Re: [boost] Re: [review] hash functions

From: JOAQUIN LOPEZ MU?Z (joaquin_at_[hidden])
Date: 2005-03-13 07:37:51

----- Mensaje original -----
De: David Abrahams <dave_at_[hidden]>
Fecha: Domingo, Marzo 13, 2005 4:52 am
Asunto: [boost] Re: [review] hash functions

> JOAQUIN LOPEZ MU?Z <joaquin_at_[hidden]> writes:
> 
> > ----- Mensaje original -----
> > De: David Abrahams <dave_at_[hidden]>
> > Fecha: Domingo, Marzo 13, 2005 0:52 am
> > Asunto: [boost] Re: [review] hash functions
> >
> >> Alberto Barbati <abarbati_at_[hidden]> writes:
> >> 
> >> > Ah! And there's one more case in which we are helpless, 
> although 
> >> I'm 
> >> > curious to see an implementation that exploits such latitude: 
> >> the 
> >> > wording used in the standard does not guarantee that the 
> mapping 
> >> from 
> >> > pointers to integers always produce the same value over time. 
> In 
> >> other 
> >> > words:
> >> >
> >> >   char* p = new char;
> >> >   size_t i1 = reinterpret_cast<size_t>(p);
> >> >   size_t i2 = reinterpret_cast<size_t>(p);
> >> >   assert(i1 == i2); // ???
> >> >   char* q1 = reinterpret_cast<char*>(i1);
> >> >   char* q2 = reinterpret_cast<char*>(i2);
> >> >   assert(p == q1 && q1 == q2);
> >> >
> >> > The standard guarantees that the second assert passes, but 
> says 
> >> nothing 
> >> > about the first assert.
> >> 
> >> There are other cases too.  reinterpret_cast could map all 
> >> pointers to
> >> the same integer.
> >
> > If you allow me to be a little pedantic, this is not the
> > case. reinterpret_cast<> guarantees back conversion, so it can't
> > possibly map every pointer into the same integer.
> 
>  4 A pointer can be explicitly converted to any integral type large
>  enough to hold it. The mapping function is implementation-defined
>  [Note: it is intended to be unsurprising to those who know the
>  addressing structure of the underlying machine. ]
> 
>  5 A value of integral type or enumeration type can be explicitly
>  converted to a pointer. 64) A pointer converted to an integer of
>  sufficient size (if any such exists on the implementation) and back
>  to the same pointer type will have its original value; mappings
>  between pointers and integers are otherwise implementation-defined.
> 
> There's no guarantee that the integer type will have sufficient size.

OK, with that provision in mind I buy your thesis now.

> 
> > As for the rest of your argument, I agree 100%.
> 
> Now you agree with all of it ;-)
> (I hope)
> 

Definitely!

Joaquín M López Muñoz
Telefónica, Investigación y Desarrollo