Re: [boost] Re: shared_ptr improvement proposal

From: troy d.straszheim (troy_at_[hidden])
Date: 2004-08-07 04:08:19

On Aug 7, 2004, at 3:14 AM, David B. Held wrote:
> troy d.straszheim wrote:
>> shared_ptr<Widget> p = shared_ptr_cast<Widget>(new Widget);
>> [...]
>
> I kinda like the idea as well.  In a sense, you *are* converting the 
> type of the pointer returned by new from a raw pointer to a 
> shared_ptr<>.  However, I wonder how much benefit would be derived
> from this change in practice.  This would break all existing 
> shared_ptr<> code, wouldn't it?

As Rob notes, there's no reason the current practice couldn't remain 
legal:

shared_ptr<Widget> p(new Widget);  // still OK but not "best practice"

but of course that falls short of the mark w.r.t. *forcing* the 
"announcement" of the conversion from raw to shared_ptr via the 
shared_ptr_cast<>.  I find Rob's observation that the shared_ptr_cast<> 
supports specialization for other source types (I knew something good 
and obvious was there, duh) compelling.

Would seem reasonable to me to allow both and have the docs reflect 
that a cast is preferred, but if the consensus is to turn off the 
constructor-from-raw-pointer, if not now, when?  As I'm new to the 
list, I don't have a good feel for the cost of this in terms of 
client-code-breakage, nor for how much emphasis is placed on 
backwards-compatibility in boost releases...  (relatively little, since 
the idea is to hammer out standards-ready designs?)

-t