https://bugzilla.redhat.com/show_bug.cgi?id=1474958 --- Comment #15 from Christoph Junghans <junghans@xxxxxxxxx> --- (In reply to Antonio Trande from comment #14) > (In reply to Christoph Junghans from comment #13) > > I saw that before for other mpi-enabled packages, if I recall correctly mock > > has no support for mpi and hence doesn't do a "module load > > mpi/openmpi-x86_64", which would prepend /usr/lib64/openmpi/lib to > > LD_LIBRARY_PATH (see module show "mpi/openmpi-x86_64") and fix the above > > issue. rpath aren't allowed for mpi packages either, so I am not sure how > > you want to fix. > > > > And setting CXX and friends won't change anything as mpicxx is internally > > just adding the same include paths and libraries to the flags: > > $ mpicxx --showme > > g++ -I/usr/include/openmpi-x86_64 -pthread -Wl,-rpath > > -Wl,/usr/lib64/openmpi/lib -Wl,--enable-new-dtags -L/usr/lib64/openmpi/lib > > -lmpi_cxx -lmpi > > All MPI files are compiled with MPI compilers as required by the guidelines; https://fedoraproject.org/wiki/Packaging:MPI doesn't list CC=mpicc as a must item and it would also be a really bad idea in general, e.g. for codes, which have mixed serial and parallel code. (Which isn't the case for LAMMPS though). > Sundials MPI libraries for example do not need to load any module in Mock: > > $ mock --no-clean -r fedora-rawhide-x86_64 --shell 'ldd > /usr/lib64/openmpi/lib/libsundials_nvecparallel.so.2' > Start: shell > linux-vdso.so.1 (0x00007fff9e309000) > libm.so.6 => /lib64/libm.so.6 (0x00007fa6b2969000) > libpthread.so.0 => /lib64/libpthread.so.0 (0x00007fa6b274a000) > libgomp.so.1 => /lib64/libgomp.so.1 (0x00007fa6b251b000) > libklu.so.1 => /lib64/libklu.so.1 (0x00007fa6b22e6000) > libHYPRE.so.0.0 => /usr/lib64/openmpi/lib/libHYPRE.so.0.0 > (0x00007fa6b1c62000) > libmpi.so.20 => /usr/lib64/openmpi/lib/libmpi.so.20 (0x00007fa6b195f000) > libc.so.6 => /lib64/libc.so.6 (0x00007fa6b157a000) > /lib64/ld-linux-x86-64.so.2 (0x0000558ae39ba000) > libdl.so.2 => /lib64/libdl.so.2 (0x00007fa6b1376000) > libamd.so.2 => /lib64/libamd.so.2 (0x00007fa6b116a000) > libcolamd.so.2 => /lib64/libcolamd.so.2 (0x00007fa6b0f62000) > libbtf.so.1 => /lib64/libbtf.so.1 (0x00007fa6b0d5e000) > libsuitesparseconfig.so.4 => /lib64/libsuitesparseconfig.so.4 > (0x00007fa6b0b59000) > libopenblas.so.0 => /lib64/libopenblas.so.0 (0x00007fa6ae77a000) > libstdc++.so.6 => /lib64/libstdc++.so.6 (0x00007fa6ae3f2000) > libsuperlu.so.5.1 => /lib64/libsuperlu.so.5.1 (0x00007fa6ae17a000) > libmpi_cxx.so.20 => /usr/lib64/openmpi/lib/libmpi_cxx.so.20 > (0x00007fa6adf5f000) > libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007fa6add48000) > libopen-rte.so.20 => /usr/lib64/openmpi/lib/libopen-rte.so.20 > (0x00007fa6adabc000) > libopen-pal.so.20 => /usr/lib64/openmpi/lib/libopen-pal.so.20 > (0x00007fa6ad808000) > librt.so.1 => /lib64/librt.so.1 (0x00007fa6ad600000) > libutil.so.1 => /lib64/libutil.so.1 (0x00007fa6ad3fd000) > libhwloc.so.5 => /lib64/libhwloc.so.5 (0x00007fa6ad1c1000) > libgfortran.so.4 => /lib64/libgfortran.so.4 (0x00007fa6acde5000) > libsatlas.so.3 => /usr/lib64/atlas/libsatlas.so.3 (0x00007fa6abfea000) > libnuma.so.1 => /lib64/libnuma.so.1 (0x00007fa6abddf000) > libltdl.so.7 => /lib64/libltdl.so.7 (0x00007fa6abbd5000) > libquadmath.so.0 => /lib64/libquadmath.so.0 (0x00007fa6ab995000) > > Understand your doubts, so you can also ask any better clarification on > devel mailing list. Ok, I got it. $ objdump -x libsundials_nvecparallel.so.2 | grep R.*PATH RUNPATH /usr/lib64/openmpi/lib which comes from the "-Wl,-rpath -Wl,/usr/lib64/openmpi/lib -Wl,--enable-new-dtags" in mpicxx, so reading https://fedoraproject.org/wiki/Packaging:Guidelines?rd=Packaging/Guidelines#Beware_of_Rpath, I am a bit unsure if only RPATH is contempt or RUNPATH as well. Anyhow, we can get the same behavior as in sundials by adding -DCMAKE_SHARED_LINKER_FLAGS_DEBUG:STRING="%{__global_ldflags} -Wl,-rpath -Wl,/usr/lib64/openmpi/lib -Wl,--enable-new-dtags". Is that ok with you? -- You are receiving this mail because: You are on the CC list for the bug. You are always notified about changes to this product and component _______________________________________________ package-review mailing list -- package-review@xxxxxxxxxxxxxxxxxxxxxxx To unsubscribe send an email to package-review-leave@xxxxxxxxxxxxxxxxxxxxxxx
