espresso-v4.1.1: MpiCallbacks_test and matrix_vector_product fail on i586 (OpenSuse Tumbleweed)

[junghans]

64-bit works, but i586 fails with:

[ 1083s]  6/68 Test  #6: MpiCallbacks_test ................***Failed    2.26 sec
[ 1083s] Running 8 test cases...
[ 1083s] Running 8 test cases...
[ 1083s] /home/abuild/rpmbuild/BUILD/espresso/src/core/unit_tests/MpiCallbacks_test.cpp(55): �[1;31;49merror: in "invoke_test": check f(i, d) == (invoke<decltype(f), int, double>(f, ia)) has failed [126.14149999999999 != 126.14149999999999]�[0;39;49m
[ 1083s] /home/abuild/rpmbuild/BUILD/espresso/src/core/unit_tests/MpiCallbacks_test.cpp(55): �[1;31;49merror: in "invoke_test": check f(i, d) == (invoke<decltype(f), int, double>(f, ia)) has failed [126.14149999999999 != 126.14149999999999]�[0;39;49m
[ 1083s] 
[ 1083s] �[1;31;49m*** 1 failure is detected in the test module "MpiCallbacks test"
[ 1083s] �[0;39;49m
[ 1083s] �[1;31;49m*** 1 failure is detected in the test module "MpiCallbacks test"
[ 1083s] �[0;39;49m-------------------------------------------------------
[ 1083s] Primary job  terminated normally, but 1 process returned
[ 1083s] a non-zero exit code.. Per user-direction, the job has been aborted.
[ 1083s] -------------------------------------------------------
[ 1083s] --------------------------------------------------------------------------
[ 1083s] mpiexec detected that one or more processes exited with non-zero status, thus causing
[ 1083s] the job to be terminated. The first process to do so was:
[ 1083s] 
[ 1083s]   Process name: [[38606,1],1]
[ 1083s]   Exit code:    201
[ 1083s] --------------------------------------------------------------------------
[ 1083s] 

and

[ 1085s] 48/68 Test #48: matrix_vector_product ............***Failed    0.01 sec
[ 1085s] Running 1 test case...
[ 1085s] /home/abuild/rpmbuild/BUILD/espresso/src/utils/tests/matrix_vector_product.cpp(34): �[1;31;49merror: in "inner_product": check result[i] == boost::inner_product(matrix[i], vector, 0.0) has failed�[0;39;49m
[ 1085s] /home/abuild/rpmbuild/BUILD/espresso/src/utils/tests/matrix_vector_product.cpp(34): �[1;31;49merror: in "inner_product": check result[i] == boost::inner_product(matrix[i], vector, 0.0) has failed�[0;39;49m
[ 1085s] /home/abuild/rpmbuild/BUILD/espresso/src/utils/tests/matrix_vector_product.cpp(34): �[1;31;49merror: in "inner_product": check result[i] == boost::inner_product(matrix[i], vector, 0.0) has failed�[0;39;49m
[ 1085s] 
[ 1085s] �[1;31;49m*** 3 failures are detected in the test module "matrix_vector_product test"
[ 1085s] �[0;39;49m
[ 1085s]

Details here

I didn't package v4.1 for OpenSuse, so I am not sure if was introduced in v4.1 or v4.1.1.

[junghans]

And:

[ 1703s] 106/143 Test  #84: collision_detection ...............................***Failed   15.09 sec
[ 1703s] .ERROR: Particle 1 moved more than one local box length in one timestep.
[ 1703s] E......
[ 1703s] ======================================================================
[ 1703s] ERROR: test_bind_at_point_of_collision (__main__.CollisionDetection)
[ 1703s] ----------------------------------------------------------------------
[ 1703s] Traceback (most recent call last):
[ 1703s]   File "/home/abuild/rpmbuild/BUILD/espresso/build/testsuite/python/collision_detection.py", line 225, in test_bind_at_point_of_collision
[ 1703s]     self.run_test_bind_at_point_of_collision_for_pos(np.array((0, 0, 0)))
[ 1703s]   File "/home/abuild/rpmbuild/BUILD/espresso/build/testsuite/python/collision_detection.py", line 140, in run_test_bind_at_point_of_collision_for_pos
[ 1703s]     self.s.integrator.run(3000)
[ 1703s]   File "integrate.pyx", line 104, in espressomd.integrate.Integrator.run
[ 1703s]   File "utils.pyx", line 264, in espressomd.utils.handle_errors
[ 1703s] Exception: Encountered errors during integrate: ERROR: Particle 1 moved more than one local box length in one timestep.
[ 1703s] 
[ 1703s] ----------------------------------------------------------------------
[ 1703s] Ran 8 tests in 14.228s
[ 1703s] 
[ 1703s] FAILED (errors=1)
[ 1703s] -------------------------------------------------------
[ 1703s] Primary job  terminated normally, but 1 process returned
[ 1703s] a non-zero exit code.. Per user-direction, the job has been aborted.
[ 1703s] -------------------------------------------------------
[ 1703s] --------------------------------------------------------------------------
[ 1703s] mpiexec detected that one or more processes exited with non-zero status, thus causing
[ 1703s] the job to be terminated. The first process to do so was:
[ 1703s] 
[ 1703s]   Process name: [[2638,1],0]
[ 1703s]   Exit code:    1
[ 1703s] --------------------------------------------------------------------------
[ 1703s] 

[jngrad]

Build log for python3-espressomd/openSUSE_Tumbleweed/i586:

[  997s]       Start 16: field_coupling_couplings
[  997s] 16/66 Test #16: field_coupling_couplings .........   Passed    0.01 sec
[  997s]       Start 17: field_coupling_fields
[  997s] 17/66 Test #17: field_coupling_fields ............***Failed    0.01 sec
[  997s] Running 7 test cases...
[  997s] /home/abuild/rpmbuild/BUILD/espresso/src/core/unit_tests/field_coupling_fields_test.cpp(397): [1;31;49merror: in "interpolated_vector_field": absolute value of (interpolated_value[0] - field_value[0]).norm(){5.7034162359677394e-16} exceeds 2.2204460492503131e-16[0;39;49m
[  997s] 
[  997s] [1;31;49m*** 1 failure is detected in the test module "AutoParameter test"
[  997s] [0;39;49m
[  997s] 
[  997s]       Start 18: field_coupling_force_field
[  997s] 18/66 Test #18: field_coupling_force_field .......   Passed    0.01 sec

The machine precision tolerance of another unit test in the same file was increased last year (213101a) due to a similar issue on i386. I'll increase the test tolerance.

[jngrad]

• MpiCallbacks_test: fixed by core: Fixed bitwise comparison of floating point numbers #3327
• matrix_vector_product: fixed by Fix Boost tests #3427
• field_coupling_fields: fixed by Fix breaking tests on 4.1.1 #3358
• collision_detection: still failing in 4.1.2, @fweik did we decide to do something?

[junghans]

Fix on its way into OpenSUSE: https://build.opensuse.org/request/show/766150

[jngrad]

Since our CI infrastructure doesn't have i586, we cannot reproduce the collision_detection failure on our side, and the error message is not sufficient to determine the cause, so we will not investigate it further. The test has been disabled for i586 (openSUSE:Factory/python3-espressomd/python3-espressomd.spec at line 102) and the package now builds on i586. I'm closing this issue.