Re: [boost] alignment trouble with boost::optional

From: David Abrahams (dave_at_[hidden])
Date: 2005-10-28 11:58:15

Martin Bonner <martin.bonner_at_[hidden]> writes:

> ----Original Message----
> From: David Abrahams [mailto:dave_at_[hidden]]
> Sent: 27 October 2005 12:57
> To: boost_at_[hidden]
> Subject: Re: [boost] alignment trouble with boost::optional
>
>> Michael Shepanski <mshepanski_at_[hidden]> writes:
>> 
>>> On 10/26/05, David Abrahams <dave_at_[hidden]> wrote:
>>>> 
>>>> 
>>>> I think the problem is that type_with_alignment can't find a type
>>>> that has an alignment of 12. Producing an error in that case is the
>>>> right behavior. We could add the type whose alignment_of<> is 12 to
>>>> the reperotoire of types being tested by type_with_alignment...
>>> 
>>> 
>>> 
>>> Don't we need to be sure that type_with_alignment<> can accept *any*
>>> number that alignment_of<> produces -- or, conversely, make sure that
>>> alignment_of<> only produces numbers that type_with_alignment<> can
>>> accept? (I'm assuming that we want boost::optional to work all the
>>> time, not just for my example.)
>> 
>> Sure, that's what we want.  Can you figure out *how*?  That's the
>> problem.
>
> In summary:
>
> 1. alignment_of<T> produces a number that is a multiple of the required
> alignment.
> 2. type_with_alignment<> cannot deal with all the numbers that
> alignment_of<> produces.
>
> Does anyone know of any platform (including platforms that are no longer in
> use), where the alignment requirement was /not/ a power of two?  If not,
> then alignment_of<> "just" needs to eliminate the factors from its result
> which are not a power of two.  (That would also allow us to eliminate the
> special case for long double under Borland).

That's an interesting heuristic.  It trades away some theoretical
safety for some practical utility.  Should we try to warn in these
cases, with a PP symbol that you can define to silence the warnings?
Can we add some config macro that tells us when it is truly safe? Is
there a config test we can run that will tell us at runtime?

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com