[netCDF #CZF-138501]: NetCDF FILL_CHAR and FILL_BYTE switched?

Hi Joe,

In the C++ documentation, we said for ncbyte:

ncbyte
Used to declare values of type ncByte, for 8-bit integer data. (This is 
currently a typedef for unsigned char, but it may be changed to a typedef for 
signed char, so don't depend on the underlying representation.)

Although it's legal in C++ to assign a signed value to an unsigned type (and 
g++ -Wall doesn't even warn about it), it does raise questions why we did it 
this way.  The answer is that originally ncbyte was just a typedef for "char", 
and different platforms interpreted it as either signed or unsigned, depending 
on their default for "char" types in C.  Then we changed from netCDF-2 
interfaces which were not type-safe to netCDF-3 interfaces which were type-safe 
and permitted requesting conversion on-the-fly of numeric values from one type 
to a different numeric type when reading data.  This required for 
interoperability that we pin down an interpretation for ncbyte in order to 
support its conversion to int, for example.  We picked "signed", but the C++ 
interface was still based on the netCDF-2 C interface, so it was not changed.

Nevertheless, since we warned users in the documentation that the underlying 
type for ncbyte would change, the time has come to make this consistent with 
the C interface.  The next beta release will define ncbyte as signed char in 
the C++ interface.  If we don't get cries of anguish about breaking anything, 
this change will be in the 3.6.2 and subsequent releases.  Thanks for your 
comments.

--Russ

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



Ticket Details
===================
Ticket ID: CZF-138501
Department: Support netCDF
Priority: Critical
Status: Closed