If a variable in an .cdl file contains a value of "-0" or "+0", ncgen
produces a syntax error. The reason is seems to be the following in ncgen.l:
[+-]?[1-9][0-9]*[lL]?|0[lL]? {
I am not familar with lex, but it seems to me, the author defined a long
constant as beginning with + or - and a number 1-9, or as a single 0 without
+ or -.
I don't know why this was done, so I don't want to change it.
The whole thing don't look like a problem, why should anyone insert a -0 in
his .cdl file?
BUT: At least on the Sun (SUN OS 4.1.3), ncdump often generates a -0!
(This seems to be a bug in SUNs output formatting routines, the MS-DOS Version
of ncdump don't have this problem).
Run the following little program:
/* ------------- Cut Here ----------------*/
#include <stdio.h>
#include <netcdf.h>
main()
{
double d=0;
int ncid, dimid, varid;
long start=0, count=1;
d *= -1;
ncid = nccreate("test.nc",NC_CLOBBER);
dimid=ncdimdef(ncid, "time",1);
varid=ncvardef(ncid, "velocity",NC_DOUBLE,1,&dimid);
ncendef(ncid);
ncvarput (ncid,varid,&start, &count,&d);
ncclose(ncid);
}
/* ------------- Cut Here ----------------*/
Now run "ncdump test.nc":
netcdf test {
dimensions:
time = 1 ;
variables:
double velocity(time) ;
data:
velocity = -0 ; // <---------- Here it is.
}
Try to run ncgen on the output and you get:
ncgen: test.cdl line 10: syntax error
Now I would like to change ncgen to accept the "-0" or ncdump to
produce at least "-0." like it does in attributes. Any ideas what's the
best solution?
Bernward Bretthauer
----- End Included Message -----
