On Fri, 2017-06-30 at 11:26 +0200, Emilio Pozuelo Monfort wrote: > Hi Drew, > > On 30/06/17 10:32, Gianfranco Costamagna wrote: > > Package: release.debian.org > > User: release.debian....@packages.debian.org > > Usertags: binnmu > > Severity: normal > > > > nmu petsc_3.7.5+dfsg1-4 . ANY . unstable . -m "Rebuild against > > openmpi 2.1.1" > > > > as said on irc, I don't know why that check is so strict, but > > better safe than sorry > > and lets the stack migrate > > Why does petsc need such a strict dependency on openmpi? If openmpi > breaks the > ABI, the SONAME and package name will change and hence a rebuild will > obviously > be necessary without this hack. But if it didn't break the ABI, do we > really > need these rebuilds? Can't we just drop the check in petsc, or will > something break?
It is a bit of a nuisance. It arises from the way PETSc handles MPI versions in /usr/lib/petsc/include/petscsys.h. In 3.7.6 they have a strict equality on the OMPI version dependency. I've contacted them about it and they've made a patch to relax it to a minor version dependency rather than a release (patch) version dependency, which should help us. That patch is held at https://bitbucket.org/petsc/petsc/commits/ca70f86ee9db8e69523e0e69f12289c6cab9b4cb?at=jed/mpi-semver but looks like it hasn't been merged to upstream master yet. We should contact upstream about it again. Drew
