[PATCH] Fix PR44555

[Richard Guenther]

On Thu, 17 Jun 2010, Dave Korn wrote:

> On 17/06/2010 13:57, Andreas Schwab wrote:
> > Dave Korn <dave.korn.cygwin@gmail.com> writes:
> > 
> >>   Yes, I was suggesting that the standard could be adjusted to say the same
> >> about &((*a).b), that the indirection and address operators are not evaluated
> >> there either,
> > 
> > The problematic part is the member selector, which results in a pointer
> > that is different from the original null pointer, especially when it's
> > not the first member.  So the expression cannot fully be a no-op.  In
> > the &*E case you have two operations that are inverse to each other.
> 
>   I see what you mean.  It would need some fairly twisted logic to argue that
> the member-selector operation actually modifies the address-of operator in the
> standard's terms, despite that that's how the code ends up being implemented.
> 
>   But are you objecting to the patch?  I think it should be allowed.

The patch has already been approved by Joseph and checked in, so you
can call stop discussing.

Richard.