[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: 951204: netCDF-2.4-beta4 comments

> Subject: 1995124 netCDF-2.4-beta4 comments
> Organization: CSIRO Division of Oceanography
> Keywords: 199512032150.AA10339

Hi Stephen,

> We are developing a complex ecological model, with many variables,
> using netCDF for data input and output. Last week we ran into trouble,
> exceeding the maximum number of attributes allowed in a cdl file.
> This became apparent when we added a few new variables to our model - each
> variable has about 10 attributes. We got up to abut 520 attributes,
> and ncgen fell over.
> 
> I down-loaded and installed NetCDF 2.4 Beta4, in an attempt to fix this,
> as I had noticed that the release notes state that:
> 
>     "We increased the sizes of MAX_NC_DIMS (from 32 to 100), MAX_NC_ATTRS 
> (from
>      512 to 2000), and MAX_NC_VARS (from 512 to 2000) to accommodate users who
>      are finding the old limits constraining."
> 
> Well, you almost did. The trouble is that the ncgen code sets its own idea
> of the maximum number of attributes, in the following line from
> ncgen/ncgen.h:
> 
>     #define MAX_NC_ATTS        500  /* max number of attributes in a netCDF */

Thanks for pointing out this inconsistency.

> I don't understand why it doesn't use the library-wide definition,
> as stated in the following line from libsrc/netcdf.h:
> 
>     #define MAX_NC_ATTRS 2000        /* max global or per variable attributes 
> */

The MAX_NC_ATTS macro used by ncgen was originally a temporary
implementation that I intended to replace with dynamic allocation of
attributes so that ncgen would have no limit.  If it's statically allocated,
it really should be more than MAX_NC_ATTRS in netcdf.h, since the latter is
*per variable*, whereas the ncgen limit is for the total of all attributes
of all variables.  Also, dynamic allocation of space for attributes would
keep the ncgen executable smaller.

> Could this minor inconsistency be cleaned up, or is there some reason for
> it which I am not aware of. In the meantime, I have edited ncgen.h for
> our purposes.

The only reason it hasn't been cleaned up yet is I forgot to put it on a
list of ncgen fixes.  I'll try to get it fixed before the final release of
netCDF 2.4, and increase MAX_NC_ATTS now to make it consistent with
MAX_NC_ATTRS.  Your fix to ncgen.h should be fine.

> Thanks for a very useful package (our model would be a nightmare without
> something like netCDF!).

You're welcome.  Thanks for the positive feedback.

______________________________________________________________________________

Russ Rew                                           UCAR Unidata Program
address@hidden                              http://www.unidata.ucar.edu