Some test fails on OpenSuse

[junghans]

From https://build.opensuse.org/request/show/634758

[  525s]   1/110 Test   #1: save_checkpoint .........................***Failed    3.00 sec
[  525s] Traceback (most recent call last):
[  525s]   File "/home/abuild/rpmbuild/BUILD/espresso-4.0.0/build/testsuite/save_checkpoint.py", line 39, in <module>
[  525s]     system.actors.add(lbf)
[  525s]   File "actors.pyx", line 186, in espressomd.actors.Actors.add
[  525s]   File "actors.pyx", line 48, in espressomd.actors.Actor._activate
[  525s]   File "lb.pyx", line 82, in espressomd.lb.LBFluid.validate_params
[  525s]   File "utils.pyx", line 334, in espressomd.utils.is_valid_type
[  525s] AttributeError: module 'numpy' has no attribute 'float128'
[  525s] -------------------------------------------------------
[  525s] Primary job  terminated normally, but 1 process returned
[  525s] a non-zero exit code.. Per user-direction, the job has been aborted.
[  525s] -------------------------------------------------------
[  525s] --------------------------------------------------------------------------
[  525s] mpiexec detected that one or more processes exited with non-zero status, thus causing
[  525s] the job to be terminated. The first process to do so was:
[  525s] 
[  525s]   Process name: [[1593,1],0]
[  525s]   Exit code:    1
[  525s] --------------------------------------------------------------------------
[  525s] 

Details here

[junghans]

@mkuron might be related to what you saw on OpenSuse 15

[mkuron]

This only happens on your 32-bit build. I think numpy just doesn't support 128-bit floats there. I have posted a patch.

Thanks for reporting all these issues. I saw that your builds for the exotic architectures (ppc64, ppc64le, s390x) are still failing. Please keep on reporting bugs and we'll try to fix them. Also, your SLE12 and openSUSE 42.3 builds should probably require openmpi instead of openmpi2 to be resolvable.

[junghans]

Once the initial rpm request went in, you can take it over an and fix all the builds ;-)

[junghans]

Oh and it seems strange, that the i686 on Fedora doesn’t have that problem!

[junghans]

opensuse_log.txt
There is something else wrong on openSUSE 42.3

[mkuron]

error: no matching function for call to 'boost::mpi::environment::environment()'
    mpi::environment mpi_env;

That comes from here:

https://github.com/boostorg/mpi/blob/e5fe21df2f42dda0ce8674a935d44e99739c4a19/include/boost/mpi/environment.hpp#L87-L101

https://github.com/boostorg/mpi/blob/56c2a212e5d9c43dab7ae873a0a8c2c7897327bd/include/boost/mpi/config.hpp#L37-L57

MPI_VERSION is unconditionally set to 3 in mpi.h, so it seems like the MPI headers are broken in your build container. It works fine in our Docker images and on our 42.3 machines. I have added a patch to #2255 though as it's inconsistent that this is the only test where we don't pass argc, argv to boost::mpi::environment.

[junghans]

It is not my build container, but the one OpenSUSE Build Service is using ;-)

[mkuron]

I know. I actually found the cause of the problem though: in Boost 1.54, it explicitly checks for MPI_VERSION == 2, which was not fixed until Boost 1.56. Can you apply #2255 and see if that fixes it?

[junghans]

I am playing around with it here: https://build.opensuse.org/package/show/home:cjunghans:branches:devel:languages:python/python-espressomd

[junghans]

Leap 15.0 and 42.3 and Tumbleweed seems to be ok now with #2255. I will post the build issues on ppc64le in a bit!