Provide a standard way to do data packing (which does not break existing, non-standard methods)

[edwardhartnett]

By packing I mean taking a 32-bit or 64-bit float array, calculating a scale and offset, and storing values as short ints. In this way a 32 or 64 bit data type can be stored in 16 bits.

Of course people have been doing this for a long time, but we have no official way to do this in the netCDF library. I suggest we add one.

So lets say I can do:

/* Define a short var. */
nc_def_var(ncid, "data", NDIM3, dimid, NC_SHORT, &varid)

/* Tell library this var is a double, packed to a short, using offset and scale factor. */
nc_def_var_packing(ncid, varid, NC_DOUBLE, offset, scale_facror)

That defines a var that is a double, stored as a short. When I call nc_put_vara_double() for this var, it automatically packs the data and stores the shorts. When I call nc_get_vara_double(), it unpacks the data. (For calls to other types, it is treated as a double, not a short.)

Then we define some attribute names to store the scale and offset.

Easy-peasy, lemon squeezy, as the kids used to say. ;-)

This is part of #1545

[edwardhartnett]

I think this feature is more appropriate for the CCR, so I will close this issue.