El 16/12/24 a las 19:24, Samuel Thibault escribió:
During a rebuild of all packages in unstable, your package failed to build:
[...]
mpirun.mpich -np 2 --bind-to none ./build-mpich/test/mpi/mpi_ping | grep "10000
16" || touch debian/failed
This program requires 2 MPI processes, aborting...
This program requires 2 MPI processes, aborting...
[...]
This is a really simple MPI smoketest without usage of eztrace, which
shows that mpich fails completely in this environment. I am thus
reassigning there.
Hi. Several openmpi packages required changes from
openmpi 4 to openmpi 5 related with the way oversubscribe is enabled.
I adjusted a few of them in debian-science, and now they build from source
even on machines with a single CPU.
Your package appears to enable oversubscribe, but it seems it's not working
as expected (the machine used to file this bug had 2 CPUs). In fact I
have a successful build of eztrace version 2.1.1-4 in trixie dated 2024-12-05
so the failure only seems to happen in unstable.
My bet is that maybe this is not a bug in mpich, but just that the
oversubscribe stuff in debian/rules is not working as expected and
just needs some fine-tuning. Unfortunately I don't know enough
about openmpi for this to be more than a bet.
btw: Does the package really uses both mpich and openmpi?
Thanks.
