NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
NOTICE: This version of the NSF Unidata web site (archive.unidata.ucar.edu) is no longer being updated.
Current content can be found at unidata.ucar.edu.
To learn about what's going on, see About the Archive Site.
Hi John Sheldon, hi netcdfgroup,
thanks for the comments and corrections. I actually encountered the Problem 2
(see appended mail at the end of this mail) right
after I submitted the patch. The tests I did before looked alright ...
I am sorry that I didn't announce the problem in the mailing list but
since there was no reaction on my posting I assumed it to be
unimportant for others.
I am sorry about that simple mistake. Naturally it was luck that it
worked with the first tests. A character byte containing a space
character is not a correctly terminated string and so the assignment
to the string pointer had to result in "wrong" nonsense strings.
Your correction is simpler than my first version so I add it at the
end of the mail as a patch for the netCDF group members.
Problem 1 is -- I assume too -- a problem for Cray.
To the importance of the patch:
I did a lot of testing especially on Cray file systems with concated
partitions and found that twiggling the switches can yield remarkable
speed gains (for a test programm writing an netCDF-array with one
statement i.e. most likely sequential). (-p and -q options of
assign). The tests aren't finished yet but if you like I can send you
the results.
Thanks for the correction and good luck
Olaf R. Heudecker
Attached:
- original mail from John Sheldon
- new corrected patch (not extensivly tested!)
----------
On Tue, 1998-4-14 11:56:09 -0400, <jps@xxxxxxxx>
John Sheldon wrote:
> Hi-
>
> We've been trying to apply Olaf Heudecker's "cray-patch" to netCDF 3.4
> on our IEEE T90 (uname -a = "sn7203 t90 9.1.0.1 gfd.11 CRAY TS"), but
> have encountered two problems during "make test":
>
> 1. With NETCDF_FFIOSPEC set to "cachea:112:7", the following cmp fails:
> cmp testfile.nc testfile_nc.sav
> testfile.nc testfile_nc.sav differ: char 391, line 3
> Make: "cmp testfile.nc testfile_nc.sav": Error code 1
> If NETCDF_FFIOSPEC is set to "cache:112:7", this test passes. Octal
> dumps of the two files show 2 bytes to be different in what appears
> to be the header part of the file; ncdump-ing the files show no
> difference. This may not be a problem in practice, but it did not
> occur with the basic 3.4 distribution.
>
> 2. In a later test, with NETCDF_XFFIOSPEC *not* set to anything, we get:
> *** Testing nc_get_var1_text ...
> lib-1503 ./nc_test: UNRECOVERABLE library error
> Attributes in the assign environment file have an unreadable format.
> Make: "./nc_test" terminated due to signal 6 (core dumped)
> If we set NETCDF_XFFIOSPEC to nothing ("setenv NETCDF_XFFIOSPEC"),
> the test passes. We think that the problem is that "xtra_assign"
> really points to a single character, not a null-terminated string as
> it should in this case. Although the resulting string "assignstr"
> seems fine (as does the ".assign" file), the fact that "xtra_assign"
> is not null-terminated somehow clobbers another area of memory when
> it is used, resulting in an eventual core dump. The fix is to change
> those two lines to:
>
> char emptystr='\0';
>
> Now "xtra_assign" will point to a null-terminated string when getenv
> returns NULL, although it's only one character long. (Thanks to Hans
> Vahlenkamp for figuring this out.)
>
> Problem #2 seems solved, but problem #1 is probably something for Cray
> to look at. In any case, we thought that you would want to know...
>
> John P. Sheldon
> (jps@xxxxxxxx)
> Geophysical Fluid Dynamics Laboratory/NOAA
> Princeton University/Forrestal Campus/Rte. 1
> P.O. Box 308
> Princeton, NJ, USA 08542
> http://www.gfdl.gov
>
> (609) 987-5053
> --
> "We are Microsoft. Resistance Is Futile. You Will Be Assimilated."
