Closed yl3dy closed 1 year ago
I'm also seeing these errors on x86-64-linux, aarch64-linux and aarch64-darwin.
Tests fail only when MPI is enabled, however it seems that the all the test scenarios require it or other specific features like CUDA (however, I'm not proficient in the test system used in PETSc). Building with either OpenMPI or MPICH produces the same errors, as does disabling p4est and BLAS/LAPACK.
I've got a strong feeling that the issue is with stale expected test results. I've ran a FEM solver utilizing PETSc 3.19.1 with doCheck = false;
, and it gave the same results as with an older version 3.14.3 (however, it only utilized CG linear solver and hypre integration). Also, deal.II's tests for PETSc integration pass with the newest version.
Link to the upstream issue: https://gitlab.com/petsc/petsc/-/issues/1386
Seems that the PETSc's test suite fails because of messages by MPI implementation complaining about disabled networking.
The current version of PETSc (3.17.4) in master and stable is broken due to make flags generation failure: for some reason
MAKEFLAGS
containsw
as target (see hydra), no actual target with this name exists. It's fixed when building the most recent version (3.19.1), however the quick tests fail. The former behavior is observed for versions prior to 3.18.6. The quick tests also fail when building 3.19.1 with earlier Nixpkgs (specificallyf096b7122ab08e93c8b052c92461ca71b80c0cc8
).make check
logs:References:
Upstream issue: https://gitlab.com/petsc/petsc/-/issues/1386
@cburstedde