[Q-e-developers] 'mpif.h' vs. 'include mpi' and regression tests

[Filippo SPIGA]

Dear Paul,

On Nov 22, 2016, at 1:10 PM, Paul Kapinos <kapinos at itc.rwth-aachen.de> wrote:
> II. When running 'make test-suite' in parallel (multiple terminals on *same node*, all working on an *own* copy of QE source in an *own* directory and having own $TMP) we see very strange failures/wrong results, which are not reproducible by subsequent runs,
> > All done. ERROR: only 13 out of 17 tests passed.
> 
> Surprize-surprize: when running the tests in different terminals one-by-one, all versions deliver almost same result (intel versions have a failure in pw_vdw - vdw-ts.in while GCC versions have none).
> 
> This lead us to assumption that different 'make test-suite' calls in different terminals on the same node *could have some data race* e.g. by overwriting some files (in /tmp? or in $HOME?).
> 
> This is not a big issue but a foul one (we build some 60 versions with different optimisation levels unless the idea came out that failures in regression tests it's not an QE issue but maybe one of the test suite itself). So we just kindly let you know about this.

test-suite is not design for run as you described. It can be made working in such way but it is not what it is design for. 

If you want to automate multiple builds and multiple runs try use JUBE (http://www.fz-juelich.de/ias/jsc/EN/Expertise/Support/Software/JUBE/_node.html). 

--
Filippo SPIGA ~ Quantum ESPRESSO Foundation ~ http://www.quantum-espresso.org