question

[John Tannahill]

Rob,

I tried your suggested changes below and still no nfmpi_xxx
entry points in libpnetcdf.a.  Again I do have a workaround by
additionally linking with libnetcdf.a in src/lib.  Still think
there are probably libnetcdf.a references in the configure
setup that should be libpnetcdf.a.  For example, after I do the
make at the very end it does a "make ranlib ../lib/libnetcdf.a".
I would have thought this would be libpnetcdf.a?

Also recall that I am working on NERSC's IBM-SP, seaborg, and
that my test C program appears to be working (an adaptation of
LBNL's csnap.c program).  My Fortran90 implementation of this code
compiles, links with the fix mentioned above, but is not yet
running properly, so I will be returning to focus my efforts
there.

I think that it is great that Tyce Mclarty here at LLNL is taking
a detailed look at some of the configuration issues.  Note that
Tyce (and Richard Hedges) are part of the ASCI parallel i/o
group here at LLNL; I am a software developer in the Atmospheric
Science Division trying to make use of your fortran parallel-
netcdf capabilities.  I had hoped to not be so much on the
bleeding edge, but the more I get into this, it looks like we
may be the first ones to be trying to do this?  Which makes it
more fun, but somewhat risky?  Hopefully we can all benefit from
this experience, as it looks like there is definitely a need
for this capability now.

Regards,
John


Robert Latham wrote:
> On Wed, Jul 23, 2003 at 09:25:05AM -0700, John Tannahill wrote:
> 
>>Please keep me posted as to the status of my problem with the
>>Fortran interface.  I will look at it some more myself, but I
>>don't have much experience with autoconf, etc.  Maybe the autoconf
>>stuff is fine and it is just something more basic I am missing?  
> 
> 
> I do all my development on a linux box, so your help in resolving this
> is greatly appreciated.
> 
> Please try modifying your configure.in to call UD_CHECK_FCALLSCSUB
> just before it calls UD_CHECK_LIB_MATH:  this should resolve the
> 'warning: Unrecognized Fortran name mapping'.  run autoconf, then
> re-run configure.  after you run configure, see which F77_NAME_* macro
> is defined.
> 
> This should make the correct fortran functions appear in
> libpnetcdf.a
> 
> Now with the underscore convention handled correctly, we can see if
> there is a problem in the way we build the library.
> 
> Thanks for your patience and help.
> ==rob
> 


-- 
       +----------------------------------------------------------------------+
       |                                          John R. Tannahill           |
       |    _/      _/      _/    _/  _/          Lawrence Livermore Nat. Lab |
       |   _/      _/      _/_/  _/  _/           P.O. Box 808, M/S L-103     |
       |  _/      _/      _/  _/_/  _/            Livermore, CA  94551        |
       | _/_/_/  _/_/_/  _/    _/  _/_/_/         Phone:  925-423-3514        |
       |                                          Fax:    925-423-4908        |
       | Lawrence Livermore National Laboratory   Email:  jrt at llnl.gov        |
       +----------------------------------------------------------------------+