automatic subtyping/intensions
Laurent Martelli
martelli@iie.cnam.fr
09 Feb 1999 00:43:52 +0100
>>>>> "Tril" == Tril <dem@tunes.org> writes:
Tril> Mozzie is still not an intensional language, because it uses
Tril> traditional data structures for objects (but it is still
Tril> very interesting). What I mean by intensional is that if
Tril> two objects have the same semantics, they are equivalent.
Tril> That is, they are treated exactly the same by the system no
Tril> matter what implementation they have. The data structure or
Tril> functions used to access and manipulate information related
Tril> to this object is irrelevant, only the meaning is important.
Tril> This meaning is encoded by declarative statements in a logic
Tril> language, a specification. In my ideal system, as long as
Tril> the statements are true of an implementation, the
Tril> implementation (automatically) "is a" subtype of the set of
Tril> all objects with the specification. TMF: Does this help
Tril> clarify? All others: please flame if you have no idea what
Tril> I am talking about, especially if you disagree or have a
Tril> better idea :)
I agree that semantics is what matters. Rather than relying on having
methods or attributes with the same name or signature à la Smalltalk.
But I think that it is wrong to talk about subtyping between a
specification and an implementation. I think that subtyping is a
behavioral notion (and so do Liskov and Wing in [1]).
1 - B. Liskov and J. Wing: "A Behavioral Notion of Subtyping", ACM
Transactions on Programming Languages and Systems, Number 16(6),
pp. 1811-1841, ACM Press, 1994.