I think I’ve found a bug in the UML 2.0 OCL spec.

It’s actually a well-known issue for C++, documented e.g. in the C++ FAQ.

It’s not so obvious in Java, since Java didn’t have Templates for a long time.

The UML 2.0 OCL spec says on page A-22, Definition A.27 (Type Hierarchy)

If t’ ≤ t then Set(t’) ≤ Set(t)

With ≤ having the meaning “conforms to” for types.
[Update: newer UML 2.0 OCL document, the latest I could find, page 38 still says the same]

Any good C++ programmer should know that this doesn’t work. From the C++ FAQ “Is a parking-lot-of-Car a kind-of parking-lot-of-Vehicle?”:

“A Bag-of-Apple is not a kind-of Bag-of-Fruit.” If a Bag-of-Apple could be passed as a Bag-of-Fruit, someone could put a Banana into the Bag, even though it is supposed to only contain Apples!

Note that the converse, Set(t) ≤ Set(t’), also doesn’t hold. While you can put Apples into a Bag-of-Fruit just fine, you may still find a Banana in there.

This has an obvious reason: when passing data to the object it can be “downcasted”, and when retrieving data it can be “downcasted”, too. But it can never be upcasted.

This should occur in OCL, too. A “Set(Apple)” doesn’t conform to “Set(Fruit)” by the usual interpretations that I can call “put(Banana)” on any “Set(Fruit)”!

Can anyone enlight me, if this really is a bug in the OCL spec, maybe has already been corrected, or if there is some magic in OCLs definition of “conforms to” or interfaces that solves this problem?

(Yes, I think there might be something in OCL that solves this problem, for example if this hierarchy is only used in condition checking, then in fact all operations are “get” operations… but I’ve just started reading on OCL, I don’t know much about its expressiveness yet…)

[Update: Onne Gorter has some good thoughts on this. He basically notes that this is not a problem if you don’t have any references to the “uncasted” list. For example if the cast is a function like in a functional language that actually returns a new set. I’ll have to consider this for OCL, but given that OCL and UML are somewhat designed for Java - and Java suffers from this problem, too - I’m not so sure about it.]