Default parallel choice -- to be or to remove

As long as SISL_NPROCS>1 the tests sometimes fails, and quite often with OMP_NUM_THREADS>1 and threaded libraries it will just simply stall for ever.

We need a more careful test of this to assert whether this is because of pytest instantiation (I have a feeling that the combination of pytest + SISL_NPROCS>1 + OMP_NUM_THREADS>1

In any case I think it is vital we don’t ship out a code that actually does not perform.

@pfebrer you also saw some of this (can’t seem to find the thread). I think we need to have a discussion here, otherwise I will simply put the default SISL_NPROCS == 1 to at least have something.

As it stands I am a bit inclined to try and default to not using multiple processors, but rely on threads for performance through BLAS. In any case many of the demanding things are based on BLAS+LAPACK calls and so adding the pool will only make it worse… 😦

[zerothi]

I think it is this one: viz/tests/test_plot.py::TestMultiplePlot::test_object_sharing, would that make sense?

I can make everything run fine with export SISL_NPROCS_VIZ=1

[zerothi]

For now everything is defaulted to 1 core.