[boost] Re: looping construct - late thought

From: David Abrahams (dave_at_[hidden])
Date: 2004-04-30 15:12:51

John Torjo <john.lists_at_[hidden]> writes:

> Just a late thought: what if we were to make the internal data of
> crange<> const?
> I guess this should give an extra hint to the compiler:
>
> template< class type> struct crange {
> protected:
>    typedef crange<type> self_type;
>    typedef void (*incrementer)(const self_type&);
>    typedef bool (*at_end)(const self_type &);
>    typedef type & (*deref)(const self_type&);
>
>    crange( incrementer inc_func, at_end at_end_func, deref deref_func)
>        : m_inc_func(inc_func), m_at_end_func(at_end_func),
>        m_deref_func(deref_func) {}
> public:
>
>    operator bool() const { return !m_at_end_func(*this); }
>    const self_type & operator++() const { m_inc_func(*this); return
>    *this; }
>    type & operator*() const { return m_deref_func(*this); }
> private:
>    const incrementer m_inc_func;
>    const at_end m_at_end_func;
>    const deref m_deref_func;
> };
>
> What do you think?

Not much; my guess is that compilers very seldom use const object
declarations for optimization, but I have no data.  At least in the
case of vtables the user really has no reasonable way of mutating it
by casting away const, but I was only speculating about that, too.
Someone should measure.

-- 
Dave Abrahams
Boost Consulting
http://www.boost-consulting.com