|[higgins-dev] Data Model: Allow zero-valued attributes?|
Mike resurrected this topic in http://dev.eclipse.org/mhonarc/lists/higgins-dev/msg03734.html
Currently we say there are 0..1 instances of a given attribute on a subject, and for any attribute, there are 1..N values. This question is: Should we allow an attribute to contain zero values, and if we do, how does a zero-valued attribute differ from the absence of an attribute?
If there's a difference between the an attr with zero values, and the absence of an attr, should we dictate that difference, or state that it's application-defined?
Drummond points out that elsewhere (XRI) they have explicitly specified similar presentations of data as equivalent.
I point out that in the face of access control, it would be extremely difficult to rely on any stated difference between the two.
There is also a side-question which is: Assuming we don't allow people to populate IdAS with a valueless attribute, should we allow a CP to return a valueless attribute? This might happen due to access control policies which state that you are allowed to know that an attribute is present, but you are not allowed to see its values.