Fortran errors

[wgnoehring]

Building atomistica fails because Fortran code cannot be compiled.
I am using gfortran version 10.2.1.

The compiler complains that REAL(8) are passed to functions which
expect INT(4). The solution would probably be to promote the INT(4)
variables to REAL(8). However, I am not sure what is going on in the
code and whether it is safe to do so.

Notice that there are also some warnings about deprecated arithmetic IF
statements.

Error message below:

atomistica/src/potentials/coulomb/fft3-public.f:165:72:

  165 |       IF (J-4) 102,102,103
      |                                                                        1
Warning: Fortran 2018 deleted feature: Arithmetic IF statement at (1)
atomistica/src/potentials/coulomb/fft3-public.f:171:72:

  171 |       IF (NR) 101,105,101
      |                                                                        1
Warning: Fortran 2018 deleted feature: Arithmetic IF statement at (1)
atomistica/src/potentials/coulomb/fft3-public.f:12:72:

   12 |       CALL CFFTB1 (N,C,WSAVE,WSAVE(IW1),WSAVE(IW2))
      |                                                                        1
Error: Type mismatch in argument ‘ifac’ at (1); passed REAL(8) to INTEGER(4)
atomistica/src/potentials/coulomb/fft3-public.f:83:72:

   83 |       CALL CFFTF1 (N,C,WSAVE,WSAVE(IW1),WSAVE(IW2))
      |                                                                        1
Error: Type mismatch in argument ‘ifac’ at (1); passed REAL(8) to INTEGER(4)
atomistica/src/potentials/coulomb/fft3-public.f:154:72:

  154 |       CALL CFFTI1 (N,WSAVE(IW1),WSAVE(IW2))
      |                                                                        1
Error: Type mismatch in argument ‘ifac’ at (1); passed REAL(8) to INTEGER(4)

[wgnoehring]

In the meantime, a workaround is to add -fallow-argument-mismatch to the gfortran flags in setup.cfg

[pastewka]

I know about this error and it is totally annoying. I don't know why they decided to break backwards compatibility this way. We can't really fix this in the code without throwing out the FFT code completely.

Another workaround is to link with FFTW (because this will then not compile the FFT code that comes with Atomistica).

[pastewka]

The simplest workaround seems to be to compile via:

FFLAGS=-fallow-argument-mismatch python3 setup.py build --force