Re: STL and Mozilla XPCOM

[Martin Sebor <sebor at roguewave dot com>]

Joe Buck wrote:
> 
> > i'm having a problem with a "&*something" idiom in the GNU STL,
> > which occurs here..
> 
> It's not specific to the GNU implementation.  You'll find common uses
> of such code; a number of authors writing on the STL recommend its use.
> 
> > .. maybe someone could explain the rationale behind this idiom?
> 
> An iterator is not necessarily a pointer, but in certain contexts
> (e.g. vectors) we are promised that the elements are in contiguous
> memory.  To get to this memory with an ordinary pointer, &*iterator
> is used.  For example, let's suppose I use a vector<char> named v to hold
> a null-terminated string, and then I want to pass that string to a
> C function, such as strcpy.  I can't use v.begin() as the argument,
> as it is not necessarily a pointer.  What I can use is &*v.begin() .

Right, except that if the iterator's value_type defines operator&()
in an unconventional way or makes it inaccessible, both of which
it is currently allowed to do, this idiom breaks. A safer approach
would be to use reinterpret_cast like so:

    extern Iterator __cur;
    typedef typename iterator_traits<Iterator>::pointer Pointer;
    construct (reinterpret_cast<Pointer>(
               &reinterpret_cast<char&>(*__cur)),
               *__first);

This must work since the iterator's operator*() is required to return
value_type&.

Regards
Martin