[Bug fortran/23349] New: -m64 gives runtime errors in combined Fortran/C programs

[Bug fortran/23349] New: -m64 gives runtime errors in combined Fortran/C programs

[georg dot denk at infineon dot com]

Combining a Fortran-routine and a C-routine gives a binary, which produces
almost all errors, like illegal instruction, segmention fault, invalid output,
when the switch -m64 is used. The program works fine using -m32 or using -m64
and an older version of GCC or using a Sun compiler.

Example:
---- snip: file test2.F -----
      PROGRAM TEST

      IMPLICIT NONE

      INTEGER wrtmsg
      EXTERNAL wrtmsg

      INTEGER I, IERROR

      I = 1
      IERROR = wrtmsg(I, 
     *     'This is the string %s no. %d: %s.'//CHAR(0),
     *     'test1'//CHAR(0), I, 'TEST 1'//CHAR(0))

      I = 2
      IERROR = wrtmsg(I,
     *     'This is the string %s no. %d: %s.'//CHAR(0),
     *     'test2'//CHAR(0), I, 'TEST 2'//CHAR(0))

      I = 3
      IERROR = wrtmsg(I,
     *     'This is the %drd string %s no. %d: %s.'//CHAR(0),
     *     I, 'test3'//CHAR(0), I, 'TEST 3'//CHAR(0))

      STOP
      END
---- snap ---

---- snip: file wrtmsg.c ----
#include <stdarg.h>
#include <stdio.h>

int
wrtmsg_(int *i, char *format, ...)
{
  char tmp_string[1000];
  va_list args;
  int *i1, *i2;
  char *s1, *s2;

  va_start(args, format);
  if (*i == 1 || *i == 2) 
    {
      s1 = (char *) va_arg(args, char*);
      i1 = (int *)  va_arg(args, int*);
      s2 = (char *) va_arg(args, char*);
      snprintf(tmp_string, 1000, format, s1, *i1, s2);
    }
  else
    {
      i1 = (int *)  va_arg(args, int*);
      s1 = (char *) va_arg(args, char*);
      i2 = (int *)  va_arg(args, int*);
      s2 = (char *) va_arg(args, char*);
      snprintf(tmp_string, 1000, format, *i1, s1, *i2, s2);
    }
  va_end(args);

  printf("%s\n", tmp_string);

  return *i+1;
}
---- snap ----

Compiler used:

% gfortran -v
Using built-in specs.
Target: x86_64-unknown-linux-gnu
Configured with: ../../gcc-4.0.1/configure --prefix=/tmp2/denk/GCC4.0.1
--exec-prefix=/tmp2/denk/GCC4.0.1/LINUXAMD64 --enable-languages=c,c++,f95,java
--with-gmp=/tmp2/denk/GCC4.0.1/LINUXAMD64 --with-mpfr=/tmp2/denk/GCC4.0.1
Thread model: posix
gcc version 4.0.1


How to reproduce:

% gfortran -c -fno-second-underscore -Wall -m64 -g test2.F
% gcc -c -Wall -m64 -g wrtmsg.c
% gfortran -o test2 -Wall -m64 -g wrtmsg.o test2.o

Results:

% ./test2
This is the string test1 no. 1: TEST 1.
This is the string test2 no. 2: TEST 2.
Illegal instruction
% ./test2
This is the string test1 no. 1: TEST 1.
This is the string test2 no. 2: TEST 2.
Segmentation fault
% ./test2
This is the string test1 no. 1: TEST 1.
This is the string test2 no. 2: TEST 2.
STOP 0

(which means that the last output line has got lost)

-- 
           Summary: -m64 gives runtime errors in combined Fortran/C programs
           Product: gcc
           Version: 4.0.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: georg dot denk at infineon dot com
                CC: gcc-bugs at gcc dot gnu dot org,georg dot denk at
                    infineon dot com
 GCC build triplet: x86_64-unknown-linux-gnu
  GCC host triplet: x86_64-unknown-linux-gnu
GCC target triplet: x86_64-unknown-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23349

[Bug fortran/23349] -m64 gives runtime errors in combined Fortran/C programs

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-08-12 12:29 -------
I think this is really just invalid.  As the ABI for var-args is different from the ABI for non var-args.  And 
fortran only passes via non var-args functions.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23349

[Bug fortran/23349] -m64 gives runtime errors in combined Fortran/C programs

[georg dot denk at infineon dot com]

------- Additional Comments From georg dot denk at infineon dot com  2005-08-12 12:45 -------
(In reply to comment #1)
> I think this is really just invalid.  As the ABI for var-args is different
from the ABI for non var-args.  And 
> fortran only passes via non var-args functions.

I don't think that this is "just invalid" --- and this is not because we use
this kind of code in our application at many places ;-)

At least it works with older gcc/g77 combination (gcc 3.4.3 does it right with
-m64), also the Sun Compiler with -xarch=v9a and the Intel Itanium compiler. So
it is at least a regression compared to gcc 3.4.3.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23349

[Bug fortran/23349] -m64 gives runtime errors in combined Fortran/C programs

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2005-08-12 22:34 -------
This is not a bug.  You are calling a var-args function basicially what is considered a non var-args 
prototype in C.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23349