[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: ncdigest V1 #770

Hi Steve,

> I'm not a heavy netcdf user (although I'm writing some netcdf code
> for reading LAPS output files) but I maintain the netcdf ebuild for
> Gentoo, as well as build rpms for RedHat/fedora/cAos, etc.
> 
> The netcdf rpm I built recently on RHAS 2.1 (which still uses gcc
> 2.9.6) seemed to build fine, and I didn't notice any warnings like
> below, however, when updating the netcdf ebuild on Gentoo, I saw a
> slew of warnings druing the fortran test.
> 
> My current Gentoo desktop has gcc 3.3.3 and g77 (no f90 compiler),
> with some of the output pasted below:
 ...
> ftest.F:187: warning:
>                call ncagt (outcdf, NCBYTE, attnam, bytval, iret)
>                     1
> ftest.F:196: (continued):
>                call ncagt (outcdf, NCGLOBAL, attnam, shval, iret) 
>                     2
> Argument #4 of `ncagt' is one precision at (2) but is some other
> precision at (1) [info -f g77 M GLOBALS] ftest.F:187: warning:
>                call ncagt (outcdf, NCBYTE, attnam, bytval, iret)
>                     1
> ftest.F:198: (continued):
>                call ncagt (outcdf, NCGLOBAL, attnam, lngval, iret)  
>                           2
> Argument #4 of `ncagt' is one precision at (2) but is some other
> precision at (1) [info -f g77 M GLOBALS]
 ...
> It seems like type/precision errors might cause a problem at some
> point, but what do I know?  Maybe they're benign...

I think these are benign and unavoidable with the netCDF-2 interface
in Fortran77, which is what this is testing.  The netCDF-2 interface
was the early Fortran interface that used the same parameter for data
of multiple types:

      SUBROUTINE NCAGT (INTEGER NCID, INTEGER VARID,
     +                  CHARACTER*(*) ATTNAM, type VALUES,
     +                  INTEGER RCODE)

where

  VALUES
    Returned attribute values. All elements of the vector of attribute
    values are returned, so you must provide enough space to hold
    them. If you don't know how much space to reserve, call NCAINQ first
    to find out the length of the attribute. Warning: neither the
    compiler nor the netCDF software can detect if the wrong type of
    data is used.

was a container for any numeric type of values (INTEGER, REAL,
DOUBLEPRECISION, ...), using knowledge that Fortran argument passing
is by address.  The equivalent C argument would be a void*.  The later
netCDF-3 interface employed type-safe interfaces for Fortran by using
lots more functions, but we still want to support the old netCDF-2
interfaces forever, and there is no way to test such interfaces
without violating picky compiler rules about types always matching on
different calls to the same function.

> PS. I'll test it with gcc 3.4 and see what happens, and the next
> version after 3.4 should have gfortran.

I suspect these potential errors will still be reported, since being
loose with types as in these tests is often a symptom of an error, but
not in this case ...

--Russ

_____________________________________________________________________

Russ Rew                                         UCAR Unidata Program
address@hidden          http://www.unidata.ucar.edu/staff/russ