This is important since it blocks the upgrade of the default version
of GCC from GCC 4.8 to GCC 4.9. If you cannot fix it, please change
to USE_GCC=4.8 for the time being.
/usr/local/mpi/openmpi/bin/mpif90 -c -O2 -ffast-math define_system_module.f
/usr/local/mpi/openmpi/bin/mpif90 -c -O2 -ffast-math basic_comms.f
Warning: Rank mismatch in argument 'aaa' at (1) (rank-1 and scalar)
/usr/local/mpi/openmpi/bin/mpif90 -c -O2 -ffast-math merge_tools.f
If the rank mismatch warning is stopping compilation, -Wno-error should get around it. I don't think this will cause any run time issues.
Reporter is Committer.
Yes, reporter is committer, but that does not mean he should be
coughing up patches for ports maintained by other people. ;-)
firstname.lastname@example.org, mind proposing a concrete patch?
The real error message here is
/usr/local/bin/ld: cannot find -lmpi_usempif08
/usr/local/bin/ld: cannot find -lmpi_usempi_ignore_tkr
This is because gcc49 supports Fortran 08, or at least enough (more than gcc48) of that to enable the f08 wrappers in openmpi (net/openmpi), and subsequently (e.g. when building dlpoly-classic) tries to use them.
Summary: the real culprit here is net/openmpi. I'll send a fix tomorrow.
Created attachment 159372 [details]
patch for net/openmpi - package f08 wrappers, too
This patch has to be applied along with the gcc 4.8 -> 4.9 upgrade (with gcc48, this would result in packaging errors).
Created attachment 159373 [details]
"make stage-qa check-orphans" output for net/openmpi gcc49 after patching
for lack of poudriere space in my 9.3 build environment
Created attachment 159374 [details]
"make stage-qa check-orphans" for science/dlpoly-classic gcc49 after fixing openmpi
for lack of poudriere space
The patch doesn't seem to work, I am doing exp-run with gcc 4.9 again and openmpi fails to package:
Created attachment 162440 [details]
patch for net/openmpi v2
(In reply to Antoine Brodin from comment #10)
That patch looks exactly like what cooking in my poudriere right now, looks like you beat me with that.
builds fine on 9.3 and 10.2. If we could have this committed before openmpi is upgraded again...
Sorry, guys. I never saw Gerald's request from 2-22. Thanks for taking care of this.
After this: https://svnweb.freebsd.org/changeset/ports/400879 the patch may no longer be necessary
Seems to work in my 9.3 VM.
Can we close this report? Looks like everything have been solved.
Indeed, let's consider this fixed.