Re: [Fwd: basic_string<> - useless because of its poor performance]

[Carlo Wood <carlo at alinoe dot com>]

On Fri, Jul 06, 2001 at 08:38:34AM +0200, Ryszard Kabatek wrote:
> size_type 
> get_new_capacity(size_type n,
>                  size_type max_len) 
> {
>   if (n > max_len) // OK, we get an exception in _Rep::_S_create
>      return n;
> 
>   size_type result = n < 17 ? 32 : 2*n;
>   return result < max_len ? result : max_len; // Avoiding an exception in _Rep::_S_create
> }

Hmm.  How about:

size_type
get_new_capacity(size_type n,
                 size_type max_len)
{
  // Not sure if this is still correct:
  if (n > max_len)					// OK, we get an exception in _Rep::_S_create
     return n;

  const size_type l2 = 5;  				// 5 == 2log(32), 32 is the minimum size.
  size_type l = (n < __page_size ? n : __page_size) + __malloc_header_size - 1;
 
  if (l >= __page_size - 1)
    n = (((n - 1) >> __log_page_size) + 1) << __log_page_size;	// __log_page_size == 2log(__page_size).
  else
    for (n = 1 << l2, l >>= l2; l > 0; n <<= 1)
      l >>= 1;

  size_type result = n - __malloc_header_size;
  return result < max_len ? result : max_len;		// Avoiding an exception in _Rep::_S_create
}

I am not sure if I understand it correctly however: does this function
return the new __capacity, or is this what is passed directly to malloc()?
This implementation assumes that the result is passed directly to malloc().
I have the feeling it should contain sizeof(CharT) and sizeof(_Rep) somehow though :/
This function is not as trivial as it looks.

-- 
Carlo Wood <carlo@alinoe.com>