Double R4/R8 link for stats.x (and more...)

[mathomp4]

This is a fun one. I will probably need @lltakacs and/or @sdrabenh to help test. (I'll also ping @tclune as it's CMake related)

I've been trying to get GEOSldas to compile with NAG. I'm using GEOSldas because it has no FMS (which NAG can never compile at present). I'm currently up to 83 files changed but stats.F90 is getting to me. The main issue now is linking stats.x:

[100%] Linking Fortran executable ../../../../../bin/stats.x
NAG Fortran Compiler Release 7.0(Yurakucho) Build 7036
duplicate symbol '_m_gfio_getfld_MP_fix_undef_in_' in:
    ../../../../../lib/libGMAO_gfio_r4.a(m_GFIO_GetFld.f.o)
    ../../../../../lib/libGMAO_gfio_r8.a(m_GFIO_GetFld.f.o)
duplicate symbol '_m_gfio_getfld_MP_gfio_get_fall_' in:
    ../../../../../lib/libGMAO_gfio_r4.a(m_GFIO_GetFld.f.o)
    ../../../../../lib/libGMAO_gfio_r8.a(m_GFIO_GetFld.f.o)

Why this is happening can be seen by looking at CMake. To wit, first we have what stats.x wants:

target_link_libraries (stats.x post_nompi GMAO_gfio_r4 MAPL ${ESMF_LIBRARIES})

So an explicit link to GMAO_gfio_r4 and to post_nompi. That has a dependency on GMAO_hermes:

esma_add_library (post_nompi
  SRCS dynamics_lattice_module.F mpi_util.F mymalloc.F timer.F
  DEPENDENCIES MAPL GMAO_hermes)
target_compile_definitions(post_nompi PRIVATE mpi)

which links in GMAO_gfio_r8:

   esma_add_library(${this} SRCS ${srcs} DEPENDENCIES GMAO_mpeu GMAO_gfio_r8)

And so the issue. stats.x wants both the r4 and r8 versions of GFIO. 😦 (NOTE: libpost.a has the same issue as well so a LOT of programs have an issue like this.)

I can obviously "fix" this by removing the explicit link to GMAO_gfio_r4 in post, but I'm not sure if that breaks stats.x...as I don't know how to run it!