Suicidal objects

[Jarrett Billingsley]

"Sean Kelly" <sean at f4.ca> wrote in message 
news:fjk58r$va5$1 at digitalmars.com...
> Jarrett Billingsley wrote:
>> "Sean Kelly" <sean at f4.ca> wrote in message 
>> news:fjjpgq$2t9c$3 at digitalmars.com...
>>
>> http://www.digitalmars.com/d/archives/digitalmars/D/Inheriting_constructors_54088.html
>>> In short, there's no easy way to perform placement new on an existing 
>>> class, and I view this as a deficiency.  My proposal would provide one 
>>> easy way to do so, but I still feel that making it automatic would be 
>>> preferable.
>>
>> With __traits, we can now get the size of a class instance and the 
>> overloads of the constructor (I think?).  I think that's about all you'd 
>> need, no?
>
> Yeah, but it would still be a bit awkward to use __traits to obtain the 
> size, allocate, then use __traits to call the proper ctor by name on the 
> new block or whatever.  Placement new is so generally useful that it 
> should really just be supported by the language.
>
>
> Sean

template Sizeof(T)
{
    if(is(T == class))
        const Sizeof = __traits(getInstanceSize, T);
    else
        const Sizeof = T.sizeof;
}

template CtorForwards(T)
{
    // use __traits to get ctor overloads, generate them
    // maybe use strings and then string mixin?  probably.
}

class Placement(T) : T
{
    new(size_t s, void* p)
    {
        return p;
    }

    mixin(CtorForwards!(T));
}

..

ubyte[Sizeof!(A)] place;
auto a = new(place.ptr) Placement!(A)(params);

This is kind of what Walter meant by __traits not having to be pretty to be 
useful; it can be hidden behind templates.  Of course, if D were perfect, 
this kind of stuff wouldn't be crufty-looking, but nothing's ever perfect..