Closed bangerth closed 2 months ago
@bangerth ugh That's a good point.
The bug is actually that OfflineData
should not use types::global_dof_index
at all because the boundary_map
and coupling_boundary_pairs
is in local (MPI rank) indexing. And we use unsigned int
for such local indices everywhere; see https://www.dealii.org/current/doxygen/deal.II/classUtilities_1_1MPI_1_1Partitioner.html#a11316c010d3d51912cdfe1939651ad72
I really need to add a 64 bit indices test for ryujin. Or, better, we should switch deal.II to always use 64bit (signed) ordinals...
The issue is that
riemann_solver.compute
takes as fourth argument the address to anunsigned int
, but we havewhere
i
ends up astypes::global_dof_index
.This patch is not elegant but will help: https://github.com/jerett-cc/ryujin/pull/32
@jerett-cc FYI