On 11/3/13, Jeffrey Yasskin <jyasskin_at_[hidden]> wrote:
> On Wed, Oct 30, 2013 at 3:06 PM, Lawrence Crowl <Lawrence_at_[hidden]> wrote:
>> Even further, we consider saying that we can access any integral type
>> with the same size and alignment.
>
> And a further generalization would be to say that we can alias any
> layout-compatible types, and then broaden "layout-compatible" to cover
> the integral cases we care about. Layout-compatible types can already
> be aliased through unions, but only when they're actually accessed
> through unions. This could provide the language side of
> http://cplusplus.github.io/LWG/lwg-active.html#2335.
>
> Thoughts?

Seems reasonable to me, but I do worried that we'll get too carried
away and unnecessarily inhibit the optimizer.

-- 
Lawrence Crowl