[netcdfgroup] variable without dimension

• To: <netcdfgroup@xxxxxxxxxxxxxxxx>
• Subject: [netcdfgroup] variable without dimension
• From: "Choonghwan Lee" <cholee@xxxxxxxxxxxx>
• Date: Fri, 11 Apr 2008 10:25:34 -0500

Hi,

While converting HDF-EOS2 files into HDF5 files that can be read by netCDF4,
I got a question. I'm implementing the converter with HDF5 C API, and
checking the generated file by using netCDF4 C API and ncdump.

Since I'm using HDF5 API, I can fabricate a variable that does not have
dimensions. I think netCDF4 API does not allow this case because variables
should be defined with dimensions. For those variables, nc_inq_var() gives
invalid dimension ids. It makes ncdump terminate with some messages.

Since those variables do not have actual scientific data, I don't want to
simply generate dimensions for each variable if I can avoid. However, I
don't want to generate a file that loses that information. I was wondering
how I can achieve this goal.

I don't think generating dimensions for each variable is the best solution
because the fact that those dimensions are fakes cannot be detected. Since
netCDF4 uses HDF5, I guess netCDF4 can detect the number of elements even
though dimensions don't exist. So, I think netCDF4 may be able to generate
fake dimensions on-the-fly at the time nc_inq_var() is called. I thought
netCDF4 may support this feature in that nc_open() does not complain when it
opens those variables; only inquiry functions give invalid ids.

Also, if there is any semantic that makes netCDF4 skip an HDF5 dataset, this
would be nice, too. I know that netCDF4 recognizes "NAME" attribute, and it
alters the way netCDF4 treats this variable. I'm wondering if I can
similarly exploit special semantics.

Any general suggestions and ideas will be appreciated. If my question was
vague, please let me know. Thank you!