Cannot build Phantom on macOS

[dmentipl]

Phantom fails to build on macOS with HDF5=yes. The failure is in building the binary in the linking stage.

My machine is macOS 10.15.6. I have gfortran (version 10.2.0) and HDF5 (version 1.12.0_1) installed via Homebrew.

I am using Phantom from the master branch plus the changes implemented in #58. I.e. @dliptai's fork.

To reproduce, assuming you have the latest gfortran and HDF5 installed from Homebrew:

$ GIT_LFS_SKIP_SMUDGE=1 git clone git@github.com:dliptai/phantom
$ cd phantom
$ make SETUP=disc SYSTEM=gfortran HDF5=yes HDF5ROOT=/usr/local/opt/hdf5

The output, at the stage where it fails:

gfortran -O3 -Wall -Wno-unused-dummy-argument -frecord-marker=4 -gdwarf-2 -finline-functions-called-once -finline-limit=1500 -funroll-loops -ftree-vectorize -std=f2008 -fall-intrinsics -fPIC -fopenmp -fdefault-real-8 -fdefault-double-8 -I/usr/local/opt/hdf5/include -o ../bin/phantom physcon.o config.o kernel_cubic.o io.o units.o boundary.o mpi_utils.o dtype_kdtree.o utils_omp.o utils_cpuinfo.o utils_allocate.o icosahedron.o utils_mathfunc.o part.o mpi_domain.o utils_timing.o mpi_balance.o setup_params.o timestep.o utils_dumpfiles.o utils_indtimesteps.o utils_infiles.o utils_sort.o utils_supertimestep.o utils_tables.o utils_gravwave.o utils_sphNG.o utils_vectors.o utils_datafiles.o datafiles.o gitinfo.o random.o eos_mesa_microphysics.o eos_mesa.o eos_shen.o eos_helmholtz.o eos_idealplusrad.o eos.o cullendehnen.o nicil.o nicil_supplement.o inverse4x4.o metric_minkowski.o metric_tools.o utils_gr.o cons2primsolver.o checkoptions.o viscosity.o options.o radiation_utils.o cons2prim.o centreofmass.o extern_corotate.o extern_binary.o extern_spiral.o extern_lensethirring.o extern_gnewton.o lumin_nsdisc.o extern_prdrag.o extern_Bfield.o extern_densprofile.o extern_staticsine.o extern_gwinspiral.o externalforces.o damping.o checkconserved.o partinject.o utils_inject.o utils_filenames.o utils_summary.o fs_data.o mol_data.o utils_spline.o h2cooling.o h2chem.o cooling.o dust.o growth.o dust_formation.o ptmass_radiation.o mpi_dens.o mpi_force.o stack.o mpi_derivs.o kdtree.o linklist_kdtree.o memory.o utils_hdf5.o utils_dumpfiles_hdf5.o readwrite_dumps_common.o readwrite_dumps_fortran.o readwrite_dumps_hdf5.o readwrite_dumps.o quitdump.o ptmass.o readwrite_infile.o dens.o force.o utils_deriv.o deriv.o energies.o sort_particles.o evwrite.o step_leapfrog.o writeheader.o step_supertimestep.o mf_write.o evolve.o checksetup.o initial.o phantom.o  -L/usr/local/opt/hdf5/lib -lhdf5 -lhdf5_fortran
Undefined symbols for architecture x86_64:
  "_GOMP_loop_maybe_nonmonotonic_runtime_next", referenced from:
      ___timestep_sts_MOD_sts_init_step._omp_fn.0 in utils_supertimestep.o
      ___densityforce_MOD_densityiterate._omp_fn.0 in dens.o
      ___forces_MOD_force._omp_fn.0 in force.o
  "_GOMP_loop_maybe_nonmonotonic_runtime_start", referenced from:
      ___timestep_sts_MOD_sts_init_step._omp_fn.0 in utils_supertimestep.o
      ___densityforce_MOD_densityiterate._omp_fn.0 in dens.o
      ___forces_MOD_force._omp_fn.0 in force.o
  "__gfortran_os_error_at", referenced from:
      ___dump_utils_MOD_read_array_real8arr in utils_dumpfiles.o
      ___dump_utils_MOD_read_array_real8 in utils_dumpfiles.o
      ___table_utils_MOD_diff in utils_tables.o
      ___table_utils_MOD_flip_array in utils_tables.o
      ___mesa_microphysics_MOD_get_eos_constants_mesa in eos_mesa_microphysics.o
      ___mesa_microphysics_MOD_read_opacity_mesa in eos_mesa_microphysics.o
      ___mesa_microphysics_MOD_get_opacity_constants_mesa in eos_mesa_microphysics.o
      ...
ld: symbol(s) not found for architecture x86_64
collect2: error: ld returned 1 exit status
make[1]: *** [phantom] Error 1
make: *** [phantom] Error 2

This is coming from the building the phantom binary stage, in the following lines in the Makefile:

phantom: checksystem checkparams $(OBJECTS) phantom.o
	$(FC) $(FFLAGS) -o $(BINDIR)/$@ $(OBJECTS) phantom.o $(LDFLAGS)

I do not have this problem on my linux machine with gfortran (version 9.3.0) and HDF5 (version 1.10.5).

