On 2024-05-25 18:31, Santiago Vila wrote:
El 25/5/24 a las 16:42, Drew Parsons escribió:
adios4dolfinx is building cleanly in reproducibility builds.
Perhaps the problem was a temporary glitch on your test system?
No, this is unlikely to be a temporary glitch:
...
My system has 2 CPUs, apparently MPI counts them as "one engine" and
fails
because the code has things like this:
ipp.Cluster(engines="mpi", n=2)
This bypasses whatever BUILD_OPTIONS=parallel=n setting the
user might set.
BTW: This bug looks similar to #1057556, which the maintainer
misdiagnosed as "fails with a single cpu" (not true!).
In such bug I proposed to run mpi with true and skip
the tests if it fails.
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1057556#34
Maybe a similar idea would work here as well.
We can try that. As you suggested in the other bug, it might be quirk of
openmpi we're facing.
openmpi distinguishs sockets, cores, slots, hwthread. On one system two
cores might be found in one socket, on another each core might have its
own socket.
To track that, what does `lscpu` report on your failing system?
(the thread,core,socket lines are probably the relevant ones)
