...
>The debate centers on two different methods for extending a "standard". The
>two methods both achieve the same end. One method (the use of multi-dimensional
>coordinate variables) is clean and elegant, but is in some sense not backwards
>compatible. The other ("referential attributes") patches the old standard in a
>backwards compatible way but is not quite as clean...
I agree that the multi-dimensional coordinate variables are cleaner.
I don't understand how either one is not backwards compatible.
Both methods are backwards compatible in that they don't require any change
to existing datasets. Existing applications also needn't be changed to deal
with *existing* datasets. Applications that would operate on generalized
coordinates would need to be modified anyway. In both schemes, one can determine
from the dataset if it is this newer case or not.