[dmentipl]

Actually this also happens without HDF5=yes.

[dmentipl]

Narrowing further (I have to do GIT_LFS_SKIP_SMUDGE=1 because the git-lfs quota is blown):

$ GIT_LFS_SKIP_SMUDGE=1 git clone git@github.com:danieljprice/phantom
$ cd phantom
$ make SETUP=empty SYSTEM=gfortran

Output (at the point where the failure is):

gfortran -O3 -Wall -Wno-unused-dummy-argument -frecord-marker=4 -gdwarf-2 -finline-functions-called-once -finline-limit=1500 -funroll-loops -ftree-vectorize -std=f2008 -fall-intrinsics -fPIC -fopenmp -fdefault-real-8 -fdefault-double-8 -o ../bin/phantom physcon.o config.o kernel_cubic.o io.o units.o boundary.o mpi_utils.o dtype_kdtree.o utils_omp.o utils_cpuinfo.o utils_allocate.o icosahedron.o utils_mathfunc.o part.o mpi_domain.o utils_timing.o mpi_balance.o setup_params.o timestep.o utils_dumpfiles.o utils_indtimesteps.o utils_infiles.o utils_sort.o utils_supertimestep.o utils_tables.o utils_gravwave.o utils_sphNG.o utils_vectors.o utils_datafiles.o datafiles.o gitinfo.o random.o eos_mesa_microphysics.o eos_mesa.o eos_shen.o eos_helmholtz.o eos_idealplusrad.o eos.o cullendehnen.o nicil.o nicil_supplement.o inverse4x4.o metric_minkowski.o metric_tools.o utils_gr.o cons2primsolver.o checkoptions.o viscosity.o options.o radiation_utils.o cons2prim.o centreofmass.o extern_corotate.o extern_binary.o extern_spiral.o extern_lensethirring.o extern_gnewton.o lumin_nsdisc.o extern_prdrag.o extern_Bfield.o extern_densprofile.o extern_staticsine.o extern_gwinspiral.o externalforces.o damping.o checkconserved.o partinject.o utils_inject.o utils_filenames.o utils_summary.o fs_data.o mol_data.o utils_spline.o h2cooling.o h2chem.o cooling.o dust.o growth.o dust_formation.o ptmass_radiation.o mpi_dens.o mpi_force.o stack.o mpi_derivs.o kdtree.o linklist_kdtree.o memory.o readwrite_dumps_common.o readwrite_dumps_fortran.o readwrite_dumps.o quitdump.o ptmass.o readwrite_infile.o dens.o force.o utils_deriv.o deriv.o energies.o sort_particles.o evwrite.o step_leapfrog.o writeheader.o step_supertimestep.o mf_write.o evolve.o checksetup.o initial.o phantom.o
Undefined symbols for architecture x86_64:
  "_GOMP_loop_maybe_nonmonotonic_runtime_next", referenced from:
      ___timestep_sts_MOD_sts_init_step._omp_fn.0 in utils_supertimestep.o
      ___densityforce_MOD_densityiterate._omp_fn.0 in dens.o
      ___forces_MOD_force._omp_fn.0 in force.o
  "_GOMP_loop_maybe_nonmonotonic_runtime_start", referenced from:
      ___timestep_sts_MOD_sts_init_step._omp_fn.0 in utils_supertimestep.o
      ___densityforce_MOD_densityiterate._omp_fn.0 in dens.o
      ___forces_MOD_force._omp_fn.0 in force.o
  "__gfortran_os_error_at", referenced from:
      ___dump_utils_MOD_read_array_real8arr in utils_dumpfiles.o
      ___dump_utils_MOD_read_array_real8 in utils_dumpfiles.o
      ___table_utils_MOD_diff in utils_tables.o
      ___table_utils_MOD_flip_array in utils_tables.o
      ___mesa_microphysics_MOD_get_eos_constants_mesa in eos_mesa_microphysics.o
      ___mesa_microphysics_MOD_read_opacity_mesa in eos_mesa_microphysics.o
      ___mesa_microphysics_MOD_get_opacity_constants_mesa in eos_mesa_microphysics.o
      ...
ld: symbol(s) not found for architecture x86_64
collect2: error: ld returned 1 exit status
make[1]: *** [phantom] Error 1
make: *** [phantom] Error 2

gfortran version:

$ gfortran --version
GNU Fortran (Homebrew GCC 10.2.0) 10.2.0
Copyright (C) 2020 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

[bjnorfolk]

Also can't build phantom on my mac. Any tips @danieljprice ?

[danieljprice]

looks like a conflict in your gfortran installations, not a phantom problem, as in you're linking against the wrong libgfortran compared to the one you used to compile the code.

[danieljprice]

this is likely related to your -L/dir/ flags which are causing a conflict with the libraries linked at linktime. This is not a problem with phantom but with your link against hdf5 libraries causing a conflict. Closing as not a phantom issue, and also a good example of why binary Fortran library dependencies cause headaches.

[dmentipl]

I don't think it's an issue with HDF5. I can't even build Phantom with make SETUP=empty SYSTEM=gfortran.