nemesis <nam### [at] gmailcom> wrote:
> more
> importantly, he forgot that objects in C++ are represented by pointers. 
> By removing all of the pointers he has removed all of the allocation but 
> also any chance of using subtypes and/or inheritance. In other words, 
> this is not OOP at all. He has just used a struct.

> Let's try an honest C++ equivalent that does support OOP, i.e. you can 
> derive a new kind of complex number from the Complex class and use it
> interchangeably because they will have the same uniform run-time 
> representation of a *pointer* to an object:

  Everything was fine until this point. However, this is starting to
sound more like he is deliberately trying to come up with a contrived,
yet plausible-sounding way of artificially making the C++ version slower.

  I find it rather ironic (and hypocritical) to do that immediately after
he argued that the straightforward iterative version in C++ had an "unfair"
advantage over the proposed Ocaml version which used recursion. In other
words, he is arguing that the first Ocaml version was slower because the
wrong (and inefficient) methodology for the given task was used. Then he goes
right ahead and gives us the completely wrong (and inefficient) methodology
in the "abstracted version" of the C++ program, just to artificially make
it slower.

  The original article didn't say anything about "object-oriented programming".
It only talked about abstracting away the complex number type. Even if you
could argue that for an implementation to be "OOP" it *must* support dynamic
binding, he is (probably deliberately) missing the point: The point was not
to make the program object-oriented, but to abstract away the complex number
type. In other words, to make it more modular. And the suggested C++ class
does exactly that, and it does so in the way that the language was designed.

  Maybe he wanted to point out that in Ocaml you cannot do the same, ie.
create classes which are *not* dynamically bound, and thus the comparison
is "unfair". However, if Ocaml lacks such capabilities, that's hardly C++'s
problem. The goal was to abstract away the complex number type as efficiently
as possible, nothing else. His argument about "OOP" is completely flawed.

-- 
                                                          - Warp

Post a reply to this message