Re: [boost] new version of smart pointer class templates

From: Peter Dimov (pdimov_at_[hidden])
Date: 2002-01-25 15:21:56

From: "Brey, Edward D" <EdwardDBrey_at_[hidden]>
> I don't think this will work.  There is no implicit conversion between 0
> shared_ptr<FILE>, which is a good thing.

Yes, you could (should?) do this

shared_ptr<FILE> my_fopen(const char* name, const char* mode)
{
    if (FILE* p = fopen(name, mode))
        return shared_ptr<FILE>(p, fclose);
    else
        throw open_error(name, errno);
}

instead. Of course this is simply an example.

> On a larger question, what is the design rational for not making the
deleter
> a template parameter?

Both approaches have their uses. It is generally expected that the Loki
smart pointer, or some other policy-based design, will cover the template
parameterized case, where you encode the deallocation policy into the type
of the pointer.

shared_ptr<> can be used in situations where you don't want to have the
deallocation policy encoded in the type. For example, if you maintain a
library that has

shared_ptr<Object> createObject();

and you want to change the allocator used for object creation (or even make
some objects static) without breaking compatibility.