Re: Question about 'ncvgt'

Hi Hui,

> I am using netcdf function in fortran to read some data out from the netcdf 
> file.  In my netcdf file, my data type is short.  So in my fortran program, 
> I set the 'integer' type and use the function 'ncvgt' to read the data I 
> want.
> I do  not know why the output is much much biger than the real values. In 
> other words, I can not output the correct data.
>   The following is my script of fortran program and I could not find the 
> problem.
> ******************************************
>   program main
>         parameter   (ndims=3)
>         parameter   (nlat=67,nlon=65)
>         integer      datain(nlon,nlat)
>         integer      ncid,iv
>         integer      rcode
>         integer      start(3), edges(3)
>         data start/1,1,1/
>         data edges/nlon,nlat,1/
>         include 'netcdf.inc'
>         iv=5
>         ncid  =  ncopn('../cal.ncep.1dom.nc',ncnowrit,rcode)
>         DO k =1,124
>         start(3)=1
>         call ncvgt(ncid,iv,start,edges,datain,rcode)
>         print *,datain(10,10)
>         enddo
>         end
> ******************************************************************
> But when I tried to get a single value instead of a array, it is fine.
> 
> Can you give me some hint?

You are using the old version 2 Fortran-77 interface, which required
that the size of the data on disk matched the size of the data in your
program.  In other words, if you use ncvgt to read in data from 16-bit
integers, you had better be reading the data into an array of 16-bit
integers in your program.  So one possible fix for your problem is to
just change
        integer      datain(nlon,nlat)
to
        integer*2    datain(nlon,nlat)
if your Fortran compiler supports that notation to indicate an array
of short (16-bit) integers.  The Fortran-77 standard did not deal with
integer sizes and provided no standard way to declare 16-bit integers,
but many Fortran compilers support notations such as "integer*2" to
indicate 2-byte (16-bit) integers.

Alternatively, you could use the later version 3 f77 or f90
interface, both of which permit on-the-fly type conversion to 32-bit
integers when you read the data.  For example, in the f77 interface,
you would leave the datain declaration alone but read the data with
a call to

  ierr = nf_get_vara_int(ncid, iv, start, edges, datain)

as documented in the Fortran-77 Users Guide

  http://www.unidata.ucar.edu/packages/netcdf/docs/netcdf-f77.html

--Russ