Re: [boost] Re: Yet another Variant class

From: Douglas Gregor (gregod_at_[hidden])
Date: 2002-06-29 08:59:15

On Friday 28 June 2002 01:50 pm, Itay Maman wrote:
> I think that variant_cast<> should be the only way for an application
> code to reach the T* pointer. Consequently, get_ptr() should be made
> private, although I must add an #ifdef-ed workaround for MSVC.

I'm happy with that. Personally, I would strongly discourage the use of 
variant_cast anyway because it is the only mechanism to access a variant that 
is not guaranteed to succeed at run-time. Visitors are typesafe as would be a 
switch-on-type construct. 
> > 1) Great Stack_Holder::swap :)
>
> I am wondering if there's a more efficient way for implementing an
> exception-safe swap() ?

For the general case, I don't think there is. There are tons of optimization 
opportunities for special cases:

  1) Same type: cast and swap
  2) Both incomplete types: swap pointers
  3) Both POD types: swap the bytes

> > 4) Is there any benefit to the use of copy/destroy function pointers
> > instead of a cascading 'if' built with a metarecursive function? The
> > function pointers impose an extra 8-byte overhead in the variant -- is
> > this overhead justified by an efficiency boost? The same question applies
> > to the visitor table.
>
> Well, we can use a single pointer for *both* clone/destroy. So the space
> overhead we should weigh here, is just 4 byte.
> As for Accept_table, it is a static local variable inside
> Variant::operator() so its cost is minimal.

I'll just try it :). My hunch is that it would be faster to use a cascading 
'if' than to perform a call through a function pointer, but only 
experimentation will tell for sure.

        Doug