when should sends be virtual?

[Rainer Blome]

maybe part of this mess is caused by the fact that my first experiences
with oop were (turbo)pascal and c++:

urs, you wrote:

> One problem with letting the sender decide whether a call is virtual
> (dispatched) or not is that it breaks encapsulation.  The receiver should
> decide what happens with a message send, and statically binding the send
> (e.g. to invoke an overriden method) is like writing instance variables
> directly.

in turbopascal/c++ it's true that virtual calls are dispatched and the
others are not (statically bound).  but that doesn't mean that virtual
equals dispatched.  dispatching or not is somewhat of an implementational
difference, i wasn't talking about that.  virtual as i see it is a matter
of semantics (see replies to guenter and randy).

> In other words, you can't reason about the object's properties and
> invariants anymore by simply looking at the object's definition.

???


rainer