Re: [netcdfgroup] nc_put_varm

• To: Wei-keng Liao <wkliao@xxxxxxxxxxxxxxxxxxxxx>
• Subject: Re: [netcdfgroup] nc_put_varm
• From: Russ Rew <russ@xxxxxxxxxxxxxxxx>
• Date: Wed, 23 Apr 2014 11:05:54 -0600

Hi Wei-keng,

> My question is about how to set imapp argument in nc_put_varm().
> 
> At first the description of nc_put_varm() in C interface guide is 
> inconsistent with the
> document generated by doxygen.
> 
> In C interface guide, it says "nc_put_varm() function will write a variable 
> of any type,
> including user defined type"
> http://www.unidata.ucar.edu/software/netcdf/docs/netcdf-c/nc_005fput_005fvarm_005f-type.html#nc_005fput_005fvarm_005f-type
> 
> But in doxygen document, it says "nc_put_varm() function will only write a 
> variable of
> an atomic type; it will not write user defined types."
> http://www.unidata.ucar.edu/software/netcdf/docs/group__variables.html#ga48533a5a1ab3c2e883e6c862720b94ab

The older C interface guide describes the intended behavior, but the
doxygen document is currently correct, as there is an explicit test in
libdispatch/dvar{get,put}.c to disallow user-defined types for _varm
functions: 

   if(vartype > NC_MAX_ATOMIC_TYPE)
        return NC_EMAPTYPE;

Since the library seems to work OK for similar _vars functions for
user-defined types, I think the intended behavior for _varm functions
can be restored without too much effort (but the fix won't be in the
upcoming 4.3.2 release).  I've created an issue to track the fix:

  https://bugtracking.unidata.ucar.edu/browse/NCF-300

> I may not fully understand the description for argument imapp in the C 
> interface guide.
> When I use nc_put_varm(), do I set the values in imapp[] to the distances in 
> number of elements
> or bytes? I tested a program (extracted below) that uses both 
> nc_put_varm_int() and nc_put_varm().
> At first, I defined a 1-D integer array variable of size 2 in netCDF file.
> The results are different between using nc_put_varm_int() and nc_put_varm().
> It looks like in nc_put_varm(), imap[] is in the units of bytes and in 
> nc_put_varm_int(), imap[]
> is in number of integers.
> 
>    int       buf[5]    = {0,1,2,3,4};
>    size_t    start[1]  = {0};
>    size_t    count[1]  = {2};
>    ptrdiff_t stride[1] = {1};
>    ptrdiff_t imap[1]   = {4};
> 
>    nc_put_varm_int(ncid, varid, start, count, stride, imap, buf);
> or
>    nc_put_varm    (ncid, varid, start, count, stride, imap, buf);
> 
> 
> Command ncdump shows the contents of variable are:
> for nc_put_varm_int()
>  VAR = 0, 4 ;
> for nc_put_varm_()
>  VAR = 0, 1 ;
> 
> Please clarify if this is correct behavior.
>
> In the C interface guide, nc_put_varm requires "the type of the data
> in memory must match the type of the variable". It sounds to me imap
> should be the distances in units of element numbers, e.g. number of 
> integers in the above example.

I think you're right, the imap[] argument for nc_put_varm() should be in
terms of number of elements, not number of bytes.  Fixing that bug has
been included in the issue above.

Thanks for reporting the problems!

--Russ

• References:
  • [netcdfgroup] nc_put_varm
    • From: Wei-keng Liao