> --
---------- patch for ffio.c to include NETCDF_XFFIOSPEC option
-- for more explanation netCDF-group Date: Tue, 17 Mar 1998
-- or email oheudeck@xxxxxxxxxxxxxxxxxx
---- patch follows: apply with patch ffio.c < patchfile
------------ start of patchfile -----------
*** ffio.c.save Wed Apr 15 11:49:54 1998
--- ffio.c Wed Apr 15 11:49:54 1998
***************
*** 3,8 ****
--- 3,10 ----
* See netcdf/COPYRIGHT file for copying and redistribution conditions.
*/
/* $Id: ffio.c,v 1.48 1998/03/07 06:46:06 davis Exp $ */
+ /* addition by O. Heudecker, AWI-Bremerhaven, 12.3.1998 */
+ /* added correction by John Sheldon and Hans Vahlenkamp 15.4.1998*/
#include "ncconfig.h"
#include <assert.h>
***************
*** 16,21 ****
--- 18,25 ----
#include <ffio.h>
#include <unistd.h>
#include <string.h>
+ /* Insertion by O. R. Heudecker, AWI-Bremerhaven 12.3.98 (1 line)*/
+ #include <fortran.h>
#include "ncio.h"
#include "fbits.h"
***************
*** 477,487 ****
{
ControlString="bufa:336:2";
}
!
if(fIsSet(ioflags, NC_NOCLOBBER))
fSet(oflags, O_EXCL);
! fd = ffopens(path, oflags, 0666, 0, &stat, ControlString);
if(fd < 0)
{
status = errno;
--- 481,515 ----
{
ControlString="bufa:336:2";
}
! /* Insertion by Olaf Heudecker, AWI-Bremerhaven, 12.8.1998
! to allow more versatile FFIO-assigns */
! /* begin new */
! {
! char assignstr[128];
! char emptystr='\0';
! char *xtra_assign;
! xtra_assign = getenv("NETCDF_XFFIOSPEC");
! if(xtra_assign == NULL)
! {
! xtra_assign=&emptystr;
! }
! if (strlen(ControlString)+strlen(xtra_assign)+21 > 127) {
! /* Error: AssignCommand too long */
! status=E2BIG;
! goto unwind_new;
! }
! sprintf(assignstr,"assign -F %s %s f:%s",ControlString,
! xtra_assign,path);
! /* printf("1 assignstr=>%s<\n",assignstr); */
! /* Aufruf der Fortran-Routine assign von C aus */
! ASSIGN(_cptofcd(assignstr,strlen(assignstr)));
! }
! /* end new */
if(fIsSet(ioflags, NC_NOCLOBBER))
fSet(oflags, O_EXCL);
! /* Orig: fd = ffopens(path, oflags, 0666, 0, &stat, ControlString); */
! fd = ffopen(path, oflags, 0666, 0, &stat);
if(fd < 0)
{
status = errno;
***************
*** 559,566 ****
{
ControlString="bufa:336:2";
}
! fd = ffopens(path, oflags, 0, 0, &stat, ControlString);
if(fd < 0)
{
status = errno;
--- 587,620 ----
{
ControlString="bufa:336:2";
}
+ /* Insertion by Olaf Heudecker, AWI-Bremerhaven, 12.8.1998
+ to allow more versatile FFIO-assigns */
+ /* begin new */
+ {
+ char assignstr[128];
+ char emptystr='\0';
+ char *xtra_assign;
+ xtra_assign = getenv("NETCDF_XFFIOSPEC");
+ if(xtra_assign == NULL)
+ {
+ xtra_assign=&emptystr;
+ }
+ if (strlen(ControlString)+strlen(xtra_assign)+21 > 127) {
+ /* Error: AssignCommand too long */
+ status=E2BIG;
+ goto unwind_new;
+ }
+ sprintf(assignstr,"assign -F %s %s f:%s",ControlString,
+ xtra_assign,path);
+ /* printf("2 assignstr=>%s<\n",assignstr); */
+ /* Aufruf der Fortran-Routine assign von C aus */
+ ASSIGN(_cptofcd(assignstr,strlen(assignstr)));
+ }
+ /* end new */
! /* Orig: fd = ffopens(path, oflags, 0, 0, &stat, ControlString); */
! fd = ffopen(path, oflags, 0, 0, &stat);
!
if(fd < 0)
{
status = errno;
------------------ snip -------- end of patch file
netcdfgroup archives: