970116: Question about NCOPN's behavior

Mary,

> To: address@hidden
> From: address@hidden (Mary Haley)
> Subject: Question about NCOPN's behavior
> Organization: UCAR/NCAR/SCD
> Keywords: 199701161959.MAA20955

In the above message you wrote:

> I just built and installed Version 2.43 of netCDF on all of NCAR's
> Crays using f90.  A user pointed out to me that previously, if she
> tried to open a netCDF file that didn't exist, she would get an ierr
> of -1 ("ierr" being the last argument of NCOPN).  Now, it's returning
> 0.
> 
> I wrote a little Fortran test program where I just open and close
> a netCDF file.  The first file I open does indeed exist, and the
> second file I open doesn't.  Here's the program:
> 
>       character*8 filename 
>       integer ncid, ierr
>       filename = 'test.nc'
> C
> C 'test.nc' does exist.
> C 
>       print *,'trying to open test.nc'
>       ncid = ncopn(filename,0,ierr)
>       print *,'ncid = ', ncid
>       print *,'ierr = ', ierr
>       call ncclos(ncid,ierr)
>  
> C
> C 'test.n' doesn't exist.
> C 
>       filename = 'test.n'
>       print *,'trying to open test.n'
>       ncid = ncopn(filename,0,ierr)
>       print *,'ncid = ', ncid
>       print *,'ierr = ', ierr
>       call ncclos(ncid,ierr)
>       stop
>       end
> 
> 
> When I run this on a J90, here's the output:
> 
>  trying to open test.nc
>  ncid = 0
>  ierr = 0
>  trying to open test.n
> Unable to open file: test.n
> Error Explanation:
>  ncid = -1
>  ierr = 0
> ncclose: -1 is not a valid cdfid
> 
> 
> Note that "ierr" is '0' in both cases, even though in the second case
> the file doesn't exist.
> 
> I ran this same program on an SGI, using the same version of netCDF,
> and here are the results:
> 
> chiquita[15] ./a.out
>  trying to open test.nc
>  ncid =            0
>  ierr =            0
>  trying to open test.n
> ncopen: filename "test.n": No such file or directory
> 
> 
> So, apparently it quits right after trying to open the non-existent
> file and doesn't reach the second set of print statements.
> 
> My question is, what should the correct behavior be?

As you use it, the netCDF library should print an error message and then
abort the program.

> It seems to me
> it's doing the right thing on the SGI by quitting after it fails on
> the file open.

That's correct.

> If so, then is this a potential bug on the Cray?  Or
> could this be something to do with the fact I'm using f90?

It's a bug in the netCDF library on a CRAY MPP system: the netCDF global
variable `ncerr' isn't getting set to NC_SYSERR when a file open failure
occurs on a CRAY MPP system.

I've enclosed a patch that should fix the problem.  You should apply the
patch to the file `xdrffio.c' in the `libsrc/' subdirectory.

Please let me know if this helps.

--------
Steve Emmerson   <address@hidden>


--------Begin patch to xdrffio.c
Index: xdrffio.c
===================================================================
RCS file: /upc/share/CVS/netcdf/libsrc/xdrffio.c,v
retrieving revision 2.7
diff -c -r2.7 xdrffio.c
*** 2.7 1996/06/27 19:38:48
--- xdrffio.c   1997/01/16 23:02:45
***************
*** 358,363 ****
--- 358,364 ----
                        strlen(_fdc_errlist[stat.sw_error - 5000]));
                write(2,mess2,1); /* end of line */
                /* nc_serror("filename \"%s\"", path) ; */
+               ncerr = NC_SYSERR;
                return (-1);
        }