp/titi && nosetests pypreprocess/tests/test_subject_data.py -m test_not_uniq
done) 2>&1 | tee test.log

In general it is not great that the tests leave files around after finishing, also you should probably use tempfile.NamedTemporaryFile rather than hard-coded files in /tmp. I have no clue about the root cause of the failure, but maybe using different filenames with tempfile.NamedTemporaryFile would fix this issue ...

[lesteve]

I debugged this further, and there is a problem in _make_sd when func_filenames has nested lists. I lack domain knowledge to understand what this function does, but here is the code and look at the comment inside for the explanation of where it goes wrong:

def _make_sd(func_filenames=None, anat_filename=None, ext=".nii.gz",
             n_sessions=1, make_sess_dirs=False, func_ndim=4,
             unique_func_names=False, output_dir="/tmp/titi"):
    if not func_filenames is None:
        n_sessions = len(func_filenames)
    func = [create_random_image(ndim=func_ndim) for _ in range(n_sessions)]
    anat = create_random_image(ndim=3)
    if anat_filename is None:
        anat_filename = '%s/anat%s' % (DATA_DIR, ext)
    _save_img(anat, anat_filename)
    if not func_filenames is None:
        for sess_func, filename in zip(func, func_filenames):
            if isinstance(filename, _basestring):
                _save_img(sess_func, filename)
            else:
                vols = nibabel.four_to_three(sess_func)
                # The problem is that sess_func (a random 4d image)
                # can have 1 as the last dimension. If that happens
                # vols has one element only.  In the test
                # func_filename is a list of list of 2 elements
                # [['img1', 'img2'], ['img3', 'img4']] so filename has
                # 2 elements. That means that the zip on the next line
                # is only over 1 element and you only write one file
                # out of the two you are supposed to write, hence the
                # FileNotFoundError
                for x, y in zip(vols, filename):
                    assert isinstance(y, _basestring), type(y)
                    _save_img(x, y)
    else:
        func_filenames = []
        for sess in range(n_sessions):
            sess_dir = DATA_DIR if not make_sess_dirs else os.path.join(
                DATA_DIR, "session%i" % sess)
            if not os.path.exists(sess_dir):
                os.makedirs(sess_dir)
            func_filename = '%s/func%s%s' % (
                sess_dir, "_sess_%i_" % sess if (
                    n_sessions > 1 and unique_func_names) else "", ext)
            _save_img(func[sess], func_filename)
            func_filenames.append(func_filename)

    sd = SubjectData(anat=anat_filename,
                     func=func_filenames,
                     output_dir=output_dir)
    return sd

[thomasbazeille]

Update on this : @lesteve did the hardwork here and his last comment is right on point. But this bug is still happening randomly sometimes, I will redesign the concerned test to make it deterministic.