Source: fenics-dolfinx Followup-For: Bug #1127800 nanobind 2.11.0 has introduced pathologies, and there appear to be more than one of them.
nanobind 2.11.0 failure on 32-bit arches is reported in Bug#1127672 The amd64 log in testing at https://ci.debian.net/packages/f/fenics-dolfinx/testing/amd64/68633180/ or unstable at https://ci.debian.net/packages/f/fenics-dolfinx/unstable/amd64/68655583/ suggests one more separate problem. There seems to be a pathology in the serial execution of MPI-parallel programs. test-dolfinx-python-unittest starts at 919s, and the first python unit tests (serial) reach test_vtkhdf.py at 2529s and finish at 2722s. So that's 30 minutes test time, consistent with what we had before (23 min) The failing MPI python unit tests started at 4380s, reached test_vtkhdf.py at 5314s (15 minutes, fine), then test_vtkhdf.py is killed at 10916s, 1.5 hours on the one unit test, not so fine. Then even worse, on my local machine, dolfinx does not finish importing at all in reasonable time with serial `python3 -c "import dolfinx"`. An MPI job runs "normally" even with 1 process, `mpirun -n 1 python3 -c "import dolfinx"`
