Re: [boost] Explicit instantiation of boost::shared_ptr<SomeClasss>

From: Peter Dimov (pdimov_at_[hidden])
Date: 2004-03-16 12:53:35

Greg Clayton wrote:
>> -----Original Message-----
>> From: Peter Dimov
>>
>> Isn't it possible for the DLL C that contains the definition
>> of SomeClass to
>> export
>>
>>     shared_ptr<SomeClass> createSomeClass();
>>
>> which DLL B can now call when it needs a SomeClass?
>>
>> Since ~SomeClass lives in C, it must be available when the last
>> shared_ptr<SomeClass> is destroyed.
>>
>
> This was my initial solution, and it does work.

I mention it because it's the recommended approach, especially if SomeClass
is most frequently allocated on the heap and used via shared_ptr<SomeClass>
(there are no automatic variables of type SomeClass, and so on).

Hiding the actual 'new SomeClass' behind an exported factory function makes
SomeClass less error-prone, because users no longer need to use 'new'
directly. You can make the constructors and the destructor of SomeClass
protected to further prevent accidental mistakes. You'll also be able to
change the way a SomeClass is allocated without changing one bit of code
outside the DLL that hosts 'SomeClass::create' or 'createSomeClass'.

> Though I was searching for the ability to just create a shared pointer
anywhere
> and had it to any other DLL by exporting the entire shared pointer
instantiation
> from the main DLL.

What does this buy you? Whether you export
shared_ptr<SomeClass>::shared_ptr<SomeClass> or createSomeClass doesn't make
much difference, and, at least in my opinion, it creates more trouble than
it's worth. But I may be missing something.