On Tue, 7 Jun 2011, Richard Guenther wrote:

> On Tue, Jun 7, 2011 at 12:27 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> > On Tue, Jun 07, 2011 at 12:19:59PM +0200, Richard Guenther wrote:
> >> On Tue, Jun 7, 2011 at 7:39 AM, Jason Merrill <jason@redhat.com> wrote:
> >> > In the testcase, fold_indirect_ref_1 won't fold *(T*)(s1+10) to an ARRAY_REF
> >> > because T != unsigned.  Even if it were just a typedef to unsigned, that
> >> > isn't close enough, but in this case it's a typedef to const unsigned.
> >> >
> >> > I'm not sure what the type coherence rules are for ARRAY_REF.  Is it really
> >> > necessary that the type of the ARRAY_REF match exactly the element type of
> >> > the array?
> >>
> >> I _think_ that you can unconditionally change the code to do
> >>
> >>   TYPE_MAIN_VARIANT (t1) == TYPE_MAIN_VARIANT (t2)
> >>   && TYPE_QUALS (t1) == TYPE_QUALS (t2)
> >>
> >> now, I'm not sure if for the testcase T and unsigned differ in qualifiers.
> >
> > I guess folding into array_ref that way is fine, but you should in the end
> > fold_convert_loc it to the expected type, while the middle-end has the
> > notion of useless type conversions, fold-const.c is also used by FEs and
> > I think it is expected to have the types exactly matching.
> > So (T)s1[10] instead of s1[10] in this case.
> 
> I'm not sure that's a good idea if the caller wants an lvalue.

Rather build the array-ref with type T directly (thus, with a mismatch
between the type of the array-ref and the element type).

Richard.