more fortran tests with pnetcdf

[Rene Redler]

hi,

    I made some more tests with the pnetcdf lib on several
systems, each one showing a different and probably wrong
behavior.

nf_test -c  and nf_test -r run on all systems. There is
possibly something wrong when testing nfmpi_open. Here
3 failures are produced during the -r test which should
probably only be one. I misinterpreted the final output,
and I need to have a closer look into it.

My first question is whether someone has managed to run
the full nf_test on a 32 Bit System successfully?

My test.nc contains for var_name ir:

  ir = -2147483648, 2147483647 ;

Then during the test of nfmpi_put_var1_real subroutine
check_var_real is called which tries to read the variables via
nfmpi_get_var1_real.

The index for accessing "ir" runs correctly from 1 to 2, but
nfmpi_get_var1_real returns two times a negative value, while
the positive second element is expected for index = 2

nf_test -v therefore gives e.g.:

expect:   2147483647.00000
    got:  -2147483648.00000

To my surprise only the real functions fails, but with slightly
different error out on Fedora Core 2 /Absoft Fortran and
Suse Linux / Intel Fortran.

On the SX (64 Bit system) everything seems to run fine but
the nfmpi_put_varm functions. When those are called a segmentation
violation is produced.

Any suggestions about what might go wrong here are more than
welcome since I am running out of ideas.

Rene