With that clarification, do you still find the option of *string* array attributes to be more objectionable than scalar *strings* with delimiters?
I do. CHAR vs StrIng is a relatively low level implementation detail for encoding text. So a netcdf lib ( with no concern with CF) can transparently convert either one into a native string type. For example py-netCDF4 can present the user with a Python string object in either case. So a delimited CHAR or String would look exactly the same to client code. And CF aware client code can now deal with the delimiters appropriately. In order for a library to present an array of strings the same way as a delimited CHAR array, it would need to be CF aware at a low level. I think the basic principle should be to not add netCDF4-only features until netCDF4 can be assumed — presumably in CF2
