Bug#996204: transition: numerical library stack
On 2021-10-22 14:35, Sebastian Ramacher wrote:
On 2021-10-12 13:09:02, Drew Parsons wrote:
...
I'd like to proceed with a transition of the numerical library stack.
...
(along with other fenics components). There is currently some problem
with fenics-dolfinx 1:0.3.0-4 on 32-bit arches i386, armel, armhf.
I've got dolfinx passing (or skipping) its own tests on 32-bit.
There's a problem with 32-bit MPI though. i386 and armhf are failing to
run pytest over python unit tests in MPI
(fenics-dolfinx/debian/tests/test-dolfinx-python-unittests). The error
is reproducible from the command line on an i386 porterbox,
$ cd fenics-dolfinx-0.3.0/python/test/unit
$ mpirun -n 2 pytest-3 -v
...
rootdir: /home/dparsons/fenics/fenics-dolfinx-0.3.0/python,
configfile: setup.cfg
plugins: mpi-0+unknown
collecting 0 items / 1 error
--------------------------------------------------------------------------
"MPI_ABORT was invoked on rank 0 in communicator MPI_COMM_WORLD with
errorcode 59."
I haven't been able to coax any more detail out of it than that.
Might be something more insidious going on with 32-bit python and MPI.
mpi4py is also failing i386 tests, see
https://ci.debian.net/packages/m/mpi4py/testing/i386/
https://github.com/mpi4py/mpi4py/issues/105
On the other hand fenics-dolfinx demo_elasticity.py is passing fine in
MPI (though I did have to disable the helmholtz, condensation and
poisson demos).
pytest-mpi is passing its tests on i386 and armhf.
What's best for us? Just skip the fenics-dolfinx python units tests in
MPI on 32-bit machines?
Or configure the migration tool to ignore these i386/armhf test
failures, so that we can keep monitoring the tests in debci?
Drew
Reply to: