Bug#1033244: unblock: armci-mpi/0.3.1~beta-7
Drew Parsons
dparsons at debian.org
Mon Mar 20 16:24:52 GMT 2023
Package: release.debian.org
Severity: normal
User: release.debian.org at packages.debian.org
Usertags: unblock
X-Debbugs-Cc: armci-mpi at packages.debian.org
Control: affects -1 + src:armci-mpi
Please unblock package armci-mpi
[ Reason ]
armci-mpi 0.3.1~beta-7 makes some small changes to test management
that will be bookworm management more reliable. Source was
occasionally and randomly failing, evidently when only 1 cpu was
available (armci-mpi usually uses 2 processes for testing). s390x
fails debci tests with mpich.
armci-mpi/0.3.1~beta-7 restricts build-time testing to 1 process if
only 1 cpu is available. It switches off mpich tests on s390x in
debian/tests.
So tests should pass more reliably and s390x won't show a "false"
failure (the failure is real but is known. No point making bookworm
continue to fail on s390x. Better to let it just monitor openmpi
operations).
[ Impact ]
If not accepted, s390x will show debci failure, when in fact openmpi
tests should be passing (an openmpi failure should not be treated as
"not a regression").
Also binNMUs may occasionally fail (if only 1 cpu is provided for the
build)
[ Tests ]
debci tests are passing as normal
s390x now reports as passing (with openmpi, skipping mpich)
[ Risks ]
(Discussion of the risks involved. E.g. code is trivial or
complex, key package vs leaf package, alternatives available.)
[ Checklist ]
[x ] all changes are documented in the d/changelog
[x ] I reviewed all changes and I approve them
[x ] attach debdiff against the package in testing
[ Other info ]
Our default MPI is openmpi. The mpich build is provided to enable an
mpich build of nwchem that supports multinode execution (with the env
variable documented in nwchem/7.0.2-4).
armci-mpi might now migrate after 20 day testing. I've filed this
unblock request to complement the unblock request for nwchem/7.0.2-4,
since nwchem/7.0.2-4 was built against armci-mpi/0.3.1~beta-7
(it uses static libraries, no problem in practice but better for
bookworm to have the matching package versions)
unblock armci-mpi/0.3.1~beta-7
-------------- next part --------------
diff -Nru armci-mpi-0.3.1~beta/debian/changelog armci-mpi-0.3.1~beta/debian/changelog
--- armci-mpi-0.3.1~beta/debian/changelog 2022-03-07 13:07:13.000000000 +0100
+++ armci-mpi-0.3.1~beta/debian/changelog 2023-03-19 14:08:54.000000000 +0100
@@ -1,3 +1,12 @@
+armci-mpi (0.3.1~beta-7) unstable; urgency=medium
+
+ * Team upload.
+ * run build-time tests on only 1 process if only 1 CPU is available.
+ Closes: #1031064.
+ * debian/tests: don't run mpich tests on s390x. Closes: #1009772.
+
+ -- Drew Parsons <dparsons at debian.org> Sun, 19 Mar 2023 14:08:54 +0100
+
armci-mpi (0.3.1~beta-6) unstable; urgency=medium
* Team upload.
diff -Nru armci-mpi-0.3.1~beta/debian/rules armci-mpi-0.3.1~beta/debian/rules
--- armci-mpi-0.3.1~beta/debian/rules 2022-03-07 13:07:13.000000000 +0100
+++ armci-mpi-0.3.1~beta/debian/rules 2023-03-19 14:08:54.000000000 +0100
@@ -49,8 +49,9 @@
ifeq (,$(filter nocheck,$(DEB_BUILD_OPTIONS)))
override_dh_auto_test:
set -e; \
+ ncpu=`nproc`; if [ $${ncpu} -gt 1 ]; then ncpu=2; fi;\
for mpi_flavor in $(TEST_MPI_FLAVORS); do \
- $(MAKE) -C $(CURDIR)/build-$${mpi_flavor} -k check MPIEXEC="mpiexec.$${mpi_flavor} -np 2" ARMCI_USE_WIN_ALLOCATE=1 || cat $(CURDIR)/build-$${mpi_flavor}/test-suite.log; \
+ $(MAKE) -C $(CURDIR)/build-$${mpi_flavor} -k check MPIEXEC="mpiexec.$${mpi_flavor} -np $${ncpu}" ARMCI_USE_WIN_ALLOCATE=1 || cat $(CURDIR)/build-$${mpi_flavor}/test-suite.log; \
done
endif
diff -Nru armci-mpi-0.3.1~beta/debian/tests/control armci-mpi-0.3.1~beta/debian/tests/control
--- armci-mpi-0.3.1~beta/debian/tests/control 2022-03-07 13:07:13.000000000 +0100
+++ armci-mpi-0.3.1~beta/debian/tests/control 2023-03-19 14:08:54.000000000 +0100
@@ -15,3 +15,4 @@
debhelper, dh-autoreconf,
gcc
Restrictions: allow-stderr
+Architecture: !s390x
More information about the debian-science-maintainers
mailing list