Re: legal ruling on aliasing

[Geoff Keating <geoffk at geoffk dot org>]

Bruce Korb <bkorb@veritas.com> writes:

> In another list, the issue was raised whether or not these
> two constructs are completely equivalent under the laws of
> aliasing:
> 
>    stumble_t s;
>    function( (mumble_t*)(void*)&s );
> 
> versus:
> 
>    stumble_t s;
>    {
>       void* v = (void*)&s;  // scope constrained to following:
>       function( (mumble_t*)v );
>    }

Yes, they are.  They're both exactly the same as

stumble_t s;
function ( (mumble_t *)&s);

for aliasing purposes.

> In other words, would the compiler be allowed to presume that
> function() did not modify `s'?  Particularly, could it make
> such an assumption in one case and not the other?

No, it can make no such assumption in either case, because 'function' could be:

void function (mumble_t *x)
{
  stumble_t * sp = (stumble_t *)x;
  *sp = *sp + 1;
}

-- 
- Geoffrey Keating <geoffk@geoffk.org>