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

From: Howard Hinnant (hinnant_at_[hidden])
Date: 2002-03-09 12:23:33

On Saturday, March 9, 2002, at 12:04  PM, rwgk wrote:

> Would you have a code sample that could help me avoid
> learning about potential pitfalls the hard way?

What?  You didn't like:

template <class F : !is_convertible<F, value_type>::value>
vector(size_type n, F init, const Allocator& = Allocator());

:-)

The "do-the-right-thing clause" refers to 23.1.1 paragraphs 9-11.  
You're in good company Ralf.  This type of bug bit the best of them just 
before the standard was ratified.

A typical implementation might look something like (warning, unchecked 
pseudo code):

template <class F>
vector(size_type n, F init, const Allocator& a = Allocator())
{
        choose_F_init(n, init, a, int2type<!is_convertible<F, 
value_type>::value>());
}

Then you've got two overloads for choose_F_init:

1.  Takes a int2type<false>.  F *is* convertible to value_type.  This is 
a mistaken binding.  Redirect to the code that implements 
vector(size_type, value_type, allocator); (converting init to a 
value_type of course).

2.  Takes a int2type<true>.  F will not convert to value_type.  This is 
the real deal, do whatever you were going to do in the first place with 
(n, init, a).

Now isn't:

template <class F : !is_convertible<F, value_type>::value>
vector(size_type n, F init, const Allocator& = Allocator());

*much* easier?! :-)  The compiler does the redirecting for you, but 
under your precise instructions.

-Howard