26_numerics fails

[Paolo Carlini]

Gabriel Dos Reis wrote:

> Ugh!  If that is really the case, then I think the compiler is
> broken.  In another words, the compiler is saying that given :
>
>     class X;
>
>     struct X { int i; };
>
>
> the following is invalid
>
>     int main()
>     {
>       X x;
>       x.i = 9;
>     }
>
> which is quite incorrect.  The compiler should be fixed.

Ok.
Therefore, roughly speaking, the current valarray regression (PR5348, that is) simply
shows that Nathan Sidwell in his dozen of checkins at the beginning of 2002 broke
something in the C++ front-end!
...in other terms, your implementation is perfectly ok as is!

What should we do? Will you take care of informing Nathan of the technical details of
this?

Thanks,
Paolo.

P.S. It would be nice if Peter's testcase could be further reduced, to understand
under which specific conditions the compiler is at fault in judging the correct
access restriction.