Builds use C-specific flags during Fortran builds

[knedlsepp]

Issue description

Building Fortran projects currently leads to a massive amount of warning messages

f951: Warning: command line option '-Wformat=1' is valid for C/C++/ObjC/ObjC++ but not for Fortran
f951: Warning: command line option '-Wformat-security' is valid for C/C++/ObjC/ObjC++ but not for Fortran

See e.g. liblapack's log output This is caused by the format hardening adding those flags to NIX_CFLAGS_COMPILE, which in turn is used also for Fortran builds.

This also leads to the breakage of a build system of one package I try to compile, as it expects gfortran not to produce any warnings for some tests. I propose thus to disable $NIX_CFLAGS_COMPILE for fortran builds or to disable the format hardening for Fortran by default.

[FRidh]

cc @globin @fpletz

[fpletz]

I'm not sure if we can disable format hardening for all fortran projects automatically but it should be done somehow. We have to look into it.

Thanks for the report!

[veprbl]

Or could we just filter these in gfortran wrapper?

[matthewbauer]

You can always add hardeningDisable = [ "format" ]; to a derivation to disable this. Otherwise, though, we don't know ahead of time what language gcc is being used to compile for. Since this isn't a fatal error, I think it is best to disable this on a case-by-case basis. Hardening is still nice to have in fortran even if it doesn't work in this case.

[Shimuuar]

I think this issue should be reopened. Current state provides absolutely horrible user experience when developing in fortran. Every fortran files gives 3 lines of spurious warnings which drown out legitimate ones. So i think

You can always add hardeningDisable = [ "format" ]; to a derivation to disable this.

isn't really a solution since it's not documented anywhere. Where to add it? What will it do? For anyone without understanding of gcc packaging it's an impenetrable mistery

[veprbl]

@Shimuuar The format issue was seemingly resolved by f42aa7e1d7ac4fb1050b73b296912fa7c7b927c2. I only see errors with -iframework, but that's on Darwin.

[Shimuuar]

It seems it isn't. On release-20.09 branch warning is still present. I tried to test it with commit 4c4a45e284cd505976618825eff9b5e44332d0cf on release-21.05 and it still persists. Test is easy. Compile hello world

      program hello
      print *, 'Hello, World!'
      end program hello

Then you get following warnings

$ gfortran hello.for 
f951: Warning: command line option '-Wformat=1' is valid for C/C++/ObjC/ObjC++ but not for Fortran
f951: Warning: command line option '-Wformat-security' is valid for C/C++/ObjC/ObjC++ but not for Fortran
f951: Warning: '-Werror=' argument '-Werror=format-security' is not valid for Fortran

[veprbl]

@Shimuuar I don't see it:

with import (builtins.fetchTarball "https://github.com/NixOS/nixpkgs/archive/4c4a45e284cd505976618825eff9b5e44332d0cf.tar.gz") {};

runCommandNoCC "fortran-test" {
  buildInputs = [ gfortran ];
  file = writeText "fortran-test.f" "
          program hello
          print *, 'Hello, World!'
          end program hello
  ";
} ''
  gfortran "$file"
  install -Dm555 ./a.out "$out"
''

[Shimuuar]

A-ha! Yes problem doesn't happen when you call gfortran inside derivation but it does happen when it's called in nix-shell:

with import (builtins.fetchTarball
  "https://github.com/NixOS/nixpkgs/archive/4c4a45e284cd505976618825eff9b5e44332d0cf.tar.gz") {};
pkgs.mkShell
  { buildInputs = [pkgs.gfortran];
  }

$ gfortran hello.for 
f951: Warning: command line option '-Wformat=1' is valid for C/C++/ObjC/ObjC++ but not for Fortran
f951: Warning: command line option '-Wformat-security' is valid for C/C++/ObjC/ObjC++ but not for Fortran
f951: Warning: '-Werror=' argument '-Werror=format-security' is not valid for Fortran

[LunNova]

This is very inconvenient if you're mixing compilers which support different flags and want to use a NIX_CFLAGS_COMPILE value that your fortran compiler errors out on, rather than just showing a warning.