iter_swapping vector<bool>::iterator..

[Felix Koehler]

I had just stumbled over this breaking my programs, when Chris
Jefferson reported the regression. Basically it makes vector<bool>
completely  unusable as a lot of the standard algorithms like
"std::reverse" stop working.

Paolo Carline wrote:
>In my opinion, *if* we agree that vector<bool>::iterator cannot be
>implemented to meet the requirements of forward iterator, there is
>no point in adding these complications. Maybe just say something in
>the docs, nothing more: a standard conforming change of behavior is
>perfectly acceptable for a major release.

>Better spend our time elsewhere...

Well, for selfish reasons I obviously  disagree ;-). vector<bool> with
most of the standard algorithms not working is a pain. Would
specialisations of swap / iter_swap help in the case of a derived
class T: public vector<bool> ? If so, please avoid breaking
vector<bool> for g++ 4.0.

Following testcase demonstrates how the breakage of iter_swap breaks 
std::reverse for vector<bool> and derived clases. (Not a minimum
testcase, but one that should make sure that a fix will have no issues
with classes derived from vector<bool>).
----------------- <testcase start> -----------------------
#include <iostream>
#include <algorithm>
#include <vector>

using namespace std;

class T: public std::vector<bool>
{
  public:
    T (int size = 0, bool value = false): std::vector < bool > (size,
            value)
    {
    }
};

inline std::ostream&
operator << (std::ostream & o, const T & v)
{
  for (int h = 0; h < v.size (); ++h)
    o << (v[h] ? '1' : '0');
  return o;
}

int main () 
{

  T a(10, true);

  cout << a << endl;

  a[0] = false;  
  a[1] = false;  
  a[2] = false;  
  a[3] = false;  
  a[4] = false;  
  
  cout << a << endl;
  
  reverse (a.begin(), a.end());
  
  cout << a << endl;
 
  return 0;
}
----------------- < testcase end> -----------------------