From: Philippe A. Bouchard (philippeb_at_[hidden])
Date: 2002-07-31 08:37:19
> > [...]
> > Ok. It's a good point and I agree. But:
> > - Allocation with operator new (size_t, xmm const &) will still be
faster
> > than:
> > - an extra allocation for a counter;
> > - amortized allocations in pool (maybe this is a negligible fact but
> > still is a fact :).
>
> It's like ripping the seats out of a Ferrari to get an extra 0.2 mph.
Sure,
> it makes it faster, but do you really want to drive it? ;)
Yeah luckily:
- that virtual destructors return the address of the real type;
- sizeof() is not virtual;
- typeid() is nearly useless... except for compiler developers of course;
- ((char *) (Y *) (Derived *) 1 - (char *) 1) is not a compile-time
constant;
It is not in my own interests to have sizeof(ptr<int>) greater than
sizeof(int *) even if the speed would be faster than shared_ptr<int> with
operator new (size_t, xmm &) allocations, so I'll go back to my
mutable_ptr<int> where sorts are not necessary... like with configuration
files...
I was great talking to you people. I learned much and I hope it is
reciprocal for the reinforcement of your arguments.
Thank you,
Philippe A. Bouchard