20030407: ncvarget "failure" under Linux pgcc

Hugh,

> To: address@hidden
> From: Hugh Ellis <address@hidden>
> Subject: ncvarget failure
> Organization: UCAR/Unidata
> Keywords: 200304021507.h32F7p7U006497

The above message contained the following:

> In attempting to build netcdf on my linux pc, I get the following error
> involving ncvarget when running make test (I use pgf90, pg cc and c++
> compilers and RedHat 8.0). Could you please tell me what might be the
> fix? The full log from make test is attached.
> 
> Thanks,
> 
> Hugh Ellis
> 
> *** Testing ncvarput1 ...     ok ***
> *** Testing ncvarget1 ...     ok ***
> *** Testing ncvarput ...      ok ***
> *** Testing ncvarget ...      *** test_slabs: ncvarget got wrong value for
> point

I was able to reproduce this error on our Linux/PC system using version
4.0 of the Portland Group's compilers.

As far as I've been able to tell, the problem lies with the Portland
Group's C compiler (pgcc).  I say this because of the following:

    1.  We haven't encountered this problem in any other development
        environment;

    2.  The primitive C types ("char", "short", etc.) of the Portland
        Group C compiler are identical to those of the GNU C compiler,
        yet the netCDF package doesn't have this problem on the same
        platform when compiled using gcc instead of pgcc; 

    3.  I've looked through the code at the place where the error occurs
        and I can't find anything wrong;

    4.  We've had problems with the Portland Group compilers before.

I imagine you didn't want to hear this, but that's what it appears to be.

As a workaround, I suggest using the GNU C compiler rather than the
Portland Group C compiler. You can do this by setting the environment
variable CC to the pathname of the GNU C compiler before executing
the configure script (after doing a "make distclean", of course).
Be sure that the environment variable CPPFLAGS contains the string
"-DpgiFortran".

Please let me know if this helps.

Regards,
Steve Emmerson