Re: -Wcast-qual broken(?) for libstdc++

[Mark Mitchell <mark at codesourcery dot com>]

On Thu, 2003-07-10 at 04:50, Gerald Pfeifer wrote:
> On Tue, 8 Jul 2003, Mark Mitchell wrote:
> > The code in question is:
> >
> >       _Link_type
> >       _M_end() const { return (_Link_type) &this->_M_header; }
> >
> > _M_header has type _Rb_tree_node_base; _Link_type is  _Rb_tree_node*.
> >
> > The type of "this" is "const _Rb_tree*" since we are in a const member
> > function.  The expression "this->_M_header" is decreed by the standard
> > to be equivalent to "(*this)._M_header".  The type of "*this" is "const
> > _Rb_tree".  The standard says that you combine cv-qualifiers on a "."
> > expression; therefore, the the type of "(*this)._M_header" is "const
> > _Rb_tree_node *".  Therefore, the cast discards cv-qualifiers.
> 
> The patch below fixes my original testcase, but the following still
> fails...
> 
>   #include <set>
>   using namespace std;
> 
>   typedef set<unsigned,less<unsigned> > S;
>   S v;
> 
>   void f(const S &w) {
>     v.find(2);
>     w.find(2);
>     }
> 
> > (This is why functions like "vector<T>::end() const" return a
> > "const_iterator".)
> 
> ...and seems to require an approach similar to const_iterator.
> 
> Clearly, this is a regression.  Just, who is going to implement a more
> complete fix then my patch below?

Hopefully one of the library people will get to it.

(I don't feel this is really a situation where my patch uncovered a
latent bug, since the problems are not in the compiler but in the
library, but maybe that's not fair.  If, however, the consensus is that
I am obligated to fix the bug, I will honor that.)

-- 
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com