Iterator "convertable to" problem

[chris jefferson]

Paolo Carlini wrote:

> Hi Chris,
>
>> However, on a related note, do you have any comments on if we have to 
>> start static_casting lots of things :)
>
>
> Probably, people reading the list would appreciate a testcase 
> demonstrating the problem in a practical way...

While I've now decided I was perhaps over-reading the standard, I shall 
provide an example anyway, borrowed from the c++std-lib mailing list.

Our (and almost everyone's) implemention of find for input iterators 
looks like this:

template<typename _InputIter, typename _Tp>
    inline _InputIter
    find(_InputIter __first, _InputIter __last,
     const _Tp& __val,
     input_iterator_tag)
    {
      while (__first != __last && !(*__first == __val))
    ++__first;
      return __first;
    }

The problem comes from the fact that as __first and __last are input 
iterators, then strictly "__first != __last" does not have to be bool. 
It only has to be convertable to bool. We could therefore write an 
overload of && for the actual type of __first != __last, which would 
totally mess up the short-cutting that we are assuming will happen (as 
obviously we only want to try !(*__first == __val) when __first!=__last).

However, this kind of pendanticness seems to be a little unnessasary :)

>
> My first instinctive reaction (I haven't really studied the issue) is 
> that hardly the designers of the C++ runtime library would have liked 
> implementations filled of static_casts... therefore, if the text 
> *really* needs that for strict conformance, then probably the text 
> should be tweaked, not the implementations... ;)
>
> Paolo.