Re: [boost] Re: container design (Was: std::complex design (Was: N-Dimensional array))

From: Howard Hinnant (hinnant_at_[hidden])
Date: 2002-03-08 17:07:23

On Friday, March 8, 2002, at 12:51  PM, Peter Dimov wrote:

> No. One way to solve it (that I see) is
>
> template<class F> vector::vector(size_type n, F f);
>
> that will allocate memory and use f(this->_Ptr, n) to initialize it.

The more I think about this, the more I'm beginning to like it.

Official Requirements on F(T* p, size_type n):

1.  It must construct and initialize all elements in the range [p, p+n).

2.  It must have strong exception safety.  vector will not be able to 
recover from an exception when the range is partially constructed.  It 
is all or nothing.

Notes:

1.  This could probably find all kinds of uses other than the one were 
aiming at.

2.  There's nothing to stop somebody from ducking initialization rules 
if that's what he really wants to do:

struct uninitialized
{
        void operator()(double*, size_t) {}
};

std::vector<double> v(10000, uninitialized());
// use v at your own risk here

As Peter said, could provide an excellent interface to legacy C code.  
Though it may require some gymnastics to map to the C interface, 
especially when there are multiple arrays.

3.  This doesn't have the allocator compatibility problems of passing 
ownership of an array.

4.  The standard would not officially sanction something like 
uninitialized_flag, and yet people can still easily manage it if they 
want to.

-Howard