main.o(.text+0x29e): undefined reference to `getarg_

[Daniel Heiserer]

sgk@troutmask.apl.washington.edu wrote:
> On Tue, Feb 15, 2005 at 09:03:15AM +0100, Daniel Heiserer wrote:
> 
>>>>I downloded a ia32-linux binary from the gfortran page.
>>>>I found out why my example did not work:
>>>>c----------------
>>>>    PROGRAM test
>>>>    implicit none
>>>>    external getarg
>>>>    character*80    argv(50)
>>>>    integer*4       argc
>>>>    do argc=1,iargc()
>>>>       call getarg(argc,argv(argc))
>>>>      write(6,'(" Argument ",I2," is ",A)') argc, argv(argc)
>>>>    enddo
>>>>    END
>>>>c----------------
>>>>
>>>>The key point is the "external getarg" command. If I remove it it works.
>>>>I do not know if this defintion is an error on my side or if it is a bug 
>>>>of the compiler, but other f90/f77 compilers required it, and g77 as 
>>>>well as g95 got along with it.
>>>>
>>>
>>>
>>>"external getarg" tells the compiler that you are linking a
>>>user-defined external procedure.  "getarg" in gfortran is an
>>>intrinsic procedure if you use the -std=gnu option, which is
>>>the default.  If you use -std=f95, then "getarg" is not an
>>>intrinsic procedure, and gfortran would issue an error during
>>>the link stage because getarg_() would be missing if you do 
>>>not supply one.
>>
>>I appreciate the fact that it is part of the delivery, but
>>as is there is no compatibility with other fortran compiler and g77.
>>
> 
> 
> What do the other fortran compilers do if you change 
> "external getarg" to "intrinsic getarg"?  If they compile
> the both codes, then IMHO something is wrong.  A procedure
> cannot be both an external and an internal procedure
> (again IMHO).

Here the results with "intrinsic":
---------------------
ia32linux/gfortran:
./getarg
Argument  1 is
---------------------
hp-ux-itanium2 f90:
./getarg
--------------------
Here the results with "external":
ERROR in both compilers.

However the iargc() counter is treated different (+1 with gfortran) in 
both compilers. What is "right"?

with gfortran I get > ./getarg 1 2 3 4
  Argument  1 is 1
  Argument  2 is 2
  Argument  3 is 3
  Argument  4 is 4
  Argument  5 is

> I'm also beginning to believe that backwards compatibility
> with g77 may be a mistake because of some of the choices 
> that author of g77 made.

I do not care about g77 combatibility as long as the gfortran works 
along the standard.

best regards, daniel