fsiesta_mpi.F90: Reopening /dev/stdout doesn't work with position="append" with gfortran

[Jonas L. B.]

As the title says: I am trying to use the fsiesta_mpi.F90 interface to use Siesta as a subroutine. When compiled with gfortran (it does not happen with ifort), I get a crash with the error "Fortran runtime error: Illegal seek".

The particular error occurs at line 501 in fsiesta_mpi.F90, where the code reads:

```
#ifdef MPI
! Go back to parent directory and reset output file
  call chdir('..')
  close(unit=6)
  open(unit=6, file=trim(mainOutFile), position='append') ! (this is line 501)
#endif
```
It closes the temporary output file and reopens the previous one (which is stdout, at least in my case). I can reproduce it in a minimal fortran example:

```
program MyTest
implicit none
close(unit=6)
open(unit=6, file="/dev/stdout", position="append")
write(6, *) "testtest"
end program
```
which shows the same behaviour of crashing when compiled with gfortran but not ifort. Removing the position="append" part works, and I have done that for myself now, but I guess that if an actual file is specified as mainOutFile, it'd be overwritten. So perhaps another solution is needed.

[Alberto Garcia (albertog)]

Which version of gfortran are you using?
Your test runs fine on my Mac with

  $ gfortran --version
  GNU Fortran (Homebrew GCC 7.2.0) 7.2.0

[Jonas L. B. (jonaslb)]

I am using 9.1.0
    $ gfortran --version
    GNU Fortran (GCC) 9.1.0

[Jonas L. B. (jonaslb)]

I can also reproduce on gfortran 8.3.0.

[Jonas L. B. (jonaslb)]

I later also reproduced it on gfortran 6.4 and 7.3, so this must be a platform dependent thing (I used different Linuxes).
A fix that should work both when main output is a file or device, is to use `call fseek(6,0,2)` rather than `position="append"`.
I can try to send an MR with this small fix on monday when everything is moved to Gitlab :)

[Nick Papior (nickpapior)]

I think we should solve this in another way than using seek.

Is this on HPC or Niflheim?

PS. I think the correct way would be to use iostat to check if the first append fails, if so do a compatible thing that might not be optimal.

[Jonas L. B. (jonaslb)]

It happens on HPC, Niflheim and my own Ubuntu PC.

I tested with iostat and it is set to 29, so we could check for that of course. But seeking seems to work regardless, so why not just do that?

[Nick Papior (nickpapior)]

fseek is not in the fortran standard. :(