Build failure: `Error: Symbol 'absolute' at (1) has no IMPLICIT type`

[barracuda156]

Build fails with gfortran 14.2.0:

/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_layer_constructors.mod CMakeFiles/neural-fortran.dir/nf_layer_constructors.mod.stamp GNU
/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_linear2d_layer.mod CMakeFiles/neural-fortran.dir/nf_linear2d_layer.mod.stamp GNU
/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_conv1d_layer.mod CMakeFiles/neural-fortran.dir/nf_conv1d_layer.mod.stamp GNU
/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_dense_layer.mod CMakeFiles/neural-fortran.dir/nf_dense_layer.mod.stamp GNU
/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_embedding_layer.mod CMakeFiles/neural-fortran.dir/nf_embedding_layer.mod.stamp GNU
/opt/local/bin/cmake -E touch CMakeFiles/neural-fortran.dir/src/nf/nf_layer_constructors.f90.o.provides.build
/opt/local/bin/cmake -E touch CMakeFiles/neural-fortran.dir/src/nf/nf_linear2d_layer.f90.o.provides.build
[ 38%] Building Fortran object CMakeFiles/neural-fortran.dir/src/nf/nf_locally_connected1d_layer.f90.o
/opt/local/bin/gfortran-mp-14   -pipe -Os -m32 -mmacosx-version-min=10.6 -Jinclude -fcoarray=single -c /opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_locally_connected1d_layer.f90 -o CMakeFiles/neural-fortran.dir/src/nf/nf_locally_connected1d_layer.f90.o
/opt/local/bin/cmake -E touch CMakeFiles/neural-fortran.dir/src/nf/nf_conv1d_layer.f90.o.provides.build
[ 39%] Building Fortran object CMakeFiles/neural-fortran.dir/src/nf/nf_linear2d_layer_submodule.f90.o
/opt/local/bin/gfortran-mp-14   -pipe -Os -m32 -mmacosx-version-min=10.6 -Jinclude -fcoarray=single -c /opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_linear2d_layer_submodule.f90 -o CMakeFiles/neural-fortran.dir/src/nf/nf_linear2d_layer_submodule.f90.o
[ 40%] Building Fortran object CMakeFiles/neural-fortran.dir/src/nf/nf_multihead_attention.f90.o
/opt/local/bin/gfortran-mp-14   -pipe -Os -m32 -mmacosx-version-min=10.6 -Jinclude -fcoarray=single -c /opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_multihead_attention.f90 -o CMakeFiles/neural-fortran.dir/src/nf/nf_multihead_attention.f90.o
/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_conv2d_layer.mod CMakeFiles/neural-fortran.dir/nf_conv2d_layer.mod.stamp GNU
[ 41%] Building Fortran object CMakeFiles/neural-fortran.dir/src/nf/nf_conv1d_layer_submodule.f90.o
/opt/local/bin/gfortran-mp-14   -pipe -Os -m32 -mmacosx-version-min=10.6 -Jinclude -fcoarray=single -c /opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_conv1d_layer_submodule.f90 -o CMakeFiles/neural-fortran.dir/src/nf/nf_conv1d_layer_submodule.f90.o
/opt/local/bin/cmake -E touch CMakeFiles/neural-fortran.dir/src/nf/nf_dense_layer.f90.o.provides.build
/opt/local/bin/cmake -E touch CMakeFiles/neural-fortran.dir/src/nf/nf_embedding_layer.f90.o.provides.build
[ 42%] Building Fortran object CMakeFiles/neural-fortran.dir/src/nf/nf_dense_layer_submodule.f90.o
/opt/local/bin/gfortran-mp-14   -pipe -Os -m32 -mmacosx-version-min=10.6 -Jinclude -fcoarray=single -c /opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_dense_layer_submodule.f90 -o CMakeFiles/neural-fortran.dir/src/nf/nf_dense_layer_submodule.f90.o
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_dense_layer_submodule.f90:132:2:

  132 | #ifdef PARALLEL
      |  1
Warning: Illegal preprocessor directive
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_dense_layer_submodule.f90:134:2:

  134 | #endif
      |  1
Warning: Illegal preprocessor directive
[ 43%] Building Fortran object CMakeFiles/neural-fortran.dir/src/nf/nf_embedding_layer_submodule.f90.o
/opt/local/bin/gfortran-mp-14   -pipe -Os -m32 -mmacosx-version-min=10.6 -Jinclude -fcoarray=single -c /opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90 -o CMakeFiles/neural-fortran.dir/src/nf/nf_embedding_layer_submodule.f90.o
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90:1:2:

    1 | #define NONE 0
      |  1
Warning: Illegal preprocessor directive
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90:2:2:

    2 | #define TRIGONOMETRIC 1
      |  1
Warning: Illegal preprocessor directive
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90:3:2:

    3 | #define ABSOLUTE 2
      |  1
Warning: Illegal preprocessor directive
/opt/local/bin/cmake -E cmake_copy_f90_mod include/nf_layernorm_layer.mod CMakeFiles/neural-fortran.dir/nf_layernorm_layer.mod.stamp GNU
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90:55:43:

   55 |       elseif (self % positional == ABSOLUTE) then
      |                                           1
Error: Symbol 'absolute' at (1) has no IMPLICIT type
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90:53:44:

   53 |       if (self % positional == TRIGONOMETRIC) then
      |                                            1
Error: Symbol 'trigonometric' at (1) has no IMPLICIT type
/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/neural-fortran-0.20.0/src/nf/nf_embedding_layer_submodule.f90:17:29:

   17 |       res % positional = NONE
      |                             1
Error: Symbol 'none' at (1) has no IMPLICIT type
make[2]: *** [CMakeFiles/neural-fortran.dir/src/nf/nf_embedding_layer_submodule.f90.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[2]: Leaving directory `/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/build'
make[1]: *** [CMakeFiles/neural-fortran.dir/all] Error 2
make[1]: Leaving directory `/opt/local/var/macports/build/_opt_local_ppcports_fortran_neural-fortran/neural-fortran/work/build'
make: *** [all] Error 2

[milancurcic]

Hi Sergey, it seems that your CMake build doesn't have preprocessing enabled. Perhaps it's because your compiler is not recognized as GNU by the CMake rules? If you add -cpp to the flags, these errors should go away.

[barracuda156]

@milancurcic Hmm, it should be recognized as GNU and normally it is. It might happen that the build picks a different preprocessor from system prefix (which is from gcc-4.2), and that does not work correctly with gcc-14 and its gfortran. I will look into this issue today, thank you for a suggestion.

[barracuda156]

@milancurcic Okay, the problem is that the build only considers Release or Debug build:

neural-fortran/cmake/compilers.cmake

Lines 18 to 19 in 2ed7b6a

	add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-cpp;-fcheck=bounds;-fbacktrace>")
	add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-cpp;-Ofast;-fno-frontend-optimize;-fno-backtrace>")

Since this is not what MacPorts uses as default https://github.com/macports/macports-ports/blob/0e39368a28b6be200d5ea67196227027e1d180ac/_resources/port1.0/group/cmake-1.1.tcl#L40-L44 the needed -cpp flag was not passed.
I think it is a bug. There should be some default for build types not mentioned explicitly, since RelWithDebInfo and MinSizeRel are valid types, not specific to MacPorts.

[milancurcic]

Sounds good, this change would be very welcome. If you know how to do it, can you go ahead and make a PR? Otherwise, I'll do a little bit of research and fix it soon.