[Bug fortran/20248] gfortran: intrinsics and std=f95, inconsistency with other compilers

["anlauf at hep dot tu-darmstadt dot de" <gcc-bugzilla at gcc dot gnu dot org>]

------- Additional Comments From anlauf at hep dot tu-darmstadt dot de  2005-03-01 08:33 -------
(In reply to comment #3)
> Read the F2003 standard.
> 
> 5.1.2.8    INTRINSIC attribute
> The INTRINSIC attribute confirms that a name is the specific name (13.6) or
> generic name (13.5) of an intrinsic procedure.

OK

> When you set -std=f95, iargc is not an intrinsic procedure.  So, when
> you ask gfortran to *confirm* that iargc is an intrinsic procedure,
> it tells that it isn't.  The default behavior for gfortran is -std=gnu,
> which makes iargc an intrinsic procedure.

It is OK to reject iargc as an intrinsic procedure in the -std=f95 case
during compilation, but the failure during linking when iargc is not
specified as an intrinsic isn't.

The Intel compiler (v8.1) resolves the problem as follows:
- it does not change the name mapping/mangling when -stand=f95 is specified
- it maps the recognized intrinsic iargc to for_iargc when it is
  requested as an intrinsic or not specified as external
- it is mangled to iargc_ when specified as external
- it provides entry points in the runtime for both iargc_ and for_iargc
  (disassembling the runtime shows that both have identical code).

The version 7.1 of the compiler had the VAX/U*ix intrinsics in a
separate portability library.

g77 recognized several options to disable special intrinsics.
In the case of iargc, g77 -funix-intrinsics-disable does the job.
g95 does not recognize this option, and having -std=f95 doing too
many things behind my back is counterproductive.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20248