On Jan 24, 2014, at 1:18 AM, Paul Hargrove <phhargrove_at_[hidden]> wrote:
> My understanding is that for (at least) pathscale and open64 compilers we have a regression because 1.7.3 automatically rejected the f08 bindings, but the current 1.7.4rc tarball fails while building usempif08 unless *manually* disabled at configure time. Is that accurate?
I just committed a change on the trunk to configure that should disqualify the older pathscale and open64 compilers from compiling the mpi_f08 module (like we did in 1.7.3 and earlier).
I trivially extended one Fortran configure test to check for BIND(C, name="foo") where strlen(foo)>32. That should make the older pathscale fail the test, and therefore we avoid the internal compiler error.
I did *not* change the OMPI C symbol names to be <= 32 characters.
It ain't perfect, but hopefully that'll do it!
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/