Re: Real OO

[ncohen@watson.ibm.com (Norman H. Cohen)]

In article <TYNOR.96Apr4110411@twratl.atlanta.twr.com>,
tynor@atlanta.twr.com (Steve Tynor) writes: 

                                                        I typically see
|> values in the 10-20% range. That means 80-90% of calls are optimized
|> to static calls

In evaluating such statistics it is important to remember that in Eiffel
ALL calls on nonfrozen features are nominally dispatching calls, even the
ones that have an obvious target.  In other words, in computing the
percentage of calls to which this optimzation applies, Steve Tynor is not
just considering calls (it's not clear whether we're counting run-time
executions of calls or call sites here) that would nominally have been
dispatching calls in Ada and counting the fraction of them that can be
optimized into nondispatching calls.  He's also taking all the calls that
would have been written as nondispatching calls in Ada (or C++) in the
first place, and including that number to both the numerator and
denominator.  It is not surprising that this gives a fraction closer to
one than the fraction that was observed in Calder and Grunwald's C++
measurements.

|>                 (and without requiring the programmer to cripple the
|> reusability of his classes by "freezing" their features).

The Ada approach, of course, is not to freeze features, but to allow
certain variables and function results to be declared as belonging to a
known DYNAMIC type.  The same (unfrozen) subprogram that is invoked as a
dispatching subprogram at call sites where the actual parameter is of
unknown dynamic type is invoked as a nondispatching subprogram at call
sites where the actual parameter is of a known dynamic type.

--
Norman H. Cohen    ncohen@watson.ibm.com