On 19/12/2019 14.33, Graham Inggs wrote:
> On Thu, 19 Dec 2019 at 15:17, Andreas Beckmann <a...@debian.org> wrote:
>> I'm suspecting openmpi (that gets loaded by the io import) somehow messes up
>> some state,
>> causing the lt_*() failures.
>
> I wonder if this is related to #946986 ?
Not sure any more.
I "fixed" this in pocl by cherry-picking a few commits from 1.4 that get
rid of ltdl and switch to plain libdl.
The underlying issues seems to be in LTDL: if OpenMPI4 is loaded, the
lt_dlopen() result is inconsistent:
* ld_dlopen() returns a non-NULL lt_dlhandle (SUCCESS)
* ld_dlerror() returns a non-NULL error string (FAILURE)
So did the lt_dlopen() call succeed?
silx will now fail with
FAIL: test_medfilt (silx.opencl.test.test_medfilt.TestMedianFilter)
----------------------------------------------------------------------
Traceback (most recent call last):
File
"/usr/lib/python3/dist-packages/silx/opencl/test/test_medfilt.py", line
115, in test_medfilt
self.assertEqual(r.error, 0, 'Results are correct')
AssertionError: 3.4028235e+38 != 0 : Results are correct
Andreas
