Re: 20050404: Poe problem with netcdf

"Dan Wright \(Home\)" <address@hidden> writes:

> Hi again Ed:
>
> I keep having problems with netcdf so I got someone else to who understand
> these things better than me to take a look.  I'm attaching three files.
> "profile_usr2" shows how I had environment variables set, "out_configure"
> shows the output generated during configure and "out_make" shows the output
> obtained when I type make after configure.  I should note that we've tried
> many different approaches, but this is perhaps the simplest.  As you'll see,
> configure seems to go smoothly, but make reveals a problem.  Our
> interpretation of this problem is that both reals and doubles are now 8
> bytes in length and the interface is unable to make a distinction between
> them.  Apparently this doesn't occur on all machines since other people have
> used
> -qautodbl=-dbl4  and not reported any problems, but I haven't been so lucky.
>
> Can you tell me if there is a simple (or not so simple) way to avoid this
> problem?
> Note that I've tried -qautodbl=dbl which promotes real to real *8 and double
> to real*16.  In that case I get an error associated with C++ having no
> equivalent to real*16, but at least a netcdf library is generated that
> *seems* to work. Unfortunately, I'm not sure if there are problems
> associated with this error that I just haven't encountered yet.
>
> I hope this makes sense to you.
>
> Regards,
> Dan
>
Sorry Dan,

I only know two ways to compile on AIX machines, and they can be found
here:

http://my.unidata.ucar.edu/content/software/netcdf/builds/index.html

In terms of changing the size of various types, well, I really don't
know how or if that will work with netCDF. I suggest you ask on the
netcdf mailing list to see if anyone else has tried it.

If you don't want the C++ interface, set CXX to '' before configuring,
and that will turn it off.

Good luck!

Ed
-- 
Ed Hartnett  -- address@hidden