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

Re: netCDF-3.5.1, CYGWIN and NAG f95

>To: address@hidden
>From: Anlauf Harald <address@hidden>
>Subject: Re: 20040409: netCDF-3.5.1, CYGWIN and NAG f95
>Organization: Deutscher Wetterdienst (DWD)
>Keywords: 200404081303.i38D3UCT005417 

Hi Harald,

Thanks for your message to netcdfgroup and the patches you provided.
I have a few comments.

> 1) The file cfortran.h coming with netCDF is quite ancient, it should be
> replaced by a more recent one or patched suitably.  By googling one can
> certainly find most necessary but scattered information.

We made some local modifications to version 4.1 of cfortran.h and had
contacted Burkhard Burow to get them incorporated into his maintained
version, but for whatever reason, they didn't get in.  I think it will
be possible to integrate our changes into cfortran.h version 4.3, but
we have problems testing the result, since we don't have access to all
the compilers and platforms it is supposed to support.  But we will
look at the change you suggest and try to include it in next version.

> If somebody knows how to correctly auto-detect the type of the Fortran 95
> compiler,
> he or she might also set suitable defaults for the compiler options, such as
> "-mismatch -w=unused" for the NAG compiler.

One of our goals for netCDF version 3.6 is to upgrade the
autoconf/automake/libtool process we use to create the configure
script and Makefiles.  We're hoping that the new version of these
tools will work better with cygwin and other platforms, and will
provide better support for newer versions of Fortran and C++
compilers.

> 2) The following patch should be applied to the source distribution
> 
> --- ./nf_test/nf_test.F.orig  1997-06-11 18:20:48.000000000 +0200
> +++ ./nf_test/nf_test.F.      2004-04-06 14:41:37.375000000 +0200
> @@ -81,6 +81,9 @@
>  !    DIGITAL Visual Fortran needs DFPORT for iargc
>       USE DFPORT
>          implicit        none
> +#elif defined(NAGf90Fortran)
> +        USE F90_UNIX_ENV, only : iargc, getarg
> +        implicit        none
>  #else
>          implicit        none
>          integer         iargc

It's not clear to me whether this change is intended to be independent
of your first patch to cfortran.h.  Should this change be made
independently of whether patch 1) is made, or only if patch 1) is also
made?  We could test that, but we don't currently have access to a NAG
f90.

> 3) In the file ncx.c there is a bug which throws an floating point exception
> with NAG f95. Running the Fortran 90 tests after compiling yields a core
> dump:
> 
> [...]
> 00400000:D:/cygwin/usr/local/src/packages/netcdf-3.5.1/src/nf_test/nf_test.e
> xe (
> symbols previously loaded)
> #0  0x00450bad in ncx_put_short_double (
>     xp=0xa0484ac, ip=0x22edb8) at ncx.c:408
> 408             ix_short xx = *ip;
> (gdb) l 408
> 403     }
> 404
> 405     int
> 406     ncx_put_short_double(void *xp, const double *ip)
> 407     {
> 408             ix_short xx = *ip;
> 409             put_ix_short(xp, &xx);
> 410             if(*ip > X_SHORT_MAX || *ip < X_SHORT_MIN)
> 411                     return NC_ERANGE;
> 412             return ENOERR;
> (gdb) print *ip
> $1 = -32769
> (gdb)
> 
> Clearly, the assignment to the short xx should be made _after_ the range
> check, not before!  I do not know what should happen when *ip is outside the
> range for shorts, so this is something for the maintainers to look into and
> to fix.

We need to look at this more carefully before we decide what to do
about it.  Thanks for pointing out the problem.

--Russ

_____________________________________________________________________

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