Bug#946422: silx: autopkgtest regression: pocl error

[PICCA Frederic-Emmanuel]

It seems that this test does not PASS

    @unittest.skipUnless(ocl, "PyOpenCl is missing")
    def testOpenCLMedFilt2d(self):
        """test cpp engine for medfilt2d"""
        res = medianfilter.medfilt2d(
            image=TestMedianFilterEngines.IMG,
            kernel_size=TestMedianFilterEngines.KERNEL,
            engine='opencl')
        self.assertTrue(numpy.array_equal(res, TestMedianFilterEngines.IMG))


testOpenCLMedFilt2d (silx.image.test.test_medianfilter.TestMedianFilterEngines)
test cpp engine for medfilt2d ... pocl error: lt_dlopen("(null)") or lt_dlsym() failed with 'can't close resident module'.
note: missing symbols in the kernel binary might be reported as 'file not found' errors.
Aborted
E: pybuild pybuild:341: test: plugin custom failed with: exit code=134: env PYTHONPATH=/home/picca/silx-0.11.0+dfsg/.pybuild/cpython3_3.8_silx/build WITH_QT_TEST=False xvfb-run -a --server-args="-screen 0 1024x768x24" python3.8 run_tests.py -vv --installed
dh_auto_test: pybuild --test -i python{version} -p "3.8 3.7" -s custom "--test-args=env PYTHONPATH={build_dir} WITH_QT_TEST=False xvfb-run -a --server-args=\"-screen 0 1024x768x24\" {interpreter} run_tests.py -vv --installed" returned exit code 13
make[1]: *** [debian/rules:70: override_dh_auto_test] Error 255
make[1]: Leaving directory '/home/picca/silx-0.11.0+dfsg'
make: *** [debian/rules:27: build] Error 2


the code of medfilt2d is there


def medfilt2d(image, kernel_size=3, engine='cpp'):
    """Apply a median filter on an image.

    This median filter is using a 'nearest' padding for values
    past the array edges. If you want more padding options or
    functionalities for the median filter (conditional filter 
    for example) please have a look at
    :mod:`silx.math.medianfilter`.

    :param numpy.ndarray image: the 2D array for which we want to apply
        the median filter.
    :param kernel_size: the dimension of the kernel.
        Kernel size must be odd.
        If a scalar is given, then it is used as the size in both dimension.
        Default: (3, 3)
    :type kernel_size: A int or a list of 2 int (kernel_height, kernel_width)
    :param engine: the type of implementation to use.
        Valid values are: 'cpp' (default) and 'opencl'

    :returns: the array with the median value for each pixel.

    .. note::  if the opencl implementation is requested but
        is not present or fails, the cpp implementation is called.

    """
    if engine not in MEDFILT_ENGINES:
        err = 'silx doesn\'t have an implementation for the requested engine: '
        err += '%s' % engine
        raise ValueError(err)

    if len(image.shape) is not 2:
        raise ValueError('medfilt2d deals with arrays of dimension 2 only')

    if engine == 'cpp':
        return medianfilter_cpp.medfilt(data=image,
                                        kernel_size=kernel_size,
                                        conditional=False)
    elif engine == 'opencl':
        if medfilt_opencl is None:
            wrn = 'opencl median filter not available. '
            wrn += 'Launching cpp implementation.'
            _logger.warning(wrn)
            # instead call the cpp implementation
            return medianfilter_cpp.medfilt(data=image,
                                            kernel_size=kernel_size,
                                            conditional=False)
        else:
            try:
                medianfilter = medfilt_opencl.MedianFilter2D(image.shape,
                                                             devicetype="gpu")
                res = medianfilter.medfilt2d(image, kernel_size)
            except(RuntimeError, MemoryError, ImportError):
                wrn = 'Exception occured in opencl median filter. '
                wrn += 'To get more information see debug log.'
                wrn += 'Launching cpp implementation.'
                _logger.warning(wrn)
                _logger.debug("median filter - openCL implementation issue.",
                              exc_info=True)
                # instead call the cpp implementation
                res = medianfilter_cpp.medfilt(data=image,
                                               kernel_size=kernel_size,
                                               conditional=False)

        return res

in our case we have engine = 'opencl' and no warning message, so medfil_opencl  should not be None.

it comes from here

from silx.opencl import medfilt as medfilt_opencl

In this code we have

        :param devicetype: type of device, can be "CPU", "GPU", "ACC" or "ALL"
 
So let's do a first test by replacing gpu by cpu to see if it change something during the test.