Re: [boost] 5 Observations - My experience with the boost libraries

Subject: Re: [boost] 5 Observations - My experience with the boost libraries
From: Emil Dotchevski (emildotchevski_at_[hidden])
Date: 2010-03-23 23:52:01

On Tue, Mar 23, 2010 at 7:57 PM, Stefan Seefeld <seefeld_at_[hidden]> wrote:
> On 03/23/2010 10:45 PM, Emil Dotchevski wrote:
>> On Tue, Mar 23, 2010 at 6:56 PM, Stefan Seefeld<seefeld_at_[hidden]>
>>  wrote:
>>> which
>>> you can not gain back "in the implementation". So what meta-programming
>>> tricks are there left to play ?
>> Pointers to incomplete types are type-safe but yes, in general, I'm
>> talking about balancing compile-time type safety and physical
>> coupling.
>> The "proper" way to do this in C++ is to use abstract base classes;
> Sorry, no. An "abstract base class" is something specifically tied to the
> Object Oriented Programming paradigm, and thus, runtime binding.
> Meta-programming, on the other hand, is all about compile-time binding,
> which you can't get with abstract base classes.

Compile-time binding is invaluable tool indeed, I'm not saying that
you can replace it with something else.

My point is that losing some physical coupling (and sometimes,
compile-time type safety) is an option that should be considered when
designing any interface.

>> except that an interface defined in terms of C-style functions is more
>> abstract than a C++ abstract base class because calling a C-style
>> function doesn't necessarily result in a virtual function call (yet it
>> *can* call a virtual function internally, if that makes sense.)
> You are comparing apples and oranges. If you want the kind of abstraction
> that abstract base classes give you, i.e. (runtime) polymorphism, you surely
> want to operate with some form of function pointers. And these are very
> similar to virtual functions.

Yes, you do need to arrive at something like a function pointer if you
want to call a function dynamically. No, the virtual function dispatch
defined by C++ is not the only way to do that.

Secondly, in a C-style interface you don't know if a particular
function is dispatched dynamically or not. This is true for virtual
function interfaces too, except that if you want to change a function
to non-virtual or vice versa you'll force the user code to recompile.

You could use pimpl but what's the point? Just so you can write
x.foo() instead of foo(x)? :)

> I can't believe we are having this discussion on this list...

Yes, the discussion is sort of pointless, because it won't lead to
anything useful.

Emil Dotchevski
Reverge Studios, Inc.
http://www.revergestudios.com/reblog/index.php?n=ReCode