https://gcc.gnu.org/bugzilla/show_bug.cgi?id=108680
Bug ID: 108680
Summary: Wrong DTIO arguments with -fdefault-integer-8
Product: gcc
Version: 13.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: albandil at atlas dot cz
Target Milestone: ---
Wrong DTIO arguments with -fdefault-integer-8
It seems that gfortran miscompiles the following simple program when
`-fdefault-integer-8` compiler option is used:
module types
type dtype
contains
procedure :: write_formatted
generic, public :: write(formatted) => write_formatted
end type
contains
subroutine write_formatted (this, unit, iotype, v_list, iostat, iomsg)
class(dtype), intent(in) :: this
integer, intent(in) :: unit, v_list(:)
character(*), intent(in) :: iotype
integer, intent(out) :: iostat
character(*), intent(inout) :: iomsg
iostat = 0
print *, 'v_list', v_list
end subroutine write_formatted
end module types
program p
use types, only: dtype
type(dtype) :: data
integer :: u
open (file = 'output.txt', newunit = u, form = 'formatted')
write (u, '(dt(1,2,3))') data
close (u)
end program p
The derived type `dtype` should be given integers 1,2,3 in the v_list parameter
of the `write(formatted)` subroutine, which only echoes them out for feedback.
This works with Intel compilers regardless of the default integer type.
However, the output is wrong if I combine gfortran `-fdefault-integer-8`. The
behaviour is the same with the current git version as well as with GCC release
11.3, so the problem has been around for some time.
$ /opt/gcc-git/bin/gfortran --version | head -n 1
GNU Fortran (GCC) 13.0.1 20230205 (experimental)
$ /opt/gcc-git/bin/gfortran main.f90
$ ./a.out
v_list 1 2 3
$ /opt/gcc-git/bin/gfortran -fdefault-integer-8 main.f90
$ ./a.out
v_list 8589934593 3 0
$ gfortran-11 --version | head -n 1
GNU Fortran (SUSE Linux) 11.3.1 20221024 [revision
bd0c76a2329e7fe6d6612c2259647bbb67f5866a]
$ gfortran-11 -fdefault-integer-8 main.f90
$ ./a.out
v_list 8589934593 3 0
$ ifx --version | head -n 1
ifx (IFORT) 2023.0.0 20221201
$ ifx -i8 main.f90
$ ./a.out
v_list 1 2 3
$ ifort --version | head -n 1
ifort (IFORT) 2021.8.0 20221119
$ ifort -i8 main.f90
$ ./a.out
v_list 1 2 3
It looks as if the subroutine was actually getting pointers to 4-byte integers
regardless of the switch, because the large value pritned first is actually
composition of the missing 2 and 1:
8589934593 ~ 00000000000000000000000000000010
00000000000000000000000000000001
~ 2 1
