Parallel-NetCDF 1.0.0 pre-release for NEC SX

[Rob Ross]

Ah hah!  Thank you very much for figuring out this "kind" bug in the 
configure scripts.  None of us here are Fortran experts and would have 
likely never spotted this.

We'll look into getting the util.F right with this new knowledge (and 
corrected configure).

Thanks!

Rob

Rene Redler wrote:
> hi all,
> 
>   after some off-line discussion and clarification with
> Jianwei I try to summarize some issues that came up when
> installing and running the pnetcdf lib and nf_test on a
> NEC SX. This concers the problem with representing int1
> and int2 in Fortran - in my oppinion due to a bug in the
> configure script.
> 
> To let the SX compiler properly built and support
> the int2 interfaces one has to set FFLAGS=-dW . -->
> Could the developers provide me with the final version
> of the pnetcdf a few days prior to the official release
> so that I can check it on the SX and update the README.SX ?
> 
> 1-byte integer is definitely not supported on the NEC SX.
> Here the configure scripts fails in detecting this
> for the following reason:
> 
> The configure first checks for integer*1 . With the
> SX compiler this test produces an error, which is correct.
> In this case it is checked whether integer(kind(1))
> is supported. But, this is not testing 1 byte integer
> but the default integer size:
> 
> The kind intrinsic returns the kind of the input argument.
> Since here the input is a default integer the default
> integer size is returned, which is 4 and corresponds the
> C int.
> 
> Instead the test should be make against integer(kind=1)
> which would correspond to integer*1.
> 
> The same happens with the int2 checks in the configure script.
> I modified the configure script and restarted the installation.
> Everything seems to work fine now. No int1 interfaces are built.
> 
> In the test/nf_test/util.F I now had to replace
> 
>       #include "No C equivalent to Fortran INTEGER1"
> 
> with some valid Fortran comment in order to get nf_test
> compiled. Having done that nf_test passed correctly with
> the expected 3 failures.
> 
> Is it reasonable for portability reasons to have the e.g. int1
> interface be included in the Library even if there is no active
> support of integer*1 or integer(kind=1) and copy the data to default
> integer in the Fortran world in this case?
> 
> Regards, Rene