Re: <string> - warning messages

[Nathan Myers <ncm@best.com>]

Vadim wrote:
> I see more problems here than that.
> It looks like an attempt to check validity of the range [__beg, end).
> 
> _M_replace contains similar piece of code (bits/string.tcc:424):
> 	size_type __dist_new = static_cast<size_type>(distance(__j1, __j2));
> 
> In the case of negative distance in both cases it is converted to big
> unsigned
> value - in the first case by _Rep::_S_create(__range, __a) call.
> 
> If it is bigger than max_size() than length_error will be thrown,
> otherwise we'll
> have undefined behavior predicted in Standard.

It's a lot worse than that.  This makes the function incompatible
with real Input iterators, because measuring the distance 
uses them up.  They're one-pass items.
 
> I would like to know what is supposed to happen in such cases.
> (e.g. 
> 	string s1("some text");
> 	string s2(s1.end(), s1.begin());
> )

It's impossible in general to check iterators, and generally a
mistake to try.  We could partial-specialize on string- and
vector- "checked iterators", but they would slow performance enough
that it would need to be a compile-time option.

Nathan Myers
ncm@cantrip.org