Re: [boost] [optional] Strict Aliasing Warnings on Trunk

Subject: Re: [boost] [optional] Strict Aliasing Warnings on Trunk
From: Peter Dimov (pdimov_at_[hidden])
Date: 2009-12-18 09:21:08

Felipe Magno de Almeida wrote:
> On Fri, Dec 18, 2009 at 6:11 AM, Emil Dotchevski
> <emildotchevski_at_[hidden]> wrote:
>> On Thu, Dec 17, 2009 at 11:17 PM, Dean Michael Berris
>> <mikhailberis_at_[hidden]> wrote:
>
> [snip]
>
>>> Really? I thought this was implementation-defined?
>>
>> Sure, but it is specified that if you reinterpret_cast from T1* to
>> T2*, then reinterpret_cast the resulting T2* back to T1*, you'll get
>> the original value (as long as T2 doesn't have stricter alignment
>> requirements.)
>
> But nothing else is guaranteed. The compiler might very well increment
> it by one byte when casting to T2 and then decrement it when casting
> it back. It is perfectly valid behavior for reinterpret_cast.

Maybe, in principle, but if you try to write such a compiler you'll find 
that it isn't as easy, even if we dismiss C compatibility as a concern and 
stick to C++03. (In the latest C++0x draft reinterpret_cast for pointers to 
standard layout types is defined to do the static_cast dance.)

> IMO we should be using static_cast<T2*>(static_cast<void*>(x)) which
> is not implementation-defined.

There isn't much difference. In practice, the two do the same thing. g++ 
should issue the same warning for both; it doesn't because the current 
frontend isn't smart enough to see through the intermediate void*.