I'm using the NetCDF interface on an SGI 4D35 running irix 4.0.4 and I recently
discovered a problem in reading and writing NC_BYTE variables through the
FORTRAN interface (sorry, NCBYTE variables - I usually use the C interface).
It seems that the fix FORTRAN_HAS_NO_BYTE in the fortran/irix.m4 file has
been set when it should not be, leading my bytes to be interpreted as
integers. The following program demonstrates the problem:
--------------------------------------
program test
#include <netcdf.inc>
integer cdf, var, rcode, dim(3), start, count, i
byte mydat(12), indat(40)
integer junk(3)
equivalence (junk, mydat)
data mydat/1,2,3,4,5,6,7,8,9,10/
type *, (junk(i), i=1,3)
start=1
count=10
cdf=nccre('test.cdf', NCCLOB, rcode)
dim(1)=ncddef(cdf, 'xdim', 10, rcode)
var=ncvdef(cdf, 'var', NCBYTE, 1, dim, rcode)
call ncendf(cdf, rcode)
call ncvpt(cdf, var, start, count, mydat, rcode)
call ncvgt(cdf, var, start, count, indat, rcode)
call ncclos(cdf, rcode)
type *, (indat(i), i=1,10)
stop
end
--------------------------------------
The output is (the first three values show that bytes are bytes) :
16909060 84281096 151650304
0 0 0 4 0 0 0 8 0 0
and ncdump gives :
netcdf test {
dimensions:
xdim = 10 ;
variables:
byte var(xdim) ;
data:
var = 4, 8, 0, 0, 6, 0, 0, 0, 0, 0 ;
}
If I modify jackets.c and remove FORTRAN_HAS_NO_BYTE, I get
16909060 84281096 151650304
1 2 3 4 5 6 7 8 9 10
which is correct (ncdump agrees).
The FORTRAN_HAS_NO_BYTE fix is also used in ultrix.m4 and unicos.m4. I know
nothing of unicos, but it would surprise me a great deal if there was no
byte type in ultrix FORTRAN.
If this is truly a bug and I am not mistaken, perhaps it can be incorporated
into the new release.
--
Peter Neelin (neelin@xxxxxxxxxxxxxxxxx)
Positron Imaging Laboratories,
Montreal Neurological Institute
